Nature Neuroscience January 2000

2000 Nature America Inc. http://neurosci.nature.
com
contents
volume 3 no 1 january 2000
http://neurosci.nature.com
Rosenthal and colleagues report
that a constitutively active form
of Smoothened (green) mimics
concentration-dependent actions
of Sonic Hedgehog in the
developing neural tube, including
activation of ventral markers
(such as Nkx2.2, red), suppression editorial
of dorsal markers and induction
2000 Nature America Inc. http://neurosci.nature.com
of multiple classes of ventral and

ventrolateral neurons. These pat- The brain within the genes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
terning activities are cell
autonomous, occurring only in
cells that express Smoothened. news and views
See page 41.
Sound and meaning: how native language affects reading strategies . . . . . . . . . . . 3

Julie A. Fiez
SEE ARTICLES, PAGE 91
Building inhibitory synapses: exchange factors getting into the act? . . . . . . . . . . . . . . 5

Justin R. Fallon
Seeing is relieving: electrical synapses between visualized neurons . . . . . . . . . . . . . . . . 7

Michael V.L. Bennett
Notch regulates dendritic
development.
Page 30.
Targeting visual motion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Jochen Braun
brief communications
Long-term visual experience recalibrates human orientation perception. . . . . . . . . . . . . . . 13
D Whitaker and PV McGraw
articles
Molecular basis of NMDA receptor-coupled ion channel modulation
by S-nitrosylation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
YB Choi, L Tenneti, DA Le, J Ortiz, G Bai, HSV Chen and SA Lipton
Collybistin, a newly identified brain-specific GEF, induces

Collybistin induces submembrane clustering of gephyrin. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
clustering of gephyrin. S Kins, H Betz and J Kirsch
Page 22. SEE NEWS AND VIEWS, PAGE 5
Nature Neuroscience (ISSN 1097-6256) is published monthly by Nature America Inc., headquartered at 345 Park Avenue South, New York, NY 10010-1707. Editorial Office: 345 Park
Avenue South, New York, NY 10010. Telephone 212 726 9200, Fax (212) 696 9635. North American Advertising: Nature Neuroscience, 345 Park Avenue South, New York, NY 10010-
1707. Telephone (212) 726-9200. Fax (212) 696-9006. European Advertising: Nature Neuroscience, Porters South, Crinan Street, London N1 9SQ. Telephone (0171) 833 4000. Fax
(0171) 843 4596. New subscriptions, renewals, changes of address, back issues, and all customer service questions in North America should be addressed to Nature Neuro-
science Subscription Department, PO Box 5054, Brentwood, TN 37024-5054. Telephone (800) 524-0384, Direct Dial (615) 377 3322, Fax (615) 377 0525. Outside North America:
Nature Neuroscience, Macmillan Magazines Ltd., Houndsmill, Brunel Road, Basingstoke, RG21 6XS, U.K.. Tel: +44-(0)1256-329242. Fax: +44-(0)1256 812358. Email:
subscriptions@nature.com. Annual subscription rates: U.S./Canada: U.S. $595, Canada add 7% for GST (institutional/corporate), U.S. $195, Canada add 7% for GST (individual
making personal payment BN: 14091 1595 RT); U.K./Europe:395 (institutional/corporate), 175 (individual making personal payment), 99 (student); Rest of world (excluding
Japan): 450 (institutional/corporate), 195 (individual making personal payment), 110 (student); Japan: Contact Japan Publications Trading Co. Ltd., 2-1 Sarugaku-cho 1 chome,
Chiyoda-ku, Tokyo 101, Japan, phone (03) 292-3755. Back issues: U.S./Canada, $45, Canada add 7% for GST; Rest of world: surface U.S. $43, air mail U.S. $45. Reprints: Nature
Neuroscience Reprints Department, 345 Park Avenue South, New York, NY 10010-1707. Subscription information is available at the Nature Neuroscience homepage at http://neu-
rosci.nature.com. POSTMASTER: Send address changes to Nature Neuroscience Subscription Department, P.O. Box 5054, Brentwood, TN 37024-5054. Application to mail periodicals
postage rate is paid at New York, NY. Executive Officers of Nature America Inc: Nicholas Byam Shaw, Chairman of the Board; Mary Waltham, President; Edward Valis, Secretary-Trea-
surer. Printed by Publishers Press, Shepherdsville, KY, USA. Copyright 1999 Nature America Inc.
nature neuroscience volume 3 no 1 january 2000 i

contents
Nuclear Notch1 signaling and the regulation of dendritic development. . . . . . . . 30

t L Redmond, SR Oh, C Hicks, G Weinmaster and A Ghosh
The seven-transmembrane receptor Smoothened cell-autonomously

induces multiple ventral cell types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
M Hynes, W Ye, K Wang, D Stone, M Murone, F Sauvage and A Rosenthal
y A locus and mechanism of action for associative morphine tolerance . . . . . . . . . . 47
JM Mitchell, AI Basbaum and HL Fields
x A model for intradendritic computation of binocular disparity . . . . . . . . . . . . . . . 54

KA Archie and BW Mel
Psychophysics of visual motion
detection.
Pages 9 and 64. Mechanisms of visual motion detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
PR Schrater, D C Knill and EP Simoncelli
SEE NEWS AND VIEWS, PAGE 9
Touch can change visual slant perception . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

MO Ernst, MS Banks and HH Blthoff
Different lateral amygdala outputs mediate reactions and actions

elicited by a fear-arousing stimulus. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
Distinct representations of P Amorapanth, JE LeDoux and K Nader
face identity and eye gaze.
Page 80.
Distinct representations of eye gaze and identity in the distributed
human neural system for face perception. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
EA Hoffman and JV Haxby
Integration of diverse information in working memory within the

frontal lobe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
V Prabhakaran1, K Narayanan, Z Zhao and JDE Gabrieli
A cultural effect on brain function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

E Paulesu, E McCrory, F Fazio, L Menoncello, N Brunswick, SF Cappa,
M Cotelli, G Cossu, F Corte, M Lorusso, S Pesenti, A Gallagher, D Perani,
C Price, CD Frith, and U Frith
SEE NEWS AND VIEWS, PAGE 3
English and Italian speakers

process language differently.
classified advertising
Pages 3 and 91.
see back pages
Editor's note
Beginning with this issue, Scientific Correspondence will be
renamed Brief Communications, to maintain consistency
with Nature and other Nature titles. The editorial content of
this section will remain unchanged.
nature neuroscience volume 3 no 1 january 2000 ii

editorial
The brain within the genes

Now is the season for predictions, so here are two. First, histori- with interactions at the molecular level) means that it should be
ans of the futureif they are still using the Christian calendar relatively straightforward to identify the genes involved, provided
will find it convenient to remember 2000 as the year when that enough data are available to achieve the necessary statistical
humankind determined its own genetic sequence. Second, they power. This power will most likely come from two emerging tech-
will regard this discovery as significant primarily because of its nologies, single-nucleotide polymorphisms (SNPs) and DNA
implications for understanding the brain. microarrays. SNPs are the most common form of human allelic
Pedants may quibble, but this enormous scientific achievement variation, and are currently being identified in very large num-
will coincide fairly precisely with the millenium. The sequence of bers, with both public and private funding. Some of these poly-
chromosome 22 was announced in December 1999, and a draft morphisms will themselves be causes of phenotypic variation,
sequence of the entire genome (which will be about 90% complete, whereas others will be linked to the causal polymorphisms and
and is likely to be accurate to better than 99%) will be available by will therefore be useful as genetic markers for association studies.
spring 2000. Celera Genomics, a private company, hopes to assem- There are also tremendous efforts being made to develop microar-
ble a full sequence using a shotgun method by 2001. There will be ray methods to the point where whole genomes can be scanned
a justifiable celebration at some point in the next three years or so, rapidly and cheaply for a comprehensive set of SNPs. These tech-
when the public consortium declares the entire sequence opera- nologies are integral to the identification of risk alleles for complex
tionally complete (meaning an error rate of less than 1 in 10,000, diseases, and they should be equally applicable to other complex
minus a few gaps), but the reality is that there will be no clear end traits such as personality and brain function.
point to the project. So the millenium will be good enough. All this will not happen overnight, but the implications for
Although the most immediate practical effects of the genome society seem potentially far-reaching. Plomin, like most thought-
project will probably be new insights into human disease, it does ful scientists, is careful to emphasize that science alone should
not seem far-fetched to suggest that the biggest impact in the long not and cannot determine social policy, which must depend on
term will be in neuroscience. It is, after all, our brains that make judgments of values as well as facts. At present, we can only spec-
us who we are, both as a species and as individuals. The genome ulate on what society will make of this new knowledge. There
project, and the work that will build on it, will provide a new will surely be individuals who will be burning with curiosity to
depth of understanding of how our brains are built and how they know their own genetic makeup and willing to pay for the infor-
differ from those of other species. But perhaps even more impor- mation, for the same reason that they buy books such as Know
tantand more challenging for societywill be the new insights Your Own I.Q. Will people also want to find reproductive part-
it will provide into how individuals differ from each other. ners based on genetic data, and should it be legal to provide a
Some of these issues are discussed1 by Robert Plomin of the service for those who do? Should parents be permitted to have
Institute of Psychiatry in London, writing in a special Nature sup- their children genotyped in order to plan their education? One
plement that is being distributed with this issue of Nature Neu- hopes that genetic testing will never become compulsory, but
roscience. Plomin focuses on intelligence, and more particularly even if the idea is anathema to most western democracies, there
on general cognitive ability (defined by factor analysis and often is no guarantee that all societies will feel the same way.
abbreviated to g), which is one of the most heritable behavioral A deeper understanding of behavioral genetics may also bring
traits. Rather than being determined by a single gene, g is a com- many benefits. It might, for instance, lead to better educational
plex trait that is influenced by many genes, each having a rela- strategies, not just for the most or least gifted, but for everyone.
tively small effect, as well as by the environment. Experts disagree One may also hope that it will help to combat racial prejudice
on the magnitude of the genetic contribution to g, but Plomin lay people are often unaware of the fact that the great majority
argues that it probably around 50%; in other words, that about of human genetic variation cuts across racial categories, and that
half of the variance in g can be attributed to genetic variation there is little evidence for systematic genetic differences between
within the population. Regardless of the exact figure, it seems races, other than in the superficial characteristics by which they
clear that there are genes that affect intelligence, as well as many are normally recognized. On the other hand, genetic informa-
other behavioral traits. tion may give new opportunities for discrimination against indi-
Many of these genes are likely to be identified as a result of the viduals. Clearly, great challenges as well as great uncertainties lie
genome project. Classical population genetics has indicated that ahead. But as Plomin says, the only thing that seems complete-
genetic effects on g are largely additive; that is, individual alleles ly clear is that nothing will be gained by ignoring the issue.
exert effects that are independent of interactions with other alle-
les. This lack of statistical interaction (which is not to be confused 1. Plomin, R. Nature 402, C25C29 (1999).
nature neuroscience volume 3 no 1 january 2000 1

news and views
Sound and meaning: how native

language affects reading strategies
Julie A. Fiez
In the Italian language, spelling is consistently related to pronunciation, whereas English

words that are spelled similarly are often pronounced differently. An imaging study now
shows that native language affects the brain regions activated by reading aloud.
How should the nonword jat be pro- dure, in which an entire written word is ciations between orthographic and
nounced? Most English speakers would represented as a set of component letters semantic (meaning-based) information,
agree that it should rhyme with bat. Yet or letter clusters, and then these ortho- and then forming associations between
what about the nonword jough? Should graphic subcomponents are associated semantic and phonological information.
it sound like cough, tough, dough or with corresponding sounds or phonolog- Lexical and semantic procedures may help
bough? This example illustrates a major ical subcomponents2. These subcompo- maximize reading speed for very common
difficulty of English and other ortho- nents can be used to build up a words, and they may help constrain the
graphically deep languages: mappings representation of an entire word sound. output of the sublexical procedure (for
between how words are spelled (their Sublexical procedures take advantage of instance, to ensure that pint is not pro-
orthography) and how they sound (their the consistent mappings between orthog- nounced so that it rhymes with mint).
phonology) are only partially consistent. raphy and phonology (such as that bad, Paulesu and colleagues1 theorized that lex-
English readers must use the consistency dad, tad, bat, dat, tat, dab and tab ical and semantic procedures are espe-
that is present to support pronunciation are different combinations of the same cially important for English readers,
of unfamiliar words (such as fint), but four letters and sounds) to avoid having because their sublexical procedure gener-
they also must accurately pronounce separate representations of the ortho- ates multiple alternative pronunciations.
words that violate these same generaliza- graphicphonological relationship for (Should fint rhyme with mint or pint?)
tions (for example, pint). Readers of Ital- each word in a language. Paulesu and col- To test their hypothesis that differences
ian and other orthographically shallow leagues theorized that the high degree of in the orthographic consistency of Italian
languages have an easier task because consistency in the Italian language per- versus English may lead subjects to place a
mappings between orthography and mits Italian readers to rely heavily on a different emphasis on sublexical versus
phonology are highly consistent: a partic- sublexical procedure for reading words lexical and semantic procedures (Fig. 1),
ular letter or letter combination is almost aloud. Paulesu and colleagues1 asked Italian and
always associated with the same sound. In Most models of word reading also con- English subjects to read aloud a variety of
this issue, Paulesu and colleagues 1 use tain one or more additional procedures words and nonwords. It took the English
behavioral measures and functional brain that do not depend on sublexical ortho- subjects significantly longer to begin read-
imaging to investigate how the orthographic structure. For instance, in one ing each word. The differences became
graphic consistency of ones native lan- class of models, a lexical procedure is even more pronounced when the stimuli
guage affects reading speed and the degree thought to permit access to stored infor- were nonwords. Control experiments
to which different language-related brain mation about the association between indicated that the English subjects were
areas are used for reading. The work whole word forms and their correspond- not just slow speakers. For instance, Eng-
demonstrates how brain imaging can be ing sounds3. In another class of models4, lish and Italian subjects took the same
used to address basic research questions phonological representations can be amount of time to produce a word for a
about mindbrain relationships and how accessed indirectly, by first forming asso- picture of an object. The faster reading
the results can be applied to questions of
clinical and educational significance.
This study1 was influenced by cogni-
Word sound Word sound
tive models of word reading. Previous (phonology) (phonology)
behavioral work indicates that skilled
readers of all alphabetic languages auto-
Lexical and/or Lexical and/or
matically use multiple procedures to semantic
Sublexical
semantic
Sublexical
translation translation translation
translate between orthography and translation
phonology, although the precise number
and nature of the procedures is hotly Word form Word form
(orthography) (orthography)
debated. Common to most models of
reading is some type of sublexical proce-
ENGLISH SUBJECTS ITALIAN SUBJECTS
Christie Albin
Julie Fiez is in the Department of Psychology, Fig. 1. Italian and English subjects place different emphasis on the two processes used to deter-
605 LRDC, University of Pittsburgh, Pittsburgh, mine how to pronounce written words. English speakers rely more heavily on lexical and/or
Pennsylvania 15260, USA. semantic translation, whereas Italian speakers rely more on sublexical translation (left). These
e-mail: fiez+@pitt.edu strategies lead to differences in the brain activation evoked by reading aloud (right).

news and views
times for Italians are consistent with the semantic processing (naming of both should help constrain interpretations of
hypothesis that Italian subjects can rely words and pictures, generating seman- activation differences within the system
almost exclusively on a sublexical proce- tic associations). and theoretical models of reading.
dure, whereas English subjects have to use Although Paulesu and colleagues 1 Indeed, these data1 already provide an
an additional lexical/semantic procedure develop a compelling interpretation for exciting glimpse into the neural basis of
to guide phonological output. their results, it will not be received with- reading. The finding that the left superi-
If the difference in reading speed for out controversy. One reason is that the or temporal, posterior inferior temporal
Italian versus English subjects reflects a existing imaging literature supports mul- and frontal areas are sensitive to the
differential emphasis upon sublexical ver- tiple interpretations. For instance, many orthographic consistency of a subjects
sus lexical/semantic procedures, then the imaging studies show that practice and native language suggests that they are crit-
two groups should have different patterns repetition can also produce relative ically involved in the derivation of
of brain activity during a reading task. To decreases in regional blood flow7. Indeed, phonology from orthography, even if the
test this idea, the authors1 used positron whether a particular region shows an specific contributions that each makes is
emission tomography to measure blood experience-related increase or decrease debatable. There is growing evidence
flow (a measure of neuronal activity) in seems to reflect several factors, such as its from reading researchers and educators
the brains of Italian and English subjects role in the learning process and the that the ability to manipulate and derive
while they read words and nonwords. As amount of practice given8. This suggests phonological representations is a funda-
expected, in both groups, this task acti- a different possible logic: that the more a mental aspect of skilled reading9. By fur-
vated a widespread network of brain sublexical, lexical or semantic procedure thering our understanding of the brain
regions previously associated with read- is used, the more efficient the brain regions and cognitive processes involved
ing5. The novel finding is that brain acti- regions that implement the procedure in the analysis of phonology, we should
vation in three areas of this larger might become, with increases in efficien- be in a better position to diagnose and
network depended on subjects native cy detected as smaller increases in activity. eventually treat reading disorders. We can
language. Italians had greater activation Thus, the smaller left frontal and posteri- already form some tantalizing links
in a left superior temporal region during or inferior temporal responses for Italian between normal and impaired perfor-
both word and nonword reading than subjects could reflect their greater skill rel- mance and the areas identified by Paule-
English subjects, who showed greater ative to English subjects at a sublexical su and colleagues. For instance, the left
activation in a left frontal and a posteri- procedure that is particularly important superior temporal region identified in
or inferior temporal region during non- for nonword reading. Similarly, the small- their study is near a region that is acti-
word reading (Fig. 1). er left superior temporal response for vated below normal levels when dyslexic
Paulesu and colleagues1 suggest that English subjects could reflect their greater subjects attempt to read words 10 . A
a sublexical procedure can be localized skill at a lexical or semantic procedure. recently developed technique for exam-
to the superior temporal region, and a This alternative interpretation is also con- ining the connections between brain
lexical/semantic procedure to the left sistent with some imaging results, such as regions indicates that this reduced acti-
frontal and posterior inferior temporal nonword versus word differences in left vation may arise from abnormal connec-
regions. This interpretation is based in frontal cortex activation5. tions between left temporal and frontal
part on the implicit logic that increasing A second issue is that the work of regions in dyslexic subjects (T. Klingberg
the emphasis on a particular procedure Paulesu and colleagues 1 leaves funda- et al. Soc. Neurosci. Abstr. 25, 654.7, 1999.
will produce more activation in regions mental differences in cognitive models of Equally exciting are the implications
that implement the procedure. Thus, word reading unresolved. We may now of this work for our understanding of
because Italians emphasize a sublexical know which brain regions are involved in how experience can shape the organiza-
procedure, the left superior temporal sublexical and lexical/semantic proce- tion of our cognitive systems. Both the
region that shows greater activation in dures, but we still do not know how these Italian and English subjects studied by
Italian subjects is likely to contribute to procedures work. For instance, do we Paulesu and colleagues1 can use sublexi-
this process. The converse is true for the develop rules that codify consistent cal and lexical procedures, but experi-
left frontal and posterior inferior tem- spelling-to-sound relationships in our lan- ence has optimized their use of these
poral regions. Neuroimaging work in guage3, or do we form distributed repre- procedures for the language they read.
other domains provides support for this sentations that capture the statistical Our perceptual systems have a remark-
logic. For instance, piano players have likelihood of both consistent and incon- able ability to self-organize based on
more activation in motor-related regions sistent pronunciations4? Such questions sensory experience 11. Reading proce-
than non-musicians when they perform have remained unresolved because exist- dures may become tuned through simi-
a simple finger-movement task6. To fur- ing behavioral data about the reading per- lar mechanisms, with much of the
ther support their interpretation, Paule- formance of normal and brain-damaged optimization occurring automatically
su and colleagues compare their results subjects is compatible with multiple mod- and continually. An interesting theoret-
to those of previous neuroimaging stud- els of reading. Paulesu and colleagues pro- ical issue with profound practical impli-
ies. They note that the left superior tem- vide a striking demonstration of how cations is how much this tuning may be
poral region is activated by other tasks neuroimaging can be used to provide a influenced by instructional strategy and
that are thought to emphasize sublexi- new type of data, in which the behavior the initial structure of reading materials.
cal phonology (reading nonwords as of individual components of the reading For instance, an early instructional strat-
compared to reading words), whereas system can be teased apart. Further inves- egy (such as phonics) that emphasizes
the left frontal and posterior inferior tigation of how the brain areas involved the consistent features of orthographic-
temporal regions are activated by tasks in reading are affected by experimental to-phonological transformation in Eng-
that are thought to emphasize lexical and and experience-dependent manipulations lish may lead children to emphasize a
4 nature neuroscience volume 3 no 1 january 2000

news and views
2. Patterson, K. & Behrmann, M. J. Exp. Psychol. 7. Raichle, M. E. et al. Cereb. Cortex 4, 826
sublexical transformation procedure that Hum. Percept. Perform. 23, 12171231 (1997). (1994).
involves the superior temporal region, 3. Coltheart, M., Curtis, B., Atkins, P. & Haller, 8. Karni, A. et al. Proc. Natl. Acad. Sci. USA 95,
compared to an early instructional strat- M. Psychol. Rev. 100, 589608 (1993). 861868 (1998).
egy (for example, whole language) that 4. Plaut, D. C., McClelland, J. F., Seidenberg, M. S. 9. Share, D. L. & Stanovich, K. E. Issues in
focuses on the relationships between & Patterson, K. Psychol. Rev. 103, 56115 Education 1, 157 (1995).
entire word forms and their corre- (1996). 10. Rumsey, J. M. et al. Arch. Neurol. 54, 562573
sponding pronunciations and meanings. 5. Fiez, J. A. & Petersen, S. E. Proc. Natl. Acad. (1997).
Sci. USA 94, 914921 (1998). 11. Recanzone, G. H., Merzenich, M. M., Jenkins,
1. Paulesu, E. et al. Nat. Neurosci. 3, 9196 6. Hund-Georgiadis, M. & von Cramon, D. Y. W. M., Grajski, K. A. & Dinse, H. R.
(2000). Exp. Brain Res. 125, 417425 (1999). J. Neurophysiol. 67, 10311056 (1992).
because the size and shape of postsy-

Building inhibitory synapses: naptic specializations, as well as
gephyrin density, depend on actin fila-
exchange factors getting ments as well as microtubules. We would

also like to know how this putative
anchorcytoskeleton link is regulated, a
into the act?
question that is likely to be central to

understanding the mechanism of synap-
Justin R. Fallon tic plasticity.
A first step toward answering these
Gephyrin is a synaptic protein that is required for clustering questions is to identify other molecules
of glycine and GABAA receptors. Gephyrin itself may be that interact with gephyrin. Kins and col-
regulated by a newly identified protein named collybistin. leagues used the yeast two-hybrid system,
a powerful genetic technique that has led
to the identification of many
For centuries, artists have depicted Atlas The precise localization of neuro- proteinprotein interactions at synapses
holding up the world, providing the transmitter receptors is essential for and elsewhere2,3,8,9. By screening a brain
ancients with a neat explanation for how synaptic function. For a neuron to cDNA library for proteins that could bind
the Earth stays in its place and making a respond to neurotransmitter release, it to gephyrin, the authors identified colly-
good story to this day. Even in ancient must maintain a high density of recep-
times, though, children must have won- tors on the postsynaptic membrane, and
dered what Atlas is standing on, and in activity-dependent modification of
GlyR GlyR
the post-Copernican age, they might also synaptic efficacy is now known to be at
ask how this arrangement can work if the least partly due to changes in number
Earth is constantly moving. Students of and density of receptors. Thus it is not
Gephyrin
the synapse face a similar conundrum. surprising that the mechanisms under-
Coll II
They learn that anchoring molecules lying receptor clustering have come Coll II activation?
hold neurotransmitter receptors in the under close scrutiny24. Many different ? Cytoskeletal
postsynaptic membrane. Yet what molecules are implicated in receptor GDP reorganization?
anchors the anchors, and if they are so clustering at various types of synapses,
firmly held in place, how can synapses and one of the major challenges now fac-
GlyR GlyR
change over time? ing the field is to determine how the
A paper by Kins and colleagues1 in this clustering machinery is regulated by
issue is likely to represent an important synaptic activity5.
Gephyrin Gephyrin
step toward answering these questions. The present story began several years Coll II Coll II
The authors describe two new guanosine ago with the identification of gephyrin6, ?
diphosphate/guanosine triphosphate a protein that is now known to be GTP
(GDP/GTP) exchange factors, collybistin involved in anchoring glycine receptors
Cytoskeleton
I and II, which may regulate the distribu- and GABA A receptors at postsynaptic
tion of glycine receptors, and perhaps also sites. Gephyrin is enriched at glyciner- Amy Center
GABAA receptors, at inhibitory synapses. gic synapses and binds the cytoplasmic
This advance is noteworthy because it domain of the glycine receptor -sub- Fig. 1. A hypothetical model for the role of
provides a potential mechanism for inte- unit as well as microtubules. Neurons collybistin II in clustering glycine receptors
grating three major elements of the clus- from mutant mice lacking gephyrin fail based on findings1 from heterologous cells.
tering machinery at this synapse: glycine to cluster glycine receptors at synapses7. Glycine receptors are expressed in the mem-
brane but are not clustered until engaged by
receptors, gephyrin and the cytoskeleton. However, although these findings con-
gephyrin. The translocation of gephyrin to the
firm that gephyrin is an anchoring mol- membrane requires interaction with the
Justin Fallon is at the Department of Neuroscience, ecule, our understanding is far from GDP/GTP exchange factor collybistin II. This
Brown University, Box 1953, 190 Thayer Street, complete. For instance, how is gephyrin protein presumably activates one or more small
Providence, Rhode Island 02912, USA. held in place? Although it binds micro- G proteins, which may in turn regulate the orga-
e-mail: justin_fallon@brown.edu tubules, this is only part of the story, nization of the cytoskeleton.

news and views
bistin I and II, two alternatively spliced A better understanding of the collybistin may be involved in the postsynap-
products of the same gene. Analysis of bistins could also shed new light on the tic response to extracellular signaling pro-
their amino-acid sequences revealed two multiple lives of gephyrin. Gephyrin is teins. Such signals might come from the
domains that identify the collybistins as necessary for clustering not only glycine presynaptic terminal, glia or even the
members of the dbl-like family of receptors but also many GABA receptors postsynaptic cell. So far, the only extra-
GDP/GTP exchange factors. This family in vivo12. It has also emerged as a likely cellular signaling molecule known to be
is characterized by dbl domains, which regulator of local mRNA translation13. involved in synapse formation is agrin,
are important for exchange activity, and Most mysteriously, gephyrin is required which is essential for neuromuscular junc-
by pleckstrin homology domains, for the synthesis of a necessary cofactor tion development. Agrin is also expressed
thought to mediate association with the for molybdenum enzymes. Such enzymes in the brain, but its role there is not
plasma membrane. Dbl-like GDP/GTP are expressed in most cell types and are understood. The identity of functionally
exchange factors (GEFs) regulate the involved in the metabolism of sulfites analogous actors at CNS synapses is large-
activity of small GTPases such as rho, rac (among other things). Mutant mice lack- ly unknown, although some candidates
and cdc42 (ref 10). These proteins act as ing gephyrin show severe neurotoxicity are now emerging14.
signaling molecules and can exist in active due to the build-up of these and other Clearly, a complete understanding of
(GTP-bound) or inactive (GDP-bound) metabolites. Finding the mechanisms by synapse formation will require the iden-
forms. Their GTPase activity brings about which collybistins are activated and identification of extracellular, transmembrane,
their inactivation, and they are reactivat- tifying the targets of the collybistins anchoring, cytoskeletal and regulatory
ed when the bound GDP is exchanged for should help elucidate gephyrins role in molecules. Once the molecular players
GTP from the cytoplasmic pool. GEFs act these diverse cellular functions. Finally, have been identified, the challenge will be
by promoting this reactivation (Fig. 1). A gephyrin and neural activity are neces- to understand how they collaborate to
major function of GTPases such as rho, sary but not sufficient to cause glycine form and maintain synapses. Among the
rac and cdc42 is to link the activation of receptor clustering on neurons. It will most interesting questions will be how the
cell-surface receptors to reorganization therefore be interesting to find out synapse maintains a structural memory.
of the actin cytoskeleton, especially at whether collybistin is somehow required In other words, how it is that the struc-
membranes11. For example, microinjec- for the activity-dependent regulation of ture of a synapse can remain stable over
tion of rho can induce the formation of synaptic structure. long periods of time, even though the
actin-based stress fibers in many cell More generally, how does this new individual molecules from which it is
types. Thus, the sequence of the colly- information advance our understanding made are constantly turning over? It
bistins, along with their ability to bind of how synaptic architecture is regulated? seems likely that this structural stability,
gephyrin, raises the possibility that they The growing list of molecules that have along with the capacity for use-dependent
may regulate GTPase signaling at been implicated in receptor clustering modification, is a function of feedback
inhibitory synapses. suggests a theme that may be character- loops that integrate networks of structur-
Kins et al.1 tested the biological activity ized as similar stories, different actors. al and regulatory molecules 15 . Thus,
of collybistins by expressing them in vari- Thus, distinct anchoring molecules have although the names may change, com-
ous combinations with gephyrin and now been found for most of the major parative studies can teach us much about
glycine receptors in non-neuronal cells. classes of ligand-gated ion channels, what kinds of molecules are needed to
When expressed individually, both gephyrin including the PSD-95/SAP90 family (for build and regulate the diverse array of
and collybistin are cytoplasmic. However, NMDA receptors)8, GRIP (AMPA-type synapses that hold up the nervous system.
when gephyrin and collybistin II (but not GluRs)3, homer (some mGluRs)9 and rap-
collybistin I) are expressed together, they syn (muscle nAChRs)4, as well as gephyrin 1. Kins, S., Betz, H. & Kirsch, J. Nat. Neurosci. 3,
2229 (2000).
form submembranous aggregates. Impor- (glycine and GABAA receptors). In addi-
2. Chen, H. J., Rojas-Soto, M., Oguni, A. &
tantly, such aggregates can also capture tion to these structural components, reg- Kennedy, M. B. Neuron 20, 895904 (1998).
glycine receptors, thereby targeting them to ulatory molecules have now been linked
3. Kim, J. H. & Huganir, R. L. Curr. Opin. Cell Biol.
clusters within the plasma membrane. to many of these complexes. One that 11, 248254 (1999).
These experiments were performed in seems particularly relevant to under- 4. Sanes, J. R. & Lichtman, J. W. Annu. Rev.
fibroblasts, but it is tempting to think that standing collybistins is SynGap, a GTPase- Neurosci. 22, 389442 (1999).
the same thing may be happening in neu- activating protein (GAP) that is associated 5. Craig, A. M. Neuron 21, 459462 (1998).
rons. A first step toward testing this idea with PSD-95 (refs. 2, 3). GAPs accelerate 6. Prior, P. et al. Neuron 8, 11611170 (1992).
will be to determine the localization of the GTPase activity of small G proteins 7. Feng, G. et al. Science 282, 13211324 (1998).
collybistin proteins in vivo. Collybistin and such as rho, rac and cdc42, thereby pro-
8. Kornau, H. C., Seeburg, P. H. & Kennedy, M. B.
its mRNA are selectively expressed in the moting their auto-inactivation. The effect Curr. Opin. Neurobiol. 7, 368373 (1997).
brain, but the available antibodies were not of GAPs is thus opposite to that of 9. Tu, J. C. et al. Neuron 21, 717726 (1998).
suitable for establishing its cellular local- exchange factors, and so it seems reason-
10. Lin, R., Cerione, R. A. & Manor, D. J. Biol.
ization. It will also be important to able to speculate that collybistin may Chem. 274, 2363323641 (1999).
demonstrate that collybistin does indeed antagonize an as-yet unidentified GAP at 11. Hall, A. Science 279, 509514 (1998).
have GDP/GTP exchange activity, to iden- inhibitory synapses, and conversely that
12. Kneussel, M. et al. J. Neurosci. 19, 92899297
tify its natural substrates, and to determine SynGap may be antagonized by another (1999).
whether the exchange activity is required exchange factor (perhaps related to colly- 13. Sabatini, D. M. et al. Science 284, 11611164
for its biological activity. Because of its bistin) at excitatory synapses. (1999).
interaction with gephyrin, one intriguing Small GTPase signaling pathways are 14. OBrien, R. J. et al. Neuron 23, 309323 (1999).
idea is that gephyrin may be required for often involved in transducing extracellular 15. Lisman, J. E. & Fallon, J. R. Science 283,
activation of its enzymatic activity. signals, raising the possibility that colly- 339340 (1999).

news and views
in response to current application (Fig. 1a

Seeing is relieving: electrical and b). The spread of applied current was
directly demonstrated between closely
synapses between visualized spaced neurons, which were examined in

several neocortical regions. More than half
the tested cell pairs were coupled. Spread
neurons of both hyperpolarizing and depolarizing
currents was observed, and transmission
Michael V.L. Bennett was reciprocal. Electrical synapses act as
low-pass filters (preferentially transmit-
Two reports of electrical synapses between identified inhibitory ting low-frequency signals; Fig. 1c), and
neurons in the neocortex suggest that many neuron types known so the postsynaptic response was delayed
to communicate chemically may also be electrically coupled. and slowed as well as attenuated com-
pared to the presynaptic input. The filter-
ing was most obvious with respect to
The controversy over whether synaptic gap junctions between neurons are too presynaptic impulses, and the peak of the
transmission is chemical or electrical primitive to be considered synapses (see postsynaptic potential was obviously
alternately raged and simmered during Box). The definition used here (which delayed relative to the peak of the presy-
the first half of the 20th century. Resolu- Cajal and Sherrington would have naptic potential when recorded at higher
tion came in the 1950s, and the clear endorsed and Bullock and Eccles did time resolution3. Coupling was specific
answer was both; there are electrical endorse) is that a synapse is a morpholog- and did not occur between inhibitory
synapses and chemical synapses1. At that ically specialized structure mediating interneurons and nearby pyramidal cells or
time, many neuroscientists (a term not yet action between neurons or between between pyramidal cells. Moreover, two
widely used) thought electrical transmis- neurons and follower cells. Use of the populations of inhibitory neurons in the
sion would be uncommon in mammals, phrase chemical synapse is implicit same region but with different firing char-
because the first electrical synapses were recognition of this view (as gin martini acteristics were coupled within popula-
found in invertebrates and lower verte- legitimizes martinis made with vodka). tions but hardly at all between them4.
brates. Furthermore, transmission at elec- The major technological advance The extent of coupling remains to be
trical synapses did not show the activity underlying the recent electrophysiologi- determined. How far does the coupling
dependence commonly observed at chem- cal demonstration of electrical synapses3,4 spread over the cortex? Are there small
ical synapses. Indeed, as ultrastructural was infrared differential interference con- domains of coupled cells or does coupling
studies of mammalian CNS became more trast microscopy applied to brain slices7. extend for long distances? Are cell pairs of
refined, almost all synapses proved to have This technique permits visualization of the same type that are not coupled to each
presynaptic vesicles and other specializa- neighboring cells in a brain slice relative- other also not coupled to any other cell?
tions characteristic of chemical transmis- ly far below the cut edge. Slices also per- This coupling does not yet have an
sion, and few synapses showed gap mit placement of two electrodes with anatomical substrate, although the pres-
junctions, the structural underpinning of much less likelihood of mechanical inter- ence of gap junctions is indicated by the
the most common form of electrical actions during the manipulations. Because uncoupling effect of octanol, a common
synapse. (Other modes of excitatory and this approach is not that new, a change in gap junction blocker4. (Octanol has mul-
inhibitory electrical transmission are perceptual set (or equipment budget) may tiple effects on neurons, and alteration of
known1,2.) Now it seems that electrical have contributed to the new observations. other functions, such as repetitive firing,
synapses may be more frequent than most In these studies3,4, the inhibitory neurons by octanol does not indicate gap junction
neuroscientists thought. In a recent issue were identified first visually by shape and involvement.) Dendrodendritic gap junc-
of Nature, two groups reported electrical then by their characteristic firing patterns tions occur between inhibitory neurons8.
coupling of inhibitory neurons in the neo-
cortex 3,4 , and other reports will soon
appear. In addition, a widely expressed, SO WHICH ARE MORE PRIMITIVE, CHEMICAL SYNAPSES OR ELECTRICAL SYNAPSES?
neuron-specific, gap junction protein has Connexins, the family of homologous proteins forming gap junctions in vertebrates,
recently been cloned5,6. are unrelated to innexins, the gap-junction-forming proteins in invertebrates of the
Over the years, the catalog of electrical protostome line16. Innexin gap junctions show an astonishing degree of convergence
synapses in adult mammals, identified with connexin gap junctions in overall morphology, in having four membrane span-
physiologically and/or anatomically, has ning regions with two extracellular loops and in being blocked by the same agents,
grown slowly1,2. However, the incidence of such as hydrogen ions, calcium and octanol. The two types differ not only in sequence,
electrical synapses compared to chemical but also in gap width, interchannel separation and channel diameter. The apparently
synapses in the adult has remained rela- independent, although convergent, evolution of gap-junction-forming proteins is in
tively low. These data would be even more marked contrast to molecules underlying action potential generation and chemically
tilted toward chemical synapses if more mediated transmission, most if not all of which have homologs in vertebrates,
researchers shared the mistaken view that Drosophila and, of course, C. elegans. Thus, it seems that gap-junction-mediated com-
munication evolved after chemical transmission and, from this point of view, is more
Michael Bennett is in the Department of advanced. Later evolutionary development of gap junctions can also be argued from
Neuroscience, Albert Einstein College of unicellular organisms, which secrete and respond to chemical signals but do not com-
Medicine, Bronx, New York 10461, USA. municate via gap junctions.
e-mail: mbennett@aecom.yu.edu

news and views
Alternatively, coupling may result from observed in one experiment. Dye coupling studies3,4, electrically coupled neurons
presynaptic fibers that make gap junctions of cortical inhibitory interneurons has could also be connected in one or both
on more than one postsynaptic neuron1. been reported9, but the rats used in these directions by inhibitory chemical synaps-
Neurobiotin injection confirmed the cell experiments may have been younger es. Stimulation of a cell that is connected
type 4 , but tracer coupling was only when dye coupling of cortical neurons is both electrically and chemically produces
more prominent10. Failure to observe dye a biphasic, depolarizinghyperpolarizing
coupling may arise from geometric fac- potential in the postsynaptic cell. Model-
a tors (relationship between junctional con- ing indicates that reciprocal or recurrent
ductance and cell volume and extent of inhibition without electrical coupling can
the coupled network) or from imperme- mediate synchronous firing, given a dis-
ability of junctions to the tracer. tributed excitatory input11. Thus, it is not
What does electrical coupling do for clear that the electrical coupling is
these cells? In both studies, impulses in required for the oscillations observed in
one cell were insufficient to excite anoth- inhibitory neurons (and their follower
er, coupled cell, but if both cells were cells), but it would make the cells fire
depolarized to near threshold, synchro- more synchronously. This synchronizing
10 m nous firing was observed. Thus, a distrib- action is not simply mutual excitation;
uted excitatory input to these cells would whereas a more depolarized cell tends to
tend to activate them together, and elec- excite a less depolarized cell, the less depo-
trical coupling would increase the degree larized cell tends to inhibit the more depo-
ba of synchronization. Electrical coupling is larized cell. The synaptic relationship
likely to contribute to the synchronous between them is inhibitory as well as exci-
oscillations that are often observed in tatory, independent of whether there are
pools of inhibitory neurons9,11. In both chemical inhibitory synapses in either or
Cell 1 both directions.
Electrical transmission is commonly
thought to be faster than chemical trans-
Fig. 1. Electrical transmission in practice and mission, and this is true in respect to the
in theory. (a) IR-DIC micrograph of two corti-
beginning of charge transfer across a gap
cal inhibitory neurons just before recording4.
The electrodes enter the image from either junction during the rising phase of the
side. (b) Depolarizing and hyperpolarizing presynaptic impulse. However, because
Cell 2 pulses applied in cell 1 (superimposed records, there is a threshold for detection, the mea-
currents on middle trace) spread to cell 2 (ref. sured postsynaptic potential is delayed
4). Impulses in cell 1 caused brief depolariza- with respect to its actual onset, and the
tions in cell 2; relative hyperpolarizations time required to charge the postsynaptic
between these depolarizations may have capacitance increases this delay. The delay
included a contribution from inhibitory can be a significant fraction of the rise time
26 mV
2 chemical synapses. Synaptic noise observed
of the presynaptic action potential
50 ms during hyperpolarization was asynchronous
in the two cells, indi
(Fig. 1c) and easily longer than the mini-
c cating that they had mum delay of chemical transmission at
separate synaptic inputs. mammalian body temperature (0.2 ms).
Vpre gj Vpost (c) Equivalent circuit There are few sites in the mammalian ner-
of transmission at an vous system where the time saved in elec-
electrical synapse1. The trical transmission is important. The
C gpost
gap junction has conduc- selection advantages of electrical synapses
tance, gj; the postsynap- seem to lie in reciprocity and transmission
tic cell has conductance, of subthreshold potentials. Where activi-
gpost, and capacitance, C.
ty of similar cells is synchronized, the syn-
The postsynaptic poten-
Voltage
tial calculated from this chronization is not so precise as to require

Presynaptic impulse circuit is slowed with the shorter latency of electrical transmis-
respect to the presynap- sion. The new observations extend previ-
Electrical
EPSP tic impulse because of ous demonstrations of coupling in
time required to charge neuronal populations where activity is syn-
and discharge the capaci- chronous but not precisely so, for exam-
tance. The postsynaptic ple, in neurons of the locus coeruleus12,
potential has a signifi hypothalamic neurosecretory cells13, pre-
cant latency because
ganglionic neurons of the spinal cord14 and
the detection threshold
introduces delay. Panels
respiratory neurons of the pre-Boetzinger
Detection threshold
(a) and (b) are re- complex (J.C. Reckling and J.L. Feldman,
Bob Crimi printed by permission personal communication).
Time
from Nature (ref. 4), Cloning and sequencing of genes
Apparent copyright 1999 Macmillan involved in chemical transmission have
delay
Magazines Ltd. made the classical criteria for chemical

news and views
transmission almost irrelevant (but we ately, phrases such as gap junction-like. ascribed to reciprocity and transmission
should not forget our roots, and new The freeze-fracture technique displays of subthreshold potentials that facilitate
chemical transmitters keep appearing). All large areas of membrane surface and can synchronization. This general physiolog-
neurons may express machinery for reveal junctions smaller in diameter than ical outcome may be achievable in other
chemical transmission, but many of them the usual 70 nm of thin sections, and ways, but electrical transmission is defi-
also transmit electrically. Identification of these junctions would be difficult to detect nitely in the common neuronal repertoire.
connexins, the family of gap junction pro- by that technique. In freeze fracture, cell
teins, should be nearing completion for type may be hard to identify, and because 1. Bennett, M. V. L. in Cellular Biology of
mammals. The sequence information per- there are many CNS gap junctions Neurons Vol. I, Sec. I, Handbook of
Physiology. The Nervous System (ed., Kandel,
mits identification by in situ hybridization between glia, neuronal localization of a E. R.) 357416 (Williams and Wilkins,
and northern analysis as well as RT-PCR gap junction must be demonstrated. The Baltimore, 1977).
from single cells. Antibodies to connexin- situation can be improved by a new and 2. Jefferys, J. G. R. Physiol. Rev. 75, 689723 (1995).
specific peptides are being used at the light laborious approach, where there is the 3. Galarreta M. & Hestrin, S. Nature 402, 7275
and electron microscope levels. A major interest and manpower to apply it15. Slices (1999).
breakthrough for mammalian electrical of CNS tissue are frozen, fractured and 4. Gibson, J. R., Beierlein, M. & Connors, B. W.
synapses is the cloning of Cx36, a (near- shadowed, and the remaining tissue and Nature 402, 7579 (1999).
ly) neuron-specific connexin5,6. In situ replica are mounted, replica down, on an 5. Condorelli, D. F. et al. Eur. J. Neurosci. 10,
hybridization and immunocytochemistry EM grid. After confocal microscopy to 12021208 (1998).
show much broader distribution of this identify cells at specific sites, the tissue is 6. Shl, G., Degen, J., Teubner, B. & Willecke, K.
FEBS Lett. 428, 2731 (1998).
connexin than was previously appreciat- dissolved away and the replica examined
7. Dodt, H. U. & Zieglgnsberger, W. Brain Res.
ed for electrical synapses (J.E. Rash, T. with knowledge of what cells were at the 537, 333336 (1990).
Yasumura, W.A. Staines, D. Patel & J.I. fracture surface. This approach has
8. Katsumaru, H., Kosada, T., Heizmann, C. W.
Nagy, Mol. Biol. Cell. 10, 404a, 1999). demonstrated a high incidence of mor- & Hama, K. Exp. Brain Res. 72, 363370
Cx36 appears abundant not only in areas phologically mixed synapses, that is, with (1988).
where many electrical synapses are known both gap junctions and transmitter release 9. Benardo, L. S. J. Neurophysiol. 77, 31343144
(retina, inferior olive, olfactory bulb, stria- sites, in the rat spinal cord. (1997).
tum and now neocortex) but also in areas Clearly, tissue slices and IR microscopy 10. Peinado, A., Yuste, R. &, Katz, L. C. Neuron 10,
where electrical coupling or gap junctions coupled with whole-cell patch clamping 103114 (1993).
are less clearly demonstrated. make feasible the characterization of the 11. Jefferys, J. G. R., Traub, R. D. & Whittington,
M. A. Trends Neurosci. 19, 202208 (1996).
Where have all these putative gap junc- microcircuitry of neighboring neurons.
tions been hiding? Very likely some of Cells can be visualized, activated directly 12. Ishimatsu, M. & Williams, J. T. J. Neurosci. 16,
51965204 (1996).
them have been seen, but adequate fixa- or synaptically, and filled with tracer for
13. Hatton, G. I. & Li, Z. Adv. Exp. Med. Biol. 449,
tion of the mammalian CNS for electron Golgi-like analysis. Immunocytochem- 7995 (1998).
microscopy is difficult. Visualization of the istry of physiologically characterized neu-
14. Nolan, M. F., Logan, S. D. & Spanswick, D. J.
diagnostic seven layers of a gap junction rons is in the offing. Neurons can express Physiol. (Lond.) 519, 753764 (1999).
in cross section requires optimum prepa- specific connexins, and it is clearer than 15. Rash, J. E. et al. Proc. Natl. Acad. Sci. USA 93,
rations, and junction diameter must be before that electrical transmission is like- 42354239 (1996).
large enough to extend through the entire ly to be found wherever it is useful. At 16. Phelan, P. et al. Trends Genet. 14, 348349
section. The literature contains, appropri- many sites, the selective advantage can be (1998).
Targeting visual motion into how the brain analyzes the motion
of complex visual patterns.
Jochen Braun An early step in understanding motion
processing was the moving plaid stimu-
A new stimulus display reveals that humans summate the lus4, which is a superposition of two sinu-
soidal gratings moving in different
motion energies of all components consistent with a single directions. Depending on the details of the
velocity, rather than optimizing sensitivity by ignoring noise. stimulus, this display can be perceived
either as two sets of gratings sliding past
In vision research, decisive advances are targeted displays is to remove all but one each other, or as a single coherent plaid
often the result of cleverly targeted stim- type of visual information, so that the pattern moving in a third direction
ulus displays. Two classic examples are stimulus activates only a very restricted (Fig. 1a). Where in the brain do these per-
Bela Julesz demonstrations that the visu- class of perceptual mechanisms. This ceptions arise? In visual cortical area V1,
al system can process both binocular dis- allows the mechanisms in question to be which represents an early stage in visual
parity1 and differences in visual textures2 studied in quasi-isolation, and in many processing, motion-sensitive neurons
at a surprisingly early level. The aim of cases it can lead to the identification of respond merely to the separate compo-
the underlying neural substrates of per- nent motions of the two gratings5. This is
Jochen Braun is at the Institute of ception within the visual cortex. On page because neurons at this level are sensitive
Neuroinformatics, Uni/ETH Zrich, 64 of this issue, Schrater et al.3 present a only to contours at a particular visual loca-
Winterthurerstr. 190, 8057 Zrich, Switzerland. new type of motion stimulus that tion and orientation, and can only signal
e-mail: achim@ini.unizh.ch promises to provide important insights the motion component that is orthogonal

news and views
to their preferred orientation (the so-called tains information at many different spatial plaids, but also from translating patterns
aperture problem). The perception of scales. Because velocity equals temporal fre- of random dots10, which have also been
coherent motion, which requires the sep- quency divided by spatial frequency, a multi- popular for studying motion. Whereas
arate motion components to be combined, scale pattern that is moving with a given Schrater et al. can target particular popu-
seems to arise in MT, a higher area that is velocity contains motion energy at many lations of V1 neurons, random dot dis-
known to be involved in motion percep- different combinations of spatial and tem- plays give rise to motion energy that is
tion, and where a proportion of neurons poral frequencies. Moreover, because of the distributed over the entire velocity plane.
responds to the coherent pattern motion aperture problem, the amount of energy at Random dot displays thus suffer from an
that is actually perceived6,7. The moving a particular combination of spatial and tem- important but rarely recognized limita-
plaid is of course a highly simplified stim- poral frequency depends also on the orien- tion: they make it difficult to study inter-
ulus, and real-world objects are typically tation preference of the detector. For this actions between the respective neural
defined by outlines composed of contours reason, motion components along the x and representations of distinct but consistent
with many different orientations. The var- y axes must be considered separately. component motions, because they do not
ious component motions from these con- Movement in a given velocity thus cre- stimulate these representations differen-
tours are expected to be represented in ates motion energy over a range of tem- tially. In other words, random dot displays
area V1, but underlying motion of the poral frequencies and spatial frequencies turn out to be a poor choice for studying
object as a whole is thought to be recov- along the x and y axes. The set of all com- the integration of component motions.
ered only in area MT (Fig. 1b). A major binations that are consistent with a par- Armed with their new displays,
open issue is how this recovery is accom- ticular velocity (speed and direction of Schrater et al. revisit the question of how
plished; that is, how do neurons in area movement) lie on a tilted plane; specifi- component motions are integrated to
MT integrate and combine the informa- cally, the tilt angle (angle between the recover pattern motion. To this end, they
tion they receive through projections from plane and the t-axis) reflects the speed, combine one or more component
neurons in area V1? and the direction of the tilt (the angle in motions and, by varying the contrast in
When moving plaid displays were first the x,y-plane) reflects the direction of their stimuli, measure how readily an
introduced, there were high hopes that they the velocity vector. observer can detect the presence of
would reveal how component motions are Figure 2 illustrates the energy distribu- motion in each case. Statistical decision
combined into pattern motion. Unfortu- tion for each of the five variants of the dis- theory predicts how an ideal observer
nately, these hopes have not been fulfilled. play used by Schrater and colleagues. The would perform in this situation: the con-
The problem is that the moving plaid dis- simplest variant is the component display, trast necessary to detect motion of n com-
play is not sufficiently targeted, in that it which contains only one component ponents decreases in proportion to 1/n.
still contains several kinds of motion infor- motion and targets one population of area By definition, an ideal observer summates
mation. Specifically, the intersections V1 neurons. A slightly more complex vari- motion energy from all relevant parts of
between the two gratings create moving ant is equivalent to a plaid display, in x,y,t space (that is, parts stimulated
discontinuities, or nodes, and these nodes which two component motions target two by a component motion) and ignores
(rather than the component gratings) can separate populations of area V1 neurons motion energy from all irrelevant parts.
govern the perception of coherent (see also Fig. 1c). The authors show that human observers
motion7,8. For this reason, moving plaids The displays of Schrater and col- generally do not conform to this prediction,
may tell us more about visual sensitivity to leagues 3 differ not only from moving which implies that they are unable to mon-
moving nodes than about the integration
of component motions. Schrater et al. now
introduce a new type of visual display that
finally overcomes this limitation3 (Fig. 1c). a
A bB c
C
To generate this display, the authors filter
dynamic random noise in a way that con-
centrates motion energy in several distinct
subregions of spatio-temporal frequency
space, each corresponding to one compo- pattern component
nent motion (Fig. 2). The stimulus is per-
ceived as an amorphous pattern that y pattern y
appears to drift in one or more directions,
but which (unlike the moving plaid dis- component
play) lacks any persistent features that t t
could serve as the basis for a feature-track-
x x Amy Center
ing mechanism of motion detection. Cru-
cially, the energy distribution within each Fig. 1. Computation of motion from its components. (a) Real-world objects contain contour seg-
subregion matches the sensitivity of V1 ments of many orientations. Object motion causes each contour segment to move orthogonally
neurons, being concentrated at one partic- to its orientation. These component motions must be pooled to recover the pattern motion of
the object as a whole. (b) Moving plaid stimulus, consisting of two gratings drifting, respectively,
ular combination of spatial and temporal
toward the upper right and the lower right of the xy plane. The pattern as a whole is perceived as
frequencies, x, y, and t5,9. moving directly rightward. The top and side surfaces show how the pattern changes over time (xt
To understand this, recall that V1 neu- plane and yt plane. (c) Stimulus of the type used by Schrater et al. This stimulus is equivalent to
rons are tuned not only for orientation but the moving plaid in (b), in that it contains motion energy in the same two directions and is per-
also for particular spatial and temporal fre- ceived as moving directly rightward. Note, however, the lack of specific features that can be
quencies. A typical real-world pattern con- tracked over longer times (compare top surfaces in b and c).

news and views
a Component b Plaid c Planar velocity when inconsistent motion compo-

nents are present12,13. Given the evidence
t t
for motion opponency in area MT neurons,
t it is puzzling that there seem to be no psy-
chophysical counterpart in the results of
Schrater and colleagues3. Perhaps the rea-
son is that opponent motions interact only
y y y in close spatial proximity12, which these spa-
tially unstructured displays do not provide.
x x x Another unresolved issue is how the
spatial organization of a visual display influ-
t
ences integration over component motions.
d t e In many situations, motion integration
depends critically on whether component
motions are aligned to form edges, corners
and so on 14,15. The importance of the
nodes in moving plaid displays5,7 provides
y x y one example. The display of Schrater et al.,
x which purposefully avoids any lasting spa-

tial structures, cannot be expected to
In plane Off plane address this important issue. Thus it

Fig. 2. Stimuli used by Schrater et al., depicted in three-dimensional spatiotemporal frequency space.
remains to be determined whether the pla-
The graphs can be understood as representing the motion energy at each point in the space. Any
nar summation rule holds true also under
given velocity is defined by a single plane. The filters used by the authors concentrate the energy at more natural stimulus conditions, which
certain locations in space, and the shape of each filter is indicated by a surface representing 65% of typically involve the dynamic transforma-
the peak response. (a) A single component is ambiguous and does not define an unique plane. tion of lasting spatial structures. Whatever
(b) Two components form a plaid and define a plane corresponding to the perceived pattern motion. the answer to this question, these innova-
(c) A series of coplanar filters allow all possible motion signals that are compatible with a particular tive displays, with their ability to target spe-
motion velocity. (d) A third component is added in the same plane as the previous two. (e) A third cific populations of area V1 neurons, will
component is added that is out of plane, and thus incompatible with a single motion velocity. surely prove to be a great boon to research
on the perception of visual motion.
1. Julesz, B. Proc. Natl. Acad. Sci. USA 61,
437441 (1968).
itor arbitrary subsets of x,y,t space. The of a third motion component on the veloc-
2. Julesz, B. Nature 290, 9197 (1981).
only situation in which human observers ity plane should reduce thresholds by 2/3,
3. Schrater, P. R., Knill, D. C. & Simoncelli, E. P.
approach ideal performance is when the whereas a third component that is off the Nat. Neurosci. 3, 6468 (2000).
displays contain 10 consistent component velocity plane should leave thresholds
4. Adelson, E. H. & Movshon, J. A. Nature 300,
motions, in which the energy is distributed almost unaffected. Both predictions are 523525 (1982).
over (almost) an entire velocity plane borne out by the results from human 5. Watson, A. B. & Ahumada, A. J. J. Opt. Soc.
(planar display, Fig. 2c). In this case, the observers, further strengthening the already Am. A 2, 322342 (1985).
factor by which human thresholds differ compelling case for planar summation. 6. Movshon, J. A., Adelson, E. H., Gizzi, M. S. &
from the one-component display is indeed Naturally, not all questions about the Newsome, W. T. Exp. Brain Res. 11, 117152
approximately 1/10. integration of component motions can be (1986).
This outcome suggests that human answered by a single study. In particular, 7. Stoner, G. R. & Albright, T. D. Nature 358,
412414 (1992).
observers summate motion energy over the the planar summation revealed by the psy-
entire planar subspace corresponding to chophysical results of Schrater et al. is only 8. Stoner, G. R., Albright, T. D. &
Ramachandran, V. S. Nature 344, 153155
any particular velocity, rather than over partially consistent with the responses of (1990).
arbitrary subspaces (as an ideal observer neurons in area MT. Planar summation is 9. Watson, A. B. & Turano, K. Vision Res. 35,
would do). To corroborate this suspicion, consistent with the sensitivity of some MT 325336 (1995).
the authors performed an additional exper- neurons to pattern velocity, which can be 10. Britten, K. H., Shadlen, M. N., Newsome,
iment in which they examined the effect of explained by assuming that they sum exci- W. T. & Movson, J. A. J. Neurosci. 12,
adding a third component motion to a dis- tatory inputs from V1 neurons that are 47454765 (1992).
play already containing two component tuned to points lying in the preferred veloc- 11. Simoncelli, E. P. & Heeger, D. J. Vision Res. 38,
743761 (1998).
motions. In one case, the third component ity plane in x,y,t space11. However, other
was consistent with the velocity of the first properties of MT neurons suggest that they 12. Qian, N. & Andersen, R. A. J. Neurosci. 14,
73677380 (1994).
two components (in-plane display, may also receive inhibitory input from V1
13. Heeger, D. G., Boynton, G. M., Demb, J. B.,
Fig. 2d), whereas in the other case the third neurons tuned to points that lie far away Seidemann, E. & Newsome, W. T. J. Neurosci.
component was inconsistent with this from the preferred velocity plane. In par- 19, 71627174 (1999).
velocity (off-plane display, Fig. 2e). If ticular, many MT neurons show a phe- 14. Lorenceau, J. Vision Res. 36, 34153427 (1996).
observers do indeed summate motion nomenon called motion opponency, that 15. Lorenceau, J. & Shiffrar, M. Visual Cognit. 6,
energy over planar subspaces, the addition is, a reduced response to their preferred 431460 (1999).

brief communications
Long-term visual experience uli were presented for 250 milliseconds twiceonce with either a
clockwise or counterclockwise reference tilt, and once with one of
recalibrates human seven possible inverted comparison tilts. Both reference-tilt direc-
tion and the presentation order of reference and comparison were
orientation perception randomly interleaved. Observers were required to indicate which
of the two trials produced the greatest impression of tilt. Reference
tilts between 5 and 20 were investigated. As a control, the same
David Whitaker and Paul V. McGraw observers were presented with non-alphanumeric stimuli using the
same procedure. These consisted of three circular Gabor patches
Department of Optometry, University of Bradford, Bradford BD7 1DP, UK
Correspondence should be addressed to D.W. (d.j.whitaker@bradford.ac.uk)
within which the sinusoidal luminance modulation was tilted either
clockwise or counterclockwise. These stimuli were chosen because
Sensory neural reorganization occurs as a result of visual experi- they are ideally suited to analysis by orientation- and frequency-
ence in the adult as well as early in life. Here we report and quanti- selective mechanisms early in the visual cortex.
fy a misperception of visual orientation: when alphanumeric From responses averaged across six observers, we calculated the
characters that are commonly tilted clockwise were presented with counterclockwise tilt that appeared perceptually equivalent to any given
a clockwise tilt, they were perceived as less tilted than the same stim- clockwise digit tilt, and vice versa (Fig. 1c). The diagonal dashed line
ulus horizontally inverted. In contrast, subjective perception of tilt represents the prediction that the matching tilt is exactly equal and
magnitude for horizontally inverted non-alphanumeric stimuli was opposite to the reference tilt. The control data for non-alphanumeric
similar to that for non-inverted stimuli. Therefore, misassessment stimuli lie almost exactly on this diagonal line. In contrast, a best-fit-
of tilt of alphanumeric characters probably reflects a persistent sen- ting linear regression to the alphanumeric data has a gradient almost
sory recalibration of orientation perception as a result of previous identical to that of the predicted function, but displays a constant error
long-term visual experience. of approximately 2.4. Thus, counterclockwise tilts are consistently per-
Considerable evidence suggests that specific visual experience early ceived as more tilted than their clockwise counterparts, providing
in life can result in a reorganization of orientation selectivity among empirical support for the perceptual observations of Fig. 1a and b.
cortical neurons; for example, the rearing of animals in an environ- It is well established that perceived orientation can be influenced
ment containing a restricted range of orientations leads to a prepon- by previous adaptation to a tilted stimulus7. This is classically known
derance of cortical neurons tuned to that orientation1,2. In addition, as the tilt aftereffect, an illusion that decays rapidly over time8. The
sensory neural reorganization can also occur throughout adulthood36. present findings, however, reflect a relatively persistent form of adap-
Alphanumeric characters such as those used in digital displays and tation within the visual system, a phenomenon probably conditioned
italicized text are commonly tilted clockwise by around 510; thereby extensive previous visual experience of alphanumeric characters
fore we asked whether long-term visual exposure to clockwise-tilted whose tilt, if present, is almost always in a clockwise direction. Restric-
alphanumeric characters influences our visual perception of tilt. tion of this misperception to alphanumeric characters implicates rel-
We found that the tilt of clockwise-slanted characters appeared atively high-level processes involved in character recognition. This
to be of much smaller magnitude than when such characters were kind of top-down regulation of perceived orientation receives con-
horizontally inverted (as when viewed in a mirror; Fig. 1a and b). To siderable physiological support from dynamic changes in the orien-
quantify this effect, we presented observers with stimuli consisting tation tuning of visual neurons via intracortical feedback
of three digits arranged side by side (888) and surrounded by a rec- mechanisms9. Our findings support the view that, in the sense that
tangular reference frame defining true vertical and horizontal. Stim- our everyday perceptions can be markedly influenced by unique prior
visual experience, we learn to perceive. A robust test of this interpre-
tation would be to establish whether the misperceptions we describe
a b exist in cultures that provide little or no exposure to italicized Latin
text or tilted digital numerals10.
Original display
ACKNOWLEDGEMENTS
P.V.M. was supported by a Vision Research Training Fellowship from the Wellcome Trust.
1. Blakemore, C. & Cooper, G. F. Nature 228, 477478 (1970).

Horizontally inverted display 2. Sengpiel, F., Stawinski, P. & Bonhoeffer, T. Nat. Neurosci. 2, 727732 (1999).
3. Merzenich, M. M. et al. J. Comp. Neurol. 224, 591605 (1984).
c 20 4. Gilbert, C. D. & Wiesel, T. N. Nature 356, 150152 (1992).
Clockwise 5. Darian-Smith, C. & Gilbert, C. D. Nature 368, 737740 (1994).
6. Gilbert, C. D., Das, A., Ito, M., Kapadia, M. & Westheimer, G. Proc. Natl. Acad.
Sci. USA 93, 615622 (1996).
10
7. Gibson, J. J. & Radner, M. J. Exp. Psychol. 20, 453467 (1937).
8. Magnussen, S. & Johnsen, T. Vision Res. 26, 661672 (1986).
Matching tilt
9. Ringach, D. L., Hawken, M. J. & Shapley, R. Nature 387, 281284 (1997).

0 10. Segall, M. H., Campbell, D. T. & Herskovitz, M. J. Science 139, 769771 (1963).
10 Fig. 1. Normal and horizontally inverted versions of a digital display (a) and
y = x italicized text (b). The horizontally inverted versions produce a greater impres-
Counter clockwise sion of tilt. This effect is quantified in the experimental data (c). For each one of
20 several clockwise (positive) and counterclockwise (negative) reference tilts, we
plotted the mirror-image tilt that appears perceptually equivalent. Solid sym-
20 15 10 5 0 5 10 15 20 bols, data for the alphanumeric characters; open symbols, data for Gabor
Reference tilt patches (spatial frequency, 8 cycles per degree). Error bars represent s.e. The
linear regression that best fits the data has the form y = 2.385 0.989x.

articles
Molecular basis of NMDA receptor-

coupled ion channel modulation by
S-nitrosylation
Yun-Beom Choi1, Lalitha Tenneti1, Dean A. Le1, Justin Ortiz1, Guang Bai1,
Huei-Sheng Vincent Chen1 and Stuart A. Lipton1,2
1 Cerebrovascular and Neuroscience Research Institute, Brigham and Womens Hospital, and Program in Neuroscience, Harvard Medical School, 221 Longwood
Avenue, LMRC First Floor, Boston, Massachusetts 02115, USA
2 Present address: Del E. Webb Center for Neuroscience and Aging Research, The Burnham Institute, 10901 North Torrey Pines Road, La Jolla, California 92037, USA
Correspondence should be addressed to S.A.L. (slipton@burnham-inst.org)
Several ion channels are thought to be directly modulated by nitric oxide (NO), but the molecular
basis of this regulation is unclear. Here we show that the NMDA receptor (NMDAR)-associated ion
channel was modulated not only by exogenous NO but also by endogenous NO. Site-directed
mutagenesis identified a critical cysteine residue (Cys 399) on the NR2A subunit whose S-nitrosylation
(NO+ transfer) under physiological conditions underlies this modulation. In cell systems expressing
NMDARs with mutant NR2A subunits in which this single cysteine was replaced by an alanine, the
effect of endogenous NO was lost. Thus endogenous S-nitrosylation can regulate ion channel activity.
The N-methyl-D-aspartate class of glutamate receptor (NMDAR) increases during neuropathological conditions including
is critical for development, learning and memory in the central ischemia, trauma and HIV-associated dementia, and may be
nervous system1,2. Excessive activation of NMDARs, however, responsible at least in part for neuronal apoptosis and necrosis
contributes to the pathological processes of stroke, epilepsy and in such neurodegenerative conditions10,2831. Second, NMDAR
several neurodegenerative disorders36. Hence, precise regulation activation stimulates neuronal nitric oxide synthase (nNOS)32,33,
of NMDAR activity is critical to normal brain function as well and endogenous NO increases in brain tissue are detected with an
as the avoidance of neuronal damage and death; accordingly, electrochemical probe34. Third, putative inhibition of NMDAR
there are several endogenous mechanisms for receptor modula- function by endogenous NO/NO-related species should serve as
tion1,2. One mechanism of control of NMDAR-associated chan- negative feedback for excessive NMDAR-associated channel activ-
nels715 and other membrane ion channels1625 involves direct ity. Here, using site-directed mutagenesis of recombinant
modulation by nitric oxide (NO) via an unknown, cyclic GMP- NMDAR subunits, we demonstrate regulation of a channel by
independent molecular mechanism. Here we show that the endogenous NO via S-nitrosylation.
NMDAR-associated ion channel was modulated by both endoge-
nous and exogenous NO. We used site-directed mutagenesis on RESULTS
the NR2A subunit to identify a cysteine residue (Cys 399) critical Initially, we examined inhibition of recombinant NR1/NR2A
to this modulation, which is S-nitrosylated (NO+ is transferred receptor responses by exogenous S-nitrosothiols (SNOs) in the
to cysteine thiol) under physiological conditions. In cells express- Xenopus laevis oocyte, as this NMDAR subunit combination
ing NMDARs with NR2A subunits in which this cysteine was shows greatest modulation by these NO-generating com-
mutated, inhibition of NMDA-evoked responses by endogenous pounds13,14. Application of S-nitrosocysteine (SNOC) decreased
NO was abolished. Moreover, reversed-phase high-performance NMDA-evoked currents by 24 2% (mean s.e., n = 11;
liquid chromatography (RP-HPLC) demonstrated S-nitrosyla- Fig. 1a). The decrease in NMDA-evoked currents persisted with-
tion of Cys 399 in an NR2A peptide after NO+ transfer. Taken out significant spontaneous recovery for over ten minutes, but
together, these results suggest that S-nitrosylation can modulate was rapidly reversed by the chemical reducing agent dithiothre-
ion channel activity via endogenous NO-thiol reactions in intact itol (DTT), suggesting covalent modification of cysteine
cell preparations. Evidence both in vitro and in vivo indicates that residue(s) on the receptor.
nitric oxide synthase (NOS) activity can affect NMDA-evoked Next, we tested methanethiosulfonate derivatives, which react
responses of primary neurons11,26, suggesting that endogenous specifically and rapidly with thiol groups to form mixed disul-
form(s) of NO-related species can inhibit NMDAR function. In fides35, for ability to mask the action of NO-related species on
single-channel recordings from primary cortical neurons, NO recombinant NMDARs. Similar to SNOC, a two-minute appli-
decreases opening frequency of the NMDAR-associated ion chan- cation of 2-aminoethylmethanethiosulfonate (MTSEA; 500 M)
nel27. We used recombinant systems to describe the molecular led to inhibition of NMDA-evoked currents in oocytes express-
basis of this effect on NMDAR responses. ing NR1/NR2A receptors; MTSEA also occluded further inhibi-
Interactions between NO and NMDARs are particularly inter- tion by subsequent SNOC application (Fig. 1b). This result
esting for three reasons. First, similar to NMDAR activity, NO suggests that MTSEA interacts with cysteine residues and blocks

articles
a b on NMDAR responses 8,10, suggest that NO

species interact with cysteine residues to exert
their effect. However, this does not constitute
direct proof, and results with some of these
agents are inconsistent36. Therefore, we used
site-directed mutagenesis to directly study the
ISNOC/Icontrol (%)
possible involvement of cysteine residues in
this phenomenon. Using the reported con-
c sensus motif for S-nitrosylation15, we inspect-
ed NR1 and NR2A subunits for candidate
nitrosylation sites. From the candidate
cysteine residues, we chose to mutate to Ala
Cys 399, which is unique to NR2A among
NMDAR subunits and, according to pro
posed NMDAR-subunit transmembrane
No MTSEA Post MTSEA topology 3739, located in the extracellular N
treatment treatment
terminus of NR2A. We then tested the effect
Fig. 1. NO species generated by S-nitrosocysteine (SNOC) inhibit NMDA-evoked currents in of SNOC on NR1/NR2A(C399A) mutant
NR1/NR2A receptors, and the cysteine modifying agent (2-aminoethyl) methanethiosulfonate receptors. Inhibition of NMDA-evoked cur-
(MTSEA) occludes NO inhibition. (a) Inhibition of NMDA-evoked currents (200 M NMDA rents by SNOC-generated NO was virtually
and 100 M glycine) after SNOC application for 3 min in Xenopus oocytes expressing abolished in NR1/NR2A(C399A) receptors
NR1/NR2A receptors. Holding potential, 80 mV; pH 7.5. Following washout of SNOC, the (Fig. 3a and b). Inhibition of NMDA-evoked
decrease in NMDA-evoked currents persisted for 10 min. (b) Modifying cysteine residues currents by methanethiosulfonate agents was
with MTSEA (0.5 M) for 2 min inhibited NMDA-evoked currents in oocytes expressing
also virtually abolished in NR1/NR2A(C399A)
NR1/NR2A receptors. Exposure to MTSEA occluded further inhibition of NMDA-evoked
currents by SNOC. (c) SNOC inhibition of NR1/NR2A receptors expressed as a percentage
receptors (Fig. 3c).
of the response to NMDA before SNOC application (mean s.e., n = 4 in each case). SNOC Inspection of NR1 and NR2 subunits revealed
inhibition after MTSEA treatment was significantly different from that before treatment (*p < an additional four cysteine residues resembling
0.01; Students t-test). the consensus motif for S-nitrosylation. In addi-
tion to NR2A Cys 399, only Cys 744 and Cys 798
(in the M3M4 extracellular linker of NR1) and
Cys 87 and Cys 320 (in the extracellular N ter-
their interaction with the NO-related species generated by SNOC. minus of NR2A) fit the consensus motif. Therefore, we also mutat-
Within one batch of oocytes, inhibition of NMDA-evoked cur- ed these cysteines to alanines, yielding the heteromeric receptor
rents by SNOC in the absence of MTSEA (21 3%, n = 4) sig- NR1(C744A,C798A)/NR2A(C87A, C320A,C399A). The crystal
nificantly differed from inhibition after MTSEA modification structure of the GluR2 ligand-binding domain reveals a disulfide
(5 3%, n = 4; Fig. 1c). Additionally, a larger, charged agent, 2- bond between cysteine residues homologous to NR1 C744 and
(trimethylammonium)ethyl methanethiosulfonate (MTSET) had C798; this may also be true for NR2A C87 and C320 (ref. 40). In
an effect similar to that of MTSEA and also occluded SNOC inhi- the form of disulfide, these additional cysteine residues would not
bition of NMDA-evoked currents. Prevention of NOs effect by possess free thiol groups and would therefore not be expected to
bath application of this permanently charged, membrane-imper- react with NO. Indeed, under physiological conditions, we could
meant compound suggested that the critical thiol group(s)
were located extracellularly.
We also considered if the modulation of NMDAR cur-
rents by NO was voltage dependent. If the critical cysteine
residue were extracellular, we would predict that the effect
of NO would be voltage independent. We found that the
currentvoltage relationship for wild-type NR1/NR2A was
approximately linear before and after exposure to SNOC
(Fig. 2), indicating that inhibition of NMDA-evoked cur-
rents by NO was not significantly voltage dependent.
Results with methanethiosulfonate agents, coupled
with our previous finding in primary rat cortical cul-
tures that N-ethylmaleimide prevents the inhibitory
effect of NO donors such as SNOC and nitroglycerin
Fig. 2. Voltage independence of NO-induced inhibition of

NMDA-evoked currents. Currentvoltage (IV) relationship for
an oocyte expressing NR1/NR2A receptors before and after
exposure to SNOC for 2 min. IV curves were generated by con-
tinuously ramping the voltage at 47 mV per s from 100 to +50
mV, and the leakage currents were subtracted. This experiment
was repeated in three oocytes with similar results.

articles
Fig. 3. Effect of NR2A(C399A) on

NMDA-evoked responses. (a) Current trac- a b
ings showing the virtual absence of inhi
bition of NMDA-evoked currents by the
NO donor SNOC in oocytes express
ing NR1/NR2A(C399A) receptors. Similar
ISNOC/Icontrol (%)
results were obtained in n = 11 oocytes.
(b) Inhibition of NMDA-evoked currents by
SNOC in wild-type or mutant NR1/NR2A
receptors expressed as a percentage of the
response to NMDA before SNOC applica-
tion. Values are mean s.e.; n = 511 oocytes
for each group. SNOC inhibition of receptors c
containing the NR2A(C399A) subunit was
significantly different from that of wild-type
NR1/NR2A receptors (*p < 0.01, ANOVA
followed by Fishers PLSD test). (c) Current
tracings showing that inhibition of NMDA-
evoked currents by MTSEA (0.5 mM) was
nearly abolished in NR1/NR2A(C399A)
receptors. Similar results were obtained in

five oocytes. (d) NMDA doseresponse e
curves in the presence of a saturating concen-
tration of glycine (100 M). (e) Glycine
d
doseresponse curves in the presence of a
Normalized response (%)
Normalized response (%)

saturating concentration of NMDA (200 M).
These doseresponse curves were deter-
mined for NR1/NR2A receptors (closed cir-
cles) and NR1/NR2A(C399A) receptors
(open circles). EC50 values s.d. were calcu-
lated from doseresponse curves generated
by fitting the data with the equation
I = Imax/[1+(EC50/[agonist])n], where n is the
Hill slope, and Imax is the maximum response
normalized to 100%. In (d, e), each point rep- [NMDA] (M) [Glycine] (M)
resents mean s.e. of the responses obtained
from 46 oocytes.
not detect a statistically significant difference in SNOC-induced (Fig. 3e). Additionally, voltage-dependent blockade of NMDA-
inhibition of NMDA-evoked currents between NR1/NR2A(C399A) evoked currents from NR1/NR2A(C399A) receptors by 100 M
and NR1(C744A,C798A)/NR2A(C87A,C320A,C399A) receptors Mg2+ was indistinguishable from that of wild-type receptors.
(Fig. 3b). However, if we pretreated NR1/NR2A(C399A) with DTT These results indicate that the agonist sites and channel pore of
to reduce the disulfide bonds to free thiol groups, we then observed the NMDA receptor-channel complex were largely unchanged
an additional small but statistically significant inhibition by NO by the NR2A(C399A) mutation.
donors (11 4%, n = 10; p < 0.05 by ANOVA). Reminiscent of Next, we sought to detect inhibition of NMDAR responses
polynitrosylation of the ryanodine receptor25, these results suggest by endogenous NO and to determine the underlying molecular
that multiple cysteine residues may react with NO depending on mechanism for this effect. For this purpose, HEK293 cells stably
the redox state of the NMDAR; this may explain divergent find- transfected with nNOS42 were transiently transfected with wild-
ings 27,41. Our results indicate, however, that even though type NR1/NR2A or mutant NR1/NR2A(C399A) receptors.
NR1(C744,C798) and NR2A(C87,C320) can interact with NO, NMDA responses were monitored by digital Ca2+ imaging with
the effect is quite small and occurs under nonphysiological reduc- fura-2. Production of endogenous NO by these HEK-nNOS cells
ing conditions. Therefore, we characterized effects of endogenous was measured with diaminofluorescein (DAF-2), a new fluores-
NO on NMDAR responses by studying NR2A(C399) under phys- cent NO indicator43. In these cells, endogenous NO increased in
iological conditions (without previous exposure to reducing or response to NMDA in the presence of L-arginine, the substrate
oxidizing agents). for NOS. NO similarly increased in cells expressing either wild-
Before considering possible differences in physiological effects type NR1/NR2A or mutant NR1/NR2A(C399A) receptors
of endogenous NO on NR1/NR2A(C399A) or wild-type (Fig. 4b and c). Quantification of DAF-2 fluorescence in three
NR1/NR2A receptors, we had to determine whether this muta- experiments showed that NO level did not statistically differ
tion produced extensive structural changes affecting other between the wild-type and mutant receptors (wild-type, 84 11;
NMDAR properties. We constructed NMDA and glycine mutant, 102 16; mean arbitrary fluorescence units s.e.,
doseresponse curves for the wild-type and mutant receptors. n = 757 cells; p > 0.4 by Students t-test). In contrast, little or no
The NMDA EC 50 for NR1/NR2A(C399A) receptors was NO was detected in cells pretreated with L-nitro-arginine, a
54.8 1.6 M, compared with 37.5 1.7 M for wild-type recep- well -known NOS inhibitor (Fig. 4df). In the presence of L-argi-
tors (Fig. 3d). The glycine EC50 for NR1/NR2A(C399A) recep- nine, repeated pulses of NMDA, which led to progressive endoge-
tors was 2.1 0.2 M, compared with 2.3 0.2 M for wild-type nous production of NO, progressively reduced the magnitude of

articles
Fig. 4. Physiological effects of endogenous

NO on NMDAR responses of HEK293 Control vector WT NR1/NR2A NR1/NR2A(C399A)
cells transfected with NMDAR subunits
and nNOS. (af) Confocal images of
DAF-2 fluorescence, reflecting NO con-
tent, in HEK-nNOS cells transfected with
vector alone (a, d), wild-type NMDARs L-Arg
(WT NR1/NR2A; b, e) or NR1/NR2A
(C399A) mutant NMDARs (c, f). Cells
were stimulated with NMDA in the pres-
ence of L-Arg (ac) or L-Nitro-Arg (df).
Pseudocolor scale represents arbitrary
fluorescence intensity values ranging from
0 to 255. Cells transfected with either
wild-type NR1/NR2A (b) or mutant
NR1/NR2A(C399A) receptors (c) gener- L-N-Arg
ated similar amounts of endogenous NO.
Similarly, in both cases, NO formation was
blocked by L-Nitro-Arg. (g) Endogenous
NO decreases NMDAR physiological
activity. HEK293 cells stably transfected

with nNOS were transiently transfected
with wild-type NMDAR subunits (WT NR1/NR2A) or mutant
NR1/NR2A(C399A) receptor subunits. NMDA-evoked
[Ca2+]i responses were monitored with fura-2. Digital images g h
(L-Arg) (L-N-Arg)
were collected before and during exposure to NMDA. Data
Relative NMDA response
points are means s.e. of 3 independent experiments, each

representing 1220 imaged cells. Values represent the increase
in [Ca2+]i expressed as a ratio of the initial response for each
cell. The maximum [Ca2+]i response in an individual cell was
500 nM. In the presence of L-Arg, WT NR1/NR2A responses
decreased with time compared to NR1/NR2A(C399) mutant
responses (*p < 0.005, t-test). The relative WT NR1/NR2A
responses were not significantly different from unity, indicating
that there was no significant temporal change. (h) In the pres-
ence of L-Nitro-Arg to inhibit the generation of NO from
nNOS, NR1/NR2A(C399) mutant responses remained stable.
WT NR1/NR2A responses, however, increased after 10 to 20
min compared to the mutant NMDAR responses, probably
reflecting relief from inhibition by endogenous NO ( p < 0.05). Time (min) Time (min)
wild-type NR1/NR2A receptor Ca2+ responses (Fig. 4g). In con- tide showed a retention time of 12.5 min, irrespective of expo-
trast, responses of mutant NR1/NR2A(C399A) receptors were sure to the NO donor. These findings strongly suggest formation
not inhibited with serial NMDA application. Moreover, pre- of S-nitrosothiol (SNO) at cysteine 399 (ref. 44). We used this
treatment with L-nitro-arginine not only prevented NMDAs inhi- chemical approach to distinguish between different possible
bition of responses elicited from wild-type receptors, but also chemical modifications of the peptide; results implicated S-nitro-
actually increased response amplitudes (Fig. 4h). This result sug- sylation rather than formation of sulfenic acid or another deriv-
gests that endogenous NO already inhibited NMDAR responses ative. This finding is consistent with the notion that NO donors
at the time of application of L-nitro-arginine, but, because gen- induce S-nitrosylation of cysteine 399 rather than some other
eration of additional NO was blocked by this agent, the inhibi- chemical modification. Although relatively stable, the reaction
tion was relieved in 1020 minutes. As expected, L-nitro-arginine may spontaneously break down over tens of minutes; we observed
had essentially no effect on mutant NR1/NR2A(C399A) respons- this reversal of NOs effect on NR2A(C399) in both chemical and
es to NMDA. Taken together, these results can be best explained physiological experiments.
by physiological reaction of endogenous NO-related species with
the cysteine residue at position 399, in the extracellular domain of DISCUSSION
the NR2A subunit, to inhibit NR1/NR2A receptor responses. This study reports a mechanism of regulation of ion channel
To characterize the chemical reaction between NO and cys- activity by S-nitrosylation resulting from endogenous NOS activ-
teine 399, we synthesized a decapeptide containing the sequence ity in an intact cell system. S-nitrosylation of the NMDAR-asso-
surrounding this residue (KSFSDCEPDD) and a mutated pep- ciated ion channel is of particular interest because the action of
tide containing an alanine residue in place of the cysteine. We NO in the central nervous system is closely linked to the
then reacted these decapeptides with an NO donor and charac- NMDAR. NO is produced in certain neurons following NMDAR
terized the products by RP-HPLC with UV detection. Resulting activation and, in turn, triggers an increase in cGMP32; further-
chromatographs of the wild-type peptide revealed a shift in the more, entry of Ca2+ through NMDAR-associated channels stim-
major peak from a retention time of 13.5 minutes to 15.0 min- ulates nNOS activity42. The NMDAR NR1 subunit is physically
utes after reaction with the NO donor (Fig. 5). The mutated pep- coupled to nNOS by an intermediary adaptor protein, PSD95,

articles
[m7G(5)P(5)G] from 1 g of linearized template in

a c vitro by T7 (for NR1) or T3 (for NR2A) RNA poly-
merase according to the Ambion (mMessage mMachine)
protocol. Concentrations of cRNA were determined by
measuring the optical density at 260 nm, and purity was
monitored by agarose gel electrophoresis.
Preparation of oocytes and injection of cRNA. As

described previously50, frog oocytes of stages V or VI
were surgically removed from Xenopus laevis ovaries.
Lumps of about 100 oocytes were incubated with 580 U
b d per ml (2 mg per ml) collagenase type I (Sigma) for 2
hours in Ca2+-free frog Ringers solution (82.5 mM NaCl,
2 mM KCl, 1 mM MgCl2 and 5 mM HEPES, adjusted to
pH 7.5 with NaOH), with slow agitation to remove the
follicular cell layer, and then washed extensively with
Ca2+-free frog Ringers solution. Oocytes were main-
tained at 18C in frog Ringers solution (96 mM NaCl,
2 mM KCl, 1.8 mM CaCl 2, 1 mM MgCl 2 and 5 mM
HEPES, adjusted to pH 7.5 with NaOH), supplemented
with 550 mg per liter sodium pyruvate as a carbon source

and 100 g per ml gentamicin. Oocytes were injected
with up to 10 ng of cRNA of each subunit using a 10-l
Drummond microdispenser under a dissecting micro-
Fig. 5. RP-HPLC analysis with UV detection of S-nitrosylation products of wild-type (or scope 24 h later.
mutant) decapeptide containing the N-terminal cysteine (or alanine substitution) at residue
399 of the NR2A subunit. Ordinate axis, relative absorbance units at 220 nm; abscissa, Two-electrode voltage-clamp recording. Under two-elec-
retention time on column (min). (a) Wild-type peptide before exposure to an NO donor. trode voltage clamp at 80 mV in frog Ringers solution,
(b) Wild-type peptide following exposure to an NO donor. (c) Mutant (C399A) peptide oocytes were recorded 27 days after injection with an
before exposure to an NO donor. (d) Mutant peptide following exposure to an NO donor. Oocyte Clamp OC-725b amplifier (Warner Instrument,
Hamden, Connecticut) using MacLab v.3.5 software
(ADI Instruments) as described50. Voltage-sensing elec-
trodes had a resistance of 14 M and current-inject-
45 ing electrodes, 0.51 M; all were filled with 3 M KCl.
through a unique PDZPDZ domain interaction . Downregu- Oocytes were continuously superfused with a solution containing
lation of NMDAR activity by reaction of endogenous NO species 90 mM NaCl, 1 mM KCl, 10 mM HEPES, 1.5 mM BaCl2 and 100 M
with the thiol group of a critical cysteine residue, as shown here, glycine (pH 7.5). Barium, rather than calcium, was used as the divalent
suggests that the juxtaposition of these proteins in the submem- cation to minimize secondary activation of the calcium-activated chloride
brane space of the postsynaptic density may facilitate this form current. Drugs dissolved in frog Ringers solution were applied by super-
of feedback inhibition. Moreover, evidence that endogenous NOS fusion at a flow rate of 2 ml per min in a 100-l chamber with an array
activity can inhibit NMDAR responses in primary neurons11,26 of pipets similar to the sewer pipe system used in patch-clamp record-
supports the premise that this is a physiologically relevant phe- ing to achieve relatively rapid solution exchange. Complete solution
nomenon, although only synapses containing the NR2A subunit exchange in the recording chamber had a time constant of 3 s. SNOC
(nominally, 500 M to 1 mM) was applied through the solution bathing
would be affected by the mechanism detailed here.
the oocytes. At pH 7.5 and room temperature, however, the half-life of
Considering the range and ubiquity of proteins whose func- SNOC is only a few seconds, so the effective concentration of NO reach-
tion can be regulated by S-nitrosylation (refs. 15, 46, 47 and cita- ing NMDARs in the 100-l chamber is far less, on the order of a few
tions therein), this protein modification may be an important micromolar, as judged from measurements with an NO-selective elec-
component in the repertoire of cell regulation. Previous reports trode10.
emphasize intracellular signaling cascades that are regulated by
S-nitrosylation. The extracellular, rather than intracellular, mod- Transfection of HEK293 cells. We used a human embryonic kidney cell
ification of protein thiol by endogenous NO that we report may line that stably expressed nNOS (HEK293-nNOS), as described previ-
have additional implications for autocrine and paracrine signal- ously42. For expression of NMDAR subunits, HEK293-nNOS cells grow-
ing on polylysine-coated cover slips were transiently transfected with
ing pathways involved in cellular control.
NR1 and NR2A or NR2A(C399A) cDNAs harbored in a pCDNA3 vector
(Invitrogen, Carlsbad, California) in a ratio of 1:3. Transfection was
METHODS accomplished using GenePORTER reagents following the manufacturers
Site-directed mutagenesis. Mutant NMDAR subunits were generated instructions (GTS, Inc.). A pCMVb construct was cotransfected in a 1:10
using the Chameleon double-stranded site-directed mutagenesis kit ratio of total plasmid. The competitive NMDAR antagonist D-2-amino-
based on T7 DNA polymerase (Stratagene). Mutants were verified by 5-phosphonovalerate (D-APV, 200 M) was added 3 hours and again
sequencing the entire coding sequence. Multiple cysteine mutants were one day after transfection to prevent glutamate-induced cell death, but
generated as necessary by restriction enzyme digestion and subcloning was washed out before experiments. After assays for intracellular NO or
of relevant fragments. Ca2+ concentration, cells were fixed with 4% paraformaldehyde in PBS
and measured for -galactosidase activity with X-gal substrate. In general,
cRNA synthesis. cDNA templates for the NMDAR subunits were NR1-1a, 80% of the cells showed a blue color, indicating -galactosidase expres-
the exclusive NR1 splice variant used in these studies (a gift from S. F. sion, and thus, transfection efficiency.
Heinemann), and NR2A (a gift from P. H. Seeburg)48,49. For cRNA syn-
thesis, the template was prepared from a circular plasmid cDNA by lin- Digital fluorescence imaging of endogenous NO and Ca2+. 4,5-Diamino-
earizing in the 3 untranslated region with NheI (for NR1) or EcoRV (for fluorescein diacetate (DAF-2 DA) is a cell-permeable derivative of the
NR2A). We transcribed cRNA incorporating 5 cap analog indicator DAF-2 whose fluorescence reflects the level of endogenous

articles
NO43. DAF-2 was used to measure nitric oxide formation in HEK-nNOS 8. Lei, S. Z. et al. Effect of nitric oxide production on the redox modulatory site
cells following NMDA stimulation. Two days after transfection of of the NMDA receptor-channel complex. Neuron 8, 10871099 (1992).
9. Hoyt, K. R., Tang, L.-H., Aizenman, E. & Reynolds, I. J. Nitric oxide
NMDAR subunits and just before the experiment, the culture medium modulates NMDA-induced increases in intracellular Ca2+ in cultured rat
was exchanged for a physiological saline based on the Hanks balanced forebrain neurons. Brain Res. 592, 310316 (1992).
salt solution (HBSS). This saline consisted of 137.6 mM NaCl, 1 mM 10. Lipton, S. A. et al. A redox-based mechanism for the neuroprotective and
NaHCO3, 0.34 mM Na2HPO4, 5.36 mM KCl, 0.44 mM KH2PO4, 2.5 mM neurodestructive effects of nitric oxide and related nitroso-compounds.
Nature 364, 626632 (1993).
CaCl2, 5 mM HEPES and 22.2 mM dextrose, adjusted to pH 7.2 with 11. Manzoni, O. & Bockaert, J. Nitric oxide synthase activity endogenously
0.3 M NaOH. To enhance responses to NMDA, experiments were per- modulates NMDA receptors. J. Neurochem. 61, 368370 (1993).
formed in nominally magnesium-free solutions. Cells were loaded with 12. Fagni, L., Olivier, M., Lafon-Cazal, M. & Bockaert, J. Involvement of divalent
2 M DAF-2 DA in HBSS at room temperature for 60 min and rinsed. ions in the nitric oxide-induced blockade of N-methyl-D-aspartate receptors
Cells were then incubated in 500 M NMDA plus 5 M glycine in the in cerebellar granule cells. Mol. Pharmacol. 47, 12391247 (1995).
13. Omerovic, A., Chen, S.-J., Leonard, J. P. & Kelso, S. R. Subunit-specific redox
presence of 1 mM L-Arg or 1 mM L-Arg plus 1 mM L-nitro-Arg. DAF-2 modulation of NMDA receptors expressed in Xenopus oocytes. J. Recept.
fluorescence was detected using a laser scanning confocal microscope Signal Transduct. Res. 15, 811827 (1995).
(Noran). For intracellular free calcium ([Ca2+]i) analysis, cells were loaded 14. Sucher, N. J., Awobuluyi, M., Choi, Y.-B. & Lipton, S. A. NMDA receptors:
with 10 M fura-2 acetoxymethylester (fura-2/AM) instead of DAF-2 from genes to channels. Trends Pharmacol. Sci. 17, 348355 (1996).
15. Stamler, J. S., Toone, E. J., Lipton, S. A. & Sucher, N. J. (S)NO signals:
DA and analyzed as described8. Briefly, fura-2 fluorescence was excited Translocation, regulation, and a consensus motif. Neuron 18, 691696 (1997).
by a 75-W xenon light source. The excitation wavelength was alternated 16. Bolotina, V. M., Najibi, S., Palacino, J. J., Pagaon, P. J. & Cohen, R. A. Nitric
between 350 10 nm and 380 10 nm, with emission at 500 nm. Images oxide directly activates calcium-dependent potassium channels in vascular
were collected using a Quantex QX-100 intensified CCD camera mount- smooth muscle. Nature 368, 850853 (1994).
ed on a Zeiss Axiovert 35 microscope following stimulation with 100 M 17. Kurenny, D. E., Moroz, L. L., Turner, R. W., Sharkey, K. A. & Barnes, S.
Modulation of ion channels in rod photoreceptors by nitric oxide. Neuron 13,

NMDA plus 5 M glycine. Agonists were superfused onto the cells for 315324 (1994).
14 s, with a 10-min interval between applications. For these experiments, 18. Gaston, B. et al. Relaxation of human bronchial smooth muscle by S-
we monitored [Ca2+]i rather than membrane current to avoid rundown nitrosothiols in vitro. J. Pharmacol. Exp. Ther. 268, 978984 (1994).
after repetitive responses. In theory, endogenous NO could affect intra- 19. Koivisto, A. & Nedergaard, J. Modulation of calcium-activated non-selective
cellular Ca2+ homeostasis by mechanisms other than NMDAR activity cation channel activity by nitric oxide in rat brown adipose tissue. J. Physiol.
(Lond.) 486, 5965 (1995).
and thereby confound the results. In practice, however, this was not a 20. Koh, S. D., Campbell, J. D., Carl, A. & Sanders, K. M. Nitric oxide activates
problem in our experiments, because we compared cells (wild-type ver- multiple potassium channels in canine colonic smooth muscle. J. Physiol.
sus mutant) differing only by a single cysteine substitution in the NR2A (Lond.) 489, 735743 (1995).
subunit; thus, only changes in NMDAR activity could account for the 21. Campbell, D. L., Stamler, J. S. & Strauss, H. C. Redox modulation of L-type
calcium channels in ferret ventricular myocytes. J. Gen. Physiol. 108, 277293
observed changes in [Ca2+]i. (1996).
22. Broillet, M.-C. & Firestein, S. Direct activation of the olfactory cyclic
RP-HPLC with UV detection for S-nitrosothiol formation. Mutations nucleotide-gated channel through modulation of sulfhydryl groups by NO
are shown in single-letter code; for instance, Cys399 to Ala is represent- compounds. Neuron 16, 377385 (1996).
ed by C399A. For the chemical evidence of S-nitrosylation of this cys- 23. Takeuchi, T., Kishi, M., Ishii, T., Nishio, H. & Hata, F. Nitric oxide-mediated
relaxation without concomitant changes in cyclic GMP content of rat
teine residue, a reaction mixture of 200 l of 4 mM sodium nitrite (as proximal colon. Br. J. Pharmacol. 117, 12041208 (1996).
the NO donor) was added dropwise to 200 l of a 2 mM solution of the 24. Yuan, X.-J., Tod, M. L., Rubin, L. J. & Blaustein, M. P. NO hyperpolarizes
wild-type decapeptide or mutant decapeptide with alanine substituted pulmonary artery smooth muscle cells and decreases the intracellular Ca2+
for cysteine (each peptide was >95% pure by HPLC analysis; synthesized concentration by activating voltage-gated K+ channels. Proc. Natl. Acad. Sci.
USA 93, 1048910494 (1996).
by Research Genetics, Huntsville, Alabama). The reaction vessel was
25. Xu, L., Eu, J. P., Meissner, G. & Stamler, J. S. Activation of the cardiac calcium
placed on ice for 30 min, and 150 l of a 1.4 M sodium hydroxide solu- release channel (ryanodine receptor) by poly-S-nitrosylation. Science 279,
tion was added to neutralize the mixture. An aliquot of 50 l containing 234237 (1998).
the reacted peptide was then analyzed by HPLC (Waters, Model #486, 26. Kendrick, K. M. et al. NMDA and kainate-evoked release of nitric oxide and
Milford, Massachusetts) at 220 nm with a NovaPak C18 column, (3.9 classical transmitters in the rat striatum: in vivo evidence that nitric oxide
may play a neuroprotective role. Eur. J. Neurosci. 8, 16191634 (1996).
150 mm; Waters) using a linear gradient of 60 min starting with water 27. Lipton, S. A., Rayudu, P. V., Choi, Y.-B., Sucher, N. J. & Chen, H.-S. V. Redox
containing 0.01% trifluoroacetic acid and ending with 80% aqueous ace- modulation of the NMDA receptor by NO-related species. Prog. Brain Res.
tonitrile containing 0.01% trifluoroacetic acid. Under these conditions, 118, 7382 (1998).
the shift observed in the RP-HPLC UV peak in Fig. 5 is indicative of SNO 28. Dawson, V. L., Dawson, T. M., London, E. D., Bredt, D. S. & Snyder, S. H.
formation, as described44. Nitric oxide mediates glutamate neurotoxicity in primary cortical cultures.
Proc. Natl. Acad. Sci. USA 88, 63686371 (1991).
29. Dawson, V. L., Dawson, T. M., Bartley, D. A., Uhl, G. R. & Snyder, S. H.
ACKNOWLEDGEMENTS Mechanisms of nitric oxide-mediated neurotoxicity in primary brain
We thank J. M. Sullivan and E. Liman for sharing cDNA clones or vectors and cultures. J. Neurosci. 13, 26512661 (1993).
30. Bonfoco, E., Krainc, D., Ankarcrona, M., Nicotera, P. & Lipton, S. A.
T. Lishnak for technical assistance. This work was supported in part by National Apoptosis and necrosis: two distinct events induced respectively by mild and
Institutes of Health grants P01 HD29587, R01 EY05477 and R01 EY09024. intense insults with NMDA or nitric oxide/superoxide in cortical cell
cultures. Proc. Natl. Acad. Sci. USA 92, 71627166 (1995).
31. Adamson, D. C. et al. Immunologic NO synthase: elevation in severe AIDS
RECEIVED 24 AUGUST; ACCEPTED 29 SEPTEMBER 1999 dementia and induction by HIV1 gp41. Science 274, 19171921 (1996).
32. Garthwaite, J., Charles, S. L. & Chess, W. R. Endothelium-derived relaxing
factor release on activation of NMDA receptors suggests role as intercellular
1. Hollmann, M. & Heinemann, S. F. Cloned glutamate receptors. Annu. Rev. messenger in the brain. Nature 336, 385388 (1988).
Neurosci. 17, 31108 (1994). 33. Bredt, D. S. & Snyder, S. H. Nitric oxide, a novel neuronal messenger. Neuron
2. McBain, C. J. & Mayer, M. L. N-Methyl-D-aspartic acid receptor structure 8, 311 (1992).
and function. Physiol. Rev. 74, 723760 (1994). 34. Shibuki, K. & Okada, D. Endogenous nitric oxide release required for
3. Choi, D. W. Glutamate neurotoxicity and diseases of the nervous system. long-term synaptic depression in the cerebellum. Nature 349, 326328
Neuron 1, 623634 (1988). (1991).
4. Meldrum, B. & Garthwaite, J. Excitatory amino acid neurotoxicity and 35. Akabas, M. H., Stauffer, D. A., Xu, M. & Karlin, A. Acetylcholine receptor
neurodegenerative disease. Trends Pharmacol. Sci. 11, 379387 (1990). channel structure probed in cysteine-substitution mutants. Science 258,
5. Dingledine, R., McBain, C. J. & McNamara, J. O. Excitatory amino acid 307310 (1992).
receptors in epilepsy. Trends Pharmacol. Sci. 11, 334338 (1990). 36. Aizenman, E., Brimecombe, J. C., Potthoff, W. K. & Rosenberg, P. A. Why is
6. Lipton, S. A. & Rosenberg, P. A. Mechanisms of disease: Excitatory amino the role of nitric oxide in NMDA receptor function and dysfunction so
acids as a final common pathway for neurologic disorders. N. Engl. J. Med. controversial? Prog. Brain Res. 118, 5371 (1998).
330, 613622 (1994). 37. Wo, Z. G. & Oswald, R. E. Transmembrane topology of two kainate receptor
7. Manzoni, O. et al. Nitric oxide-induced blockade of NMDA receptors. subunits revealed by N-glycosylation. Proc. Natl. Acad. Sci. USA 91,
Neuron 8, 653662 (1992). 71547158 (1994).

articles
38. Hollmann, M., Maron, C. & Heinemann, S. F. N-glycosylation site tagging 44. Stamler, J. S. & Feelisch, M. in Methods in Nitric Oxide Research (eds.
suggests a three transmembrane domain topology for the glutamate receptor Feelisch, M. & Stamler, J. S.) 521539 (Wiley, Chichester, 1996).
GluR1. Neuron 13, 13311343 (1994). 45. Brenman, J. E. et al. Interaction of nitric oxide synthase with the postsynaptic
39. Wood, M. W., VanDongen, H. M. & VanDongen, A. M. Structural density protein PSD95 and 1-syntrophin mediated by PDZ domains. Cell
conservation of ion conduction pathways in K channels and glutamate 84, 757767 (1996).
receptors. Proc. Natl. Acad. Sci. USA 92, 48824886 (1995). 46. Lander, H. M. et al. A molecular redox switch on p21(ras). Structural basis for
40. Armstrong, N., Sun, Y., Chen, G.-Q. & Gouaux, E. Structure of a glutamate- the nitric oxide-p21(ras) interaction. J. Biol. Chem. 272, 43234326 (1997).
receptor ligand-binding core in complex with kainate. Nature 395, 913917 47. Mannick, J. B. et al. Fas-induced caspase denitrosylation. Science 284,
(1998). 651654 (1999).
41. Aizenman, E. & Potthoff, W. K. Lack of interaction between nitric oxide and 48. Monyer, H. et al. Heteromeric NMDA receptors: molecular and functional
the redox modulatory site of the NMDA receptor. Br. J. Pharmacol. 126, distinction of subtypes. Science 256, 12171221 (1992).
296300 (1999). 49. Sullivan, J. M. et al. Identification of two cysteine residues that are required
42. Bredt, D. S. et al. Cloned and expressed nitric oxide synthase structurally for redox modulation of the NMDA subtype of glutamate receptor. Neuron
resembles cytochrome P450 reductase. Nature 351, 714719 (1991). 13, 929936 (1994).
43. Kojima, H. et al. Detection and imaging of nitric oxide with novel 50. Choi, Y.-B. & Lipton, S. A. Identification and mechanisms of action of two
fluorescent indicators: diaminofluoresceins. Anal. Chem. 70, 24462453 histidine residues underlying high-affinity Zn2+ inhibition of the NMDA
(1998). receptor. Neuron 23, 171180 (1999).

articles
Collybistin, a newly identified brain-

specific GEF, induces submembrane
clustering of gephyrin
Stefan Kins1,2, Heinrich Betz1 and Joachim Kirsch1,3
1 Department of Neurochemistry, Max-Planck-Institute for Brain Research, Deutschordenstr. 46, D-60528 Frankfurt, Germany
2 Present address: Department of Psychiatric Research, University of Zrich, August-Forel-Str. 7, CH-8008 Zrich, Switzerland
3 Department of Anatomy and Cellular Neurobiology, University of Ulm, Albert-Einstein-Allee 11, D-89069 Ulm, Germany
Correspondence should be addressed to H.B. (neurochemie@mpih-frankfurt.mpg.de)
The formation of postsynaptic GABAA and glycine receptor clusters requires the receptor-associated
peripheral membrane protein gephyrin. Here we describe two splice variants of a novel gephyrin-binding
protein, termed collybistin I and II, which belong to the family of dbl-like GDP/GTP exchange factors
(GEFs). Co-expression of collybistin II with gephyrin induced the formation of submembrane gephyrin
aggregates that accumulate hetero-oligomeric glycine receptors. Our data suggest that collybistin II regu-
lates the membrane deposition of gephyrin by activating a GTPase of the Rho/Rac family. Therefore, this
protein may be an important determinant of inhibitory postsynaptic membrane formation and plasticity.
Information processing in the CNS depends on fast neurotrans- A high-affinity gephyrin-binding motif in the cytoplasmic
mission, which is mediated by receptor proteins that are selec- loop region of the GlyR subunit1618 suggests that heterologously
tively concentrated at distinct postsynaptic membrane expressed gephyrin is not localized in submembrane areas, but
specializations. Mechanisms that establish these specialized mem- forms large intracytoplasmic aggregates16. This suggests that
brane domains during synaptogenesis and determine the posi- important regulatory elements of gephyrin targeting to synaptic
tions, size and packing densities of postsynaptic neurotransmitter sites are still unknown. Here we used the yeast two-hybrid sys-
receptor clusters are poorly understood. Receptor-associated pro- tem19 to search for gephyrin-binding proteins that may be involved
teins are assumed to be important in these processes13. in regulating the subcellular distribution of gephyrin. We describe
Postsynaptic clustering of the inhibitory glycine receptor (GlyR)4 a new dbl-like GDP/GTP exchange factor that induced the for-
involves a multi-step process requiring the local release of neuro- mation of submembrane gephyrin-rich microdomains. These
transmitter from apposed nerve terminals followed by activation submembrane gephyrin clusters were capable of accumulating
of GlyRs and depolarization of the postsynaptic membrane5,6. The hetero-oligomeric GlyRs. Our findings suggest that the novel GEF,
latter is thought to trigger Ca2+ influx through voltage-gated L-type for which we propose the name collybistin (from Greek
calcium channels; this influx then induces submembrane accumu- : to exchange), is an important determinant of gephyrin
lation of the GlyR-associated protein gephyrin5. The primordial localization in neurons, and thus of both GlyR and GABAAR clus-
gephyrin clusters are proposed to trap GlyRs that diffuse laterally tering at developing postsynaptic sites.
in the neuronal plasma membrane, thereby concentrating recep-
tors beneath active presynaptic terminals5. RESULTS
Studies on cultured spinal neurons7 and genetic approaches8 Collybistins I and II, two dbl-like GEFs
underscore the importance of gephyrin in GlyR clustering. Both To identify proteins that interact with gephyrin, we screened a yeast
attenuation of gephyrin expression by antisense oligonucleotides7 two-hybrid cDNA library from newborn rat brain using the
and the targeted disruption of the gephyrin gene8 prevent forma- gephyrin clone p1 (ref. 20) as bait. This resulted in the isolation of
tion of postsynaptic GlyR clusters. Moreover, immunohistochem- five cDNAs representing two alternatively spliced mRNAs of 3786
istry9,10, experiments with cultured neurons11 and analysis of bp and 3068 bp (Fig. 1a). The corresponding open reading frames
gephyrin-knockout mice12 also implicate gephyrin in the synaptic encode proteins of 493 and 413 amino acids with calculated mole-
localization of -aminobutyric acid type A receptors (GABAARs). cular weights of 58.2 kDa and 48.6 kDa. In the 5 UTR, several stop
In addition, gephyrin has non-synaptic functions as a coenzyme- codons were found in frame with the putative start codon, indi-
synthesizing protein in different peripheral tissues8 and species13. cating that full-length cDNAs were identified. A BLAST search of
Overlay and copolymerization assays demonstrate high-affin- the GenBank database revealed high homology to the human com-
ity binding of gephyrin to polymerized tubulin14. Furthermore, plete cDNA clone KIAA0424, with 93% identity throughout the
the morphology of postsynaptic gephyrin/GlyR clusters is regu- coding region and 6070% identity in the 3 UTR, but no significant
lated by cytoskeletal elements15. These findings are consistent homology in the 5 UTR. The homology between the rat and
with a model in which gephyrin serves as both an anchor and an human cDNAs starts abruptly 27 bp downstream from the first
organizing molecule linking postsynaptic receptors to the sub- potential start codon at the second methionine codon, suggesting
membrane cytoskeleton15. that the latter may serve as translational start signal. The predict-

articles
ed primary structures of the rat and human a SH3 DH PH CC

proteins are 93% identical (Fig. 1b). Fluores-
collybistin I
cent in-situ hybridization to murine and
human metaphase chromosomes localized the
genes from which these mRNAs were tran- collybistin II
scribed to region D of the mouse and
q12.2q13 of the human X chromosome (data
not shown).
Analysis of the deduced amino-acid rat
sequences revealed a modular structure of the b human
40
proteins with several characteristic domains. rat 80

Notably, both splice variants harbor combined human:
dbl (DH) and pleckstrin (PH) homology rat 120
human:
domains connected by a short linker sequence
rat 160
(Fig. 1a). Such DH/PH tandem domains are
human:
found in all known dbl-like GEFs and are con- rat 200
sidered a signature of this family of nucleotide human:
exchange factors (Fig. 2a). In particular, the
rat 240
DH moiety (Fig. 2c) mediates the GDP/GTP human:
exchange activity of dbl-like oncoproteins21. rat 280
The role of the poorly conserved PH domain human:
rat 320
(Fig. 2d) is unclear and might involve the human:
proper subcellular localization of GEFs22 and rat 360
binding of the second messenger phos- human:
phatidylinositol 4,5-bisphosphate (PIP2)23. rat 400
Because these homologies suggest that our human:
cDNAs encode a GEF for small GTPases of the rat 440
human:
Rho/Rac-family, we named the two splice vari-
rat 480
ants collybistin I and II. human:
The N-terminal region of the longer colly- 520
bistin splice variant contains an additional Src rat
human:
homology 3 (SH3) domain24. The primary 560
human:
structures of SH3 domains, known to mediate
interactions with proline-rich regions of other Fig. 1. Primary structures of rat and human collybistin. (a) Schematic representation of colly-
bistin I and II cDNAs. Nucleotide sequences encoding src homology 3 (SH3, gray box), Dbl
polypeptides24, are highly variable among dif-
homology (DH, black box), Pleckstrin homology (PH, black box) and the predicted coiled-coil
ferent protein families (Fig. 2b). The SH3 domain (CC, hatched box) are indicated. The cDNAs of collybistin I and II differ only by
domain identified here displays a higher nucleotide sequences encoding SH3 and coiled-coil domains, absent from the collybistin II
homology to SH3 domains of cytoskeletal cDNA. (b) Comparison of the primary structures of rat and human collybistin. Identical amino
proteins (Fig. 2b) than to those of the PAK- acids are marked by an asterisk, nonidentical residues with (). The homology domains shown in
interacting exchange factors (PIXs)25. A sec- (a) are boxed, and the predicted coiled-coil domain is underlined. Collybistin II has a unique C
ond region of variation is found in the terminus (VTQRKWHY*), starting with residue 462 of the collybistin I sequence.
C-terminal region of the longer variant, which
contains a segment predicted to form a coiled
coil, a structure known to mediate protein
interactions (Fig. 1a). the adult rat brain (Fig. 4a and b). Hybridization signals were
localized to neuronal somata (not shown). Distribution patterns
Collybistin transcripts expressed predominantly in brain of collybistin transcripts resembled those of gephyrin mRNAs
Northern hybridizations of polyA+ mRNA from several rat tissues (Fig. 4c), although hybridization intensities differed in various
with oligonucleotides specific for either the collybistin I or both the brain regions. Most prominently, collybistin expression was near-
collybistin I and II mRNAs (Fig. 3a) and with a random-primed ly equal in the granular layer of the cerebellum and in the dien-
collybistin cDNA revealed predominant expression of 5.5 kb and cephalon, whereas gephyrin expression was considerably weaker
6.0 kb transcripts in brain (Fig. 3b). In heart and skeletal muscle, in diencephalon than in cerebellum.
weak signals were detected at 6.0 kb and 6.5 kb and at 6.5 kb and To express collybistin proteins, we generated collybistin I and
6.8 kb respectively upon prolonged exposure (not shown); none of II cDNAs that encoded a hemagglutinin- (HA) epitope tag at
the other tissues examined showed significant hybridization their C-terminal ends and subcloned both constructs in a mam-
(Fig. 3b). The oligonucleotide probes used and the random-primed malian expression vector. Expression of the recombinant pro-
collybistin cDNA produced identical results, indicating similar dis- teins in human embryonic kidney (HEK) 293 cells transfected
tribution and abundance of collybistin I and II mRNAs. These data with these constructs was demonstrated using a polyclonal anti-
show that the collybistin gene is predominantly transcribed in brain HA antibody and an antibody generated against a peptide posi-
and, to a rather low extent, in heart and skeletal muscle. tioned amino-terminal to the DH domain and common to both
In-situ hybridization using either antisense oligonucleotide collybistin splice variants (Fig. 3a). In western blots of lysates
on horizontal and parasagittal sections revealed widespread from cells transfected with the HA-tagged collybistin I or II
expression of the collybistin gene throughout the gray matter of cDNAs, proteins migrating at 62 or 55 kDa, respectively, were

articles
aa
bb
Collybistin I
Stac
Myosin
Spectrin
Vav
DrhoGEF
rat PIX
c
Collybistin I
Dbl
Tiam 1
PIX
Vav
Collybistin
Dbl
Tiam 1
PIX
Vav
d
Collybistin
Dbl
Tiam 1
PIX
Vav
Fig. 2. Domain comparison of collybistin and other members of the GEF family. (a) Comparison of the modular organization of collybistin I, Dbl
(P10911), Tiam 1 (Q60610), DrhoGEF (E1264107), rat PIX (D25304), Dbs (Q63406) and Vav (P54100). Accession numbers of the respective pro-
tein sequences are given in brackets. The domains are drawn roughly to scale. In addition to the DH, PH, SH3 and coiled-coil (slashed box) domains,
myosin-like motifs and, in Vav, a helix-loop-helix motif (HLH), a putative diacylglycerol-binding motif (DAG) and a SH2 domain are indicated.
(bd) Multiple sequence alignments of the SH3 (b), DH (c) and PH (d) domains of collybistin and related proteins. Residues present in at least 50%
of the sequences are highlighted by bold letters. Only amino-acid sequences displaying comparatively high homology to the respective domains of
collybistin are shown. In addition, the SH3 domains of myosin and spectrin are presented in (b); note that the SH3 domains of these cytoskeletal pro-
teins are more homologous to the SH3 domain of collybistin than those of other GEF family members.
specifically stained with either the anti-HA or the anti-collybistin- Collybistin associates with gephyrin in HEK 293 cells
peptide antibody (Fig. 5a). These findings are consistent with the To detect interaction of collybistin with gephyrin in intact mam-
predicted molecular weights of 58 kDa and 49 kDa, respective- malian cells, we used an assay previously employed to character-
ly. However, no specific immunolabeling was observed in brain ize the interaction of the GlyR subunit with gephyrin16,17. This
sections and cultured spinal neurons with this or a second anti- assay is based on the observation that recombinant gephyrin forms
peptide antibody (data not shown). Furthermore, an anti-HA large intracytoplasmic aggregates that trap hetero-oligomeric
antibody immunoprecipitated a complex containing gephyrin GlyRs16 as well as mutant GABAARs17 and NMDA receptors27 or
and collybistin I from extracts of HEK 293 cells co-expressing green fluorescent protein carrying a gephyrin-binding motif18.
these proteins (Fig. 5b) but not of cells transfected with cDNAs of Immunostaining of HA-tagged collybistins I and II expressed in
collybistin II (not shown). These data confirm the interaction of transfected HEK 293 cells localized both collybistin isoforms to
collybistin with gephyrin detected in our two-hybrid screen. We the cytoplasm (Fig. 6c and data not shown). However, when co-
also tried to co-immunoprecipitate collybistin solubilized from expressed with gephyrin, collybistin I was redistributed to
brain and spinal cord homogenates using anti-gephyrin anti- gephyrin-rich cytoplasmic regions as previously observed for
bodies. However, no specific band immunoreactive with the anti- polypeptides carrying a gephyrin-binding motif (Fig. 6d), again
collybistin peptide antibodies could be detected (not shown). demonstrating binding of the two proteins. By contrast, two addi-
This may reflect the poor reactivity of anti-gephyrin antibodies tional polypeptides identified as putative gephyrin-binding pro-
with detergent-solubilized gephyrin26. teins during the two-hybrid screen, BS69, a suppressor of

articles
Fig. 3. Collybistin mRNAs a terns. In 37 7% of the transfected cells, HA-tagged collybistin II

are mainly expressed in probe 1 probe 2 immunoreactivity accumulated in gephyrin-rich intracytoplas-
brain. (a) Schematic repre- 5' 3' mic aggregates. In 22 8% of the cotransfected cells, the intra-
sentation of the collybistin cytoplasmic collybistin II/gephyrin aggregates had long, thin
I mRNA. The different peptide
extensions not observed upon cotransfection with collybistin I
domains are symbolized as
described for Fig. 1. Bars
(data not shown). Interestingly, collybistin II and gephyrin
n le y s
indicate the relative posi- b t
ar ain lee ng er sc ne sti immunoreactivities colocalized in small (diameter, 0.20.5 m)
he br sp lu liv mu kid te submembrane microaggregates in 29 9% of the cells but dis-
tions of the nucleo [kb]
tide sequences used to 9.5 played a mixed distribution pattern with both intracytoplasmic
derive 32P-labeled anti 7.5 and submembrane collybistin II/gephyrin aggregates in 11 3%
sense oligonucleotides and of the cotransfected cells (Fig. 6e and f). Thus collybistin II
of a synthetic peptide 4.4 expression can induce submembrane gephyrin microaggregates
used to generate anti resembling those observed during GlyR clustering at differenti-
bodies. Probe 1 is ating postsynaptic membrane specializations15,30.
expected to hybridize 2.4
only to collybistin I tran-
scripts, whereas probe 2 Collybistin II/gephyrin microaggregates recruit GlyRs
1.35 Intracytoplasmic gephyrin aggregates can retarget GlyRs hetero-
should hybridize to both
oligomeric GlyRs16 and mutated receptor polypeptides carrying
collybistin I and II mRNAs.

(b) A multiple-tissue 2.4 a gephyrin-binding motif1618,27, thus preventing their transloca-
mRNA blot was hybrid tion to the plasma membrane. Here we investigated whether sub-
ized with oligonucleotide membrane collybistin II/gephyrin microaggregates could bind
probe 1, revealing two 1.35 hetero-oligomeric GlyRs and thus induce local accumulations of
transcripts of 6 kb and 5.5 GlyRs in the plasma membrane, as seen in differentiating neu-
kb expressed predominantly in brain RNA (top). Hybridizations with
rons. To this end, we co-expressed collybistin II and gephyrin
probe 2 gave identical results (data not shown). Rehybridization with a
random-primed probe derived from a 2-kb human -actin cDNA probe
with GlyR 1 and subunits in HEK293 cells and examined their
demonstrated that comparable amounts of mRNA were loaded (bottom). subcellular distributions. Peripheral collybistin II/gephyrin
microaggregates clearly showed immunoreactivity with the GlyR-
specific monoclonal antibody 4a (ref. 26; Fig. 7 and b). These
local accumulations of receptor polypeptides were absent when
collybistin II and gephyrin were co-expressed with only GlyR 1
E1A-activated transcription28, and the receptor for the axonal subunits (Fig. 7c and d). This confirmed the interaction of het-
guidance molecule netrin29, were not redistributed when co- ero-oligomeric 1/, but not homo-oligomeric 1, GlyRs with
expressed with gephyrin (Fig. 6a and b and data not shown), gephyrin and corroborated the identification of a gephyrin-bind-
demonstrating specificity of collybistingephyrin interactions. ing motif in the receptors subunit16,17. We therefore conclude
that collybistin II causes the formation of submembrane gephyrin
Collybistin II induces submembrane gephyrin clustering clusters that can recruit GlyRs to the plasma membrane.
Whereas HA-tagged collybistin I consistently associated with
intracellular gephyrin aggregates, the overlapping distributions DISCUSSION
of HA-tagged collybistin II and gephyrin immunoreactivity var- The receptor-associated protein gephyrin is pivotal in the for-
ied. In particular, submembrane clusters were frequently found. mation and anchoring of GlyR7,8 and GABAAR11,12 clusters at
(Fig. 6e and f). Analysis of 700 cotransfected HEK 293 cells in 3 postsynaptic membrane specializations. Four proteins bind
independent experiments revealed 4 distinct redistribution pat- gephyrin, namely tubulin14, the subunit of the GlyR16,17, the
actin-binding protein profilin 31 and the
rapamycin and FKBP 112 target 1
a collybistin I b collybistin I/II c gephyrin (RAFT1) 32. Here, we identify two splice
variants of a gephyrin-binding protein we
name collybistin. Its modular primary
structure, including a DH/PH tandem
domain characteristic of the dbl-like GEF
family, is consistent with a GDP/GTP
exchanger function. We therefore tenta-
tively classify collybistin I and II as GEFs.
Importantly, collybistin II induced sub-
membrane collybistin II/gephyrin micro-
clusters, which by size and location
resemble postsynaptic gephyrin aggregates
observed in differentiating spinal neurons.
These collybistin II/gephyrin co-aggregates
Fig. 4. Distribution of collybistin and gephyrin mRNAs in the adult rat brain. Autoradiographs of
parasagittal (top) and horizontal (bottom) brain sections hybridized with 35S-labeled antisense oligonu- were capable of recruiting hetero-oligomer-
cleotides specific for collybistin I (a), collybistin I and II (b) or gephyrin (c) transcripts. Control ic but not homo-oligomeric GlyRs to plas-
hybridizations were performed on parallel sections using the corresponding labeled sense oligonu- ma membrane microdomains. Based on
cleotides, or by antisense hybridizations in the presence of an excess of the unlabeled oligonucleotide these data, we propose that activation of a
(data not shown). Control hybridizations showed only very weak background labeling (not shown). Rho/Rac-family GTPase by collybistin reg-

articles
Fig. 5. Western blot of recombinant collybistin and co-

d d
immunoprecipitation of recombinant gephyrin with
a te I II te I II b I /
ec in in ec in in in in in I
f st st st yr tin
collybistin. (a) Western blot of HEK 293 cells expressing
ns b ist bi ns
f
bi bi b ist hy
r h
p bis
a a
HA-tagged collybistin I with antibodies against HA or a tr lly lly tr lly lly lly p ge lly
un co co un co co co ge
collybistin peptide revealed that both antibodies recog- co
nize a single band of 62 kDa (lane 2), whereas a 55 kDa
protein (lane 3) is detected from cells transfected with a
cDNA encoding HA-tagged collybistin II. These protein anti-
bands were not found in extracts of untransfected cells gephyrin
(lane 1). (b) From extracts of HEK 293 cells co-express-
ing gephyrin and HA-tagged collybistin I, a complex con-
taining these two proteins was immunoprecipitated with
an anti-HA antibody (lane 3). In contrast, immunoprecip-
itates obtained with extracts prepared from singly trans- anti-HA anti-collybistin anti-HA
fected cells contained only minute amounts of
endogenous or recombinant gephyrin (lanes 1 and 2).
ulates postsynaptic GlyR-cluster formation at a step crucial for domains and thereby initiating GlyR and GABAAR clustering. In
gephyrin deposition at presumptive postsynaptic sites. addition, it may localize subsynaptic signaling cascades to post-
synaptic sites34 by GEF-dependent activation of Rho-family
Collybistin is a GEF expressed in brain GTPases. Additional domains found in many GEFs include,
GEFs activate GTPases of the Rho/Rac family by rapidly exchang- among others, one or several SH2 or SH3 domains and regions
ing GDP for GTP21. Nucleotide exchange is mediated by the DH predicted to form coiled-coil structures22. In line with this mod-
moiety of the DH/PH tandem domain present in all GEFs ular-building principle, collybistin I harbors a SH3 domain in
known22. From sequence comparison with the DH domains of the aminoterminal region and a putative coiled-coil domain at
other GEFs, Rac and Cdc 42 seem the most likely candidates for the carboxyterminal end in addition to the central DH/PH
activation by collybistin. The PH region is thought to be required domain. The SH3 domain of collybistin I displays highest homol-
for the intracellular localization of GEFs33 and binds PIP2 (ref. ogy to SH3 domains of cytoskeletal proteins (Fig. 2b) and not to
23). The latter property may be important for recruiting colly- those of the PIX family25; therefore, signaling through PAK kinas-
bistin/gephyrin aggregates to PIP 2-rich plasma-membrane es seems unlikely. Both SH3 and coiled-coil domains mediate
interactions with other polypeptides, but there is no evidence for
proteins binding to the corresponding domains of collybistin I.
Notably, submembrane collybistin/gephyrin microclusters do
a b not form when collybistin II is replaced with collybistin I, which
differs only by the presence of the SH3 and coiled-coil domains.
This suggests that proteins recognizing these collybistin I-spe-
cific domains might regulate GEF function and/or interaction
with gephyrin. Conversely, the absence of such domains in colly-
bistin II indicates that gephyrin binding must be mediated by
segments adjacent to or within the DH/PH domains.
c d Fig. 6. Co-expression of gephyrin alters the distribution of collybistin I

and II in HEK 293 cells. HEK 293 cells were transfected with cDNAs
encoding HA-tagged BS69 (a, b), HA-tagged collybistin I (c, d) or HA-
tagged collybistin II (e, f), either alone (a, c) or together with gephyrin
clone p1 (b, df). The subcellular distribution of the recombinant pro-
teins was then visualized using polyclonal anti-HA antiserum (red) and
the gephyrin-specific monoclonal antibody 5a (green). Upon single
transfection, the immunoreactivities of HA-tagged BS69 (a), HA-tagged
collybistin I (c) and HA-tagged collybistin II (not shown) were diffusely
distributed in the cytoplasm of transfected HEK cells. Co-expression of
gephyrin (green) did not change the subcellular distribution of BS69
e f (red, b). In contrast, gephyrin co-expression redistributed both HA-
tagged collybistin I (d) and II (e, f) to gephyrin-rich domains, as indicated
by the overlap (yellow). Collybistin I was targeted to gephyrin-rich intra-
cellular aggregates (d), as described for other gephyrin-binding proteins.
Co-expression of collybistin II (red) and gephyrin (green) induced the
formation of coaggregates (yellow) of different morphologies. In some
cotransfected cells, collybistin II was found in gephyrin-rich intracellular
aggregates, as observed for collybistin I (e, and not shown), whereas in
other cotransfected cells, collybistin II and gephyrin formed numerous
submembrane clusters (f). Mixed morphologies of collybistin II and
gephyrin co-aggregates were also observed (e). Scale bars, 5 m.

articles
a b The molecular mechanisms underlying the collybistin II-

induced redistribution of heterologously expressed gephyrin are
unknown, but probably involve the activation of a Rho-family
GTPase. These enzymes are implicated in the regulation of many
cellular processes35, but most prominently, in the regulation of
actin-based cytoskeletal dynamics36. Rac and Cdc42, in particu-
lar, act as intermediates of signal-transduction pathways37 con-
trolling actin-microfilament organization35,38. As we could not
detect cytoskeletal rearrangements indicative of activation of
Rho-family GTPases in NIH/3T3 fibroblasts expressing either
c d collybistin variant (J.K., unpublished data), it is possible that GEF
activity may require gephyrin. Gephyrin may regulate signaling by
the tyrosine kinase RAFT1, which requires binding to gephyrin
for its activation 32. Additional roles of Rho-family GTPases
include initiation of intracellular signaling cascades by activat-
ing phosphatidylinositol-specific phospholipase C, which in turn
generates PIP2. This second messenger molecule is known to reg-
ulate intracellular Ca2+ homeostasis. Interestingly, microdomains
of elevated Ca2+ are implicated in the generation of primordial

gephyrin clusters in differentiating spinal neurons5.
SynGAP, a Ras GTPase-activating protein (inhibitor of Ras
signaling) in excitatory synapses39,40 associates with the PDZ
domains of PSD95/SAP90 and SAP102, scaffolding proteins
assumed to organize various components of the postsynaptic sig-
Fig. 7. Collybistin II/gephyrin co-aggregates recruit hetero-oligomeric naling machinery at excitatory synapses (reviewed in ref. 34). It is
but not homo-oligomeric GlyRs. HEK 293 cells were quadruple-trans- believed that, in response to NMDA receptor activation, SynGAP
fected with cDNAs encoding GlyR 1 and subunits, HA-tagged colly- modulates neurotrophin-induced Ras signaling, and thus down-
bistin II and gephyrin (a, b), or triple-transfected with cDNAs encoding regulates the downstream activation of mitogen-activated kinase
the GlyR 1 subunit, HA-tagged collybistin II and gephyrin (c, d). The signaling cascades, implicated in synaptic plasticity and hip-
subcellular distributions of the recombinant proteins were then visual-
pocampal LTP41. Conversely, genetic deletion of the neuron-spe-
ized using polyclonal anti-HA antiserum (green) and the GlyR-specific
monoclonal antibody 4a (red). (a) Immunoreactivities of hetero- cific Ras GEF, RAS-GRF/CDC25Mm, was shown to disrupt LTP
oligomeric GlyR (red) and collybistin II (green)/gephyrin co-aggregates in the basolateral amygdala and impair memory consolidation
overlap (yellow) in quadruple-transfected cells. (b) Single-channel visual- in mutant mice40. By analogy, collybistin could similarly modu-
ization of the submembrane collybistin II/gephyrin co-aggregates shown late receptor function at inhibitory synapses by regulating the
in (a). (c) Homo-oligomeric GlyR composed of 1 subunits (red) is density of gephyrin and GlyRs or GABA A Rs. Cytoskeletal
evenly distributed at the plasma membrane of triple-transfected cells. rearrangements alter GlyR density at glycinergic synapses15, and
(d) Single-channel tangential section shows submembrane collybistin changes in receptor density modify the agonist responses of the
II/gephyrin co-aggregates in the cell shown in (c). Scale bars, 5 m. GlyR42. Thus, collybistin might control the postsynaptic response
and thereby tune the efficacy of inhibitory synapses. Generation
of GlyR clusters by heterologous expression of collybistin II,
gephyrin and GlyR polypeptides as demonstrated here should
Northern blots and in-situ hybridizations revealed predomi- enable electrophysiologists to examine this hypothesis.
nant expression of the collybistin gene in brain. Transcripts were
confined to neurons and overlapped with gephyrin transcripts. METHODS
No significant differences were found between distributions of Yeast two-hybrid screening. For yeast two-hybrid screening19, a library
mRNAs encoding collybistin I and II. Although some expression was prepared by subcloning oligo-dT-primed cDNA from newborn rat
was seen in heart and skeletal muscle, the cells producing colly- brain into the EcoRI/XhoI sites of pADGAL4 phagemid vector (Strata-
gene, La Jolla, California) encoding the GAL4 activation domain. Approx-
bistin in these organs are unknown.
imately 1.8 106 clones were screened using the pAS2-1 vector43 encoding
the GAL4 DNA-binding domain fused in frame to the coding sequence of
Role of collybistin in receptor clustering gephyrin clone p1 (ref. 44). The plasmids were used to cotransform Y190
The most surprising result of this study is that co-expression of yeast cells (Clontech, Palo Alto, California), and positive clones were
collybistin II and gephyrin in HEK 293 cells induced submem- selected by growth on triple-minus plates (Leu, Trp, His) and con-
brane collybistin/gephyrin microaggregates that accumulated co- firmed by -galactosidase assay. Positive clones were cotransformed with
expressed hetero-oligomeric GlyR. These structures resembled either the bait vector or the pAS2-1 vector without insert into Y190 yeast
GlyR clusters in spinal neurons in both size and subcellular loca- cells to confirm a specific interaction with the bait protein. All DNA
tion. Consistent with previous data17, these receptor aggregates sequences were analyzed by dideoxy-sequencing of both strands using
an automated sequencer (Applied Biosystems, Foster City, California).
formed only when the receptor complex harbored a gephyrin-
binding subunit. Furthermore, GlyR cluster formation in cul- 5-rapid amplification of complementary ends (RACE). The 5 end of
tured spinal cord neurons could be efficiently suppressed by the collybistin mRNA was determined by 5-RACE45 using Marathon-
attenuating collybistin expression with addition of specific colly- Ready cDNA from rat brain (Clontech) as template and the nested reverse
bistin antisense oligonucleotides (J.K., unpublished data). These primers col4as (5-GTGTTGATAGCGGCTGTCCT-3) and col1as
observations indicate a crucial role for collybistin in regulating (5-CACGAATTCCAGCTTCTGTGAGGCCGGGTG-3) designed from
the cell-surface distribution of GlyRs. the collybistin cDNA. The reaction products of approximately 370 bp

articles
and 240 bp were subcloned into pBluescript SK- (Stratagene) and RECEIVED 3 SEPTEMBER; ACCEPTED 3 NOVEMBER 1999
sequenced. Three independent clones were analyzed. The protein and
nucleotide sequences reported here have been submitted to the 1. Froehner, S. C. Regulation of ion channel distribution at synapses. Annu. Rev.
EMBL/GeneBank data library under accession number AJ250425. Neurosci. 16, 347368 (1993).
2. Ehlers, M. D., Mammen, A. L, Lau, L.-F. & Huganir, R. L. Synaptic targeting
Antibodies. Monoclonal antibodies mAb 5a against gephyrin and mAb of glutamate receptors. Curr. Opin. Cell Biol. 8, 484489 (1996).
4a against the GlyR are characterized26,46. The monospecific rabbit poly- 3. Kirsch, J., Meyer, G. & Betz, H. Synaptic targeting of ionotropic
neurotransmitter receptors. Mol. Cell. Neurosci. 8, 9398 (1996).
clonal anti-hemagglutinin antibody (anti-HA) was purchased from Santa 4. Kuhse, J., Betz, H. & Kirsch, J. The inhibitory glycine receptor:
Cruz Biotechnologies (Santa Cruz, California). Rabbits were immunized architecture, synaptic localization and molecular pathology of a
with a custom-synthesized peptide (Eurogentech, Seraing, Belgium) postsynaptic ion-channel complex. Curr. Opin. Neurobiol. 5, 318323
matching the amino-acid sequence between the SH3 and DH domain of (1995).
collybistin I (EDGVEEGPSDVQNGHLDPNSDC) coupled to keyhole 5. Kirsch, J. & Betz, H. Glycine receptor activation is required for receptor
clustering in spinal neurons. Nature 392, 717720 (1998).
limpet hemocyanin. The resulting antisera were purified using protein A 6. Lvi, S., Vannier, C. & Triller, A. Strychnine-sensitive stabilization of
sepharose. Primary antibodies were used at dilutions of 1:100 (mAb 5a), postsynaptic glycine receptor clusters. J. Cell Sci. 111, 335345 (1998).
1:250 (anti-collybistin) or 1:500 (anti-HA) in phosphate-buffered saline. 7. Kirsch, J., Wolters, I., Triller, A. & Betz, H. Gephyrin antisense
Fluorescein isothiocyanate (FITC)- or carboxymethylindocyanine 3 (Cy3)- oligonucleotides prevent glycine receptor clustering in spinal neurons.
Nature 366, 745748 (1993).
coupled goat anti-mouse and anti-rabbit IgG (Dianova, Hamburg) were 8. Feng, G. et al. Dual requirement for gephyrin in glycine receptor clustering
used as secondary antibodies (dilution 1:100 to 1:500). and molybdoenzyme activity. Science 282, 13211324 (1998).
9. Sasso-Pognetto, M. et al. Colocalization of gephyrin and GABAA-receptor
Cell culture, transfections and immunocytochemistry. Culture conditions subunits in the rat retina. J. Comp. Neurol. 357, 114 (1995).
for human embryonic kidney 293 (HEK 293) cells (ATCC#CRL1537), the 10. Todd, A. J., Watt, C., Spike, R. C. & Sieghart, W. Colocalization of GABA,
glycine and their receptors at synapses in the rat spinal cord. J. Neurosci. 16,
protocols for transfections and the immunocytochemical procedures used 974982 (1996).
are described16, as are cDNAs encoding gephyrin (p1) GlyR 1 and sub- 11. Essrich, C., Lorez, M., Benson, J., Fritschy, J. M. & Lscher, B. Postsynaptic
units44,47,48. A hemagglutinin epitope was introduced by PCR-based muta- clustering of major GABAA receptor subtypes requires the 2 subunit and
genesis at the 3 end of coding region of the collybistin I and II cDNAs, gephyrin. Nat. Neurosci. 1, 563571 (1998).
12. Kneussel, M. et al. Loss of postsynaptic GABAA receptor clustering in
and constructs were subcloned into the eukaryotic expression vector pCIS2. gephyrin-deficient mice. J. Neurosci. 19, 92899297 (1999).
All DNA sequences were verified by dideoxy-sequencing of both strands. 13. Stallmeyer, B. et al. The neurotransmitter-anchoring protein gephyrin
reconstitutes molybdenum-cofactor biosynthesis in bacteria, plants and
Northern blot analysis. A multiple-tissue mRNA blot, loaded with approx- mammalian cells. Proc. Natl. Acad. Sci. USA 96, 13331338 (1999).
imately 2 g of polyA+ RNA from different rat tissues (Clontech), was 14. Kirsch, J. et al. The 93-kDa glycine receptor-associated protein binds to
tubulin. J. Biol. Chem. 266, 2224222245 (1991).
prehybridized for 4 h at 50C in Express Hybridization buffer (Clontech) 15. Kirsch, J. & Betz, H. The postsynaptic localization of the glycine receptor-
with single-stranded salmon testis DNA (100 g per ml) and sequential- associated protein gephyrin is regulated by the cytoskeleton. J. Neurosci. 15,
ly hybridized with the 32P-labeled antisense oligonucleotides 5-CCT- 41484156 (1995).
CATCGTCGATCTGGCCCCACCACCAATCCTTGTTGGAAGCA-3, 16. Kirsch, J., Kuhse, J. & Betz, H. Targeting of glycine receptor subunits to
gephyrin-rich domains in transfected human embryonic kidney cells. Mol.
5-TCTCTTTCGGCACTGCTTCAGGTACCCCTCGCAAATGTC-
Cell. Neurosci. 6, 450461 (1995).
CTTGAG-3 and a random-primed probe derived from a 2-kb human 17. Meyer, G., Kirsch, J., Betz, H. & Langosch, D. Identification of a gephyrin
-actin cDNA probe for 4 h at 50C. The blot was sequentially washed binding motif on the glycine receptor subunit. Neuron 15, 563572
under increasingly stringent conditions. The final wash was for 30 min in (1995).
0.1 SSC containing 0.1% SDS at 65C. The washed membrane was used 18. Kneussel, M., Hermann, A., Kirsch, J. & Betz, H. Hydrophobic interactions
mediate binding of the glycine receptor -subunit to gephyrin. J. Neurochem.
to expose Biomax film (Kodak, Rochester, New York) for 1060 h at 70C. 72, 13231326 (1999).
19. Fields, S. & Song, O. K. A novel genetic system to detect proteinprotein
In-situ hybridization. The 3 end-labeling of antisense oligonucleotides interactions. Nature 340, 245246 (1988).
was the same as for those used in Northern blots, and in-situ hybridiza- 20. Akagi, H. & Miledi, R. Heterogeneity of glycine receptors and their
tion on adult rat brain sections was as described49. Control hybridiza- messenger RNAs in rat brain and spinal cord. Science 242, 270272
(1988).
tions included the corresponding sense probes labeled to identical specific 21. Hart, M. J. et al. Cellular transformation and guanine nucleotide exchange
activities and antisense hybridizations in the presence of an 20-fold excess activity are catalyzed by a common domain on the DBL oncogene product. J.
of the unlabeled oligonucleotide. No signal was detected in brain sec- Biol. Chem. 269, 6265 (1994).
tions hybridized under these conditions. Chromosomal mapping of the 22. Cerione, R. A. & Zheng, Y. The Dbl family of oncogenes. Curr. Opin. Cell Biol.
8, 216222 (1996).
collibystin gene was done by fluorescent in-situ hybridization (FISH) at 23. Harlan, J. E., Hajduk, P., Sup Yoon, H. & Fesik, S. W. Pleckstrin homology
SeeDNA Biotech, North York, according to published procedures50. domains bind to phosphatidylinositol 4,5-bisphosphate. Nature 372,
375379 (1994).
Western blots. Cell and tissue extracts were prepared according to stan- 24. Lim, W. A., Richards, F. M. & Fox, R. O. Structural determinants of peptide-
dard procedures. The extracts were separated by SDS-PAGE using a 10% binding orientation and of sequence specificity in SH3 domains. Nature 372,
375379 (1994).
gel followed by electrophoretic transfer onto a nitrocellulose membrane. 25. Manser, E. et al. PAK kinases are directly coupled to the PIX family of
The membrane was blocked and reacted with primary and HRP-con- nucleotide exchange factors. Mol. Cell 1, 183192 (1998).
jugated secondary antibodies as described46. Antigen/antibody com- 26. Pfeiffer, F., Simler, R., Grenningloh, G. & Betz, H. Monoclonal antibodies
plexes were visualized using enhanced chemoluminescence (Amersham, and peptide mapping reveal structural similarities between the subunits of
the glycine receptor of rat spinal cord. Proc. Natl. Acad. Sci. USA 81,
Freiburg) and Kodak Biomax film. 72247227 (1984).
27. Kins, S., Kuhse, J., Laube, B., Betz, H. & Kirsch, J. Incorporation of a
gephyrin-binding motif targets NMDA receptors to gephyrin-rich domains
ACKNOWLEDGEMENTS in HEK 293 cells. Eur. J. Neurosci. 11, 740744 (1999).
28. Hateboer, G. et al. BS69, a novel adenovirus E1A-associated protein that
The authors thank Maria Fischer for the construction of the bait vector, Ina inhibits E1A transactivation. EMBO J. 14, 31593169 (1995).
Bartnik for technical assistance with in-situ hybridizations and 29. Keino-Masu, K. et al. Deleted in colorectal cancer (DCC) encodes a netrin
immunocytochemistry and M. Baier and H. Reitz for help in preparing the receptor. Cell 87, 175185 (1996).
30. Bchade, C., Colin, I., Kirsch, J., Betz, H. & Triller, A. Glycine receptor
manuscript. Expression vectors encoding BS69 and the netrin receptor were gifts subunit and gephyrin expression in cultured spinal neurons: a quantitative
from R. Bernards and M. Tessier-Lavigne, respectively. This study was supported analysis. Eur. J. Neurosci. 8, 429435 (1996).
by Deutsche Forschungsgemeinschaft (Ki339/7-1, SFB 474-B7 (JK), and 31. Mammoto, A. et al. Interactions of drebrin and gephyrin with profilin.
Biochem. Biophys. Res. Comm. 243, 8689 (1998).
Ki339/9-1), Stifterverband fr die Deutsche Wissenschaft, BMBF (01KV9539/3) 32. Sabatini, D. et al. RAFT1 signaling requires interaction with the clustering
and Fonds der Chemischen Industrie. protein gephyrin. Science 284, 11611164 (1999).

articles
33. Musacchio, A., Gibson, T., Rice, P., Thompson, J. & Saraste, M. The PH 43. Harper, J. W., Adami, G. R., Wei, N., Keyomarsi, K. & Elledge, S. J. The p21
domain: a common piece in the structural patchwork of signalling proteins. Cdk-interacting protein Cip1 is a potent inhibitor of G1 cyclin-dependent
Trends Biochem. Sci. 18, 343348 (1993). kinases. Cell 75, 805816 (1993).
34. Kirsch, J. Assembly of signaling machinery at the postsynaptic membrane. 44. Prior, P. et al. Primary structure and alternative splice variants of gephyrin, a
Curr. Opin. Neurobiol. 9, 329335 (1999). putative glycine receptor-tubulin linker protein. Neuron 8, 11611170
35. Hall, A. Rho GTPases and the actin cytoskeleton. Science 279, 509514 (1998). (1992).
36. Tapon, N. & Hall, A. Rho, Rac and Cdc42 GTPases regulate the organization 45. Frohmann, M. A., Dush, M. K. & Martin, G. R. Rapid production of full-
of the actin cytoskeleton. Curr. Opin. Cell Biol. 9, 8692 (1997). length cDNAs from rare transcripts: amplification using a single gene-
37. Minden, A., Lin, A., Claret, F. X., Abo, A. & Karin, M. Selective activation of specific oligonucleotide primer. Proc. Natl. Acad. Sci. USA 85, 89989002
the JNK signaling cascade and c-Jun transcriptional activity by the small (1988).
GTPases Rac and Cdc42Hs. Cell 81, 11471157 (1995). 46. Kirsch, J. & Betz, H. Widespread expression of gephyrin, a putative
38. Nobes, C. & Hall, A. Rho, Rac and Cdc42 GTPases regulate the assembly of receptor-tubulin linker protein, in rat brain. Brain Res. 621, 301310
multimolecular focal complexes associated with actin stress fibers, (1993).
lamellipodia and filopodia. Cell 81, 5362 (1995). 47. Grenningloh, G. et al. The strychnine-binding subunit of the glycine receptor
39. Chen, H.-J., Rojas-Soto, M., Oguni, A. & Kennedy, M. B. A synaptic ras- shows homology with nicotinic acetylcholine receptors. Nature 328, 215220
GTPase activating protein (p135/SynGAP) inhibited by CaM kinase II. (1987).
Neuron 20, 895904 (1998). 48. Grenningloh, G. et al. Cloning and expression of the 58 kd subunit of the
40. Kim, J. H., Liao, D., Lau, L.-F. & Huganir, R. L. SynGAP: a synaptic RasGAP that inhibitory glycine receptor. Neuron 4, 963970 (1990).
associates with the PSD-95/SAP90 protein family. Neuron 20, 683691 (1998). 49. Kirsch, J., Malosio, M.-L., Wolters, I. & Betz, H. Distribution of gephyrin
41. Brambilla, R. et al. A role for the ras-signalling pathway in synaptic transcripts in the adult and developing rat brain. Eur. J. Neurosci. 5,
transmission and longterm memory. Nature 390, 281286 (1997). 11091117 (1993).
42. Taleb, O. & Betz, H. Expression of the human glycine receptor 1 subunit in 50. Heng, H. H. Q., Squire, J. & Tsui, L.-C. High resolution mapping of
Xenopus oocytes: apparent affinities of agonists increase at high receptor mammalian genes by in situ hybridization to free chromatin. Proc. Natl. Acad.
densities. EMBO J. 13, 13181324 (1994). Sci. USA 89, 95099513 (1992).

articles
Nuclear Notch1 signaling and the

regulation of dendritic
development
Lori Redmond1, Sang-Rog Oh1, Carol Hicks2, Gerry Weinmaster2 and Anirvan Ghosh1
1 Department of Neuroscience, Johns Hopkins University School of Medicine, Baltimore, Maryland 21205, USA
2 Department of Biological Chemistry, University of California Los Angeles School of Medicine, Los Angeles, California 90095, USA
Correspondence should be addressed to A.G. (aghosh@jhmi.edu)
To understand the function of Notch in the mammalian brain, we examined Notch1 signaling and its
cellular consequences in developing cortical neurons. We found that the cytoplasmic domain of
endogenous Notch1 translocated to the nucleus during neuronal differentiation. Notch1 cytoplasmic-
domain constructs transfected into cortical neurons were present in multiple phosphorylated forms,
localized to the nucleus and could induce CBF1-mediated transactivation. Molecular perturbation
experiments suggested that Notch1 signaling in cortical neurons promoted dendritic branching and
inhibited dendritic growth. These observations show that Notch1 signaling to the nucleus exerts an
important regulatory influence on the specification of dendritic morphology in neurons.
Although the role of cellcell interactions in the regulation of Four mammalian homologs of Notch (Notch1, Notch2, Notch3
mammalian neural development is not well understood, it has and Notch4) are expressed in the developing nervous sys-
become increasingly apparent from studies in invertebrates that tem2529. Mammalian homologs of the Notch ligands Delta and
such interactions can exert an important regulatory influence on Serrate are expressed in the developing brain and spinal
differentiation. The role of cellcell interactions has been exten- cord3033. Mice lacking Notch1 die embryonically, before the
sively studied in the context of lateral specification in inverte- period of neuronal differentiation, and show a high occurrence
brates. Such interactions among developmentally equivalent cells of cell death in the central nervous system34,35. In addition,
allow a single cell within the group to adopt a cell fate distinct transgenic mice overexpressing Notch3 have neural tube
from the neighboring cells and have an important role in neu- defects36. These observations suggest that, as in other verte-
roblast specification1,2. Molecular and genetic studies indicate brates, mammalian Notch homologs regulate early neural
that the cell-surface protein Notch is a central mediator of later- development.
al specification in invertebrates35. Protein localization studies also suggest a role for mammalian
Notch is a type I cell-surface protein, approximately 300 kDa Notch in neuronal differentiation. In the ventricular zone of devel-
in size, which functions as a receptor. Proteolytic processing of oping ferret cortex, Notch1 is localized to the basal pole of divid-
full-length Notch generates two fragments that seem to associ- ing progenitor cells37. Because the basal daughter is thought to
ate at the membrane to form a receptor complex69. One of the become postmitotic in asymmetric cell divisions, this observa-
fragments (p180) contains most of the extracellular domain, tion suggests that Notch1 is preferentially inherited by the post-
and the other fragment (p120) contains the transmembrane mitotic cell. This study suggests that Notch may continue to
and cytoplasmic domains. An important intracellular target of function in differentiated neurons after cell fate has been decided.
Notch signaling in Drosophila is the transcription factor, Sup- To explore the role of Notch in mammalian neural develop-
pressor of Hairless [Su(H)]10,11. There is also considerable evi- ment, we examined the molecular and cellular consequences of
dence that the mammalian homolog of Su(H), called CBF1, is a Notch signaling in developing cortical neurons. We report that
target of Notch signaling in mammalian cells1215. Although the neuronal Notch1 signals to the nucleus during differentiation and
mechanisms by which Notch signaling leads to transactivation regulates the dendritic development of cortical neurons.
via Su(H)/CBF1 are poorly understood, Notch-receptor acti-
vation seems to involve cleavage and nuclear translocation of RESULTS
the cytoplasmic domain of the receptor1619. Developmental regulation of Notch1 localization
Several observations suggest that vertebrate Notch proteins To characterize expression of Notch1 in cortical tissues, we used
are involved in regulating neural development in nonmam- affinity-purified antibodies generated against the intracellular
malian vertebrates. For example, overexpression of the Notch domain of Notch1 (-Notch.IC). These antibodies recognize a
ligand Delta or a constitutively active form of Notch inhibits truncated Notch1 construct when expressed in COS cells, and
primary neurogenesis in Xenopus2022. Similarly, Notch-medi- this interaction can be specifically blocked by preincubation of
ated interactions seem to negatively regulate neuronal differ- the affinity-purified antibodies with recombinant Notch1 pro-
entiation in Xenopus and chick retina23,24. It is likely that Notch tein (Fig. 1a). In addition, these antibodies recognize full-length
is also involved in regulating neural development in mammals. Notch1 but not Notch2, Notch3 or Notch4 in western blots of

articles
Fig. 1. Notch1 expression in the developing

a b cortex. (a, b) Characterization of -Notch.IC
(93-4). (a) On western blots, -Notch.IC rec-
ognizes FCDN1.HA expressed in COS cells
and immunoprecipitated using anti-HA.
Staining is blocked by preincubation of -
Notch.IC with the intracellular domain Notch1
fusion protein. (b) -Notch.IC recognizes
Notch1, but not Notch2, 3 or 4, expressed in
293 cells. Expression of Notch2, Notch3 and
Notch4 in the 293 cells was confirmed by west-
ern blotting with anti-Notch2 (93-3) or with
anti-HA (to detect tagged Notch3 and
Notch4). (c) Western blot of proteins isolated
c d from cortical neurons in culture at E18 + 8 DIV
and from cortical tissue from animals at ages
indicated, probed with -Notch.IC. The major
forms of Notch1 detected by the intracellular-
domain antibody are the full-length protein
(p300) and a cleaved form of the protein
(p120). (d) Characterization of the cell-surface

forms of Notch1 in cortical neurons.
Biotinylated cell-surface proteins on cortical
neurons were separated by polyacrylamide gel
electrophoresis (PAGE) and probed with anti-
bodies to the intracellular (Notch.IC) or extra-
cellular (Notch.EC) domains of Notch1.
(e) Immunohistochemical localization of
Notch1 using -Notch.IC in cortical coronal
sections taken from E18, P0, P7 and P14 rats.
e The P14 image is taken from the cortical plate.
Notch1 immunoreactivity is present both in
the ventricular zone (which contains proliferat-
ing cells) and in the cortical plate (which con-
tains postmitotic neurons) at all ages shown.
CP, cortical plate; IZ, intermediate zone; VZ,
ventricular zone; WM, white matter.
recombinant proteins (Fig. 1b). Western blots showed that the To determine Notch1 protein localization in developing cor-
two previously described forms of Notch1, full-length p300 and tex, we performed immunohistochemical staining with
cleaved p120, were present in cortical neurons at all ages -Notch.IC on cortical sections from late embryonic and early
(Fig. 1c). Proteins migrating at about 140 kDa and 100 kDa were postnatal rats. In agreement with previous studies25,30,37, we
found at embryonic and early postnatal ages, but not at later ages. detected Notch1 protein in cells of the ventricular zone of E18
This pattern suggests that Notch1 might regulate developmental cortex (Fig. 1e). In addition, at both embryonic and postnatal
events both during and after neurogenesis. ages, we detected significant levels of Notch1 immunoreactiv-
To define the cell-surface form of Notch1 in developing cor- ity in the cortical plate, suggesting that Notch1 may regulate
tical neurons, we did surface biotinylation experiments in E18 the development of postmitotic neurons.
cortical neurons in culture. After surface biotin labeling, We performed confocal microscopy on immunofluorescent-
biotinylated proteins were precipitated using streptavidin- ly labeled sections through the developing cortex to determine
agarose and separated by SDS-PAGE. Western blots using anti- the subcellular localization of Notch1 in cortical cells. In the ven-
bodies to the intracellular and extracellular domains of Notch1 tricular zone, Notch1 was concentrated near the cell membranes
indicated that the p120 form of the Notch1 protein was the and seemed largely excluded from the nucleus (Fig. 2ac). By
dominant cell-surface form of the receptor recognized by the contrast, Notch1 was localized predominantly in the nucleus in
intracellular-domain antibody, and the p180 form of the pro- cortical plate neurons (Fig. 2 di). Because cells migrate from
tein was the dominant cell-surface form recognized by the extra- the ventricular zone to the cortical plate once they become post-
cellular-domain antibody (Fig. 1d). This is consistent with a mitotic, these observations suggest that neuronal differentiation
model in which the cell-surface Notch1 receptor is a het- is accompanied by a translocation of Notch1 to the nucleus.
erodimer of p120 and p180 forms. We also found a high-mole- To examine the relationship between Notch 1 localization and
cular-weight form (Fig. 1d) that may represent uncleaved, neuronal differentiation more carefully, we carried out experiments
full-length Notch1 at the cell surface. in E18 dissociated cortical cell cultures, in which the transition from

articles
Fig. 2. The subcellular localization of

Notch1 changes as cells migrate from the
ventricular zone to the cortical plate.
Confocal images of sections taken from the
ventricular zone at E18 (ac), the cortical
plate at P0 (df) and the cortical plate at
P14 (gi). The sections were labeled with
-Notch.IC (red; a, d, g) and with Hoechst
33258 (Sigma) to reveal the nucleus (green;
b, e, h). (c, f, i) Merged images showing
localization of Notch1 immunofluorescence
with respect to the nucleus. Notch1 is
largely excluded from the nucleus of ventric-
ular zone cells (ac; arrowheads), but is
predominantly nuclear in postmitotic neu-
rons in the cortical plate (dg; arrows).
a dividing to postmitotic state could be

examined closely. Western blots indi-

cated that the same forms of Notch1
were expressed in vitro as in vivo
(Fig. 1c). To determine if -Notch.IC
could be used to detect Notch1 distrib-
ution in dissociated neurons, we trans-
fected E18 cortical neurons with a
hemagglutinin (HA)-tagged, truncated
Notch1 construct (FCDN1.HA). Trans-
fected neurons were strongly
immunopositive for Notch1 (Fig. 3a and b). Notch1 immunoflu- To examine the relationship between subcellular Notch1 local-
orescence was eliminated when -Notch.IC was preincubated with ization and the differentiation state of individual neurons, we
recombinant Notch1 (Fig. 3c and d). Immunofluorescence for labeled E18 cortical cultures with -Notch.IC and differentiation
endogenous Notch1 revealed that many cortical neurons in culture stage-specific antibodies (Fig. 4). In dividing cells (immunoposi-
had high nuclear levels of Notch1 (Fig. 3e and f). Endogenous stain- tive for PCNA), Notch1 immunofluorescence was primarily local-
ing was blocked by preincubating the antibody with recombinant ized to cytoplasmic and membrane compartments, indicating that
Notch1, indicating that the immunofluorescence signal reflects the Notch1 is largely excluded from the nucleus in undifferentiated
localization of a Notch1 epitope (Fig. 3; compare e with g). cells (Fig. 4ac). In contrast, Notch1 was detected mainly in the
nucleus in TuJ1-positive
(postmitotic) neurons
(Fig. 4df). To further ana-
lyze the distribution of
Notch1 with respect to cell
cycle withdrawal, E18 cells
were labeled with BrdU for
various durations before
being fixed and processed
for immunofluorescence. In
cultures labeled for 6 hours,
all BrdU-labeled cells should
be actively dividing, and in
cultures labeled for 24 hours
(longer than the cell-cycle
duration), all BrdU-negative
cells should be postmitotic.
Immunofluorescence using
anti-BrdU and -Notch.IC
indicated that Notch1 was
Fig. 3. Distribution of Notch1 in cortical cells in culture. (ad) Characterization of -Notch.IC for immunofluores- excluded from the nucleus
cence. Cortical neurons were transfected with HA-tagged FCDN1 and processed for immunofluorescence using -
in actively dividing cells,
Notch-IC (a, c) and anti-HA (b, d). A strong nuclear Notch1 signal is detected in transfected neurons (a) that is
blocked by preadsorption of -Notch.IC with Notch1 fusion protein (c). FCDN1.HA expression is detected in both
whereas Notch1 was
conditions with anti-HA (b, d). (eh) Distribution of endogenous Notch1 in cortical cultures visualized using - concentrated in the nucleus
Notch.IC (e). Notch1 immunofluorescence is detected in the nucleus (revealed by Hoechst staining in f, h), and is in postmitotic neurons
blocked by preadsorption of the antibody with Notch1 fusion protein (g). Images of cultures with or without pread- (Fig. 4gl). This relationship
sorption were obtained under identical confocal settings. was also confirmed by con-

articles
Fig. 4. Neuronal differentiation is associated with a relative increase

in nuclear Notch1 immunofluorescence. E18 cortical neurons in cul-
ture were double labeled for expression of Notch1 in red (b, e, h, k)
and PCNA (a), TuJ1 (d) or BrdU (g, j) in green. (c, f, i, l) Merged
images. (ac) Notch1 immunofluorescence is predominantly cytoplas-
mic in PCNA-positive undifferentiated cortical cells (arrowheads) and
predominately nuclear in PCNA-negative cells (arrows). (df) Notch1
immunofluorescence is predominantly nuclear in TuJ1-positive postmi-
totic cortical neurons (arrow). (gi) Following BrdU labeling for 6 h,
Notch1 immunofluorescence is predominantly extranuclear in BrdU-
positive (dividing) cells (arrowheads). (jl) Following BrdU labeling for
24 h, Notch1 immunofluorescence is predominantly extranuclear in
BrdU-positive cells (arrowheads), and predominantly nuclear in BrdU-
negative (postmitotic) neurons (arrows). (m) Quantification of the
relative intensity of Notch1 immunofluorescence in the nucleus versus
the cytoplasm in dividing (BrdU-positive after 6-h labeling) and postmi-
totic (BrdU-negative after 24-h labeling) cells in E18 cortical cultures.
Dividing cells identified by either PCNA or Nestin immunoreactivity
and postmitotic neurons identified by either TuJ1 or MAP2 immunore-
activity also demonstrate a predominance of Notch1 in the nucleus of
postmitotic neurons.
P3 cytosolic fraction.) These results strengthen the conclusion

that the intracellular domain of endogenous Notch1 is present
in the nucleus of cortical neurons.
To characterize these two Notch1-immunoreactive proteins,
m we tagged truncated Notch1 constructs (Fig. 6a) with a hemag-
glutinin epitope, transfected them into cortical neurons,
immunoprecipitated with anti-HA and loaded them alongside a
P3 nuclear extract for western blots. ZEDN1.HA and FCDN1.HA
were about the same size as p120 and p100 bands detected in P3
nuclear extract, whereas CDN1.HA had a smaller molecular
weight than either band (Fig. 5b). (Although the ZEDN1 con-
struct is larger than FCDN1, ZEDN1 protein is cleaved intracel-
lularly to generate a cytosolic fragment the size of FCDN1).
Comigration of nuclear Notch1 and FCDN1.HA bands suggests
that endogenous nuclear Notch1 probably corresponds to the
cytoplasmic domain of Notch1 (cleaved at about aa1744)18.
Detection of p120 and p100 proteins in cortical nuclear
extracts and in ZEDN1- and FCDN1-transfected cells suggested
that the proteins might represent differentially phosphorylated
forms of the cleaved intracellular domain of Notch1. To examine
this possibility, FCDN1.HA and ZEDN1.HA immunoprecipitates
focal microscopy (data not shown). The relative levels of Notch1 were treated with phosphatase (a nonspecific phosphatase), sep-
immunofluorescence measured in the nuclei and cytoplasm of arated by SDS-PAGE, and probed with -Notch.IC (Fig. 5c). For
dividing cells and postmitotic neurons revealed that neuronal dif- both constructs, phosphatase treatment led to a decrease in the
ferentiation was associated with an increase in nuclear Notch 1 lev- presence of the p120 band and an increase in the p100 form, sug-
els (Fig. 4m). gesting that the p120 and p100 forms represent differentially phos-
As nuclear localization of endogenous Notch 1 is difficult to phorylated forms of Notch intracellular domain. This mobility
demonstrate, we also determined the subcellular localization shift was attenuated by the phosphatase inhibitor sodium ortho-
of endogenous Notch 1 by biochemical cell fractionation. Tissue vanadate. To directly determine if p120 was a phosphorylated
extracts from P3 and adult cortex were separated into cyto- form of the intracellular domain of Notch1, we transfected E18
plasmic and nuclear components on a sucrose-density gradi- cortical cultures with FCDN1.HA and metabolically labeled them
ent, separated by SDS-PAGE, blotted onto nitrocellulose and with [32P]orthophosphate. FCDN1-HA was then immunopre-
probed with -Notch1.IC. These western blots revealed two cipitated with anti-HA, separated by PAGE and transferred to
bands with approximate molecular weights of 120 kDa and 100 nitrocellulose. This blot was first exposed to film to detect phos-
kDa in the nuclear extracts (Fig. 5a). Probing the same blots phoproteins and then incubated with -Notch.IC for western
with an antibody to the -1 subunit of Na+/K+ ATPase (an inte- blotting. Anti-HA antibodies immunoprecipitated only one [32P]-
gral cell-membrane protein) verified that the nuclear Notch1- labeled protein from FCDN1.HA-transfected neurons (Fig. 5d).
immunoreactive bands did not represent contamination by Western analysis with -Notch.IC revealed that the phosphory-
membrane proteins (Fig. 5a). (During development, some lated band corresponded to p120 (Fig. 5d, left blot). Thus, the
Na+/K+ ATPase activity is present on intracellular organelles38, intracellular domain of Notch1 can exist in two forms, p120 and
which probably accounts for the Na+/K+ ATPase signal in the p100, the larger of which is phosphorylated .

articles
a b
c d e
Fig. 5. Characterization of nuclear forms of Notch1. (a) The distribution of Notch1 in whole tissue homogenates (H), the cytosolic fraction (C) and
the nuclear fraction (N) in extracts prepared from P3 and adult cortex as determined by western blotting using -Notch.IC. The nuclear extract con-
tains two Notch1-immunoreactive proteins that migrate at 120 kDa and 100 kDa (arrows). Probing the same blot with an antibody to a membrane
protein (Na+/K+ ATPase) does not give a signal in the nuclear fraction, indicating that the nuclear Notch1 signal does not represent membrane-asso-
ciated Notch1. (b) Western blot of nuclear fraction from P3 cortex loaded alongside immunoprecipitates of HA-tagged truncated Notch1 constructs
and probed with -Notch.IC. The upper and lower bands present in P3 nuclear fraction comigrate with the bands found with ZEDN1.HA and
FCDN1.HA. CDN1.HA lacks sequences in FCDN1 and migrates lower than the two bands in the P3 nuclear fraction. (c) Phosphatase treatment of
ZEDN1.HA and FCDN1.HA immunoprecipitates indicates that the p120 form of these proteins is phosphorylated. E18 cortical neurons were trans-
fected with either ZEDN1.HA or FCDN1.HA, immunoprecipitated with anti-HA, treated with PPase or 10 mM Na3VO4 and separated by PAGE.
The western blot was then probed with -Notch.IC. Treatment with PPase causes a reduction in the p120 band compared to the p100 band. The
residual p120 signal in the ZEDN1.HA-transfected cells may correspond to the p120 transmembrane form of Notch1 that has not yet undergone
cleavage in the intracellular domain. (d) Metabolic labeling of FCDN1.HA-transfected neurons with [32P]orthophosphate followed by immunoprecip-
itation with HA antibodies identifies the p120 form of FCDN1.HA as a phosphorylated protein. Immunoprecipitates from [32P]-labeled FCDN1.HA
(+) and control () transfected neurons were separated by PAGE, transferred to nitrocellulose and exposed to film to detect [32P] radioactive signal
(right). Subsequently, the blot was probed with -Notch.IC to reveal the p100 and p120 forms of FCDN1 (left). Lanes 3 and 4 in the western blot
(left) are the same two lanes shown in the film exposure (right). Comparison of the two indicates that the p120 form of FCDN1 is phosphorylated
(arrow). (e) Relative mobility of nuclear Notch1 bands (P3 nuclear) and the biotinylated transmembrane p120 form of Notch1 (E18 + 5DIV biotiny-
lated). The upper (phosphorylated) nuclear band comigrates with the transmembrane p120 form of Notch1.
To determine if the phosphorylated form of nuclear Notch1 in postmitotic cortical neurons, we transfected a series of Notch1-
was distinguishable from the truncated transmembrane form by deletion constructs into cortical neurons (Fig. 6a) together with a
size, we ran cortical nuclear extracts next to biotinyated cell-sur- reporter driven by CBF1 binding sites14 (CBF1-CAT). By immuno-
face proteins and probed the blot with -Notch.IC. The upper fluorescence, transfected full-length Notch1 was excluded from
nuclear Notch1 band comigrated with p120 transmembrane the nucleus and was present mainly in intracellular organelles (and
Notch1 (Fig. 5d). Thus, neurons had two processed forms of perhaps at the cell surface, to some extent). Transfected 0CDN1
Notch1 that migrated at about 120 kDa: the truncated trans- was present mainly on the cell surface and in the cytoplasm. In
membrane receptor and the phosphorylated intracellular domain contrast, CDN1 and CDCN1T were present both in the cytoplasm
of Notch1. These two forms could not be distinguished by west- and in the nucleus, and ZEDN1 and FCDN1 were present mainly
ern blots of whole-cell lysates, and their identification required in the nucleus (Fig. 3 and data not shown).
surface biotinylation and cell-fraction experiments. In neurons transfected with reporter together with the par-
ent vector (pBos), there was a low basal level of reporter activ-
Regulation of transcription by nuclear Notch1 ity (Fig. 6b and c). Transfection with the various Notch1
In non-neuronal cell lines, the cytoplasmic domain of Notch1 can constructs led to markedly different levels of transactivation
translocate to the nucleus and induce CBF1-mediated transacti- (Fig. 6b and c). Whereas transfection with full-length Notch1
vation. To determine whether this signaling pathway is functional led to a weak but significant activation of the reporter, trans-

articles
Fig. 6. Notch1 can induce

transactivation via CBF1 a
binding sites in cortical
neurons. (a) Full-length
and truncated Notch1
cDNA constructs used in
transfection experiments.
The location of a HA-tag
in the CDCN1T construct
is indicated by a T. (b, c)
Relative CAT activity in
E18 cor tical cultures b
transfected with indi
cated Notch1 constructs
together with CBF1-CAT. c
(d, e) Relative CAT activ-
ity in E18 cortical cultures
transfected with vector
Fold induction
(pBos) or full-length
Notch1 (N1) together
w i t h C B F 1 - C AT a n d
increasing amounts of
0CDN1. 0CDN1 attenu-
ates Notch1 activation of
CBF1-CAT.
fection with ZEDN1 and

FCDN1 led to robust
activation of the
d e
reporter (Fig. 6b and
c). Transactivation was
only weakly activated
Fold induction
by CDN1, a construct
lacking the RAM
domain, a region
between the trans-
membrane domain and
the ankyrin repeats.
Thus the cytoplasmic
domain of Notch1 can
strongly activate CBF1-
mediated transcription
in cortical neurons,
and Notch1 activation
of CBF1-mediated
transcription requires
the RAM domain.
A form of Notch lacking the intracellular domain (0CDN1) (Fig. 7). In this experiment, Notch1 was likely activated by lig-
can act as a dominant-negative receptor39. When expressed in ands expressed by the cultured cortical neurons. In contrast to
cortical neurons, 0CDN1 did not activate the CBF1-CAT the CBF1-CAT reporter, which is strongly activated by ZEDN1
reporter and inhibited activation of CBF1-CAT by full-length and FCDN1, the NSE-CAT reporter was not activated by trun-
Notch1, indicating that it could inhibit Notch1 signaling to cated Notch1 constructs. The failure of truncated Notch1 con-
CBF1 (Fig. 6d and e). structs to activate NSE.CAT suggests that Notch1 signaling to
We next asked if Notch1 signaling could regulate neuron-spe- NSE-CAT may be CBF1 independent. This is not entirely unex-
cific gene expression. The neuron-specific enolase (NSE) gene is pected, as alternate Notch1 signaling pathways have been
expressed exclusively in terminally differentiated neurons and described both in Drosophila development and in myogenesis42.
neuroendocrine cells40, and an 1800-bp fragment of the promoter
can drive neuron-specific expression of a LacZ reporter in trans- Notch1 signaling regulates dendritic morphology
genic mice41. A CAT-reporter construct driven by this 1800-bp The Notch1 localization and signaling experiments suggested
fragment (NSE-CAT) transfected into cortical neurons gave a that Notch1 might regulate aspects of neuronal differentiation.
low level of reporter activity (Fig. 7). Cotransfection of full-length We thus examined the effects of inhibiting Notch1 signaling on
Notch1 led to increased reporter expression, indicating that one of the clearest manifestations of neuronal differentiation,
Notch1 signaling can lead to activation of the NSE promoter the development of dendrites. The dendritic morphology of cul-

articles
a b both process number (Fig. 9c) and den-

dritic branching (Fig. 9d and f). Notch1
antisense treatments also increased aver-
age dendritic length (Fig. 9e). To demon-
strate the specificity of the antisense
Fold induction
treatments, we rescued the effects by over-
expressing an activated Notch1 construct
(FCDN1). The effects of antisense treat-
ment on various dendritic parameters were
partially or completely reversed by overex-
pressing FCDN1 (Fig. 9gj). These obser-
vations support the results of the
transfection experiments, and suggest that
Notch1 signaling exerts a positive effect on
dendritic branching and a negative effect
on dendritic growth.
DISCUSSION
Our experiments demonstrate an impor-

Fig. 7. Notch1 induces gene expression via the neuron-specific enolase (NSE) promoter. (a) tant regulatory role for Notch1 signaling in
Relative CAT activity in E18 cortical cultures transfected with vector (Bos) or full-length Notch1 the differentiation of newly postmitotic
(N1) together with either the parent vector (0NSE) or an NSE-CAT construct containing 1800 bp neurons. We found that neuronal differen-
of the NSE promoter (-1800 NSE). (b) Relative CAT activity in E18 cortical cultures transfected tiation was associated with a change in the
with indicated Notch constructs and the -1800 NSE reporter.
subcellular localization of Notch1 from
extranuclear compartments to the nucle-
us. In cortical neurons, the intracellular
domain of Notch1 translocated to the
tured neurons can be clearly visualized by transfecting them with nucleus and could induce transactivation of a CBF1-driven
-galactosidase followed by immunocytochemistry for -galac- reporter construct. Attenuation of Notch1 signaling, either by
tosidase49. We transfected cortical cultures with -galactosidase expression of a dominant-negative construct or by treatment
and Bos (parent vector), 0CDN1 (dominant negative receptor) with Notch1 antisense oligonucleotides, reduced both the num-
or ZEDN1 (constitutively active receptor) at two days in vitro ber of primary processes and dendritic branching. Increasing
(DIV) and examined the morphology of transfected neurons Notch1 signaling decreased average dendritic length. These obser-
three days later. vations indicate that, in addition to previously identified roles
Bos-transfected and ZEDN1-transfected neurons had com- in cell-fate specification, mammalian Notch1 continues to func-
plex dendritic morphologies characterized by several primary tion in postmitotic neurons, and is important in the specifica-
processes and highly branched dendritic trees (Fig. 8a and c). In tion of dendritic morphology.
contrast, 0CDN1-transfected neurons had noticeably simpler den- Although Notch1 is largely excluded from the nucleus in
dritic morphologies (Fig. 8b). Processes were confirmed to be undifferentiated cells, it is present in the nucleus of postmitotic
dendrites by MAP2 immunofluorescence (data not shown). neurons both in vivo and in vitro. A possible regulatory mecha-
Quantitative analysis indicated that Notch1 signaling had distinct nism in this process is receptor activation at the onset of corti-
effects on process number, dendritic length and dendritic branch- cal differentiation. As the cytoplasmic domain of Notch1 seems
ing. Inhibiting Notch1 signaling decreased the number of process- to translocate to the nucleus once neurons reach the cortical plate,
es and the number of dendritic branch points per neuron it is possible that a Notch1 ligand present in the cortical plate
(Fig. 8d and e). In contrast, increasing Notch1 signaling reduced induces intracellular cleavage of the receptor. Although we have
average dendritic length (Fig. 8f). Although constitutively active not yet identified the relevant Notch ligand, Delta is a likely can-
Notch1 did not affect the total number of branches, it increased didate, as it is expressed in the cortical plate (L.R. and A.G.,
the dendritic branching index (number of branches per unit den- unpublished observation).
dritic length; Fig. 8g). Although there is evidence that Notch signaling might involve
As an alternative method to evaluate the role of Notch1 sig- nuclear translocation of the cytoplasmic domain, there has been
naling in dendritic development, we inhibited Notch1 by treating controversy about such a mechanism because of a lack of evidence
cortical cultures with previously characterized Notch1-antisense for endogenous Notch in the nucleus in vivo. Our immunofluo-
oligonucleotides24 (Methods). Antisense treatment decreased rescence and cell-fractionation experiments clearly indicated the
Notch1 protein levels approximately 30% compared to sense presence of the cytoplasmic domain of endogenous Notch1 in the
oligonucleotide-treated controls (data not shown). Cultures were nucleus of postmitotic neurons. We detected two forms of nuclear
transfected with -galactosidase one day before oligonucleotide Notch1, p120 and p100, which correspond to differentially phos-
treatment, and transfected neurons were visualized by -galac- phorylated forms of the cleaved intracellular domain. A major
tosidase immunocytochemistry after two days of oligonucleotide nuclear target of Notch1 is CBF1, and it will be interesting to
treatment. Whereas neurons treated with sense (control) oligonu- determine if phosphorylation affects Notch1CBF1 interactions or
cleotides had complex dendritic morphologies, neurons treated Notch1-induced transcription in neurons.
with either of the two Notch1 antisense oligonucleotides had Our experiments suggest that Notch1 signaling influences
markedly simpler morphologies (Fig. 9a and b). Like dominant- dendritic development in cortical neurons. We found that
negative Notch1 transfections, antisense treatments decreased Notch1 signaling exerts a positive effect on dendritic branch-

articles
Fig. 8. Effects of altering

Notch1 signaling on dendritic a
morphology. (ac) Video cam-
era-lucida drawings of the den-
dritic tree of cortical neurons
transfected with -galactosidase
together with pBos (parent vec-
tor; a), 0CDN1 (b) or ZEDN1
(c) at 2 DIV, and processed for
-galactosidase immunoreactiv-
ity at 5 DIV. Scale bar is 20 m.
(d) Number of primary
processes per cell in pBos-, b
0CDN1- and ZEDN1-trans-
fected neurons. (e) Number of
dendritic branch points per cell
in pBos-, 0CDN1- and ZEDN1-
transfected neurons. (f) Average
dendrite length (total dendrite
length/number of primary den-
drites) per cell in pBos-,

0CDN1- and ZEDN1-trans-
fected neurons. (g) Branching
index (see Methods) in pBos-, c
0CDN1-, and ZEDN1-trans-
fected neurons.
ing and a negative effect on

dendritic growth. The effects
we report are consistent with
findings of two other
groups 4345 . Activation of d e f g
Notch1 is associated with a
decrease in total neurite
Branchpoints per cell
Avg. dendrite length (m)
length45, similar to effects we

Branching index
processes per cell
see on dendritic length. We

also saw a positive effect of
Notch1 signaling on den-
dritic branching. Although
dendritic growth and
branching might be mecha-
nistically linked, the branch-
ing effects were probably not
secondary to an effect on dendritic growth, as inhibition of tions supporting a role for Notch1 in dendritic development,
Notch1 had opposite effects on dendritic length and dendrit- suggest that Notch-mediated interactions may be critical in
ic branching, and certain perturbations (such as 0CDN1 regulating the patterning of connections well after cell fate has
expression) affected the number of dendritic branches without been decided.
affecting dendritic length. We should also note that our results
do not rule out the possibility that Notch1 signaling favored METHODS
Immunohistochemistry and immunofluorescence. Animals were anes-
differentiation of neurons into a subtype characterized by thetized and perfused with phosphate-buffered saline (PBS) followed
reduced dendritic length and greater dendritic branching. by 4% paraformaldehyde in PBS. The brains were removed, embedded
However, such a role of Notch1 in postmitotic neurons seems in gelatin, cryoprotected in 4% paraformaldehyde/30% sucrose in PBS
unlikely, as most cell-fate decisions in the cortex are made and sectioned at 2040 m. Cultures were fixed with 4% paraformalde-
before the final cell division. hyde/4% sucrose in PBS and washed with PBS. Following fixation, tis-
Despite the dominant view that Notch signaling functions sue sections and cultures were blocked with 3% bovine serum albumin,
to inhibit differentiation and restrict cell fates, observations 3% goat serum and 0.3% Triton X-100 in PBS and incubated in the
in Drosophila development suggest a role for Notch in regu- primary antibody (diluted in blocking solution) overnight. For BrdU
lating developmental decisions subsequent to specification of immunostaining, cultures were postfixed in 70% ethanol, additionally
permeabilized with 0.4% Triton X-100 and 2 N HCl and neutralized
cell fate. For example, Notch is present on the growth cones
with 0.1 M NaB2O7 before incubation in blocking solution. After incu-
of developing axons, and mutations in Notch affect axonal bation in primary antibody, tissue sections and cultures were incubat-
growth 4647. Similarly, Kuzbanian (a protein implicated in ed in peroxidase-conjugated (ABC Elite, Vector Labs, Burlingame,
Notch and Delta processing) is required for axon extension in California) or fluorescent (Cy-2 and Cy-3, Jackson Immunoresearch,
Drosophila48. These observations, together with our observa- West Grove, Pennsylvania; Pacific blue and Oregon green, Molecular

articles
Fig. 9. Effects of Notch1 antisense

a oligonucleotide treatment on den-
dritic morphology.(a, b) Video
camera-lucida drawings of cortical
neurons transfected with -galac-
tosidase at 2 DIV, treated with
either 2 M CDC sense
(a) or 2 M CDC antisense
(b) oligonucleotides at 3 DIV, and
processed for -galactosidase
immmunoreactivity at 5 DIV. Scale
bar, 20 m. (cf) Number of pri-
b mary processes (c), number of
dendritic branch points (d), aver-
age dendritic length (e) and
branching index (f) in cortical neu-
rons treated with 2 M Notch1
sense (CDC.S, LN.S) or antisense
(CDC.AS, LN.AS) oligonu-
cleotides. (gj) Effects of FCDN1
expression on number of primary

processes (g), number of dendritic
branch points (h), average dendrite
c d e f length (i) and branching index (j) in
Notch1 sense-(CDC.S) and anti-
processes per cell

Branching index
sense-(CDC.AS) treated neurons.
poly-D-lysine- and laminin-coat-

ed dishes at 1 106 cells per well
(12-well dishes), at 1.5 106 cells
g h i j per well (6-well and 35-mm dish-
es), or at 3 106 cells per 60-mm
processes per cell
Branching index
dish in glutamine-free Basal Medi-

um Eagle (Sigma), 1 mM gluta-
mine, 1% N2, and fetal bovine
serum (Gibco) and maintained at
37C in 95%/5% O 2 /CO 2 . The
double-labeling experiments
shown in Fig. 4 were performed in
35-mm glass bottom dishes, and
bFGF (10 ng per ml) was added at
the time of plating. For bromod-
Probes, Eugene, Oregon) secondary antibodies. Samples processed for eoxyuridine (BrdU) labeling, 10 M BrdU was added to the cultures 6 h
immunocytochemistry without primary antibody incubation did not or 24 h before fixation.
show detectable staining (data not shown).
The Notch1 antibodies used in this study were the affinity-purified Constructs and transfections. Wild-type and truncated Notch1 cDNA
rabbit polyclonal antiserum 93-4, generated against the cytoplasmic sequences were engineered in the mammalian expression vector pEF1-
domain of Notch1, and affinity-purified rabbit polyclonal antiserum BOS or pcDNA3 to encode the following amino acids: N1, 12531;
5261, generated against the extracellular domain of Notch1 (ref. 42). 0CDN1, 11759; ZEDN1, 124, 17122531; FCDN1, 17412531; CDN1,
Notch1-glutathione S-transferase (GST) used for affinity purification 18482531; CDCN1T, 18732078. All constructs were confirmed by DNA
was engineered and purified using Pharmacia protocols as described42. sequencing42. Additional ZEDN1, FCDN1 and CDN1 constructs were
The 93-4 affinity-purified antiserum was used in all of the Notch1 engineered with a 3 HA tag in pEF1-BOS. The CBF1-CAT reporters,
immunofluorescence and immunocytochemistry experiments. In con- with four copies of the wild-type or mutant CBF1 binding elements
trol experiments, 93-4 affinity-purified antiserum was preadsorbed with cloned upstream of a chloramphenicol acetyltransferase (CAT) gene in
an excess (200 g per ml) of Notch1 fusion protein at 4C before incu- the mammalian expression vector pSG5, were a gift from S. D. Hayward
bation with cells. We also used the following additional antibodies: mouse (Johns Hopkins). The previously described NSE-CAT reporter40 con-
anti--galactosidase (Promega, Madison, Wisconsin); rabbit anti--galac- tains 1800 bp of NSE promoter sequences upstream of a CAT reporter.
tosidase (53, Boulder, Colorado); mouse anti-Nestin (Pharmingen, The control reporter (0-NSE-CAT) was generated by removing the NSE
San Diego, California); mouse anti-TuJ1 (Sigma); mouse anti-MAP2 promoter sequences between the SacI and HindIII sites of NSE-CAT and
(Sigma); mouse anti-BrdU (Becton Dickinson); mouse anti-HA (clone religating the plasmid.
12CA5, Boehringer Mannheim) and mouse anti-PCNA (Boehringer Cells were transfected by a modified calcium phosphate procedure as
Mannheim). Confocal images of tissue sections and cultured cells were described49. This procedure exclusively transfects postmitotic neurons.
obtained on a Zeiss LSM410 or Zeiss LSM510 microscope equipped with The expression of various Notch1 constructs was confirmed by western
argon (red), HeNe (green) and UV (blue) lasers. blots of cell lysates from cultures of transfected cortical cells and 293T
cells, and also by immunofluorescence of transfected cortical neurons.
Primary cell cultures. E18 cortical cells from Long-Evans rats were cul- To demonstrate attenuation of full-length Notch1 signaling by 0CDN1,
tured as described 49 . Dissociated cortical cells were plated onto we transfected cultures with CBF1-CAT, either full-length Notch1 or

articles
pBos and either 0CDN1 or pBos at molar ratios of 1:1:1 and 1:1:3. The trol and FCDN1.HA-transfected cultures with 200 Ci per 60-mm dish
final amount of DNA transfected was equalized with pBluescript for each of [32P]orthophosphate (H3PO4; ICN) in phosphate-free DMEM (Gibco)
condition. For morphological analysis, E18 cortical cultures were trans- for 4 h at 37C. Cells were lysed, immunoprecipitated with anti-HA and
fected at 2 DIV with RSV--galactosidase together with the parent expres- transferred to nitrocellulose as above. The blot was first exposed to film
sion vector (control), the 0CDN1 expression plasmid or the ZEDN1 to detect radioactivity and then immunoprecipitated. FCDN1.HA was
expression plasmid in a molar ratio of 1:2 (gal construct) and were fixed detected with -Notch.IC as described above.
and processed for -galactosidase immunocytochemistry at 5 DIV before For the biotinylation experiments, E18 cortical cultures in 60-mm
morphology of transfected neurons was analyzed (see Analysis below). dishes were washed with dextrose (1 mg per ml) in PBS, then incubated
at 4C in 2 mM EZ link Sulfo-NHS-LC-Biotin (Pierce) in dextrose (1 mg
Western blots and cell fractionation. We used 10 g of Notch1, Notch2, per ml) in PBS for 30 min. Cells were rinsed multiple times with dex-
Notch3-HA, Notch4-HA or parent vector (pBos) to transfect 293 cells trose (1 mg per ml) in PBS and lysed as above for immunoprecipitation.
using a calcium phosphate procedure. Cells collected from the same trans- Lysis supernatants were incubated with immobilized Streptavidin (Pierce)
fection were lysed in RIPA buffer, pooled and then split for immunopre- overnight at 4C, washed and separated by PAGE. Proteins were trans-
cipitation. Lysates were immunoprecipitated with 1:100 of rabbit ferred to nitrocellulose and incubated with -Notch.IC (93-4) or an
polyclonal antibodies to either Notch1 (93-4) or Notch2 (93-3) or with extracellular-domain antibody (5261) and detected as above.
monoclonal antibody to HA (12CA5) and captured with Protein-A
agarose. Immunoprecipitates were analyzed by western blot using 93-4 CAT assays. Cells were harvested 1 (NSE-CAT) or 2 (CBF1-CAT) days
(1:5000) to detect Notch1, 93-3 (1:3000) to detect Notch2 and 12CA5 after transfection in isotonic TNE (10 mM Tris, 150 mM NaCl and 1 mM
(1:1000) monoclonal to detect the HA tag present on Notch3 and Notch4. EDTA at pH 7.8). Cells were spun down gently and subjected to three
Antibodyprotein interactions were detected by ECL. cycles of freeze-thaw lysis. Lysis supernatants were incubated with 0.5
For tissue westerns, cortices were dissected from embryonic (E13 and Ci 14C-labeled chloramphenicol (Amersham) and 0.8 mmol acetyl CoA
E18) and postnatal rats (P0, P7, P14 and adult). Tissues were homogenized (Boehringer Mannheim) at pH 7.8 and at 37C for 1 h. Reaction mix-
with a dounce homogenizer in lysis buffer (137 mM NaCl, 20 mM Tris, 1% tures were extracted with ethyl acetate, speed-vacuumed, resuspended
NP40, 10% glycerol, 1 mM PMSF, 10 g per ml aprotinin, 1 g per ml pep- in chloroform, spotted onto thin-layer chromatography (TLC) plates
statin A and 1g per ml leupeptin at pH 8.0). To disrupt nuclei, (J.T. Baker) and separated by ascending chromatography for 2 h (95%
homogenates were sonicated with 2 bursts (15 s), homogenized with pestel chloroform, 5% methanol). To normalize for transfection efficiency,
B and mixed for 1 h at 4C. E18 cultured cells were rinsed in TBS, harvest- equal amounts of a RSV--galactosidase plasmid were cotransfected in
ed and homogenized as for whole tissue. BCA Protein Assay (Pierce, Rock- all reporter experiments, and -galactosidase activity was used for nor-
ford, Illinois) was used to determine protein concentration, and 10 g of malization. Data from any single histogram were quantified from exper-
each homogenate was separated by PAGE. Proteins were transferred to nitro- iments on cells that were simultaneously cultured, transfected and assayed
cellulose using a Genie Electroblotter system (Idea Scientific, Minneapolis, to minimize variability due to subtle differences in experimental proce-
Minnesota). Blots were blocked with 4% BSA in TBST and incubated with dure. Fold induction was calculated relative to control transfected cul-
-Notch.IC (93-4) diluted 1:10,000 in blocking solution overnight at 4C. tures. Shown are representative examples of experiments performed three
In control experiments, -Notch.IC was incubated at 1:15,000 with an excess times. For measurements of relative CAT activity, levels of [14C] emis-
of Notch1 fusion protein (20 g per ml) overnight at 4C before incuba- sions on TLC plates were quantified by phosphorimager scans. Statisti-
tion with blots of FCDN1.HA immunoprecipitates. Blots were incubated cally significant differences (Students t-test, p < 0.05) in reporter activity
with peroxidase-conjugated anti-rabbit secondary antibody (Amersham) are indicated by an asterisk.
diluted 1:20,000 in blocking solution and visualized using a chemilumi-
nescent detection system (SuperSignal Substrate, Pierce). Antisense oligonucleotides. Oligonucleotides were designed against two
Cell fractionation and isolation of nuclear proteins was carried out distinct regions of rat Notch1 as reported24, a portion of the 5 intra-
as described50. Cortices from P3 and adult rats were minced and homog- cellular cdc10/ankyrin repeat region (CDC) and the extracellular
enized by two-component dounce homogenization, and nuclei were lin12/Notch repeat region (LN). The CDC antisense sequence, 5-CCTC-
separated on a sucrose density gradient. Isolated nuclei were resuspended CGCTGCAGGAGGCAATCAT-3, and the LN antisense sequence,
in lysis buffer and disrupted by sonication. Protein concentrations were 5-CCAGCACTGCAGGGACTGAGTGC-3, were used. For each anti-
determined by BCA protein assay, and 10 g of each fraction sense oligonucleotide, corresponding sense oligonucleotides were syn-
homogenate, cytoplasmic and nuclearwere separated by PAGE and thesized and used in parallel in each experiment. All oligonucleotides
transferred to nitrocellulose. Notch1 was detected using -Notch.IC as were 23 bp in length, with phosphothioate linkages added between all
described above. Parallel blots were blocked with 3% nonfat dry milk bases. Oligonucleotides were synthesized at the Johns Hopkins School
in PBS and incubated overnight with an antibody to the -1 subunit of of Hygiene and Public Health DNA Synthesis Core Facility. Optical den-
the Na+/K+ ATPase (Upstate Biotechnology) at 1:1000 in blocking solu- sity (OD260) readings were taken to determine concentration. For anti-
tion at 4C. Blots were incubated with peroxidase-conjugated anti-rab- sense treatment experiments, E18 cortical neurons were cultured as
bit secondary antibody (Amersham) diluted 1:10,000 in blocking described on 12-well tissue culture plates. Cells were transfected with
solution and were visualized by chemiluminescence, as above. RSV--galactosidase at 2 DIV. The day after transfection, sense and anti-
sense oligonucleotides were added to a final concentration of 2 M and
Immunoprecipitation, metabolic labeling and cell-surface biotinylation. maintained for an additional two days before fixation. The morphology
E18 cortical cultures in 60-mm dishes were transfected with 6 g per of the transfected neurons was revealed by -galactosidase immunocy-
dish of ZEDN1.HA, FCDN1.HA or CDN1.HA. Three days after trans- tochemistry and analyzed as described below.
fection, cells were lysed in lysis buffer. Lysed cells were homogenized with
pestel B of a dounce homogenizer. Lysis supernatants were separated Analysis. Transfection experiments were carried out in duplicate wells,
from fragmented cells by centrifugation, incubated with anti-HA anti- and all of the experiments described here were repeated multiple times in
bodies and Gamma Bind G Sepharose (Pharmacia), washed and sepa- several independent experiments with comparable results. To analyze the
rated by PAGE. For the analysis of protein phosphorylation, cells from effects of antisense treatment, ZEDN1 and 0CDN1 expression on dendritic
the same transfection condition were lysed as above, pooled and then morphology, images of -galactosidase-transfected neurons were captured
split for immunoprecipitation. After binding to Gamma Bind G using a digital CCD camera (Dage, MTI, Michigan City, Indiana) attached
Sepharose, immunoprecipitates were incubated for 1 hour at 30C with to the side port of a Nikon Eclipse TE300 inverted microscope driven by
50 U Phosphatase (NEB) or 10 mM sodium orthovanadate (Na3VO4) in IP Lab Spectrum 3.1.1 image-acquisition software (Scanalytics, Fairfax,
50 mM TrisHCl, 5 mM DTT, 2 mM MnCl2, 100 g per ml BSA and pro- Virginia). Captured cells were then traced using ClarisDraw software. To
tease inhibitors as above at pH 7.8 and washed and separated by PAGE. assess the effects of ZEDN1 and 0CDN1 expression on dendritic mor-
After separation by PAGE, proteins were transferred to nitrocellulose and phology, the dendritic trees of at least 50 Bos-(control), ZEDN1- and
detected as above. Metabolic labeling was performed by incubating con- 0CDN1-transfected neurons were reconstructed and scored for number

articles
of primary processes, dendritic branch points, average dendritic length 20. Coffman, C., Harris, W. & Kintner, C. Xotch, the Xenopus homolog of
and branching index (number of branch points divided by average den- Drosophila notch. Science 249, 14381441 (1990).
21. Coffman, C. R., Skoglund, P., Harris, W. A. & Kintner, C. R. Expression of an
dritic length, expressed as percent of control). For quantitative analysis of extracellular deletion of Xotch diverts cell fate in Xenopus embryos. Cell 73,
the effects of oligonucleotide treatments on dendritic morphology, the 659671 (1993).
dendritic trees of at least 100 transfected neurons from each treatment 22. Chitnis, A., Henrique, D., Lewis, J., Ish-Horowicz, D. & Kintner, C. Primary
condition per experiment were reconstructed and scored for various den- neurogenesis in Xenopus embryos regulated by a homologue of the
dritic parameters. Independent blind analysis of dendritic morphology Drosophila neurogenic gene Delta. Nature 375, 761766 (1995).
23. Dorsky, R. I., Rapaport, D. H. & Harris, W. A. Xotch inhibits cell
under various experimental conditions showed similar results. Data are differentiation in the Xenopus retina. Neuron 14, 487496 (1995).
shown as mean standard error. Statistically significant differences (Stu- 24. Austin, C. P., Feldman, D. E., Ida, J. A. Jr. & Cepko, C. L. Vertebrate retinal
dents t-test; p < 0.05) are indicated by an asterisk. To analyze the relative ganglion cells are selected from competent progenitors by the action of
amounts of intracellular domain Notch1 in the nucleus and cytoplasm of Notch. Development 121, 36373650 (1995).
25. Weinmaster, G., Roberts, V. J. & Lemke, G. A homolog of Drosophila Notch
double-labeled cells (Fig. 4m), we captured images of Notch1-immunos- expressed during mammalian development. Development 113, 199205 (1991).
tained neurons as above. The intensity of Notch1 immunofluorescence 26. del Amo, F. F. et al. Expression pattern of Motch, a mouse homologue of
was then measured in identical volumes of the nucleus and cytoplasm from Drosophila Notch, suggests an important role in early postimplantation
the same cell using IP Lab Spectrum 3.1.1 image software. mouse development. Development 115, 737744 (1992).
27. Reaume, A. G., Conlon, R. A., Zirngibl, R., Yamaguchi, T. P. & Rossant, J.
Expression analysis of a Notch homologue in the mouse embryo. Dev. Biol.
ACKNOWLEDGEMENTS 154, 377387 (1992).
We thank Diane Hayward for providing us with CBF1 reporter constructs, 28. Weinmaster, G., Roberts, V. J. & Lemke, G. Notch2: a second mammalian
Carrie Shawber and Libby Walker for the pCDNA3.FCDN1 plasmid, Donna Notch gene. Development 116, 931941 (1992).
29. Lardelli, M., Dahlstrand, J. & Lendahl, U. The novel Notch homologue
Nofziger for affinity purification of the 93-4 antisera, Greg Sutcliffe for the NSE-
mouse Notch3 lacks specific epidermal growth factor-repeats and is

CAT construct, Yuh-Nung Jan, Weimin Zhong and Connie Cepko for discussions expressed in proliferating neuroepithelium. Mech. Dev. 46, 123136
and members of the Ghosh lab for comments on the manuscript. This work was (1994).
30. Lindsell, C. E., Boulter, J., diSibio, G., Gossler, A. & Weinmaster, G.
supported by NIH grant NS36176 and the Pew Scholars Program (A.G.). Expression patterns of Jagged, Delta1, Notch1, Notch2, and Notch3 genes
identify ligand-receptor pairs that may function in neural development. Mol.
RECEIVED 20 OCTOBER; ACCEPTED 22 NOVEMBER 1999 Cell. Neurosci. 8, 1427 (1996).
31. Bettenhausen, B., de Angelis, M. H., Simon, D., Guenet, J. L. & Gossler, A.
Transient and restricted expression during mouse embryogenesis of Dll1, a
1. Greenwald, I. & Rubin, G. M. Making a difference: the role of cellcell murine gene closely related to Drosophila Delta. Development 121,
interactions in establishing separate identities for equivalent cells. Cell 68, 24072418 (1995).
271281 (1992). 32. Lindsell, C. E., Shawber, C. J., Boulter, J. & Weinmaster, G. Jagged: a
2. Ghysen, A., Dambly-Chaudiere, C., Jan, L. Y. & Jan, Y. N. Cell interactions and mammalian ligand that activates Notch1. Cell 80, 909917 (1995).
gene interactions in peripheral neurogenesis. Genes Dev. 7, 723733 (1993). 33. Shawber, C., Boulter, J., Lindsell, C. E. & Weinmaster, G. Jagged2: a serrate-
3. Artavanis-Tsakonas, S., Matsuno, K. & Fortini, M. E. Notch signaling. Science like gene expressed during rat embryogenesis. Dev. Biol. 180, 370376 (1996).
268, 225232 (1995). 34. Swiatek, P. J., Lindsell, C. E., del Amo, F. F., Weinmaster, G. & Gridley, T.
4. Kimble, J. & Simpson, P. The lin-12/Notch signaling pathway and its Notch1 is essential for postimplantation development in mice. Genes Dev. 8,
regulation. Annu. Rev. Cell Dev. Biol. 13, 333361 (1997). 707719 (1994).
5. Greenwald, I. Lin-12/Notch signaling: lessons from worms and flies. Genes 35. Conlon, R. A., Reaume, A. G. & Rossant, J. Notch1 is required for the
Dev. 12, 17511762 (1998). coordinate segmentation of somites. Development 121, 15331545 (1995).
6. Pan, D. & Rubin, G. M. Kuzbanian controls proteolytic processing of Notch 36. Lardelli, M., Williams, R., Mitsiadis, T. & Lendahl, U. Expression of the
and mediates lateral inhibition during Drosophila and vertebrate Notch 3 intracellular domain in mouse central nervous system progenitor
neurogenesis. Cell 90, 271280 (1997). cells is lethal and leads to disturbed neural tube development. Mech. Dev. 59,
7. Blaumueller, C. M., Qi, H., Zagouras, P. & Artavanis-Tsakonas, S. 177190 (1996).
Intracellular cleavage of Notch leads to a heterodimeric receptor on the 37. Chenn, A. & McConnell, S. K. Cleavage orientation and the asymmetric
plasma membrane. Cell 90, 281291 (1997). inheritance of Notch 1 immunoreactivity in mammalian neurogenesis. Cell
8. Logeat, F. et al. The Notch1 receptor is cleaved constitutively by a furin-like 82, 631641 (1995).
convertase. Proc. Natl. Acad. Sci. USA 95, 81088112 (1998). 38. Gonatas, J. O., Gonatas, M. K., Stieber, A. & Fleischer, B. Isolation and
9. Kidd, S., Lieber, T. & Young, M. W. Ligand-induced cleavage and regulation characterization of an enriched Golgi fraction from neurons of developing
of nuclear entry of Notch in Drosophila melanogaster embryos. Genes Dev. rat brains. J. Neurochem. 45, 497507 (1985).
12, 37283740 (1998). 39. Rebay, I., Fehon, R. G. & Artavanis-Tsakonas, S. Specific truncations of
10. Fortini, M. E. & Artavanis-Tsakonas, S. The suppressor of hairless protein Drosophila Notch define dominant activated and dominant negative forms of
participates in notch receptor signaling. Cell 79, 273282 (1994). the receptor. Cell 74, 319329 (1993).
11. Lecourtois, M. & Schweisguth, F. The neurogenic Suppressor of Hairless 40. Schmechel, D. E., Brightman, M. W. & Marangos, P. J. Neurons switch from
DNA-binding protein mediates the transcriptional activation of the non-neuronal enolase to neuron-specific enolase during differentiation.
Enhancer of split Complex genes triggered by Notch signaling. Genes Dev. 9, Brain Res. 190, 195214 (1980).
25982608 (1995). 41. Forss-Petter, S. et al. Transgenic mice expressing -galactosidase in mature
12. Jarriault, S. et al. Signalling downstream of activated mammalian Notch. neurons under neuron-specific enolase promoter control. Neuron 5, 187197
Nature 377, 355358 (1995). (1990).
13. Tamura, K. et al. Physical interaction between a novel domain of the receptor 42. Shawber, C. et al. Notch signaling inhibits muscle cell differentiation through
Notch and the transcription factor RBP-J kappa/Su(H). Curr. Biol. 5, a CBF1independent pathway. Development 122, 37653773 (1996).
14161423 (1995). 43. Qi, H. et al. Processing of the Notch ligand Delta by the metalloprotease
14. Hsieh, J. J. et al. Truncated mammalian Notch1 activates CBF1/RBPJk- Kuzbanian. Science 283, 9194 (1999).
repressed genes by a mechanism resembling that of Epstein-Barr virus 44. Berezovska, O. et al. Notch1 inhibits neurite outgrowth in postmitotic
EBNA2. Mol. Cell. Biol. 16, 952959 (1996). primary neurons. Neuroscience 93, 433439 (1999).
15. Lu, F. M. & Lux, S. E. Constitutively active human Notch1 binds to the 45. Sestan, N., Artavanis-Tsakonas, A. & Rakic, P. Contact-dependent inhibition
transcription factor CBF1 and stimulates transcription through a promoter of cortical neurite growth mediated by Notch signaling. Science 286, 741746
containing a CBF1responsive element. Proc. Natl. Acad. Sci. USA 93, (1999).
56635667 (1996). 46. Giniger, E., Jan, L. Y. & Jan, Y.-N. Specifying the path of the intersegmental
16. Kopan, R., Schroeter, E. H., Weintraub, H. & Nye, J. S. Signal transduction by nerve of the Drosophila embryo: a role for Delta and Notch. Development 117,
activated mNotch: importance of proteolytic processing and its regulation by 431440 (1993).
the extracellular domain. Proc. Natl. Acad. Sci. USA 93, 16831688 (1996). 47. Giniger, E. A role for Abl in Notch signaling. Neuron 20, 667681 (1998).
17. Struhl, G. & Adachi, A. Nuclear access and action of Notch in vivo. Cell 93, 48. Fambrough, D., Pan, D., Rubin, G. M. & Goodman, C. S. The cell surface
649660 (1998). metalloprotease/disintegrin Kuzbanian is required for axonal extension in
18. Schroeter, E. H., Kisslinger, J. A. & Kopan, R. Notch1 signalling requires Drosophila. Proc. Natl. Acad. Sci. USA 93, 1323313238 (1996).
ligand-induced proteolytic release of intracellular domain. Nature 393, 49. Threadgill, R., Bobb, K. & Ghosh, A. Regulation of dendritic growth and
382386 (1998). remodeling by Rho, Rac, and Cdc42. Neuron 19, 625634 (1997).
19. Lecourtois, M. & Schweisguth, F. Indirect evidence for Delta-dependent 50. Thormodsson, F. R., Redmond, L. & Hockfield, S. Identification of nuclear
intracellular processing of Notch in Drosophila embryos. Curr. Biol. 8, proteins that are developmentally regulated in embryonic rat brain.
771774 (1998). J. Neurochem. 64, 19191927 (1995).

articles
The seven-transmembrane receptor

Smoothened cell-autonomously
induces multiple ventral cell types
Mary Hynes1, Weilan Ye1, Kevin Wang1, Donna Stone1, Maximilien Murone2, Frederic de Sauvage2
and Arnon Rosenthal1
1 Departments of Neuroscience and 2Molecular Oncology, Genentech, Inc., South San Francisco, California 94080, USA
Correspondence should be addressed to M.H. (mah@gene.com) and A.R. (ar@gene.com)
Sonic Hedgehog (Shh) is a secreted protein that controls cell fate and mitogenesis in the develop-
ing nervous system. Here we show that a constitutively active form of Smoothened (Smo-M2)
mimics concentration-dependent actions of Shh in the developing neural tube, including
activation of ventral marker genes (HNF3, patched, Nkx2.2, netrin-1), suppression of dorsal
markers (Pax-3, Gli-3, Ephrin A5) and induction of ventral neurons (dopaminergic, serotonergic)
and ventrolateral motor neurons (Islet-1+, Islet-2+, HB9+) and interneurons (Engrailed-1+,
CHX10+). Furthermore, Smo-M2s patterning activities were cell autonomous, occurring
exclusively in cells expressing Smo-M2. These findings suggest that Smo is a key signaling
component in the Hh receptor and that Shh patterns the vertebrate nervous system as a
morphogen, rather than through secondary relay signals.
Sonic hedgehog (Shh) is a vertebrate homolog of the secreted and basal cell carcinomas21. In contrast, Smo acquires activating
segment-polarity gene, Hedgehog (Hh), which is produced by mutations in sporadic forms of this disorder22.
the embryonic notochord and floor plate. Genetic studies in Taken together, these findings lead to the hypothesis that the
human1,2 and mouse3 as well as studies with cultured chick and receptor for Shh is composed of both Ptc and Smo15,20,23. How-
rodent neural-tube explants4 revealed that Shh controls multiple ever, although Smo mediates simple mitogenic signals of Shh in
cell-patterning events along the ventral aspect of the neural tube. vertebrate skin22, the question of whether it is involved in the
An early activity of Shh is to restrict the expression of multiple coordination of the complex patterning actions of Shh in the
transcription factors (for example, Pax3, Pax7, Msx1 and Msx2)4 neural tube remains open. To directly examine this issue, we
and secreted proteins (for instance, Ephrin-A5)5 to the dorsal expressed a constitutively active form of Smo (Smo-M2) in the
aspect of the neural tube. Concomitant with the exclusion of neural tubes of mice and chick embryos. We show that ectopic
future dorsal cell markers, Shh upregulates the expression of expression of Smo-M2 resulted in downregulation of dorsal cell
ventral transcription factors [for example, HNF3, Nkx 2.2 (ref. markers, induction of ventral cell markers and the ectopic for-
6) Pax 6 (ref. 7) and Gli-1 (ref. 5)] as well as cell surface (Ptc)8 mation of ventral neurons.
and secreted (for example, Netrin-1)9,10 proteins in the ventral Having obtained evidence that Smo-M2 is a key signaling com-
neural tube, thus committing the affected neural precursors to a ponent of the Shh receptor, we asked whether Shh induces distinct
general ventral cell fate. Shh further induces the specification of cell types in the neural tube directly or through a second wave of
these generic ventral progenitors to distinct neuronal and non- secreted inducers. We assumed that if Shh specified neural prog-
neuronal cell types that typify the ventral aspect of the mature enitors directly, its activated receptor would suppress dorsal cell
vertebrate nervous system. These include floor plate cells, Islet- markers and upregulate ventral cell markers only in the cells in
1, 2 and HB9-positive motor neurons, Chx10+ and Engrailed-1+ which it was expressed. On the other hand, if a relay signal were
(En-1) interneurons4, serotonergic (5HT)5,11, dopaminergic involved, cells that did not express Smo-M2 would also be affect-
(DA)12,13 and multiple forebrain6 neurons. ed, as they would respond to secreted relay signals produced by
The mechanism of Shh signal transduction is not fully Smo-M2-expressing cells. Studies in non-neuronal tissues in
understood. However, biochemical studies demonstrate that Drosophila and vertebrates suggest that, depending on the partic-
Shh binds the 12-transmembrane (TM) protein Ptc14,15, and ular system, Hh can mediate its activities directly or via a relay
that Ptc in turn forms a physical complex with Smo15. Gene mechanism. For example, the mitogenic and patterning effects of
ablation and overexpression experiments in Drosophila as well Shh in the vertebrate limb are executed via members of both the
as gene ablation of Ptc in mice further suggest that Smo is fibroblast growth factor (FGF) and the bone morphogenic factors
required for Hh signal transduction16,17. In contrast, Ptc seems (BMP) protein families24,25, which act as relay signals. Likewise,
to be a negative regulator in this signaling system1820. Consistent the cell-patterning effects of Hh in the anterior compartment of
with these suggestions, Ptc is inactivated in hereditary basal cell the Drosophila leg imaginal disc are thought to be mediated indi-
nevus syndrome (BCNS), a disease associated with develop- rectly via the induction of BMP, dorsally, and wingless, ventrally26.
mental abnormalities and high incidence of medulloblastomas In addition, Hh seems to influence cell polarity in the Drosophila

articles
Fig. 1. Smo-M2 ventralizes the neural tube in transgenic mice. (a, b)

Induction of Shh-responsive reporter gene by Smo and Smo-M2. WT Smo
is inhibited by Ptc (a), whereas Smo-M2 is not (b). Smo activity is indicated
by the luciferase reporter gene. Data represent the mean s.d. of duplicate
Relative light units
determinations from one of three representative experiments. (c, d)

Schemata for transgenic mice. The region in which WT Smo and Smo-M2
were expressed in the transgenic mice is indicated in pink. Blue line indi-
cates the ventral midline. (e, f) Pax-3, which is normally expressed along the
length of the dorsal midbrain and hindbrain (e) is suppressed in the region
flanked by the red arrowheads in the Smo-M2 TG embryos (f). Anterior is
to the left. (gj) Ptc (g, h) and HNF3 (i, j), normally present at high levels
only in the ventral mid-/hindbrain of E12 WT embryos, are each ectopically
induced (marked by red arrowheads) in the dorsal mid-/hindbrain of Smo-
Ptc/Smo (g) Ptc/Smo (g) M2 transgenic embryos. D, dorsal; V, ventral brain. Scale bar, 1.0 mm (ei).
RESULTS
Smo-M2 alters cell pattern in the neural tube
We expressed cDNAs encoding wild-type (WT) Smo15 or a con-
stitutively active form (Trp 535 to Leu)22 of Smo (Smo-M2) in

transgenic mice under the En-2 promoter-enhancer30 (Fig. 1c and
d). These two receptor forms seemed to have similar signaling
properties but differed in their susceptibility to negative regula-
tion. Thus, in the absence of Ptc, both WT Smo and Smo-M2 were
capable of activating a Gli-BS luciferase reporter assay31 (Fig. 1a
and b). However, whereas Ptc suppressed the activity of WT Smo31
(Fig. 1a), Smo-M2 was refractory to this suppression (Fig. 1b).
Visual inspection of Smo-M2 embryos at days 12 and 14 (E12
and E14) revealed excessive overgrowth of dorsal neural tissue and
displacement of the skull bones in the midbrain and hindbrain
region (data not shown). A similar phenotype is reported for trans-
genic embryos that ectopically express Shh32, Shh-N5, HNF-3 (an
inducer of Shh)30 or the zinc-finger transcription factor Gli-1(a
candidate mediator of the Shh signal)5. Consistent with the aber-
rant overgrowth, the expression domain of the homeobox-con-
taining transcription factors En-1 and En-2, normally found in
regions of midbrain and hindbrain33,34, was expanded in these ani-
mals (data not shown). In contrast, transgenic mice that expressed
the WT Smo protein appeared normal (data not shown).
We next examined whether ectopic expression of Smo-M2
affected dorsal gene expression normally inhibited by Shh. Dor-
sally, WT embryos strongly expressed Pax-3 and Ephrin-A5 (refs.
35 and 36, respectively; Fig. 1e and data not shown). In Smo-M2
E12 and E14 embryos, these markers were not expressed in dor-
sal midbrain and hindbrain regions where Smo-M2 was overex-
pressed (Fig. 1f and data not shown). Downregulation of the
zinc-finger transcription factor Gli3, which takes place in response
to Shh signaling in the chick limb14 and in the mouse and frog
neural tube5,37, also occurred here (data not shown).
abdomen indirectly, through the induction of additional secreted Another important function of Shh in the developing neural
factors27. On the other hand, Hh directly specifies multiple types of tube is the induction of early ventral cell markers, such as Ptc,
cuticular structures and cell types in the Drosophila abdomen27 HNF3 4 and Shh itself. At E12, Ptc8,15,38,39, HNF3 and Shh32,4042,
and dorsal epidermis28. Moreover, Shh seems to have direct, long- normally expressed at high levels only in the ventral aspect of the
range actions on sclerotome differentiation24. neural tube, were ectopically induced in the dorsal midbrain/hind-
We found that Smo-M2 induced ventral markers and cell brain of the Smo-M2 embryos (Fig. 1gj and data not shown).
types requiring high concentrations of Shh29 that develop adja- Ectopic induction of these genes was maintained in the E14
cent to Shh-producing floor plate and notochord cells, as well Smo-M2 embryos (data not shown), indicating that dorsal prog-
as cell types induced by low concentrations of Shh4 that devel- enitors in the midbrain and hindbrain retained a ventral cell fate.
op at a distance from the Shh source. Each of the examined cell In contrast, no ectopic induction of ventral genes was detected in
types were induced cell autonomously in Smo-M2-expressing transgenic embryos expressing WT Smo (data not shown).
cells but not in adjacent or distant cells. These findings strong-
ly support the hypothesis that Smo-M2-expressing cells pat- Smo-M2 specifies ventral cell types
tern the ventral neural tube directly, and that Shh therefore, We next examined whether Smo-M2 could mimic the ability of
normally acts as a morphogen. Shh to specify multiple ventral cell types. Three classes of neu-

articles
ficial aspect of the ventral hindbrain and a site of lower expression

in the deep hindbrain (Fig. 2g). In Smo-M2-transgenic (TG)
embryos, netrin-1 was ectopically induced in the dorsal midbrain
and hindbrain, with the two regions of expression in the hind-
brain faithfully duplicated in mirror image (Fig. 2h).
TH
Thus, Smo-M2 reproduced multiple actions of Shh in the
developing nervous system, including induction of cell prolifer-
ation, suppression of dorsal markers, induction of ventral mark-
ers and formation of mature ventral neurons.
Mid-/hindbrain Smo-M2 activity is cell autonomous

Having established that Smo-M2 can reproduce multiple actions
5HT of Shh, we next used this receptor to examine whether Shh pat-
terns the neural tube directly or through a relay mechanism.
Examination of WT embryos revealed endogenous Smo expres-
sion along the length of the dorsal midbrain and hindbrain at a
consistent, low level (data not shown). In contrast, transgenic
Smo-M2 was intensely expressed in an irregular pattern (Fig. 3a).
The ectopic induction of ventral (for instance, HNF3, Shh and

Ptc ) and the suppression of dorsal (for example, Ephrin-A5)
islet-1 markers seemed restricted to Smo-M2-expressing cells (Fig. 3bd
and data not shown), indicating that Smo-M2, and therefore Shh,
patterned the neural tube directly, and not via a relay mechanism.
Because it was impossible to determine the precise boundary
of the Smo-M2 expression domain at the single-cell level in the
transgenic mice, it remained possible that cells at the border of
Smo-M2-expressing cells were patterned via a short acting or mem-
brane-bound relay signal. To address this question, stage-10 chick
embryos were electroporated with a single DNA construct to co-
netrin-1
express Smo-M2 and green fluorescent protein (GFP; via an inter-
nal ribosomal entry site; Fig. 3eg). The chicks were allowed to
develop for three days in ovo, and adjacent sections from the mid-
brain/hindbrain region were analyzed for ectopic expression of ven-
tral cell markers. Netrin-1 (Figs. 3h and 4f), Ptc (Fig. 3i) and HNF3
Fig. 2. Induction of multiple neuronal classes and netrin-1, in Smo-M2
TG mice. (ah) Sections of E14 WT (a, c, e and g) and Smo-M2 TG
(Figs. 3j and 4i) were each induced and co-expressed in ectopic
(b, d, f and h) embryos stained for TH (a, b) or 5HT (c, d), or probed locations. More importantly, ectopic expression of each of these
for islet-1 (e, f) or netrin-1 (g, h). TH immunoreactivity, 5HT immunore- ventral markers was observed only in Smo-M2/GFP-expressing cells.
activity, islet-1 and netrin-1 are all induced dorsally in the Smo-M2 TG We next studied whether Smo-M2 could cell-autonomously
mice. Blue arrowheads point to ventral expression and in the case of specify ventral midbrain/hindbrain (motor) neurons. Islet-1+
islet-1, normal ventral and dorsal expression. Red arrowheads point to motor neurons are thought to be induced in response to low con-
ectopic expression. Scale bars, 0.65 mm (a, b), 0.3 mm (c), 0.8 mm centrations of Shh in vivo at a distance from the Shh-producing
(d), 1.1 mm (e), 0.9 mm (f), 1.1 mm (g, h). floorplate and notochord4. Individual sections from Smo-M2/GFP-
expressing chick embryos were immunostained for Islet-1 (red flu-
orescence) and examined for co-expression of Smo-M2/GFP
(green fluorescence). In all cases, Islet-1 was induced exclusively
rons were examined, DA neurons5,12,13,43,44, located in the ventral in cells that co-expressed Smo-M2/GFP (Fig. 4ac). Individual
midbrain45, 5HT neurons, present in the ventral hindbrain5,11,44,46 cells expressed different combinations of markers; some expressed
and islet-1-expressing, putative motor neurons in the midbrain, Smo-M2/GFP, netrin-1 and Islet-1 (Fig. 4df; arrows), Smo-
hindbrain and spinal cord11,47,48. In addition, we tested for ectopic M2/GFP and Islet-1 but not netrin-1 (Fig. 4df; arrowheads) or
induction of the ventral marker and axon-guidance protein, Smo-M2/GFP and netrin-1 but not Islet-1. Cells co-expressing
Netrin-1 (refs. 9, 10), by Smo-M2. Islet-1 and HNF-3 were not observed (Fig. 4gi). Taken togeth-
In WT animals, tyrosine hydroxylase-positive DA neurons er, these findings show that Smo induces stereotypic combinations
were restricted to the ventral part of the midbrain (Fig. 2a). In of ventral markers in a cell-autonomous manner.
contrast, Smo-M2 embryos showed two regions of tyrosine
hydroxylase immunoreactivity, one ventral and the second in an Smo-M2 induces ventral cell types in the spinal cord
ectopic, dorsal midbrain location expressing Smo-M2 (Fig. 2b). The ventral markers and cell types examined in midbrain/hind-
Further, whereas WT embryos developed 5HT neurons only in brain normally reside close to the source of Shh. Thus, it
the ventral hindbrain (Fig. 2c), large numbers of ectopic 5HT remained possible that neurons, which develop in more dorsal
neurons were observed in the Smo-M2-transgenic dorsal hind- aspects of the ventral neural tube, were induced by relay signals.
brain (Fig. 2d). Islet-1 was also ectopically induced in the dorsal We therefore examined the consequences of overexpressing Smo-
midbrain, showing expanded expression in the dorsal mid- and M2 in the spinal cord where multiple, well defined cell types are
hindbrain of Smo-M2 animals (Fig. 2e and f). Examination of induced in distinct positions along the dorsoventral axis of the
netrin-1 in WT embryos revealed intense expression in the super- neural tube in response to different concentrations of Shh 4.

articles
Fig. 3. Changes in gene expression in the midbrain/hind-

brain region are confined to the Smo-M2 expression
domain. (ad) Photomicrographs of adjacent sections
from an E14 Smo-M2 transgenic embryo showing expres-
sion of Smo/Smo-M2 (a, b), HNF-3 (c) and Ephrin-A5 (d).
Note that the induction and inhibition of the Shh-depen-
dent genes seem restricted to the domain of Smo-M2-
expressing cells. (eg) Visualization of GFP signal in
electroporated chick embryos. (e) Chick embryos elec-
troporated with GFP show high expression of GFP mRNA
(pink) spanning the midbrain and hindbrain one day after
electroporation, as detected by in-situ hybridization. Blue
signal marks the midbrain/hindbrain boundary (MHB; in-
situ hybridization with fgf8). (f) High expression of Smo-
M2/GFP is retained in the midbrain and hindbrain three
days after electroporation. (g) Sagittal cryostat sections
show GFP expression in patches (arrows point to some)
along the length of the midbrain and hindbrain. Clustering
of GFP-positive cells was not seen following electropora-
tion of WT Smo/GFP or GFP alone or in animals injected
with five other genes, suggesting that these patches may

result from proliferative effects of Smo-M2. (hj) The
Shh-dependent genes netrin-1 (h), Ptc (i) and HNF3
(j) were induced in the dorsal hindbrain (black arrows)
within a region of Smo-M2/GFP-expressing cells (see
inserts; white arrows). These genes were always induced
in a subset of the Smo-M2/GFP-expressing cells. Scale
bars, 0.3 mm (a), 0.1 mm (bd), 0.12 mm (e), 1.0 mm
(f), 1.2 mm (g), 0.3 mm (hj).
Specifically, we followed the expression of Nkx2.2, Islet 1 and 2 DISCUSSION

and HB9 (expressed in different subtypes of motor neurons or By ectopically expressing Smo-M2 in the dorsal mid- and hindbrain
their progenitors), the Chx10 protein (a marker of the more dor- of transgenic mice and chicks, we demonstrated that this protein
sal V2 interneurons) and the homeodomain protein En-1 could mimic multiple actions of Shh. These include the induction
(expressed even further dorsal in V1 interneurons). These mark- of Shh-responsive genes Ptc, HNF3, Shh, Nkx2.2 and netrin-1, the
ers, characteristic of cell types lying close the source of Shh as suppression of dorsally restricted markers Pax-3, Ephrin-A5 and Gli-
well as those at a distance from the floor plate and notochord, all 3 and the induction of DA-, 5HT-, d-MN and v-MN motor neu-
showed cell-autonomous, ectopic induction in Smo-M2-express-
ing cells (Fig. 5). Interestingly, induction of the characteristical-
ly more dorsal markers Chx10 and En-1 was more robust in
Smo-M2-transgenic chick embryos, and expression of these
markers in cells that expressed lower amounts of Smo-M2/GFP
was more prevalent. Thus, overexpression of a single gene, Smo-
M2, resulted in cell-autonomous induction of multiple, distinct,
ventral markers characteristic of cells induced over a wide con-
centration range of Shh activity.
Fig. 4. Smo-M2 induces Islet-1+ motor neurons cell autonomously in the

mid/hindbrain region. (ac) Sagittal section through the chick dorsal
midbrain showing ectopic Islet-1 expression that is completely confined
to Smo-M2/GFP+ cells (ectopic Islet-1 expression was never observed in
Smo-M2/GFP negative cells in multiple sections, derived from 6 animals).
By confocal analysis, cytoplasmically localized GFP was seen surrounding
nuclear Islet-1 in a single cell (not shown). (df) Sequential photography
of a single section shows that the ectopic induction of Islet-1 and netrin-1
are both confined to Smo-M2/GFP+ cells. In some cases, Islet-1 colocal-
izes with netrin-1 (arrows), but in other cases Islet-1 is induced in the
absence of netrin-1 (arrowheads). Likewise, netrin-1 can be detected in
some cells that are Islet-1-negative. (gi) Sequential photography of a
single section for Smo-M2/GFP, Islet-1 and HNF3 shows that Islet-1
(arrows) and HNF3 (arrowheads) are both induced in subsets of Smo-
M2/GFP+ cells, but co-expression of Islet-1 and HNF3 is not seen. Scale
bar, 0.35 mm (ac), 0.4 mm (df), 0.5 mm (gi).

articles
Fig. 5. Smo-M2 induces ventral and ventrolateral markers and Ventral markers SmoM2/GFP Overlap
cell types in the spinal cord. (ao) Smo-M2 induces the marker
for ventral spinal cord and d-MN progenitor, Nkx2.2 (ac), the
v-MN motor neuron markers HB9 (df) and islet-2 (gi), and
the progressively more dorsal V1 and V2 interneuron markers
Chx10 (jl) and En-1 (mo). Each of these markers and cell
types is induced in a cell-autonomous manner in the electropo-
rated (right) but not control (left) side of the neural tube
(examples marked by arrowheads). Scale bar, ~0.5 mm (ao).
rons as well as V1 and V2 interneurons (for nomencla-

ture, see ref. 49). We further show that these patterning
activities are cell autonomous, occurring only in cells that
express Smo-M2. Together with the available genetic and
biochemical data, these findings support the ideas that
Smo is the key signaling component of the Shh receptor
and that Shh patterns the entire ventral aspect of the neur-
al tube directly as a morphogen, and not via relay signals.
Smo as a signaling coreceptor for Shh

Genetic studies in Drosophila as well as biochemical stud-
ies in vertebrates prompted the hypothesis that Smo is a sig-
naling subunit in a multi-component receptor for Shh.
However, biological evidence for this hypothesis in verte-
brates has been diffuse. Known activities of Smo included
its ability to activate a single reporter gene in culture and to
elicit basal cell carcinoma in the skin of transgenic mice, but
the relationship between Smo and the patterning activities
of Shh in the developing nervous system were not exam-
ined. Thus, the Shh signal in the neural tube could be medi-
ated by a Shh receptor other than Smo and/or indirectly by
an entirely distinct relay signal. Here we show that Smo can coordi- dent manner, in explant culture, strongly argues that Shh acts as
nate complex patterning signals involving the activation or suppression a morphogen to pattern the ventral aspect of the neural tube.
of at least 15 different genes including Ptc, HNF3, Shh, netrin-1, All the observed activities of Smo-M2 seemed cell
Nkx2.2, En1, Chx10, HB9, Islet-1 and 2, TH, tryptophan hydroxy- autonomous, even though Shh was induced and conceivably could
lase, Pax-3, Ephrin-A5 and Gli3. These findings demonstrate that Smo have acted on neighboring cells. One possible explanation for the
can transduce the full spectrum of Shh responses in the vertebrate failure of the induced Shh to act at a distance is that its induction
nervous system and can interpret and execute a graded Shh signal. occurred after the developmental period during which neural
Unlike Smo-M2, WT Smo did not elicit any Shh response in progenitors can change their fate. Alternatively, Shh may have
transgenic mice or in chick embryos. Similarly, whereas expres- failed to diffuse in the dorsal neural tube. Additionally, it is for-
sion of Smo-M2 in the skin of transgenic mice under the Keratin- mally possible that Shh does not normally diffuse but instead only
5 promoter leads to basal cell carcinoma-like skin lesions 22, patterns cells in which it is transiently expressed. In sum, our find-
expression of WT Smo under the same promoter produced no ings support the hypotheses that Smo is the key signaling co-
detectable phenotype (F. de S., A.R.; unpublished observations). receptor for Shh and that Shh is a morphogen in the neural tube.
These differences seem to stem from differential susceptibility to
negative regulation by Ptc31 (Fig. 1), the Shh coreceptor, which is METHODS
Transgenic mice. Complementary cDNAs encoding either the WT Smo
present15 and active19 throughout the mouse nervous system.
or Smo-M2 were placed under the control of the En-2 enhancer pro-
moter30,43, and the resulting constructs were injected into SJL/C57b/6
Smo and Shh pattern the neural tube directly zygotes. Embryos were harvested at either E12 or E14 (plug, D0) and
The availability of Smo-M2 allowed us to directly discern whether were tested for the transgene by PCR. Twenty-seven WT Smo and seven
Shh patterns the vertebrate neural tube as a morphogen or via Smo-M2 mutant animals were identified and analyzed.
additional patterning molecules. Whereas non-cell-autonomous
activities would indicate relay signals, cell-autonomous actions Histology and in-situ hybridization. Embryos were fixed in 4%
of Smo-M2 demonstrated that Shh patterned the neural tube paraformaldehyde overnight, cryoprotected in 30% sucrose in 0.12 M phos-
directly. We cannot exclude the possibility that ventral markers phate buffer, pH 7.4, embedded in the sagittal plane and stored frozen until
and neurons not examined here arise by non-cell-autonomous sectioning on a cryostat (14-M sections). Prepared sections were dried
(50oC, 30 min) and re-stored at 70oC until further use. Sections were sub-
actions of Smo. However, all examined changes in gene expres-
jected to in-situ hybridization performed with 500700 bp digoxygenin-
sion and cell fate, including induction of motor neurons and labeled probes, as described12,43. Immunohistochemistry was carried out as
interneurons that normally arise in response to low concentra- described43 using antisera to HNF-3 (provided by A. Ruiz i Altaba), rab-
tions of Shh and at a distance from its sources, were strikingly bit anti-Islet-1, 4G11 (anti-En-1, anti-HB9 and anti-CHX-10, provided by T.
restricted to cells expressing Smo-M2. This, combined with the Jessell and S. Brenner-Morton), 39.4D5 (mouse anti-Islet-1), 51.4H9 (mouse
induction of multiple cell types by Shh in a concentration-depen- anti-Islet-2 and anti-Nkx 2.2; purchased from the Developmental Studies

articles
Hybridoma Bank, University of Iowa), anti-rabbit TH (1:500; Chemicon, 16. Alcedo, J., Ayzenzon, M., Von Ohlen, T., Noll, M. & Hooper, J. E. The Drosophila
Temecula, California) and anti-5HT (Incstar, Stillwater, Minnesota). smoothened gene encodes a seven-pass membrane protein, a putative receptor
for the hedgehog signal. Cell 86, 221232 (1996).
17. van den Heuvel, M. & Ingham, P. W. smoothened encodes a receptor-like
Luciferase assay. The luciferase experiments were reported as serpentine protein required for hedgehog signalling. Nature 382, 547551 (1996).
described31,50. C3H10T1/2 cells were seeded in 15-cm plates and trans- 18. Ingham, P. W., Taylor, A. M. & Nakano, Y. Role of the Drosophila patched gene in
fected 24 h after plating with different ratios of gD-WT-Smo or gD-M2- positional signalling. Nature 353, 184187 (1991).
Smo to Ptc-GFP (normalized to 8 g of total DNA with control GFP 19. Goodrich, L. V., Milenkovic, K. M., Higgins, L. & Scott, M. P. Altered cell fates
and medulloblastoma in mouse patched mutants. Science 277, 11091113 (1997).
vector), 8 g of reporter plasmid, and 0.01 g Renilla Luciferase reference 20. Chen, Y. & Struhl, G. In vivo evidence that Patched and Smoothened constitute
plasmid (pRL-TK). Forty-eight hours after transfection, 1 106 cells were distinct binding and transducing components of a Hedgehog receptor complex.
labeled for FACS analysis with an anti-gD monoclonal antibody and Development 125, 49434948 (1998).
gD/GFP double-positive cells were sorted with an Epics Elite cell sorter 21. Johnson, R. L. et al. Human homolog of patched, a candidate gene for the basal
cell nevus syndrome. Science 272, 16681671 (1996).
(Coulter, Hialeah, Florida). Equal numbers of double-positive cells were 22. Xie, J. et al. Activating Smoothened mutations in sporadic basal-cell carcinoma.
then directly lysed and processed for luciferase activity31. Luciferase activ- Nature 391, 9092 (1998).
ity was measured with the Dual-Luciferase Reporter Assay System 23. Chen, Y. & Struhl, G. Dual roles for patched in sequestering and transducing
(Promega, Madison, Wisconsin), and values were normalized using Renil- Hedgehog. Cell 87, 553563 (1996).
la Luciferase activity. All the experiments were repeated at least three times. 24. Johnson, R. L., Grenier, J. K. & Scott, M. P. Patched overexpression alters wing disc
size and pattern: transcriptional and post-transcriptional effects on hedgehog
targets. Development 121, 41614170 (1995).
Electroporation. Fertilized E0 chicken eggs were incubated at 38C in a 25. Yang, Y. et al. Relationship between dose, distance and time in Sonic Hedgehog-
humidified incubator for 20 hours until the embryos reached HH stages mediated regulation of anteroposterior polarity in the chick limb. Development
810. A small window was made in the egg shell, and DNA (58 mg per 124, 43934404 (1997).
ml) encoding Smo-M2/GFP or WT-Smo/GFP with an internal riboso- 26. Perrimon, N. Hedgehog and beyond. Cell 80, 517520 (1995).
27. Struhl, G., Barbash, D. A. & Lawrence, P. A. Hedgehog acts by distinct gradient
mal entry site or GFP alone, was microinjected into the central canal of and signal relay mechanisms to organise cell type and cell polarity in the
the neural tube at the mid-/hindbrain or spinal cord levels. A pair of plat- Drosophila. Development 124, 21552165 (1997).
inum electrodes flanking the neural tube delivered 6 pulses (28 V; dura- 28. Heemskerk, J. & DiNardo, S. Drosophila hedgehog acts as a morphogen in
tion, 50 ms; interpulse interval, 100 ms). DNA was introduced into one cellular patterning. Cell 76, 449460 (1994).
29. Roelink, H. et al. Floor plate and motor neuron induction by different
side of the neural tube after the electric shock. The manipulated egg was concentrations of the amino-terminal cleavage product of Sonic hedgehog
then sealed with Scotch tape (3M), and the embryos were allowed to autoproteolysis. Cell 81, 445455 (1995).
develop for three days before harvesting and processing for immunohis- 30. Sasaki, H. & Hogan, B. L. M. HNF-3 as a regulator of floor plate development.
tochemistry and in-situ hybridization. For colocalization of GFP, in-situ Cell 76, 103115 (1994).
hybridization and antibody staining signals, photos of the GFP signal 31. Murone, M., Rosenthal, A. & de Sauvage, F. J. Sonic hedgehog signaling by the
patched-smoothened receptor complex. Curr. Biol. 9, 7684 (1999).
were obtained before antibody incubation and in-situ hybridization. 32. Echelard, Y. et al. Sonic hedgehog, a member of a family of putative signaling
molecules, is implicated in the regulation of CNS polarity. Cell 75, 14171430 (1993).
ACKNOWLEDGEMENTS 33. Davis, C. A. & Joyner, A. L. Expression patterns of the homeobox containing
We thank Kenji Shimamura for help with the electroporation, Susan Brenner-Morton genes En-1 and En-2 and the proto-oncogene int-1 diverge during mouse
development. Genes Dev. 2, 17361744 (1988).
and T. Jessell for the Islet-1, En-1, Chx-10 and HB9 antibodies and A. Ruiz i Altaba
34. Davis, C. A., Noble-Topham, S. E., Rossant, J. & Joyner, A. L. Expression patterns
for the HNF-3 antiserum. We also thank E. Berry and A. Bruce for help with prepa- of the homeobox containing gene en-2 delineates a specific region in the
ration of the manuscript and A. Shih and K. Poulsen for technical assistance. developing mouse brain. Genes Dev. 2, 361371 (1988).
35. Tremblay, P., Pituello, F. & Gruss, P. Inhibition of floor plate differentiation by
pax3: evidence from ectopic expression in transgenic mice. Development 122,
RECEIVED 16 SEPTEMBER; ACCEPTED 12 NOVEMBER 1999 25552567 (1996).
36. Winslow, J. W. et al. Cloning of AL-1, a ligand for an Eph-related tyrosine kinase
receptor involved in axon bundle formation. Neuron 14, 973981 (1995).
1. Belloni, E. et al. Identification of Sonic hedgehog as a candidate gene responsible 37. Ruiz i Altaba, A. Combinatorial Gli gene function in floor plate and neuronal
for holoprosencephaly. Nat. Genet. 14, 353356 (1996). inductions by Sonic hedgehog. Development 125, 22032212 (1998).
2. Roessler, E. et al. Mutations in the human Sonic Hedgehog gene cause holoprosen- 38. Marigo, V., Davey, R. A., Zuo, Y., Cunningham, J. M. & Tabin, C. J. Biochemical
cephaly. Nat. Genet. 14, 357360 (1996). evidence that Patched is the Hedgehog receptor. Nature 384, 176179 (1996).
3. Chiang, C. et al. Cyclopia and defective axial patterning in mice lacking Sonic 39. Marigo, V., Scott, M. P., Johnson, R. L., Goodrich, L. V. & Tabin, C. J.
hedgehog gene function. Nature 382, 407413 (1996). Conservation in hedgehog signaling: induction of a chicken patched homolog by
4. Tanabe, Y. & Jessell, T. M. Diversity and pattern in the developing spinal cord. Sonic hedgehog in the developing limb. Development 122, 12251233 (1996).
Science 274, 11151123 (1996). 40. Roelink, H. et al. Floor plate and motor neuron induction by vhh-1, a vertebrate
5. Hynes, M. et al. Control of cell pattern in the neural tube by the zinc finger homolog of hedgehog expressed by the notochord. Cell 76, 761775 (1994)
transcription factor and oncogene Gli-1. Neuron 19, 1526 (1997). 41. Krauss, S., Concordet, J.-P. & Ingham, P. W. A functionally conserved homolog of
6. Ericson, J. et al. Sonic hedgehog induces the differentiation of ventral forebrain the drosophila segment polarity gene hh is expressed in tissues with polarizing
neurons: a common signal for ventral patterning within the neural tube. Cell 81, activity in zebrafish embryos. Cell 75, 14311444 (1993).
747756 (1995). 42. Riddle, R. D., Johnson, R. L., Laufer, E. & Tabin, C. Sonic hedgehog mediates the
7. Ericson, J. et al. Pax6 controls progenitor cell identity and neuronal fate in polarizing activity of the limb. Cell 75, 14011416 (1993).
response to graded Shh signaling. Cell 90, 169180 (1997). 43. Hynes, M., Poulsen, K., Tessier-Lavigne, M. & Rosenthal, A. Control of neuronal
8. Goodrich, L. V., Johnson, R. L., Milenkovic, L., McMahon, J. A. & Scott, M. P. diversity by the floor plate: contact-mediated induction of midbrain
Conservation of the hedgehog/patched signaling pathway from flies to mice: dopaminergic neurons. Cell 80, 95101 (1995).
induction of a mouse patched gene by Hedgehog. Genes Dev. 10, 301312 (1996). 44. Ye, W., Shimamura, K., Rubenstein, J. L. R., Hynes, M. A. & Rosenthal, A. FGF8
9. Serafini, T. et al. The netrins define a family of axon outgrowth-promoting and Shh signals control dopaminergic and serotonergic cell fate in the anterior
proteins homologous to C. elegans UNC-6. Cell 78, 409424 (1994). neural plate. Cell 93, 755766 (1998).
10. Kennedy, T. E., Serafini, T., de la Torre, J. R. & Tessier-Lavigne, M. Netrins are 45. Specht, L. A., Pickel, V. M., Joh, T. H. & Reis, D. J. Light-microscopic
diffusible chemotropic factors for commissural axons in the embryonic spinal immunocytochemical localization of tyrosine hydroxylase in prenatal rat brain.
cord. Cell 78, 425435 (1994). II. Late ontogeny. J. Comp. Neurol. 199, 255276 (1981).
11. Yamada, T., Placzek, M., Tanaka, H., Dodd, J. & Jessell, T. M. Control of cell 46. Parent, A. in Serotonin Neurotransmission and Behavior (eds. Jacobs, B. L. &
pattern in the developing nervous system: polarizing activity of the floor plate Gelperin, A.) 334 (MIT Press, Cambridge, Massachusetts, 1981).
and notochord. Cell 64, 635647 (1991). 47. Pfaff, S. & Kintner, C. Neuronal diversification: development of motor neuron
12. Hynes, M. et al. Induction of midbrain dopaminergic neurons by Sonic subtypes. Curr. Opin. Neurobiol. 8, 2736 (1998).
Hedgehog. Neuron 80, 95101 (1995). 48. Tsuchida, T. et al. Topographic organization of embryonic motor neurons
13. Wang, M. Z. et al. Induction of dopaminergic neuron phenotype in the midbrain defined by expression of LIM homeobox genes. Cell 79, 957970 (1994).
by Sonic hedgehog protein. Nat. Med. 1, 11841188 (1995). 49. Thaler, J. et al. Active suppression of interneuron programs within developing
14. Marigo, V., Johnson, R. L., Vortkamp, A. & Tabin, C. J. Sonic hedgehog motor neurons revealed by analysis of homeodomain factor HB9. Neuron 23,
differentially regulates expression of Gli and Gli3 during limb development. Dev. 659674 (1999).
Biol. 180, 273283 (1996). 50. Sasaki, H., Hui, C.-C., Nakafuku, M. & Kondoh, H. A binding site of Gli proteins
15. Stone, D. M. et al. The tumour-suppressor gene patched encodes a candidate is essential for HNF-3 floor plate enhancer activity in transgenics and can
receptor for Sonic hedgehog. Nature 384, 129134 (1996). respond to Shh in vitro. Development 124, 13131322 (1997).

articles
A locus and mechanism of action

for associative morphine tolerance
Jennifer M. Mitchell1, Allan I. Basbaum1,2 and Howard L. Fields1,3
1 Department of Physiology, University of California, San Francisco, San Francisco, California 94143-0444, USA
2 Department of Anatomy, Box 0452, University of California, San Francisco, San Francisco, California 94143-0452, USA
3 Department of Neurology, Box 0453, University of California, San Francisco, San Francisco, California 94143-0453, USA
Correspondence should be addressed to H.L.F. (hlf@itsa.ucsf.edu)
Repeated administration of an opioid in the presence of specific environmental cues can induce
tolerance specific to that setting (associative tolerance). Prolonged or repeated administration of an
opioid without consistent contextual pairing yields non-associative tolerance. Here we demonstrate
that cholecystokinin acting at the cholecystokinin-B receptor is required for associative but not non-
associative morphine tolerance. Morphine given in the morphine-associated context increased Fos-
like immunoreactivity in the lateral amygdala and hippocampal area CA1. Microinjection of the
cholecystokinin B antagonist L-365,260 into the amygdala blocked associative tolerance. These
results indicate that cholecystokinin acting in the amygdala is necessary for associative tolerance to
morphines analgesic effect.
The role of cholecystokinin (CCK) as an anti-opioid peptide is erance (group A), we used a balanced crossover design in which all
firmly established. CCK suppresses the analgesic effect of mor- rats received i.v. morphine in one of two distinct environments
phine and other -opioid receptor agonists1,2. CCK antagonists and, on alternate days, i.v. saline in the other environment. After
potentiate the analgesic effects of exogenous and endogenous several administrations, morphine was ineffective when given in
-opioid receptor agonists38 and enhance the inhibitory effect the environment in which the rat had previously received mor-
of morphine on nociceptive dorsal horn neurons9,10. Addition- phine (Fig. 1a, black circles). However, the same dose of mor-
ally, CCK antagonists impede the acquisition of morphine toler- phine was fully effective when administered in the context
ance yet have no effect on morphine dependence or previously paired with saline (Fig. 1b). Thus, using this behav-
withdrawal3,1116. Although CCK antagonists potentiate the effects ioral protocol and method of drug administration, the analgesic
of opioids, they do not alter nociceptive threshold levels when tolerance to morphine could be accounted for by learning factors
administered independently8,16. There are two distinct CCK and therefore was associative. To produce non-associative toler-
receptors (CCK-A and CCK-B) in the CNS1720. ance (group NA), morphine pellets were implanted in rats that
Learning can contribute to drug tolerance2127. When mor- then received i.v. saline in either environment used to condition
phine administration is paired with specific environmental cues, rats in the associative tolerance protocol. Group NA animals were
these cues can act as conditioned stimuli that elicit a conditioned tolerant to morphine in all experimental environments (data not
response opposing the agonist effect of morphine (associative shown), indicating that the morphine tolerance observed in these
tolerance). Although associative analgesic tolerance following animals was largely independent of context (non-associative).
repeated opioid administration is well established, the underlying
neurobiological mechanisms are unknown. RESULTS
The -opioid receptor is necessary and sufficient for mor- Systemic administration of the CCK-B antagonist L-365,260,
phine analgesia28. Animals that become non-associatively tolerant but not the CCK-A antagonist MK-329, blocked the expression
to morphine are selectively tolerant to other -opioid receptor of associative tolerance (Fig. 1c and d). Neither L-365,260 nor
agonists. In contrast, associative morphine tolerance generalizes MK-329 affected the expression of non-associative tolerance
such that the effects of agonists acting at other opioid receptors (Fig. 1e and f).
are also reduced29. This suggests that these two types of morphine Although the same dose of morphine and same baseline tail-
tolerance involve different CNS circuitry. However, direct evi- flick latency were used to assess analgesic tolerance, it is possible
dence for such circuitry differences has yet to be demonstrated. that the non-associative tolerance was of greater magnitude and
Because of its role as an anti-opioid peptide, we tested the thus obscured a possible morphine-potentiating effect of
hypothesis that CCK is required for associative but not non-asso- L-365,260. To address this possibility, we studied a second group
ciative morphine tolerance. In addition, we investigated poten- of non-associatively tolerant rats following three instead of five
tial differences in the CNS circuitry underlying these two types days of morphine pelleting. These rats were only partially toler-
of morphine tolerance. ant to morphine, so any potentiating effect of L-365,260 on mor-
CCK-A and CCK-B receptor antagonists were administered phine analgesia should have been easy to demonstrate. However,
i.v. following the acquisition of either associative or non-associa- L-365,260 had no effect on morphine analgesia in these partial-
tive analgesic tolerance to morphine. To produce associative tol- ly but non-associatively morphine-tolerant rats (Fig. 1g and h).

articles
Fig. 1. Associative tolerance is

reversed by CCK-B receptor antago- aA b
B
nism. (a) Acquisition of associative tol- 12 Morphine (morphine-paired context) 12 **
erance (n = 24) with administration of Saline (saline-paired context)
saline on even-numbered days and
Tail-flick latency (s)

10 10
morphine (2.5 mg per kg) on odd-num-

bered days in their associated environ- 8 8
ments. (b) Associative tolerance is
context dependent. Tail-flick latency 6 6
(n = 19) in morphine-paired and saline-
paired environments before (Baseline) 4
4
and after (Morphine) the acquisition of
associative tolerance. Baseline tail-flick
2 2
latencies were measured on experi- 3 4 5 6 7 8 9 10 11 12 13 14 Baseline Morphine Baseline Morphine
mental days 1 and 2, respectively. Training session
Animals were tolerant to morphine Morphine-paired Saline-paired
context context
when tested in the morphine-paired
environment by the sixth administration (day 13) but c
C Associative d
D Associative
showed no evidence of analgesic tolerance to the same
12 12
dose of morphine given in the saline-paired environment **
(**p < 0.01). (c, d) Associative tolerance is reversed by 10 10

the systemic administration of the CCK-B antagonist
L-365,260 (0.2 mg per kg, n = ll, p < 0.01) but not by the 8 8
CCK-A antagonist MK-329 (0.1 mg per kg, n = 13).
6 6
(e, f) Non-associative tolerance is not reversed by the sys-
temic administration of L-365,260 (0.2 mg per kg, n = 8) 4 4
or MK-329 (0.1 mg per kg, n = 4). Following the full devel-
opment of non-associative tolerance, L-365,260 and MK- 2 2
329 were administered 7 min before systemic morphine. Morphine L-365,260 + Morphine MK-329 +
morphine morphine
(gj) Neither pre- nor post-treatment with CCK antago-
nists reversed partial non-associative tolerance. To avoid a e
E Non-associative f
F Non-associative
ceiling effect, two groups of animals were implanted with
12 12
subcutaneous morphine pellets every day for three
instead of five days. This produced partial analgesic toler- 10 10
ance, which was unaffected by either L-365,260 (0.2 mg
per kg, n = 13) or MK-329 (0.1 mg per kg, n = 13). (g, h) 8 8
L-365,260 given 45 min after an additional challenge dose
6 6
of morphine (2.5 mg per kg, i.v., n = 6, gray bar) does not
potentiate morphine in partially tolerant rats (black bar, 4 4
g). As a fall in morphine concentration could account for
this failure, we controlled for this possibility by adminis- 2 2
L-365,260 + MK-329 +
tering L-365,260 7 min before systemic morphine adminis- Morphine Morphine
morphine morphine
tration (2.5 mg per kg, i.v., n = 7; h). No potentiation was
observed (black bar in h not significantly different from
gray bar in g). (i, j) MK-329 had no significant effect on G
g Non-associative
h
H Non-associative
10 10
analgesia in partially tolerant rats, whether administered
45 min after (i; n = 6) or 7 min before (j; n = 7) a challenge 9 9
dose of morphine (2.5 mg per kg, i.v.). 8 8
7 7
6 6
5 5
These data strongly support the specificity of the 4 4
CCK-B antagonist in the reversal of associative mor- 3 3

phine tolerance. 2 2
After establishing the effect of the systemically Baseline Challenge L-365,260 Baseline L-365,260 +
challenge
administered CCK-B antagonist L-365,260 on asso-
ciative tolerance, we investigated the potential CNS
I i Non-associative Jj Non-associative
sites of its action using Fos immunohistochemistry. 10 10
Animals were again divided into two groups, asso- 9 9
ciative (A) and non-associative (NA). Following the 8 8

acquisition of tolerance, animals trained in the asso- 7 7
ciative contexts were given either i.v. morphine in the 6 6
morphine-paired environment or saline in the saline- 5 5
paired environment. Non-associatively tolerant (pel- 4 4
let-implanted) rats were given either morphine in the 3 3
environment in which associatively tolerant rats were 2 2
given morphine or i.v. saline in the context in which Baseline Challenge MK-329 Baseline MK-329 +
challenge
the associatively tolerant rats had received saline.

articles
Fig. 2. Mean number of Fos-positive cells

NAcc Core NAcc Shell per section (50 m) per brain region for
Number Fos-positive cells

240 300 associatively and non-associatively tolerant
animals (n = 4 animals per group).
180 225 Increases in Fos expression are observed
in amygdala and hippocampus during
120 150 expression of associative morphine toler-
ance. NAcc, nucleus accumbens; VTA, ven-
60 75 tral tegmental area; CAl, hippocampal area
CAl; CeA, central amygdala; BLA, basolat-
eral amygdala; LA, lateral amygdala.
AS AM NAM NAS AS AM NAM NAS (*p < 0.05; **p < 0.01).
VTA CA1
p < 0.05, respectively). In contrast, no
Number Fos-positive cells 420

** significant differences were found
30
among these four groups for the other
22 315 brain regions examined (VTA, NAcc
*
core, NAcc shell and Ce amygdala;
15 210
Fig. 2). These results indicate that
7 105 activity of neurons in both lateral and
basolateral amygdala and in CA1 of the
hippocampus was elevated in associa-
AS AM NAM NAS AS AM NAM NAS
tively tolerant animals receiving mor-
phine in a morphine-associated
context as compared with activity
CeA BLA under conditions of non-associative
240 tolerance. Region CA1 of the hip-

120 * pocampus was also activated in asso-
180 ciatively tolerant rats exposed to saline
90
in the saline-paired context.
120 Two additional control groups of
60
associatively tolerant animals were
30 60
examined to determine the specificity
of the increase in amygdala fos activa-
AS AM NAM NAS AS AM NAM NAS tion. One control group received saline
in the morphine-associated context,
and the other group received morphine
LA in the saline-associated context. When
these two groups were included in the
120
overall statistical analysis, there was a
** significant main effect of treatment, but
90
only within the lateral amygdala
60 (p < 0.0001). Post-hoc tests revealed sig-
nificant differences between animals
30 receiving morphine in the morphine
context and those receiving either saline
in the morphine context (p < 0.01) or
AS AM NAM NAS
morphine in the saline context
Associative Non-associative (p < 0.01). Specifically, when given
saline in the morphine context, animals
had 12.6 12.2 (mean s.e.) cells in
Associatively tolerant rats receiving morphine in the mor- the lateral amygdala, whereas those given morphine in the saline
phine-paired environment had significantly more Fos-positive context had 7.5 4.9 cells. In contrast, rats given morphine in the
cells in the lateral amygdala (LA) than did animals in the three morphine context had 76.4 24.6 cells in the lateral amygdala.
other groups (Figs. 2 and 3; p < 0.01). These same rats also had a Thus, the appropriate morphinecontext pairing was necessary
significantly greater number of Fos-positive cells in the basolat- for fos activation within the lateral amygdala, and the activation
eral amygdala than either group of non-associatively tolerant ani- was not a nonspecific effect of morphine administration.
mals (p < 0.05). Associatively tolerant animals receiving either To determine if the reversal of associative tolerance fol-
morphine in the morphine-paired context (AM) or saline in the lowing the systemic administration of the CCK-B antagonist
saline-paired context (AS) had significantly more Fos-positive L-365,260 was due to blockade of an action of CCK in either
cells in area CA1 than did either group of non-associatively tol- the amygdala or hippocampus, L-365,260 was microinjected
erant animals (NAM and NAS groups; p < 0.01 and into L/BL amygdala and CA1 following the acquisition of asso-

articles
ciative tolerance. Injections were also made

into Ce amygdala, VTA and NAcc core
because these nuclei also express the CCK-
B receptor18,20,30.
Microinjection of L-365,260 into L/BL
and Ce amygdala significantly attenuated
associative morphine tolerance (p < 0.01 for
both nuclei; Fig. 4). However, no significant
change in associative morphine tolerance
was apparent following the microinjection
of L-365,260 into CA1, VTA or NAcc core.
The effect of L-365,260 in L/BL amygdala
was not significantly different from that in
Ce amygdala and, importantly, was also not
significantly different from the effect of sys-
temic L-365,260. In contrast, the effect of L-
365,260 in Ce amygdala was significantly
different from the systemic effect (p < 0.05).
This indicates that CCK in L/BL amygdala

was required for the expression of associa-
tive morphine tolerance and suggests that
the attenuation of associative morphine tol-
erance by systemic administration of L-
365,260 was through an action in L/BL Fig. 3. Photomicrographs of Fos-positive cells in the amygdala of associatively and non-associa-
amygdala. Given the proximity of the L/BL tively tolerant animals. AS, associatively tolerant animals administered saline in the saline-paired
amygdala to the Ce amygdala and the data environment on experimental day 15; AM, associatively tolerant animals administered morphine
from our Fos study, it seems most likely that in the morphine-paired environment on experimental day 15; NAS, non-associatively tolerant
the effect of L-365,260 in Ce amygdala is due animals administered saline on experimental day 6; NAM, non-associatively tolerant animals
to diffusion of the drug into L/BL amygdala. administered morphine on experimental day 6.
Taken together, the evidence indicates that
neurons in the L/BL amygdala are activated when morphine is compensatory action does not occur in non-associative tolerance
given in the morphine-associated context and that CCK acting because the required temporal relationship between contextual
in this brain region is necessary for the expression of associa- stimuli and drug administration has not been established through
tive morphine tolerance. conditioning. Our results raise the possibility that the contextu-
To confirm that L-365,260 alone did not have an analgesic al cues that signal impending drug administration are necessary
effect, control injections of L-365,260 were made into L/BL for the release of CCK in the L/BL amygdala. Another possibili-
amygdala in the saline-paired environment. Whereas the admin- ty is that morphine elicits CCK release in the L/BL amygdala in
istration of L-365,260 into the L/BL amygdala in the morphine- associatively tolerant rats but that this does not occur in non-
paired environment significantly attenuated morphine tolerance associatively tolerant animals.
(Fig. 4), injections of L-365,260 into the L/BL amygdala in the The contribution of the hippocampus to learning and mem-
unpaired environment had no effect (Fig. 4b). This indicated that ory is clearly established4244. Furthermore, CCK seems to be
L-365,260 in the L/BL amygdala did not have an antinociceptive involved in the acquisition and retrieval of emotionally based
effect independent of morphine administration. memory through an action in the hippocampus45. Although we
did not establish a contribution of hippocampal CCK, the Fos
DISCUSSION analysis suggests that CA1 neurons contribute to the expression
The amygdala is implicated in the encoding of affective memo- of associative morphine tolerance. The hippocampus seems to
ries3133 and in the expression of stimulusreward associations34,35. serve as a locus for memory consolidation and to perform the
Additionally, neurons in the amygdala change their pattern of matchmismatch discriminations necessary for memory recog-
firing in response to emotionally significant stimuli36,37. Our cur- nition 46. Moreover, there are direct, reciprocal connections
rent results demonstrate that the L/BL amygdala is necessary for between the hippocampus and L/BL amygdala47,48. We suggest
the expression of the compensatory response that results from that the projection from CA1 to the L/BL amygdala contributes to
the association between particular environmental cues and the associative tolerance.
effect of morphine. Neurons in the hippocampus and amygdala are postulated to
When injected into the amygdala, opioid antagonists enhance participate in a mechanism of learning and memory necessary
memory retention38, and the intra-amygdaloid administration for processing and assigning value to drug-associated cues, which
of CCK agonists facilitates learning and memory39,40. Further- contributes to drug addiction49. It is intriguing that this circuit-
more, CCK mRNA expression increases in the amygdala follow- ry overlaps with that involved in associative tolerance. We hypoth-
ing repeated morphine administration, suggesting that the esize that during the acquisition phase of associative tolerance,
amygdala is a critical locus for a CCK-mediated compensatory CA1 is important for the matchmismatch discrimination that
mechanism that contributes to opioid tolerance 41. We have occurs when an animal is exposed to the morphine-associated
demonstrated that CCK release in the amygdala is required for environment. Therefore, we predict that acquisition of associa-
a contextually triggered compensatory action that maintains pain tive tolerance will be impaired when hippocampal connections
sensitivity in the presence of morphine. We propose that this are disrupted.

articles
METHODS
a
A Subjects. Male Sprague-Dawley rats (Charles River, Wilmington, Mass-
achusetts) weighing 300325 g were individually housed. The colony
12
room was kept constant at 21C and followed a standard 12-h light/dark
**
** cycle with food and water available ad libitum. Animals were tested at
10
the same time during their light cycle each day.
8
*
Drugs. Morphine sulfate powder (supplied by the National Institute on
6 Drug Abuse) was dissolved in physiological saline for micro- and i.v.
injections. Subcutaneous morphine pellets (75 mg morphine base) were
4
obtained from the National Institute on Drug Abuse. MK-329 was
2
obtained from Merck Pharmaceuticals (Terlings Park, UK) and was dis-
solved in distilled water. The CCK-B antagonist L-365,260 (a gift from
0 Les Iversen, Panos Pharmaceuticals) was dissolved in 90% ethanol.
SYST L/BL Ce A VTA Core CA1 Surgery. Animals were anesthetized with 50 mg per kg Nembutal,
implanted with jugular catheters (PE 50), and allowed a one week recov-
ery period. Group A animals in the microinjection study were implant-
ed bilaterally with stainless steel 26 gauge chronic guide cannula (Plastics
One, Roanoke, Virginia). Group NA animals were anesthetized with

B
b halothane (14%) and administered 2 subcutaneous morphine pellets
(75 mg morphine base, NIDA) every 24 h for 4 days following the recov-
12.0
ery period.
**
9.0 Behavioral apparatus. Tail-flick latency was measured using a radiant

heat source below the glass floor of the experimental chamber. The source
was positioned beneath the rats tail. A photocell was affixed to the radi-
6.0 ant heat source so that movement of the tail terminated the trial. A cut-
off time of 12 s was used to avoid tissue damage. The behavioral chambers
measured 15 15 15 inches and consisted of either white or black Plex-
3.0
iglas walls and a Plexiglas ceiling.
0.0 Behavioral protocol. Group A (associative tolerance) rats were exposed

to two distinct environments on alternating days for 16 days. Environ-
Morphine Saline ments differed in size, color, smell, tactile stimulation and lighting. On the
context context
first day in each environment, rats were administered 0.05 ml physio-
logical saline i.v. before being placed in the experimental chamber. Once
inside, animals were allowed a 15-min habituation period. Tail-flick laten-
Fig. 4. Microinjection of the CCK-B receptor antagonist L-365,260 cy was then measured every 10 min for 60 min. On alternate days, ani-
into CNS sites. Abbreviations as in Fig. 2. (a) Comparison of tail- mals were given 2.5 mg per kg morphine i.v. in 0.05 ml physiological
flick latencies following systemic administration (SYST; n = 11) or saline in one environment or 0.05 ml physiological saline in the other
microinjections of L-365,260 into the L/BLA (n = 12), CeA (n = 7), environment. Following i.v. administrations, animals were allowed a 15-
VTA (n = 7), NAcc core (n = 10) or CA1 (n = 12). White bars, base- min habituation period; subsequently tail-flick latency was measured as
line in the morphine-paired context on experimental day 1 or 2; before.
black bars, L-365,260 administered 15 min before morphine on
experimental day 15 in the morphine-paired context (**p < 0.0l, Fos study. On the last experimental day (day 15), animals were divided
*p 0.05). (b) Comparison of tail-flick latencies following the into four groups of four animals: saline in the saline- or morphine-asso-
administration of L-365,260 or vehicle in either the morphine- ciated context and morphine in the morphine or saline context. Ninety
paired or saline paired-context. White bars, baseline in the saline- min after drug administration on the last experimental day, all animals
paired and morphine-paired contexts; black bars, L-365,260 were perfused, and their brains were removed and processed for Fos.
administered either 15 min before morphine in the morphine-paired
context on experimental day 15 or 15 min before saline in the Systemic drug administration. On experimental day 15 or 16, animals
saline-paired context on experimental day 17; gray bar, vehicle (90% received either L-365,260 (0.2 mg per kg, i.v.) or MK-329 (0.1 mg per
ethanol) administered 15 min before saline in the saline-paired con- kg, i.v.) in their home cages 7 min before morphine administration. Fol-
text on experimental day 17. lowing morphine administration, animals were allowed to habituate to
the experimental environment for 15 min before tail-flick latency was
measured. On experimental day 15 or 16, morphine was administered
in the saline-paired environment to confirm that the tolerance that devel-
oped was associative. Experimental environments and testing days were
In summary, we demonstrated that the CCK-B receptor counterbalanced. Thus, animals that received L-365,260 or MK-329 plus
antagonist L-365,260 reverses associative but not non-associa- morphine in the morphine context on day 15 received morphine alone in
tive morphine tolerance. Additionally, our findings indicate that the saline-paired environment on day 16 and vice versa.
associative and non-associative tolerance involve different pat-
Microinjection study. The injection and testing protocol was the same
terns of neuronal activity in the L/BL amygdala and in CA1 of as for the systemic drug administration, but only L-365,260 (15 ng in
the hippocampus. Furthermore, we demonstrated that the 0.5 l per brain site) was used. Animals that received microinjections
learned compensatory mechanism contributing to associative into the L/BL amygdala participated in one additional experimental day.
morphine tolerance requires an action of CCK at the CCK-B On day 17, these animals were divided randomly into two groups. Half of
receptor in the L/BL amygdala. the animals were microinjected with L-365,260 (15 ng in 0.5 l per brain

articles
site) in the saline-paired environment while the other half received vol- and 0.3% TritonX before being incubated for 1 h at room temperature
ume-matched microinjections of vehicle (90% ethanol) in the saline in a blocking solution of 0.05 M Tris-phosphate-buffered saline with 3%
paired environment. Tail-flick latency was then measured as before. normal goat serum and 0.3% TritonX. Primary antibody (rabbit poly-
clonal, provided by Dennis Slamon, University of California, Los Ange-
Group NA (non-associative tolerance). On the first experimental day, les) was then applied to the tissue. Tissue was incubated overnight at
animals received 0.05 ml saline i.v. followed by a 0.2-ml heparinized saline room temperature. Sections were thoroughly washed before the appli-
flush. Animals were allowed a 15-min habituation period in one of the cation of secondary antibody (goat anti-rabbit IgG and avidin-biotin-
two experimental environments used to condition rats in the associative peroxidase complex, Vector Labs, Burlingame, California). Secondary
tolerance study. Tail-flick latency was then measured every 10 min for antibody was left on the tissue for a period of 2 h. A nickel/diaminoben-
60 min. Animals were then returned to their home cages. Sixty min later, zidine (DAB) reaction was used to visualize Fos immunoreactivity50. Tis-
animals were implanted with two subcutaneous morphine pellets and sue was processed simultaneously for Group A and NA animals.
returned to their home cages. Thus all non-associatively tolerant animals
were exposed to high levels of morphine in their home cages as well as Fos quantification. Sections were examined under dark-field illumination
in the two contexts used for saline and morphine conditioning (for the to identify the boundaries of each brain region of interest. To simplify quan-
associatively tolerant rats). titation, regional boundaries were defined (NAcc, AP 2.2 to AP 0.48; VTA,
AP 5.2 to AP 6.8; CA1, AP 2.3 to AP 4.16; CeA, AP 1.8 to AP 3.0;
Systemic study. Animals were pellet-implanted as above. On the last BLA, AP 1.8 to AP 3.3; LA, AP 2.5 to AP 3.3). Fos-positive cells were
experimental day, animals were divided into 2 groups. Seven min before quantified at 20 magnification using light-field illumination. An observ-
i.v. administration of 2.5 mg per kg morphine, group 1 received either er blind to the experimental treatment counted Fos-positive cells. Once
MK-329 or L-365,260. Forty-five min following the i.v. administration quantitation was completed, the total number of cells for each animal in
of 2.5 mg per kg morphine, group 2 received either MK-329 or L-365,260. each brain region was divided by the number of section per region, ren-
Following morphine administration, group 1 animals were placed in one dering a mean number of Fos cells for each animal for each brain nucleus.
of the experimental environments and allowed a 15-min habituation
period. Tail-flick latency was measured every 10 min for 60 min. Fol- Statistical analysis: Fos study. A one-way randomized ANOVA was used
lowing morphine administration, group 2 animals were placed in one of to analyze the data collected from each brain region. Tukey-Kramer post-
the experimental environments and allowed a 15-min habituation peri- hoc tests were conducted for pairwise comparisons.
od before tail-flick latency was measured every 5 min for 30 min. Group
2 animals were then removed from the experimental environment, Statistical analysis: systemic and microinjection studies. The mean of
administered either MK-329 or L-365,260 and then placed back in the the six tail-flick latencies measured for each animal was calculated for
experimental environment. Animals were allowed a 15-min habituation each test day. As the data did not follow a standard distribution (due to
period. Tail-flick latency was measured every 5 min for 30 min. Care was the use of a 12-s cut-off) non-parametric statistics were used for data
taken to treat NA animals in such a way as to prevent the association of analysis. Wilcoxon two-group signed-rank tests were used to compare
morphine administration with any environmental cues. the mean tail-flick latencies on the different test days. Statistical signifi-
cance was set at p < 0.05. All statistical tests were conducted using GB-
Microinjections. All injection sites were determined based on the atlas STAT, version 6.5. All figures were created using StatView, version 4.0,
of Paxinos and Watson (1986). Bilateral microinjections were made into (SAS Institute, Cory, North Carolina) and Adobe Illustrator, version 5.5.
the following brain structures: VTA (n = 7; AP, 5.8; ML, 0.5; DV, 8.5),
NAcc core (n = 10; AP, 2.5; ML, 1.2; DV, 6.5), CA1 (n = 12; AP, 3.6;
ML, 1.5; DV, 3.0), CeA (n = 7, AP, 2.5; ML, 4.3; DV, 8.3), L/BLA ACKNOWLEDGEMENTS
(n = 12; AP, 2.8; ML, 5.2; DV, 8.5). Each injection was made using a We wish to thank our colleagues Mary Dallman, Michael Silver and Larry Tecott
1 l syringe (Hamilton, Reno, Nevada) attached to 10 cm of PE 50 tub- for reading this manuscript. This work was supported by NINDS grant # 21445,
ing connected to a 33-gauge injection cannula (Plastics One). Microin- NIDA grant # 01949, the Wheeler Center for the Neurobiology of Addiction and
jection was conducted at a rate of 0.5 l per min. The injection cannula
by an NSF student fellowship.
extended 2 mm past the guide cannula and was left in place for 1 min
following microinjection to minimize the backflow of drug solution.
At the conclusion of the experiment, animals were anesthetized with RECEIVED 27 SEPTEMBER; ACCEPTED 3 NOVEMBER 1999
pentobarbital, injected with 0.5 l pontamine sky blue per brain site and
perfused intracardially through the ascending aorta. Brains were blocked
and sectioned coronally at 50 m, mounted and stained with neutral red. 1. Faris, P. L., Komisaruk, B. R., Watkins, L. R. & Mayer, D. J. Evidence for the
neuropeptide cholecystokinin as an antagonist of opiate analgesia. Science
219, 310312 (1983).
Fos study. On days 2 and 4, animals were allowed a 75-min habituation 2. Magnuson, D. S., Sullivan, A. F., Simonnet, G., Roques, B. P. & Dickenson, A.
period in one of the 2 experimental environments, whereas the other H. Differential interactions of cholecystokinin and FLFQPQRF-NH2 with
context was used on days 3 and 5. Animals were then returned to their mu and delta opioid antinociception in the rat spinal cord. Neuropeptides 16,
213218 (1990).
home cages for 60 min, after which animals were anesthetized and admin- 3. Watkins L. R., Kinscheck, I. B. & Mayer, D. J. Potentiation of opiate analgesia
istered 2 subcutaneous morphine pellets as before. On the last experi- and apparent reversal of morphine tolerance by proglumide. Science 224,
mental day (day 6), half (n = 4) of the animals received morphine in the 395396 (1984).
context in which associative animals (group A) had received saline, and 4. Katsuura, G. & Itoh, S. Potentiation of -endorphin effects by proglumide in
half received morphine in the context in which associative animals had rats. Eur. J. Pharmacol. 107, 363366 (1985).
5. Dourish, C. T. et al. The selective CCK-B receptor antagonist L365,260
received morphine. Ninety min after drug administration on the last enhances morphine analgesia and prevents morphine tolerance in the rat.
experimental day, all animals were perfused, and their brains were Eur. J. Pharmacol. 176, 3544 (1990).
removed and processed for Fos. Care was taken to prevent the associa- 6. Zhou, Y., Sun, Y. H., Zhang, Z. W. & Han, J. S. Increased release of
tion of morphine administration with any environmental cues. immunoreactive cholecystokinin octapeptide by morphine and potentiation
of -opioid analgesia by CCKB receptor antagonist L-365,260 in the rat
spinal cord. Eur. J. Pharmacol. 234, 147154 (1993).
Immunohistochemistry. Animals were perfused with 150 ml 0.1 M phos- 7. Valverde, O., Maldonado, R., Fournie-Zaluski, M. C. & Roques, B. P.
phate-buffered saline followed by 400 ml of 4% formaldehyde in 0.1 M Cholecystokinin B antagonists strongly potentiate antinociception mediated
phosphate buffer. Brains were then removed and post-fixed for 2 h at by endogenous enkephalins. J. Pharmacol. Exp. Ther. 270, 7788 (1994).
8. Vanderah, T. W., Bernstein, R. N., Yamamura, H. I., Hruby, V. J. & Porecca, F.
room temperature. Tissue was transferred to a 30% sucrose solution and Enhancement of morphine antinociception by a CCKB antagonist in mice is
left overnight. Tissue was cut in 50-m sections and washed in a solu- mediated via opioid delta receptors. J. Pharmacol. Exp. Ther. 278, 212219
tion of 0.05 M tris-phosphate-buffered saline with 1% normal goat serum (1996).

articles
9. Chapman, V., Honore, P., Buritova, J. & Besson, J. M. Cholecystokinin B 30. Reum, T., Schafer, U., Marsden, C. A., Fink, H. & Morgenstern, R. CCK
receptor antagonism enhances the ability of a low dose of morphine to reduce increases extracellular dopamine overflow in the anterior nucleus accumbens
c-fos expression in the spinal cord of the rat. Neuroscience 67, 731739 (1995). via CCK(B) receptors in the VTA assessed by in vivo voltammetry.
10. Kellstein, D. E., Price, D. D. & Mayer, D. J. Cholecystokinin and its antagonist Neuropeptides 31, 8288 (1997).
lorglumide respectively attenuate and facilitate morphine-induced inhibition 31. Le Doux, J. E. in Mind and Brain: Dialogues in Cognitive Neuroscience (eds.
of C-fiber evoked discharges of dorsal horn nociceptive neurons. Brain Res. LeDoux, J. E. & Hirst, W.) 301354 (Cambridge Univ. Press, Cambridge, UK,
540, 302306 (1991). 1986).
11. Tang, J., Chou, J., Iadarola, M., Yang, H. Y. T. & Costa, E. Proglumide prevents 32. LeDoux, J. E., Iwata, J., Cicchetti, P. & Reis, D. J. Different projections of the
and curtails acute tolerance to morphine in rats. Neuropharmacology 23, central amygdaloid nucleus mediate autonomic and behavioral correlates of
715718 (1984). conditioned fear. J. Neurosci. 8, 25172529 (1988).
12. Dourish, C. T., Hawley, D. & Iversen, S. D. Enhancement of morphine 33. McGaugh, J. L. & Cahill, L. Interaction of neuromodulatory systems in
analgesia and prevention of morphine tolerance in the rat by the modulating memory storage. Behav. Brain Res. 83, 3138 (1997).
cholecystokinin antagonist L-364,718. Eur. J. Pharmacol. 147, 469472 34. Jones, B. & Mishkin, M. Limbic lesions and the problem of stimulus-
(1988). reinforcement associations. Exp. Neurol. 36, 362377 (1972).
13. Xu, X. J., Wiesenfeld-Hallin, Z., Hughes, J., Horwell, D. C. & Hokfelt, T. 35. Spiegler, B. J. & Mishkin, M. Evidence for the sequential participation of
CI988, a selective antagonist of cholecystokinin B receptors prevents inferior temporal cortex and amygdala in the acquisition of stimulus-reward
morphine tolerance in the rat. Br. J. Pharmacol. 105, 591596 (1992). associations. Behav. Brain Res. 3, 303317 (1981).
14. Panerai, A. E., Rovati, L. C., Cocco, E., Sacerdote, P. & Mantegazza, P. 36. Rolls, E. T. in The Amygdaloid Complex (ed. Ben-Ari, Y.) 383393 (Elsevier,
Dissociation of tolerance and dependence to morphine: a possible role for New York, 1981).
cholecystokinin. Brain Res. 410, 5260 (1987). 37. Rolls, E. T. in Emotions: Neuronal and Chemical Control (ed. Oomur, Y.)
15. Baber, N. S., Dourish, C. T. & Hill, D. R. The role of CCK, caerulein, and CCK 325344 (Japan Scientific Societies Press, New York, 1986).
antagonists in nociception. Pain 39, 307328 (1989). 38. Introini-Collison, I. B., Nagahara, A. H. & McGaugh, J. L. Memory
16. Maldonado, R., Valverde, O., Derrien, M., Tejedor-Real, P. & Roques, B. P. enhancement with intra-amygdala post-training naloxone is blocked by
Effects induced by BC 264, a selective agonist of CCK-B receptors, on concurrent administration of propranolol. Brain Res. 476, 94101 (1989).
morphine-dependent rats. Pharmacol. Biochem. Behav. 48, 363369 (1994). 39. Belcheva, I., Belcheva, S., Petkov, V. V. & Petkov, V. D. Asymmetry in
17. Branchereau, P., Champagnat, J. & Denavit-Saubie, M. Cholecystokinin- behavioral responses to cholecystokinin microinjected into rat nucleus
gated currents in neurons of the rat solitary complex in vitro. J. Neurophysiol. accumbens and amygdala. Neuropharmacology 33, 9951002 (1994).
70, 25842595 (1993). 40. Huston, J. P. et al. Modulation of memory, reinforcement and anxiety
18. Shigeyoshi, Y. et al. Distribution of mRNA for CCK-B receptor in the brain of parameters by intra-amygdala injection of cholecystokinin-fragments Boc-
Mastomys natalensis: abundant expression in telencephalic neurons. Brain CCK-4 and CCK-8s. Peptides 19, 2737 (1998).
Res. 640, 8192 (1994). 41. Pu, S., Zhuang, H., Lu, Z., Wu, X. & Han, J. Cholecystokinin gene expression
19. Crawley, J. N. & Corwin, R. L. Biological actions of cholecystokinin. Peptides in rat amygdaloid neurons: normal distribution and effect of morphine
15, 731755 (1994). tolerance. Brain Res. Mol. Brain Res. 21, 183189 (1994).
20. Mercer, L. D. et al. On the distribution of cholecystokinin B receptors in 42. Alvarez, P., Zola-Morgan, S. & Squire, L. R. Damage limited to the
monkey brain. Brain Res. 738, 313318 (1996). hippocampal region produces long-lasting memory impairment in monkeys.
21. OBrien, C. P., OBrien, T. J., Mintz, J. & Brady, J. P. Conditioning of narcotic J. Neurosci. 15, 37963807 (1995).
abstinence symptoms in human subjects. Drug Alcohol Depend. 1, 115123 43. Morris, R. G. & Frey, U. Hippocampal synaptic plasticity: role in spatial
(1975). learning or the automatic recording of attended experience? Phil. Trans. R.
22. Siegel, S. Morphine analgesic tolerance: Its situation specificity supports a Soc. Lond. B Biol. Sci. 352, 14891503 (1997).
pavlovian conditioning model. Science 193, 323325 (1976). 44. Reed, J. M. & Squire, L. R. Impaired recognition memory in patients with
23. Siegel, S., Hinson, R. E. & Krank, M. D. Morphine-induced attenuation of lesions limited to the hippocampal formation. Behav. Neurosci. 111, 667675
morphine tolerance. Science 212, 15331534 (1981). (1997).
24. Siegel, S., Hinson, R. E., Krank, M. D. & McCully, J. Heroin overdose death: 45. Lemaire, M., Bohme, G. A., Piot, O., Roques, B. P. & Blanchard, J. C. CCK-A
contribution of drug-associated environmental cues. Science 216, 436437 and CCK-B selective receptor agonists and antagonists modulate olfactory
(1982). recognition in male rats. Psychopharmacology 115, 435440 (1994).
25. Siegel, S. Pavlovian conditioning and heroin overdose: reports by overdose 46. Young, B. J., Otto, T., Fox, G. D. & Eichenbaum H. Memory representation
victims. Bull. Psychonomic Soc. 22, 428430 (1984). within the parahippocampal region. J. Neurosci. 17, 51835195 (1997).
26. OBrien, C. P., Ehrman, R. N. & Ternes, J. W. in Behavioral Analysis of Drug 47. Amaral, D. G. in Excitatory Amino Acids and Epilepsy (eds. Schwarcz, R. &
Dependence (eds. Goldberg, S. R. & Stolerman, I. P.) 329356 (Academic, Ben-Ari, Y.) 317 (Plenum, New York, 1986).
Orlando, 1986). 48. Saunders, R. C. & Rosene, D. L. A comparison of the efferents of the
27. OBrien, C. P., Childress, A. R., Ehrman, R. & Robbins, S. J. Conditioning amygdala and the hippocampal formation in the rhesus monkey: I.
factors in drug abuse: can they explain compulsion? J. Psychopharmacol. 12, convergence in the entorhinal, prorhinal, and perirhinal cortices. J. Comp.
1522 (1998). Neurol. 271, 153184 (1988).
28. Matthes, H. W. D. et al. Loss of morphine-induced analgesia, reward effect 49. White, N. M. Addictive drugs as reinforcers: multiple partial actions on
and withdrawal symptoms in mice lacking the mu-opioid-receptor gene. memory systems. Addiction 91, 921949 (1996).
Nature 383, 819823 (1996). 50. Llewellyn-Smith, I. J., Pilowsky, P. & Minson, J. B. The tungstate-stabilized
29. Carter, B. L. & Tiffany, S. T. Cross-tolerance of associative and nonassociative tetramethylbenzidine reaction for light and electron microscopic
morphine tolerance in the rat with mu- and kappa-specific opioids. immunocytochemistry and for revealing biocytin-filled neurons. J. Neurosci.
Psychopharmacology 123, 289296 (1996). Methods 46, 2740 (1993).

articles
A model for intradendritic

computation of binocular disparity
Kevin A. Archie1 and Bartlett W. Mel2
1 Neuroscience Program, University of Southern California, Los Angeles, California 90089-2520, USA
2 Department of Biomedical Engineering, University of Southern California, Los Angeles, California 90089-1451, USA
Correspondence should be addressed to K.A.A. (karchie@lnc.usc.edu)
Many complex cells in mammalian primary visual cortex are finely tuned to binocular disparity. In
the prevailing model, several binocular simple cells drive each disparity-tuned complex cell.
However, some cat complex cells receive direct LGN input, and binocular simple cells are rare in
macaque. In our biophysically detailed compartmental model, active dendrites of a single neuron
perform the multiple simple-cell-like subunit computations that underlie both orientation and
disparity tuning. The responses of our detailed model could be predicted by a simple algebraic
formula closely related to an energy model. Adding inhibitory synapses led to sharper, more
contrast-invariant tuning curves. Thus active dendrites could contribute to binocular-disparity
tuning in complex cells.
Neurons in mammalian primary visual cortex are grouped into computations19,2227. Also, dendrites of neocortical pyramidal cells
two broad categories: simple cells, sensitive to stimulus orienta- contain voltage-dependent channels that could profoundly influ-
tion and position (or phase), and complex cells, relatively insen- ence their integrative behavior, including NMDA channels and
sitive to the spatial phase of an oriented stimulus. Many complex voltage-dependent Na+ and Ca2+ conductances that can amplify
cells receive input from both eyes and are tuned to binocular dis- synaptic inputs and generate both fast and slow dendritic spikes28,29.
parity, the horizontal offset of features between the two eyes that Based on this evidence, two modeling studies show that a sin-
conveys stimulus depth1,2. Whereas simple cells are often approx- gle neuron with active dendrites can reproduce the defining spatial
imated as linear3, complex cells are fundamentally nonlinear: nonlinearities of complex cells, including position-invariant ori-
their responses to combinations of stimuli cannot be predicted entation tuning17 and binocular-disparity tuning30. The morpho-
by summing the responses to individual stimuli. The form of the logically realistic pyramidal cell and the random subunit allocation
response nonlinearity suggests a hierarchy in which outputs of used in these studies, however, make it difficult to quantify con-
multiple simple-cell-like subunits are pooled by complex cells4,5. tributions of active dendritic currents, or to compare the behav-
In a disparity-tuned complex cell, those subunits seem to be ior of the biophysical model to existing mathematical models.
binocular and oriented6,7. Here we developed a simplified model, patterned after an
An early proposal held that simple cells receive input from existing energy model for disparity tuning6,7, in which four
unoriented cells in the lateral geniculate nucleus (LGN) in cat or binocular simple-cell-like subunits were mapped onto four basal
cortical layer 4C in monkey, and complex cells pool over the ori- branches of a dendritic tree. In addition, although it was previ-
ented simple-cell outputs4,8. In similar models for disparity tuning, ously shown that excitatory inputs onto active dendrites can pro-
a set of binocular simple cells performs the requisite subunit com- duce nonlinear boosting interactions, we asked whether observed
putations6,7,9,10. However, evidence rules out such a strict hierarchy. suppressive interactions between stimuli57,15 could likewise result
In cat, some first order complex cells receive direct LGN input, from intradendritic computations, by assuming more realistic
circumventing the simple-cell stage11,12. In macaque, the binocu- (nonzero) background firing rates of inputs, or by including
lar simple cells required by conventional models are rare or nonex- synaptic inhibition16,19.
istent. Although many disparity-tuned neurons are found in
macaque V1, a recent study found no binocular simple cells: a few RESULTS
had spatially offset light and dark subregions (accounting for the We used compartmental simulations to study the response of a
small proportion of binocular cells classified as simple in earlier simplified pyramidal neuron with active dendrites (Fig. 1). Low
studies13,14), but all binocular cells showed position-invariant dis- input resistance at the soma (87 M) meant that voltage signals
parity tuning15. generated within the thin basal dendrites were strongly attenuated
If the simple-cell stage is bypassed, where might the required when recorded at the cell body, electrically isolating the branch-
subunit computations be carried out? It has been suggested that es from one another such that a synaptic input changed voltage
simple-cell-like subunit computations might be performed direct- within its associated branch much more than in the other three.
ly within the dendrites of complex cells8,16,17. In theoretical stud- Nonlinear interactions between synaptic inputs were thus lim-
ies, spatially extended dendrites electrotonically compartmentalize ited primarily to those among synapses on the same branch and
synaptic potentials18,1921, potentially allowing different dendrites of mediated by dendritic Na+ and K+ currents and voltage-depen-
a single neuron to perform quasi-independent nonlinear subunit dent NMDA-type synaptic currents.

articles
Fig. 1. Connectivity between LGN centersurround cells and the

100 m
model complex cell. The cortical neuron had a simplified pyramidal-cell
morphology, with a spherical soma 20 m in diameter (1 compartment),
4 basal dendrites 1.5 m in diameter and 200 m long (16 compart-
ments each), and one apical dendrite 5 m in diameter and 500 m long
L L (8 compartments). Visual images were presented to the LGN, which
R R
consisted of four arrays of spatial filters (ON- and OFF-center cells in
ON-center cell
left and right eyes). Each of the 4 basal dendrites received input from 16
excitatory synapses (small black circles on dendrites), each provided by
a different LGN cell. The collection of LGN cells providing input to each
branch corresponded to one of the simple-cell-like subunits in a binocu-
lar energy model7. The type and spatial layout of LGN cells driving each
1
-1
branch were identical for each eye (insets).
OFF-center cell
total input to the cell. This modulation was distinct from that
L L seen in responses of typical simple cells in that the peaks and
R R valleys of the modulation occurred at identical locations for
-1 1 light and dark stimuli, confirming precise registration rather

than spatial segregation of ON and OFF subfields. To verify
the importance of the subunit organization, we randomly
scrambled the 64 afferents over the basal dendrites. Orienta-
tion tuning was abolished by spatial scrambling (Fig. 3c,
squares) or by blocking NMDA and dendritic Na+ channels
(Fig. 3c, crosses), replicating findings using a morphological-
To determine the inputoutput relation of each branch and ly realistic cell17 and showing that a small number of excita-
the degree of crosstalk between branches, we measured the mean tory afferents projecting to a few dendritic branches could
firing rate of the cell as varying numbers of synapses were stim- produce clear complex-cell-like responses.
ulated at 100 Hz on 2 of the 4 basal branches (Fig. 2). Under the
conditions tested, the single-branch inputoutput relation was
expansively nonlinear with a threshold of approximately nine Table 1. Biophysical simulation parameters.
active synapses (Fig. 2a). The cell showed significant functional Parameter Value
compartmentalization, as indicated by the drop in the cells firing
Rm 10 kcm2
rate when a given number of active synapses on a single branch
were dispersed over two branches. The branches were not per- Eleak 70 mV
fectly independent (compare Fig. 2a with response predicted by Ra 250 cm2
the sum of two single-branch responses in Fig. 2b; difference in Cm 1.0 F/cm2
Fig. 2c). Significant facilitative crosstalk peaked when both Somatic gNa, gDR 0.02, 0.012 S/cm2
branches were driven by 1012 synapses. The two-branch Dendritic gNa, gDR 0.006, 0.0012 S/cm2
response plot was compared to that of a hypothetical cell with gAMPA 0.180.56 nS
perfectly isolated squaring subunitsthe building blocks of an AMPA (on, off) 0.1 ms, 2 ms
energy model (Fig. 2d). Whereas the compartmental model gNMDA 0.361.13 nS
responds differently than a quadratic subunit model overall, many
NMDA (on, off) 0.1 ms, 80 ms
iso-synapse-count contours (constant levels of synaptic activity)
Esyn (excitory) 0 mV
are similar in the two cases. When the number of activated
synapses is held constant across stimulus conditions, as in most gGABA A
0.33 nS
of our studies, the biophysical model closely mimics a quadratic GABA (on, off)
A
0.1 ms, 50 ms
subunit model. In contrast, an electrically compact neuron does EGABA A
65 mV
not support independent subunits and would therefore respond gGABA B
0.099 nS
at a constant level along every iso-synapse-count contour AMPA (on, off)
B
62.5 ms, 213 ms
(Fig. 2e). EGABA 95 mV
B
To test the orientation tuning of the model cell, we pre-
sented sinusoidal gratings at various orientations and phases Hodgkin-Huxley-style voltage-dependent Na+ and K+ channels were distrib-
uted throughout the cell, with higher concentrations in the soma than in the
to both eyes at zero horizontal disparity. A grating at the cells dendrites. The channel implementation27 was adapted from ref. 48. This set
optimal (vertical) orientation directed most synaptic input to of conductances led to a somatic input resistance of 87 M for small
one branch, giving a high output firing rate (37 Hz, Fig. 3a). A perturbations around a steady-state voltage of 70 mV, comparable to the
measured input resistance of visual cortical neurons in vivo49. Synapses were
grating at the null (horizontal) orientation led to a more dif- modeled according to a described kinetic scheme50, with activation and
fuse arrangement of the same synaptic activation values, and inactivation time constants as shown. Excitatory synapses had both voltage-
a far weaker overall response (15 Hz, Fig. 3b). Based on this independent (AMPA/kainate) and voltage-dependent (NMDA) components;
peak conductances were scaled inversely by the local input resistance to
spatial concentration, the cell showed clear orientation tun- reduce the dependence of local EPSP size on synapse location. The local
ing (Fig. 3c, diamonds). The modest variation in the cells peak of an isolated EPSP varied from 0.94 mV near the soma to 2.15 mV at
response with grating phase (Fig. 3d) was due to subsampling the distal tips, with the corresponding somatic peak varying from 0.81 mV
(proximal synapse) to 0.23 mV (distal synapse).
of the LGN array, yielding 25% peak-to-peak modulation in

articles
aa bb two bars were presented at corresponding loca-

tions in the two eyes (zero horizontal disparity).
80 80 The peak response region was elongated along
60 60 the zero-disparity diagonal, indicating that the
40
Hz
40
Hz cells disparity preference was invariant to the
absolute positions of the bars within the recep-
20 20
16 16 tive field. When the bars were in either of the
0 0
12 16 12 16 opposite-contrast conditions (light and dark,
8 12 8 12
8 8 or dark and light), the response pattern was
n1 4 4 n2 n1
4
4
0 0 0 0 n2 complementary, with weak responses along the
zero-disparity diagonal, and strong responses in
the nonzero-disparity flanks.
c Our cells measured response differed from
a linear prediction based on the sum of the
80
responses to the same stimuli presented monoc-
60
ularly (Fig. 4b). Responses to same-contrast con-
16 Hz ditions showed a modest nonlinear
12 40 enhancement along the zero-disparity diagonal
n1 8 and somewhat stronger suppression at nonzero

4 20
disparities, whereas responses to opposite-con-
0 0
0 4 8 12 16 trast cases showed a complementary pattern of
n2 strong suppression at zero disparity and weak
boosting at nonzero disparities. This pattern of
dd e boosting and suppression is consistent with
e recordings from disparity-tuned complex cells
80 80
in macaque V1 (ref. 15).
60 60 A binocular kernel was obtained for the
Hz Hz model cell by subtracting the mean opposite-
40 40
20 20
contrast response from the mean same-contrast
16
0
16
0
response (Fig. 5a). The diagonally elongated
12 12
8 12
16
8 12
16 structure with alternating excitatory and
4
8
4
8 inhibitory lobes closely resembled kernels gen-
n1 4
n n1 4
n erated from binocular complex cells in cat visu-
0 0 2 0 0 2
al cortex (Fig. 5b), which are often well fit by an
Fig. 2. Interaction between synaptic inputs on two branches. Each frame shows a cells mean
energy model consisting of four simple-cell-like
firing rate (averaged over 64 simulations, 1 s each) as the number of fully active (100 Hz)
synapses was increased on the 2 branches. Within each branch, synapse locations were ran-
subunits (negative and positive parts of even-
7
dom. The other 2 branches each contained 16 synapses firing at a background rate of 15 Hz. and odd-symmetric Gabor filters) . We attempt-
Cell parameters were as in Table 1. (a) Response of the model cell. Solid and dashed lines ed to predict the binocular kernels of our bio-
show two iso-synapse-count contours (see text; solid, 12 total synapses, dashed, 16 total physically modeled cell using a simple algebraic
synapses). Dips in these contours reflect compartmentalization of cell. (b) Response of an formula inspired by the structure of the energy
idealized cell with single-branch nonlinearity as in (a), but with perfect independence of the model. To do this, we calculated the linear stage
two branches. Each branch was assumed to produce an independent nonlinear response f(n), of each of the four subunits by adding the mean
where n was the number of active synapses on that branch. f(n) was taken as the average of firing rates of the 16 afferent inputs to each
the values on the branch 1 and branch 2 axes in (a), which differed slightly because of the ran- branch. Each of the subtotals was then squared,
domization of spike train inputs used in these simulations. The idealized cell output was the
sum of the individual branch responses, f(n1) + f(n2). (c) Difference between (a) and (b).
and the four resulting values were added to pre-
Raised portions represent cell responses above linear prediction. (d) Model cell whose out- dict the cells overall response. The resulting
put is proportional to n12 + n22. The result is scaled for comparison with other frames. Iso- binocular disparity kernel (Fig. 5c) resembled
synapse-count contours are similar to those shown in (a). (e) Response of a cell with branch corresponding plots generated by the compart-
nonlinearity f(n) as in (a), but with no subunit isolation, giving a response f(n1 + n2). All iso- mental simulations (Fig. 5a) as well as experi-
synapse-count contours are flat, indicating that the cell is insensitive to the arrangement of mental data (Fig. 5b).
synapses across the two branches. To assess the importance of the squaring
nonlinearity in determining the basic form of
the binocular kernel, we applied three other
expansive nonlinear functions to the branch-
We tested the binocular-disparity tuning of the cell using bars subunit outputs in the algebraic model: x1.05 (Fig. 5d), x3(Fig. 5e)
at its optimal orientation. Following experimental protocols used and ex/10(Fig. 5f). Binocular kernels produced by all cases were
in cat7 and monkey15, we ran four stimulus conditions consist- similar, and on this basis would be difficult to distinguish exper-
ing of light and dark bars presented simultaneously to the left imentally. However, the magnitude of the nonlinear response
and right eyes in each of the four possible combinations (Fig. 4). components that make up the kernel depended strongly on the
For each condition, we presented the two bars at 64 horizontal subunit exponent: the output of the x1.05 model, for instance,
locations in each of the two eyes to generate a two-dimensional contained a nonlinear component that reached only 3% of the
binocular-response plot (Fig. 4a). In the same-contrast condi- peak cell response, whereas the contribution rose to 38% in the
tions (both light or both dark), the cell responded best when the quadratic model. For comparison, the magnitude of the nonlin-

articles
optimal orientation null orientation postsynaptic compartment. As in a previ-

a b ous proposal16, the functional role of the
synaptic inhibition in this model was to
define a subunit both in terms of a stimulus
that should be present (carried by excita-
25 Hz tion), and a stimulus that should be absent
50 Hz (carried by inhibition).
0 0 We examined two different spatial dis-
75 Hz mV mV tributions of inhibitory synapses to test a
100 Hz 60 60 biophysical principle formulated assuming
0 ms 250 0 ms 250
passive dendritic integration19: that shunt-
ing inhibition is most suppressiveserv-
ing as a veto mechanismwhen physically
interposed between a distal site of synaptic
excitation and the axosomatic site of spike
initiation. In the proximal condition, the
c
c 50 d d 50 16 inhibitory contacts impinging on each
branch were distributed over the first 20
m of the branch where it emerged from

Response (Hz)
40 active, ordered 40
30 30 the cell body. In the distal condition,
passive, ordered inhibitory contacts were distributed uni-
20 20 formly over the distal 180 m of the
active, scrambled branch, avoiding the 20 proximal m.
10 10
Counter to expectations, proximal inhi-
0 0 bition weakened orientation tuning relative
90 45 0 45 90 0 120 240 360 to that seen in the excitation-only condition
Grating orientation Grating phase (compare Fig. 6a with Fig. 3c). Blunting of
orientation tuning was particularly evident
Fig. 3. Response of the model cell to sinusoidal gratings. (a) Presynaptic firing rates are indicated at higher contrasts, and was due to both an
by the size of the synapse marker; synapses with firing rate of 0 Hz are not shown. An input at the
elevation of responses at null orientations
preferred (vertical) orientation leads to grouping of the most active synapses on a single branch.
This grouped input leads to robust firing at the cell body (inset voltage trace). The visual stimulus is and an anomalous suppression of respons-
shown above the voltage trace, with the grid of LGN cell centers indicated by black dots. One LGN es to optimal stimuli. Binocular-response
receptive field is shown for scale. (b) An input at the null orientation leads to dispersal of the most plots (Fig. 6b and c) were qualitatively sim-
active synapses throughout the basal dendrites; the cell responds only weakly to this diffuse pattern ilar to those in the excitation-only condi-
of input. (c) Firing rate as a function of grating orientation shows clear orientation tuning (dia- tion, though monocular responses were
monds) that is abolished when afferents were randomly reassigned to the four branches (squares) very weak. The binocular interaction plots
or when the NMDA and dendritic Na+ channels were blocked (crosses). Higher synaptic conduc- differed from the excitation-only case in that
tances were used in the latter case to boost firing rates. Each data point is an average over 1-s runs they consisted almost entirely of boosting,
at each of 18 different phases. (d) Firing rate for an optimally oriented grating as a function of grat-
but not suppresive nonlinearities (Fig. 6d
ing phase (diamonds). Modulation is due to spatial variation in total LGN input to cell (see text).
Each data point represents a single 1-s run. Two lower curves are for conditions as in part (c).
and e). Although suppressive binocular
interactions were almost absent in this cell,
the binocular disparity kernel appeared
normal in form, containing both positive
and negative lobes (Fig. 6f).
ear component of the biophysically modeled cell response was In contrast to proximal inhibition, distal inhibition produced
in the range of 2560% of the peak response of the cell (1015 both sharper orientation tuning than was seen with excitation
Hz boost over a 40-Hz peak above background in the same-con- alone, and approximately multiplicative scaling of the tuning
trast conditions; 20-Hz suppression relative to peak of 35 Hz in curve with increased contrast (Fig. 6g). The cell lacked monocu-
different-contrast conditions), values comparable to those report- lar responses almost entirely, but showed strong, disparity-tuned
ed for a disparity-tuned complex cell in monkey V1 (ref. 15). binocular responses (Fig. 6h and i). Binocular interactions were
We added synaptic inhibition to the model to investigate again exclusively boosting (Fig. 6j and k), and binocular kernels
whether it could also contribute to the dendritic subunit com- seemed normal for a disparity-tuned complex cell, with pro-
putations underlying disparity tuning. Through an inhibitory nounced positive and negative lobes (Fig. 6l).
interneuron, each excitatory LGN afferent was assumed to drive To identify the source of differences between cases with and
a single GABAA-type synapse placed on one of the four dendrit- without inhibition, we recorded simultaneously from each of the
ic compartments. This shunting inhibitory synapse was mapped four dendritic branches and the soma while presenting an opti-
onto a pyramidal cell dendrite at random and balanced so that mally oriented grating (Fig. 7a). Somatic action potentials were
every branch received an identical number of inhibitory contacts most often triggered when two or more dendritic branches fired
and direct excitatory and indirect inhibitory inputs driven by any in rapid succession, one spike riding piggyback on the other.
given LGN afferent never fell on the same branch. This wiring For an optimally oriented stimulus, the initial spike was most
constraint reflected the assumption that a Hebb-type rule gov- often generated within the primary subunit, followed quickly by
erning synapse survival would discourage stabilization of corre- a secondary spike in another branch. Isolated dendritic spikes
lated excitatory and inhibitory synapses within the same that did not trigger a somatic spike led to EPSP-like somatic depo-

articles
same contrast opposite contrast one subunit far more than the
L R L R L R L R other three (Fig. 3a). In this con-
figuration, spikes continued to be
generated frequently within the pri-
mary subunit, but weak activation
of the other branches reduced the
a incidence of secondary spikes in
2 45 Hz these branches and increased spike
failures at the cell body.
In contrast to proximal inhibi-
0 tion, an identical set of inhibitory
Right stimulus position
synapses distributed distally led to

2 0 Hz a substantial reduction in the rate
of spike initiation within the
branch itself, particularly when
2 +25 Hz
b
b excitation to the branch was rela-
tively weak. This accounted for the
0 0 Hz strong suppression of responses to
stimuli at non-preferred orienta-

2 tions, which tended to produce a
25 Hz
2 0 2 2 0 2 2 0 2 2 0 2 mixture of middling excitation
Left stimulus position
and middling inhibition within
each of the four branches. The
Fig. 4. Response of the cell to optimally oriented pairs of bars. The four stimulus conditions are shown at near elimination of dendritic
top. Each gray box represents a full array of LGN center locations; stimulus bars are shown to scale. (a) Firing spikes in this condition meant that
rate of the cell as a function of bar position in the left and right eyes. The elongated band of strong responses
somatic action potentials were
along the zero-disparity diagonal in the two same-contrast conditions indicates position-invariant disparity
tuning. In opposite-contrast cases, strongest responses are in symmetrical off-diagonal flanks. Responses rarely triggered. Optimally orient-
were smoothed by convolution with a Gaussian filter ( = 0.08) before contours were generated. ed stimuli continued to drive the
Conventional disparity-tuning curve is shown above each frame, derived by integrating the binocular- cell strongly in this case, however,
response plot along iso-disparity diagonals7. Monocular responses were computed in separate runs with bars as most of the excitation was con-
presented to a single eye, but were nearly identical to responses shown here along the edges of the plot (in centrated within the primary sub-
stimulus configurations with a bar at either extreme end of the receptive field in one eye). (b) Nonlinear unit, whereas inhibition was
component of the firing rate, computed by subtracting the linear prediction based on the monocular largely relegated to distal portions
responses. Positive contours are solid, negative contours are dashed. Both suppressive and boosting interac- of the other three branches.
tions are seen in complementary patterns for same-contrast and opposite-contrast conditions.
Similar mechanisms accounted
for the difference in the pattern of
binocular interactions seen in cases
with and without inhibitionin
larizations of several millivolts. The precise size of these events particular, the disappearance of suppressive nonlinear interac-
depended on many factors, including the assumed concentrations when inhibition was included in the model. As for stim-
tions of Na+ and K+ channels in the dendritic branches. In gen- uli at non-optimal orientations, monocular stimuli were poorly
eral, however, the impedance mismatch between the thin suited to evoke dendritic spikes because they produced only
dendritic branches and the much larger cell body led to a pro- half the excitation of binocular stimuli within their respective
nounced attenuation of dendritic spike height as measured at the subunits; even at high contrast, monocular responses with dis-
soma. The magnitude of this attenuation was illustrated by plot- tal inhibition peaked at 3 Hz. Weak monocular responses led, in
ting the voltage along a dendritic branch during a dendritic spike turn, to a floor effect: binocular responses had little room to
followed by a somatic spike (Fig. 7b). drop below the monocular prediction. For such a purely binoc-
Observed interactions between dendritic and somatic spikes ular cell, any nonzero response indicated a boosting interac-
largely accounted for the differential effects of proximal versus tion between the two eyes. As a result, the binocular-interaction
distal inhibition. Whereas we expected proximal shunting- plots were nearly identical to the corresponding binocular-
inhibitory synapses to produce powerful branch-specific sup- response plots under both inhibitory conditions. Because many
pression, we found that they did little to impede production of alternative assumptions could produce more powerful monoc-
dendritic spikes in the unclamped distal portions of their ular responses in this model, the balance of boosting versus
respective branches. Rather, the girdle of perisomatic inhibi- suppressive interactions in the cells binocular receptive field
tion in the proximal case led mainly to an overall suppression of should be viewed as a weakly specified parameter of the model,
spike rate at the cell body (compare lowest curve in Fig. 6a with rather than a central feature.
excitation-only tuning curve in Fig. 3c). We found this sup- In other conditions tested, we found that distributing
pression was due to an increased dependence of somatic spike inhibitory inputs over the full length of the dendrites yielded
generation on close temporal summation of dendritic spikes in results intermediate to those for proximal and distal conditions.
two or more branches. We also obtained qualitatively similar results in simulations using
This effect accounted for the peculiar dip at the peak of the slow, hyperpolarizing (GABAB) inhibitory synapses, although
cells orientation-tuning curve seen at high contrast: optimally the differences between proximal and distal conditions were less
oriented stimuli provided lopsided excitation to the cell, driving pronounced in this case.

articles
mal inhibition or with no inhibi-

aa Biophysical simulation b
b Experimental data7 cc Subunit nonlinearity: x2 tion (data not shown) were more
additive in nature. However, binoc-
ular-disparity kernels were similar
with or without proximal or distal
inhibition, consistent with the very
weak dependence of this represen-
tation on the form of the underly-
ing subunit nonlinearity (Fig. 5).
The lack of subunit specificity of
proximally distributed inhibition
suggests that the on-path condition
d e f for maximally effective shunting
d x1.05 e x3 f ex/10
inhibition, derived in the context of
passive dendritic integration19, may
have limited applicability when den-
drites are physiologically active. In
our model, perisomatic inhibition
primarily controlled output firing

rates of the cell as a whole, but had
practically no influence on dendrit-
ic spike initiation. We also found
that, in the present model, proximal
shunting inhibition in the absence
Fig. 5. Binocular disparity kernels, showing the deviation of the cells response from linearity. Kernels were
of an expansive boosting mecha-
calculated by subtracting the mean opposite-contrast response from the mean same-contrast response7.
Axes in (a, cf) were shifted and scaled to match data in (b). (a) Kernel generated from our simulation nism (when NMDA and sodium
results. The outward curvature of the nonzero-disparity flanks is due to strong side lobes in the subunit ker- channels were blocked) could not
nel. (b) Kernel from a disparity-tuned complex cell in cat (reprinted from ref. 7). (c) Kernel based on alge- produce orientation or disparity-
braic subunit model: for each dendritic branch, the firing rates of all 16 synaptic terminals were summed, and tuned complex-cell-like responses.
this sum was squared. The complex-cell response was then calculated by summing the four subunit When pure AMPA-type excitatory
responses. (df) Binocular kernels using alternative expansive nonlinearities applied to each subunit output. conductances were powerful enough
By comparison, a cell with linear subunits would result in a uniform value of zero. to evoke reasonable firing rates in
response to an optimally oriented
grating, the large synaptically
induced depolarizations within the
DISCUSSION dendritic branches resulted in a significant loss of driving forcean
We demonstrate that a small number of excitatory, monocular, effect that opposed and, in fact, overwhelmed the veto-like non-
centersurround inputs to a few weakly excitable dendritic linearity provided by shunting inhibition. The balance in this com-
branches could produce the phase-invariant orientation and dis- petition between the shunting veto and classical synaptic saturation
parity tuning characteristic of many binocular complex cells in could probably be altered by spreading the excitatory and inhibito-
cats and monkeys. In this model, each of the four dendrites of ry synapses on a more elaborate dendritic arbor, but this change
the complex cell plays the role conventionally assigned to a binoc- seems unlikely to generate a nonlinear effect as powerful as that
ular simple cell8. In particular, each of the subunits binds togeth- associated with inhibitory control of dendritic spiking.
er, under a separate threshold, swaths of the left and right visual Surprisingly, binocular responses showed both boosting and
fields at a fixed horizontal offset, thus ensuring that the cell suppressive nonlinearities when our model cell was driven exclu-
responds best to a consistent horizontal disparity over the entire sively by excitatory synapses (Fig. 4). A boosting interaction was
width of its binocular receptive field. The within-branch non- seen when two light bars were presented to the two eyes at the
linearity is threshold-like in that it involves a rapid expansion in same location. This binocular facilitation occurred because, for
branch output as synaptic input intensity is increased (Fig. 2). this stimulus configuration, synaptic input was spatially con-
The expansive nonlinear regime in the excitation-only version of centrated within one or two dendritic branches, effectively dri-
our model results from the combination of NMDA receptors, ving these branches up through their expansive nonlinear
whose voltage-dependence amplifies synaptic currents primari- inputoutput regimes (Figs. 2 and 3). Suppressive binocular
ly in the subthreshold range, and voltage-dependent sodium and interactions, which peaked when bars of opposite polarity were
potassium channels, which generate dendritic spikes at a rate that presented at zero disparity (Fig. 4), result from this expansive
increases supralinearly with locally injected synaptic current. dendritic nonlinearity operating in reverse: each synaptic input to
When distal synaptic inhibition was included, orientation- the cell fired at a spontaneous background rate of 15 Hz, so each
tuning curves were significantly sharpened, indicating that the branch received significant input even without a visual stimulus.
combination of excitatory pushing and inhibition pulling led to When a stimulus bar was presented, the firing rate of some of the
a more powerful branch nonlinearity than was seen with excita- synaptic afferents was driven towards zero because of the oppo-
tion alone. With both excitation and distal inhibition, orienta- nency inherent in the underlying centersurround receptive fields.
tion-tuning curves were boosted in a quasi-multiplicative fashion Such removal of excitation drove a branch-subunit down its
as contrast was increased, in agreement with empirical data31,32, expansive operating curve, thereby reducing the efficacy of other
whereas the corresponding increases in responses under proxi- excitatory inputs onto the same branch. A non-zero background

articles
Proximal inhibition Distal inhibition zero disparity. There are several

aa 50 gg 50 other types of disparity-tuned cells2;
40 contrast 40
all could be reproduced with minor
Response (Hz)
(arbitrary units) modifications of the present model.

30 25 30 A nonzero-disparity preference
20 20 20 (near or far) would result if the
15 receptive-field centers of the affer-
10 12 10
ents from one eye were shifted hori-
0 0 zontally. A tuned-inhibitory cell
90 45 0 45 90 90 45 0 45 90
would result if the afferents from
Grating orientation Grating orientation
one eye only were swapped, so that
45 Hz every ON-center input was replaced
b
b cc hh ii
by an OFF-center input with the
responses
Binocular
same receptive field location and vice

versa. This manipulation produces
same-contrast responses identical to
0 Hz the opposite-contrast responses
shown in Fig. 4. In addition, some
+25 Hz
dd e
e jj kk neurons in macaque V1 have spa-
interactions
tially offset light and dark subre-

Binocular
0 Hz gions, consistent with a simple-cell

receptive field, but diagonal binoc-
ular interaction maps characteristic
25 Hz
of complex cells15. We were able to
ff ll
replicate responses of this kind by
modifying the model so that affer-
Binocular
ent contacts from left and right eyes

kernels
were mapped preferentially to prox-

imal versus distal dendritic sites.
Our model critically depends on
the assignment of synaptic contacts
Fig. 6. Responses of two cells with different configurations of inhibitory synapses. (a) In the proximal con- to distinct dendritic subunits. This
dition, 16 inhibitory synapses were placed within the proximal 20 m of each branch as described in
leads to the prediction that in visu-
Methods. Orientation tuning is shown at four contrasts (12, 15, 20 and 25, in arbitrary units, where a con-
trast of 12 was equivalent to that used in excitation-only case of Fig. 3). Each data point is an average over
al cortex, afferents with unoriented
18 runs. Half width at half maximum varied from 30 at low contrast to 90 at high contrast. (b, c) Binocular monocular receptive fields, wherev-
responses in same-contrast (mean of light-light and dark-dark) and different-contrast (mean of light-dark er they impinge on complex-cell
and dark-light) conditions. Monocular responses were reduced relative to excitation-only case. dendrites, should, as an outcome of
(d, e) Binocular interaction maps corresponding to (b, c), showing deviations from linear prediction based a Hebb-type developmental mecha-
on monocular firing rates. (f) Corresponding binocular disparity kernel. (g) In distal condition, 16 inhibitory nism, be appropriately grouped into
synapses were distributed at random along the distal 180 m of each branch. Half width at half maximum distinct postsynaptic compartments
varied from 20(low contrast) to 45(high contrast). Binocular results (hl) are defined as in (af). consistent with the orientation and
disparity preferences of the cell. In
the present model, each such group
contained 16 synapses distributed
firing rate thus enabled suppressive nonlinear interactions in the throughout a dendritic branch 200 m in length, which would
absence of inhibitory synaptic input. account for only a small fraction of the excitatory contacts nor-
The simplified pyramidal-cell morphology developed for these mally present on a branch of this length33. Similar results were
experiments allowed us to examine in a controlled fashion both generated with a wide variety of synapse arrangements, demon-
the nonlinear inputoutput behavior of a single dendritic branch strating that the effect is robust to changes in the number and
and nonlinear interaction between branches. We found that a spacing of synapses25,27,30. The spatial grouping of afferents might
model cell with two active dendrites behaved in key respects like be visualized physiologically in vivo using calcium- and/or volt-
a quadratic function with two input variables (Fig. 2). This con- age-sensitive dyes while driving a complex cell with optimal ver-
nection was echoed in the striking resemblance between binocular- sus non-optimal visual stimuli.
disparity kernels generated by the biophysically detailed Our model also depends on boosting nonlinear interactions
compartmental model, which involved numerical integration of between nearby synapses, leading to the prediction that, in complex
hundreds of nonlinear differential equations, and those produced cells receiving a significant component of unoriented monocular
by a simple algebraic formula akin to standard binocular energy input from LGN (in cat) or layer 4C (in monkey), the nonlinear
models. This successful reduction of the inputoutput behavior tuning properties of the cell, including both orientation and dis-
of a neuron with complex intradendritic physiology to a simple parity tuning, should be significantly degraded by an intracellu-
algebraic formula could provide a method for computationally lar blockade of excitatory voltage-dependent currents (including
efficient study of large networks of dendritically realistic neurons. Na+, Ca2+ and NMDA-type synaptic currents). Conversely, a uni-
The pattern of afferent connectivity in Fig. 1 was designed to form reduction of cellular responses under such a blockade with-
replicate the responses of a tuned excitatory cell selective for out a loss of tuning would constitute evidence against the present

articles
aa in the receptive field properties that result

from intracortical feedback36. The impor-
0
tance of cortical feedback should not be
Somatic underestimated: one modeling study
voltage
demonstrates that even the classical com-
(mV)
plex-cell receptive field may fundamen-
70 tally depend on intracortical feedback, so
0 50 100 200 250
that cells classified as simple based on their
afferent drive would actually respond like
complex cells by virtue of their network
interactions37.
0
soma Our model included two simplifying
Dendritic primary branch assumptions regarding the temporal
voltage
(mV)
other branches dynamics of neuronal and synaptic
responses. First, each centersurround
70 afferent providing input to the cell was
50 60 70 80 90 100
time (ms) modeled as a stationary Poisson process
with mean firing rate determined by the
response of a half-wave-rectified linear fil-

ter applied to the input image. This
approach ignored the transient compo-
bb nent of the responses of LGN cells and
most cortical cells when a stimulus is
flashed on or off. Second, the modeled
0 synapses lacked any form of activity-
Voltage 20 dependent dynamics, whereas excitatory
(mV) synapses in neocortex typically show
40
some depression during trains of
0 200
input38,39. Our model must therefore be
1 150 viewed as representing the response of a
2 100 cortical neuron during a short period
Time (ms) distance from over which the cells input rates are stable
3 50 soma (m) and the synaptic transfer functions can be
4 0
adequately approximated by a mean
value. As typical visual cortex recording
Fig. 7. Mechanisms of intradendritic spiking with distal inhibition. (a) Somatic voltage trace in protocols involve counting only a few
response to an optimal grating with simultaneous voltage recordings from the four dendritic branches spikes per stimulus7,15, this assumption of
and the cell body, shown on an expanded time scale. Bold trace, primary subunit (that which received stationarity seems unlikely to have impor-
maximal excitation); dashed trace, somatic voltage; thin traces, other dendritic branches. Somatic
tant consequences for our results.
action potentials occurred most often following two or more closely spaced dendritic spikes, usually
evoked first within the primary subunit, followed closely by a secondary, piggyback spike in another
The set of somatic and dendritic con-
branch. Not all dendritic spikes evoked somatic spikes, giving rise instead to small EPSP-like depolar- ductances used in our model does not
izations at the cell body. (b) Spacetime plot of membrane potential from the soma to a distal dendritic include all channels known to exist in the
tip showing a dendritic spike followed closely by a somatic spike. Time increases to the right; dendritic dendrites of cortical pyramidal cells, omit-
branch length is mapped into depth (soma at front, distal tip at back). Dendritic spike is largest at the ting, for example, persistent Na+ channels,
distal tip and strongly compressed at the somadendritic interface. Branch voltage moves up and down several types of voltage-dependent Ca2+
essentially as a unit (within 1 ms from leading to trailing edge). Attenuation of back-propagated somatic channels, calcium-dependent K+ channels
spike was explained by Na+-channel inactivation by preceding dendritic spike in the branch. and so on28. However, simulation studies
show that expansive nonlinear boosting
of synaptic inputs can arise from several
different kinds of voltage-dependent cur-
model. All species of voltage-dependent conductances must be rents acting in various combinations17,25,27,30. Pilot simulations for
blocked in such an experiment: although NMDA and Na+ con- this study indicated that tuning persists over a large range of con-
ductances in this model both contributed to the expansive non- ductance values, for example, spanning two orders of magnitude in
linearity on which the computation was based, we found in Na+-channel density, with physiological estimates falling near the
simulations that NMDA, Na+ or Ca2+ conductances alone (or in middle of this range40. Insensitivity to the specific choice, density
any combination) could supply a suitable expansive nonlinearity. and implementation of ionic channels present in such models is
Intracortical feedback, both excitatory and inhibitory, is impor- consistent with our observation that the nonlinear component of
tant in shaping the responses of cortical neurons34,35. Our model, the disparity-tuned responses depended little on the particular
driven exclusively by feedforward connections, is therefore unable functional form of the expansive nonlinearity present in the den-
to account for all the response properties of complex cells in visu- dritic branches (Fig. 5).
al cortex. For example, our model does not account for the effects Our results lend further support to the possibility that active
of stimuli outside the classical receptive field, presumably carried in dendrites provide compartmentalized interactions among synap-
part by long-range horizontal connections, or for dynamic changes tic inputs, where the the cell as a whole computes a sum over the

articles
outputs of these nonlinear subunits. Note that the basic struc- strained by available data. The values of these parameters were
ture of this nonlinear computationsometimes expressed as a thus set within reasonable ranges, then adjusted until the inhibi-
sum of product termsoccurs frequently in analyses of both tion produced strong, but not total, suppression of cell firing rates.
classical and extraclassical receptive-field properties in visual cor- Stimulus contrasts were then adjusted to drive the cell through a
tex41. For example, beyond our illustration of phase-invariant reasonable range of output rates.
orientation and disparity tuning, similar computations are per- The cell began each simulation uniformly polarized at a rest-
formed by cells whose responses are facilitated by stimuli pre- ing potential near 70 mV. The firing rate typically stabilized after
sented in the extraclassical receptive field42, by cells that respond a brief initial period of depolarization from rest. Each stimulus
to illusory contours43 and by cells that show multiplicative boost- was presented for 1.05 s, with all response rates calculated over a
ing under the influence of focal visual attention44. This com- 1-s interval following the initial 50 ms simulation transient.
monality could indicate that active dendritic processing
contributes widely to the information-processing functions of ACKNOWLEDGEMENTS
cortical neurons. This work was supported by the National Science Foundation. We thank Dan
Ruderman for contributions to early stages of this work, and Margaret
METHODS Livingstone and Gary Holt for comments.
Responses of a single cortical neuron were modeled using the
NEURON simulation environment45. The cell morphology is RECEIVED 29 JUNE; ACCEPTED 5 NOVEMBER 1999
described in Fig. 1; dendritic spines were not modeled, but pre-
vious simulations27,46 suggest that the inclusion of spines would 1. Barlow, H. B., Blakemore, C. & Pettigrew, J. D. The neural mechanism of
not significantly affect the results. Biophysical parameters are binocular depth perception. J. Physiol. (Lond.) 193, 327342 (1967).
2. Poggio, G. F. & Fischer, B. Binocular interaction and depth sensitivity in
summarized in Table 1. striate and prestriate cortex of behaving rhesus monkey. J. Neurophysiol. 40,
Stimulus images were presented to an LGN model consist- 13921405 (1977).
ing of 4 arrays (64 64) of spatial filters modeling ON- and 3. Movshon, J. A., Thompson, I. D. & Tolhurst, D. J. Spatial summation in the
OFF-center cells for left and right eyes and covering a 5 5 receptive fields of simple cells in the cats striate cortex. J. Physiol. (Lond.)
283, 5377 (1978).
region of visual space. The spatial filter kernel used was a dif- 4. Hubel, D. H. & Wiesel, T. N. Receptive fields, binocular interaction and
ference-of-Gaussians, adapted from the spatiotemporal LGN functional architecture in the cats visual cortex. J. Physiol. (Lond.) 160,
106154 (1962).
model of Wrgtter and Koch 47 , with Gaussian kernels 5. Movshon, J. A., Thompson, I. D. & Tolhurst, D. J. Receptive field
G(x, y) = (K/22) exp[(x2 + y2)/22] and parameter values organization of complex cells in the cats striate cortex. J. Physiol. (Lond.) 283,
center = 10.6, surround = 31.8 and Kcenter/Ksurround = 17/16. The 7999 (1978).
6. Ohzawa, I., DeAngelis, G. C. & Freeman, R. D. Stereoscopic depth
resulting center diameter was approximately 0.8. discrimination in the visual cortex: neurons ideally suited as disparity
The firing rate of each LGN unit was calculated by applying detectors. Science 249, 10371041 (1990).
the local spatial filter to the input image to obtain the stimulus- 7. Ohzawa, I., DeAngelis, G. C. & Freeman, R. D. Encoding of binocular
disparity by complex cells in the cats visual cortex. J. Neurophysiol. 77,
driven component of the firing rate, adding a spontaneous fir- 28792909 (1997).
ing rate of 15 Hz, and setting all negative values to 0 Hz. The 8. Hubel, D. H. Exploration of the primary visual cortex, 19551978. Nature
resulting value specified the mean rate of a Poisson train of presy- 299, 515524 (1982).
9. Fleet, D. J., Wagner, H. & Heeger, D. J. Neural encoding of binocular
naptic action potentials. A regular 4 4 grid covering 0.7 0.7 disparity: energy models, position shifts and phase shifts. Vision Res. 36,
was centered on each of the 4 LGN arrays (see Fig. 3). Each of 18391857 (1996).
the 16 grid points specified the receptive field center of an LGN 10. Zhu, Y.-D. & Qian, N. Binocular receptive field models, disparity tuning, and
characteristic disparity. Neural Comput. 8, 16111641 (1996).
cell forming a single excitatory synapse onto one of the four 11. LeVay, S. & Gilbert, C. Laminar patterns of geniculocortical projection in the
branches of the cortical neuron. The LGN afferents were grouped cat. Brain Res. 113, 119 (1976).
into 4 subunits of 16, where each group projected to one of the 4 12. Alonso, J.-M. & Martinez, L. M. Functional connectivity between simple cells
and complex cells in cat striate cortex. Nat. Neurosci. 1, 395403 (1998).
basal branches (Fig. 1). Odd-symmetric subunits (Fig. 1, upper 13. Hubel, D. H. & Wiesel, T. N. Receptive fields and functional architecture of
left and right insets) included four ON-center and four OFF-cen- monkey striate cortex. J. Physiol. (Lond.) 195, 215243 (1968).
ter cells from each eye, and even-symmetric subunits (Fig. 1, 14. Poggio, G. F., Doty, R. W. Jr & Talbot, W. H. Foveal striate cortex of behaving
monkey: single-neuron responses to square-wave gratings during fixation of
lower left and right insets) included eight ON- or eight OFF-cen- gaze. J. Neurophysiol. 40, 13691391 (1977).
ter cells from each eye. In each case, the 16 afferents projecting 15. Livingstone, M. S. & Tsao, D. Y. Receptive fields of disparity-selective neurons
in macaque striate cortex. Nat. Neurosci. 2, 825832 (1999).
to a given branch were drawn from LGN cells of the same type 16. Koch, C. & Poggio, T. in Synaptic Function (eds. Edelman, G. M., Gall, W. E.
and in the identical spatial layout in the left and right eyes. & Cowan, W. M.) 637697 (Wiley, New York, 1987).
In cases involving inhibition, a GABAA-type inhibitory synapse 17. Mel, B. W., Ruderman, D. L. & Archie, K. A. Translation-invariant
orientation tuning in visual complex cells could derive from intradendritic
was created to mirror the activity of each of the 64 excitatory computations. J. Neurosci. 18, 43254334 (1998).
synapses. Each synapse was presumed to arise from an inhibitory 18. Rall, W. & Rinzel, J. Branch input resistance and steady attenuation for
interneuron with tuning similar to that of the corresponding exci- input to one branch of a dendritic neuron model. Biophys. J. 13, 648688
(1973).
tatory input. The inhibitory afferents thus had center-surround 19. Koch, C., Poggio, T. & Torre, V. Retinal ganglion cells: a functional
receptive fields; orientation-tuned inhibition has yielded similar interpretation of dendritic morphology. Philos. Trans. R. Soc. Lond. B Biol.
results in pilot studies. Each of the 64 inhibitory synapses was Sci. 298, 227264 (1982).
20. Woolf, T., Shepherd, G. & Greer, C. Local information processing in dendritic
assigned to a branch at random and subject to 2 constraints: each trees: Subsets of spines in granule cells of the mammalian olfactory bulb. J.
branch received 16 inhibitory synapses, and corresponding exci- Neurosci. 11, 18371854 (1991).
tatory and inhibitory connections never fell on the same branch. 21. Zador, A. M., Agmon-Snir, H. & Segev, I. The morphoelectrotonic
transform: a graphical approach to dendritic function. J. Neurosci. 15,
Each inhibitory synapse was driven by an independent Poisson 16691682 (1995).
spike train, with mean rate one half that of its corresponding exci- 22. Koch, C., Poggio, T. & Torre, V. Computations in the vertebrate retina: Gain
tatory afferent. Three parameters determining the magnitude of enhancement, differentiation and motion discrimination. Trends Neurosci. 9,
204211 (1986).
inhibitionthe number of inhibitory synapses per branch, their 23. Rall, W. & Segev, I. in Synaptic Function (eds. Edelman, G. M., Gall, W. E. &
peak conductances and their firing rateswere only weakly con- Cowan, W. M.) 605636 (Wiley, New York, 1987).

articles
24. Mel, B. W. in Advances in Neural Information Processing Systems Vol. 4 (eds. 38. Markram, H. & Tsodyks, M. Redistribution of synaptic efficacy between
Moody, J. E., Hanson, S. J. & Lippman, R. P.) 3542 (Kaufmann, San Mateo, neocortical pyramidal neurons. Nature 382, 807810 (1996).
California, 1992). 39. Abbott, L. F., Varela, J. A., Sen, K. & Nelson, S. B. Synaptic depression and
25. Mel, B. W. NMDA-based pattern discrimination in a modeled cortical cortical gain control. Science 275, 220224 (1997).
neuron. Neural Comput. 4, 502517 (1992). 40. Destexhe, A. & Par, D. Impact of network activity on the integrative
26. Borg-Graham, L. J. & Grzywacz, N. M. in Single Neuron Computation (eds. properties of neocortical pyramidal neurons in vivo. J. Neurophysiol. 81,
McKenna, T., Davis, J. & Zornetzer, S. 347375 (Academic, San Diego, 15311547 (1999).
1992). 41. Mel, B. W. in Dendrites (eds. Stuart, G., Spruston, N. & Husser, M.) 271289
27. Mel, B. W. Synaptic integration in an excitable dendritic tree. J. Neurophysiol. (Oxford Univ. Press, Oxford, 1999).
70, 10861101 (1993). 42. Kapadia, M., Ito, M., Gilbert, C. D. & Westheimer, G. Improvement in visual
28. Johnston, D., Magee, J. C., Colbert, C. M. & Christie, B. R. Active properties sensitivity by changes in local contextparallel studies in human observers
of neuronal dendrites. Annu. Rev. Neurosci. 19, 165186 (1996). and in V1 of alert monkeys. Neuron 15, 843856 (1995).
29. Stuart, G., Spruston, N., Sakmann, B. & Husser, M. Action potential 43. Peterhans, E. & von der Heydt, R. Mechanisms of contour perception in
initiation and backpropagation in neurons of the mammalian CNS. Trends monkey visual cortex. II. Contours bridging gaps. J. Neurosci. 9, 17491763
Neurosci. 20, 125131 (1997). (1989).
30. Mel, B. W., Ruderman, D. L. & Archie, K. A. in Advances in Neural 44. McAdams, C. J. & Maunsell, J. H. R. Effects of attention on orientation-
Information Processing Systems Vol. 10 (eds. Jordan, M. I., Kearns, M. J. & tuning functions of single neurons in macaque cortical area V4. J. Neurosci.
Solla, S. A.) 208214 (MIT Press, Cambridge, Massachusetts, 1998). 19, 431441 (1999).
31. Sclar, G. & Freeman, R. D. Orientation selectivity in the cats striate cortex is 45. Hines, M. L. & Carnevale, N. T. The NEURON simulation environment.
invariant with stimulus contrast. Exp. Brain Res. 46, 457461 (1982). Neural Comput. 9, 11791209 (1997).
32. Skottun, B. C., Ohzawa, I., Sclar, G. & Freeman, R. D. The effects of contrast 46. Koch, C. & Zador, A. The function of dendritic spines: Devices subserving
on visual orientation and spatial frequency discrimination: a comparison of biochemical rather than electrical compartmentalization. J. Neurosci. 13,
single cells and behavior. J. Neurophysiol. 57, 773786 (1987). 413422 (1993).
33. Beaulieu, C. & Colonnier, M. A laminar analysis of the number of round- 47. Wrgtter, F. & Koch, C. A detailed model of the primary visual pathway in
asymmetrical and flat-symmetrical synapses on spines, dendritic trunks, and the cat: comparison of afferent excitatory and intracortical inhibitory
cell bodies in area 17 of the cat. J. Comp. Neurol. 231, 180189 (1985). connection schemes for orientation selectivity. J. Neurosci. 11, 19591979
34. Vidyasagar, T. R., Pei, X. & Volgushev, M. Multiple mechanisms underlying (1991).
the orientation selectivity of visual cortical neurones. Trends Neurosci. 19, 48. Bernander, O., Douglas, R., Martin, K. & Koch, C. Synaptic background
272277 (1996). activity influences spatiotemporal integration in single pyramidal cells. Proc.
35. Sompolinksy, H. & Shapley, R. New perspectives on the mechanisms for Natl. Acad. Sci. USA 88, 1156911573 (1991).
orientation selectivity. Curr. Opin. Neurobiol. 7, 514522 (1997). 49. Borg-Graham, L. J., Monier, C. & Frgnac, Y. Visual input evokes transient
36. Gilbert, C. D., Das, A., Ito, M., Kapadia, M. & Westheimer, G. Spatial and strong shunting inhibition in visual cortical neurons. Nature 393,
integration and cortical dynamics. Proc. Natl. Acad. Sci. USA 93, 615622 369373 (1998).
(1996). 50. Destexhe, A., Mainen, Z. & Sejnowski, T. J. Synthesis of models for excitable
37. Chance, F. S., Nelson, S. B. & Abbott, L. F. Complex cells as cortically membranes, synaptic transmission and neuromodulation using a common
amplified simple cells. Nat. Neurosci. 2, 277282 (1999). kinetic formalism. J. Comput. Neurosci. 1, 195230 (1994).

articles
Mechanisms of visual motion

detection
Paul R. Schrater1, David C. Knill2 and Eero P. Simoncelli3
1 Department of Psychology, University of Minnesota, N218 Elliott Hall, 75 E. River Dr., Minneapolis, Minnesota 55455, USA
2 Department of Psychology, University of Pennsylvania, 3815 Walnut St., Philadelphia, Pennsylvania 19104, USA
3 Center for Neural Science, New York University, 4 Washington Place, New York, New York 10003, USA
Correspondence should be addressed to P.R.S. (schrater@eye.psych.umn.edu)
Visual motion is processed by neurons in primary visual cortex that are sensitive to spatial
orientation and speed. Many models of local velocity computation are based on a second stage that
pools the outputs of first-stage neurons selective for different orientations, but the nature of this
pooling remains controversial. In a human psychophysical detection experiment, we found
near-perfect summation of image energy when it was distributed uniformly across all orientations,
but poor summation when it was concentrated in specific orientation bands. The data are consistent
with a model that integrates uniformly over all orientations, even when this strategy is sub-optimal.
When a viewer moves relative to the environment, the visual ral frequency bands that lie on a common plane (Fig. 1c)810. The
image projected onto the retina changes accordingly. Within small population response of a family of detectors built in this way can
regions of the retina and for short durations, this motion is com- determine the presence and velocity of local translating patterns,
monly approximated as a two-dimensional translation. The field and the responses of individual detectors are well matched to the
of velocity vectors associated with each such region is referred to behaviors of a subset of neurons in visual area MT10.
as optic flow. Physiological and psychophysical experiments Alternatively, the visual system could use an adaptive integra-
demonstrate that mammalian cortex uses mechanisms sensitive tion rule, selectively combining only those first-stage detectors
to local image motion1. These mechanisms are generally consid- that are tuned to the spatial structure of the image. For example,
ered to form the neural substrate for representing optic flow. one type of model makes initial robust estimates of the compo-
Neurons in primary visual cortex perform the first stage of cor- nents of a pattern by selecting those detectors responding to the
tical motion processing. They are selective for both the spatial ori- stimulus but ignoring those responding to noise11. These one-
entation and speed of the visual input within a spatially localized dimensional estimates are then combined to produce an estimate
region. Because of their orientation specificity, however, these neu- of the two-dimensional pattern velocity that is most consistent
rons are impaired by an ambiguity commonly known as the aper- with the measured one-dimensional velocity components. In gen-
ture problem: each neuron can signal the speed only of motion eral, such adaptive pooling rules produce more efficient motion
perpendicular to the orientation to which it is tuned2; thus, it is detectors than fixed pooling rules because the detector is better
insensitive to the velocity component parallel to this orientation. It matched to the signal. Human observers adapt their spatial pool-
is suggested that a second stage of processing, commonly associ- ing to improve the detection of static images12. Thus, it is plausi-
ated with visual area MT, computes an unambiguous representa- ble that they may do the same for moving images.
tion of local pattern velocity by selectively combining the outputs We designed a set of psychophysical detection experiments to
of the first-stage detectors25. Although such an integration stage rigorously test the predictions of three models for pooling of spec-
seems largely consistent with psychophysical and physiological tral energy: (1) a planar power detector that sums the signal ener-
data, the precise form of the combination rule remains unclear. gy over all orientations in a fixed planar region of spatiotemporal
Here we describe psychophysical experiments designed to test the frequency, (2) an adaptive planar power detector that sums ener-
predictions of several general combination rules. gy only over those regions of a plane containing the signal and
The simplest combination strategy is to integrate responses of (3) an adaptive unrestricted power detector that can sum energy
all those first-stage mechanisms consistent with a particular two- over arbitrary regions of spatiotemporal frequency. Experimental
dimensional velocity2. It is convenient to describe this construction stimuli were constructed by summing dynamic random noise
in the three-dimensional spatiotemporal frequency domain, where patterns that were band-limited using spatiotemporally oriented
the spectral energy of a rigidly translating image is concentrated on filters analogous to those used to describe first-stage motion
an oblique plane (Fig. 1a and b). The orientation of this plane mechanisms. Each of the three models makes distinct subthresh-
uniquely specifies the translational velocity (both speed and direc- old summation predictions for the detection of such stochastic
tion) of the luminance pattern6. First-stage motion detectors in the signals embedded in white noise. The planar power detector pre-
mammalian visual system can be described as computing the spec- dicts optimal summation performance when signal energy is dis-
tral energy within limited bands of spatiotemporal frequency7. Thus, tributed uniformly over all spatial orientations in a plane.
a pattern-velocity detector may be constructed by summing the However, this model predicts suboptimal performance when the
weighted outputs of first-stage mechanisms tuned to spatiotempo- signal energy is concentrated in a subset of planar orientation

articles
lus in the plane). A two-alterna-

a
(a) b
(b) c
(c)
t t t tive forced choice task was used
to measure detection thresholds
for each of the three stimuli.
Stimuli were presented in two
consecutive intervals: one con-
tained the signal added to
y broadband white noise and the
y y were asked to judge which inter-
other only white noise. Subjects
x
x x
val contained the signal. Thresh-
olds were estimated by fitting a
psychometric function to the
Fig. 1. A translational motion detector. (a) Space-time luminance pattern of an image translating to the right. This is a data and selecting the contrast
representation of the intensity information in the retinal image (the xy plane) over time (t). The rightward motion can
energy producing 81.1%
be inferred from the oriented pattern on the xt face. (b) The Fourier amplitude spectrum of the luminance pattern,
represented by the intensity of points in a three-dimensional spatiotemporal frequency domain. Non-zero Fourier detectability. Different stimulus
amplitudes are constrained to lie on a plane through the origin. The orientation of this plane uniquely specifies the types were presented in distinct
direction and speed of translation. (c) Construction of a translation detector10, illustrated in the Fourier domain. Pairs experimental sessions, and sub-
of balls symmetric about the origin indicate the Fourier amplitude spectra of band-pass filters whose peak frequencies jects were told which stimulus
lie in the plane. A translation detector can be constructed by summing the squared outputs of such filters. type was being used. Before data
collection, subjects were given
several hours practice for each
stimulus type over the course of
bands or distributed over non-coplanar regions of spatiotempo- several days, until performance saturated (see Methods).
ral frequency, because the model sums both signal and noise in To compare detection performance across stimuli, we assumed
these cases. Similarly, the adaptive planar power detector predicts that performance for the component stimulus reflects the detec-
optimal performance for any signal whose energy is concentrated tion efficiency of a band-limited, first-stage motion-energy detector
in a plane, regardless of orientation content, but sub-optimal per- mechanism. (The component signal has a bandwidth similar to
formance for stimuli composed of non-coplanar distributions of psychophysically optimal bandwidths13). Specifically, we assumed a
signal energy. Finally, the adaptive unrestricted power detector model in which subjects made detection judgments for the com-
predicts optimal performance for any distribution of signal ener- ponent stimulus based on output of a filter matched to the com-
gy, with no improvement in performance for planar distributions. ponent signal; this output was then corrupted by additive internal
noise. The internal noise represented the added uncertainty creat-
RESULTS ed by neural noise, filter/signal mismatch and central-decision noise.
In the first experiment, we tested these predictions by measuring We estimated internal noise by comparing subjects thresholds with
detection thresholds for three types of spatially localized stochastic that of an ideal observer for the component stimulus. Assuming
signals embedded in white noise. Signals were generated by filter- approximately constant internal noise across all stimulus condi-
ing spatiotemporally white noise with different sets of band-pass fil- tions, we derived optimal summation predictions for the plaid and
ters. Component stimuli (Fig. 2a) had spectral energy confined to planar stimuli. These predictions reflected the performance of ideal
a single pass-band. Plaid stimuli (Fig. 2b) had two component observers that sum energy over only the spatiotemporal frequency
bands on the same plane. Planar stimuli (Fig. 2c) were constructed bands containing the signal in a stimulus and were corrupted by a
using a set of component filters arranged in a ring on an oblique constant level of internal noise. (Such a model gives the common-
plane (so that signal energy was uniformly distributed over an annu- ly cited square-root law for contrast summation.)
a Component b Plaid c Planar

Fig. 2. Filter sets and examples of their correspond-
ing signals. Top row, level surfaces (65% of peak
t t t
response) of the three different filter sets used to
generate stimuli. Bottom row, spacetime lumi-
nance patterns of signals produced by passing spa-
tiotemporal Gaussian white noise through the y y y
corresponding filter sets. (a) The component stim-
ulus, constructed from a spatially and temporally x x x
band-pass filter. The xy face of the stimulus shows
structures that are spatially band-pass and oriented
along the x-axis. The orientation of structures on
the xt face indicates rightward motion. (b) The t t t
plaid stimulus, constructed from two component
filters lying in a common plane. The xy face of the
stimulus shows a mixture of spatial structures with
dominant orientations close to the y-axis. (c) The
planar stimulus, constructed from a set of ten
y y y
component filters lying in a common plane. The
stimulus is spatially band-passed and isotropic (x-y
face) and moves rightward (xt face). x x x

articles
2 0.16
Fig. 3. Detection thresholds. (a) Per
Component Component formance of three subjects for the three
Plaid Plaid stochastic signals of Fig. 2. Threshold
Planar 0.14 Planar
signal-to-noise ratio (SNR) for 81.1%
detectability. SNR is calculated as the
1.5 0.12 ratio of the signal power to the back-
ground-noise power. Heavy black lines
Threshold SNR
0.1 indicate predictions for ideal summa-
Efficiency
tion, derived from the component con-
1 0.08 dition thresholds. (b) Detection
efficiencies for the three stimulus types.
0.06 Efficiencies are plotted in proportions,
with 1.0 reflecting perfect performance
0.5 0.04 (matching that of an ideal observer
tuned to the signal structure for that
particular stimulus). The differences
0.02
between the efficiencies of the pattern
stimuli (plaid and planar stimuli) and the
0 0 component stimulus provide a quantita-
a
(a) PS ML AS (b
b) PS ML AS tive measure of summation of the pat-
Subject Subject terns components.
Figure 3a shows subjects detection thresholds for the compo- ence between those for plaid and non-planar-triplet stimuli
nent, plaid and planar stimuli. The heavy black lines on the plaid (Fig. 4b). Figure 4c shows subjects efficiencies for detecting each
and planar plots represent predictions of ideal summation derived of the three signals used in the experiment. Detection efficiency
from the component data. The plaid data show little or no sum- was significantly better for the planar triplet than for the plaid,
mation, whereas thresholds for the planar stimulus fell well with- implying improved summation for the triplet stimulus (Fig. 4c).
in the range predicted by perfect summation. Figure 3b shows an Whereas detection efficiencies for the planar triplet were sub-
alternative characterization of the results. We plotted subjects stantially better than for the plaid stimulus, subjects efficiencies
detection efficiencies for each of the three types of stimuli. Detec- for the planar triplet remained lower than for the planar stimu-
tion efficiencies were computed by comparing subjects detection lus in experiment 1 (Fig. 3). Again, this is consistent with the pla-
thresholds with those of ideal observers optimally tuned to the nar power-detector model, which predicts progressively better
signals contained in the stimuli. (Note that the ideal observers are summation as signal energy is distributed more broadly over a
different for each of the three stimulus types.) Efficiency provides plane in spatiotemporal frequency, attaining ideal summation
a measure of the fraction of stimulus information effectively used only when the energy is distributed uniformly over the plane
by subjects in performing the detection task. Because perfect sum- a stimulus that best matches the putative detector.
mation is ideal for these experiments, efficiency is also a measure
of summation. The significantly lower detection efficiencies for DISCUSSION
the plaid stimulus than for the component stimulus reflect poor We can summarize the qualitative results of experiments 1 and
summation for the plaid. Conversely, equal detection efficiencies 2 as follows: subthreshold summation of signal-contrast energy
for the planar stimulus and the component stimulus reflect near- improved as the energy was distributed more and more broadly
ly optimal summation for the planar stimulus. around different orientations in a plane in spatiotemporal fre-
Of the three detection models described above, the data are quency, but did not improve when the energy was distributed
clearly most consistent with the planar power-detector model. into non-planar regions of spatiotemporal frequency. The lack
The experiment, however, only compared summation perfor- of summation for non-planar regions agrees with previous stud-
mance for two types of pattern stimuli. Motivated by the obser- ies suggesting a lack of generic summation for components mov-
vation that the planar power-detector model predicts ing in opposite directions1416. Stated more plainly, detection
progressively improved summation when signal energy is dis- efficiencies improve as more orientations are added to a moving
tributed more broadly across spatial orientations in a plane, but pattern, as long as the motion of those orientation components
not when the energy is distributed out of a plane, we ran a second are consistent with the velocity of the pattern. Of the three detec-
experiment. We generated a new set of stochastic signals to test tion models proposed in the introduction, these results are most
this prediction. Signals for the stimuli were created by passing consistent with the planar power-detector model.
spatiotemporal white noise through three configurations of fil- We extended the analysis one step further by comparing per-
ters (Fig. 4a). The first was a plaid signal, similar to the plaid used formance of a particular planar power detector on the five dif-
in experiment 1. The second was a planar triplet, created by ferent pattern stimuli used in the experiments with that of the
adding a component band to the plaid, in the same plane as the subjects. The detector optimally summed energy over the band of
plaid. The third was a non-planar triplet, created by adding a frequencies contained in the planar stimulus from experiment 1
component band out of the plane of the plaid. Detection thresh- (using a matched power filtersee Methods). We assumed this
olds were measured using the same method described for exper- detectors output to be corrupted by levels of internal noise esti-
iment 1. The planar power-detector model predicts improved mated from subjects detection thresholds for the component
summation for the planar triplet relative to the plaid, but no stimulus in experiment 1. Figure 5 shows the model predictions
improved summation for the non-planar triplet. for the five pattern stimuli used. Given the assumptions built
Detection thresholds for the planar triplet were, as predicted, into the model concerning the exact spatiotemporal frequency
lower than for the plaid, whereas there was no significant differ- band covered by the planar power detector, the match is sur-

articles
t tiotemporal frequency. Such mechanisms have

a
(a) b
(b) c
(c) been plausibly linked to the subclass of pat-
Plaid Plaid tern neurons in the middle temporal (MT)
In plane In plane visual area10. The model may be contrasted
Off plane Off plane
x y 0.6 0.12 with adaptive velocity-estimation models
dynamically tuned to the spatial-orientation
Plaid 0.5 0.1 content of the image. Our subthreshold sum-
t mation results favor the fixed pooling model
Threshold SNR
0.4 0.08 over adaptive models.
Efficiency
As with all psychophysical studies, we had
0.3 0.06
to make certain assumptions to draw con-
x y clusions about the underlying neural mech-
0.2 0.04
anisms. In particular, we assumed that
In plane
t 0.1 0.02
detection of single component signals is
mediated by responses of first-stage energy
0 0 mechanisms, but that detection of compound
PS ML PS ML signals is mediated by second-stage respons-
y Subject Subject es. In addition, we assume that observers can
x learn (through training) to attend to the most

Off plane sensitive mechanism for each stimulus.
Fig. 4. Experiment 2. (a) Filters used to generate stimuli. (b) Threshold SNRs for detecting the Although our experiments strongly argue
three types of stimuli in. Black bands indicate the predictions of ideal summation, based on sub- for second-stage mechanisms that pool over
jects detection thresholds for the component stimulus used in experiment 1. (c) Detection effi- orientation, further processing is required to
ciencies for the three stimulus types. The results show a large increase in efficiency for the account for the full range of motion-percep-
in-plane triplet over the plaid and a decrease for the off-plane triplet, indicating increased summation phenomena. For example, although pla-
tion with additional power on a common plane. nar pooling models are capable of
representing more than one motion in a
given local region10, they are unable to pro-
vide a consistent account of whether stimuli
prisingly good. That is, not only do the qualitative results follow appear to move coherently. Additional computations are neces-
the predictions of the planar power-detector model, but the quan- sary to account for such phenomena. Athough our experimen-
titative results are well fit by a pre-defined instantiation of the tal results suggest fixed pooling over orientation, they do not
model (without fitting the parameters of the model to the data). preclude the possibility of adaptive integration over spatial posi-
Our experiments critically test the predictions of a simple model tion, time or spatial frequency (scale). Adaptive integration over
for velocity-tuned motion detectors in the mammalian visual sys- spatial position, for example, would be a useful strategy for pre-
tem. The model detectors sum outputs of early motion-energy venting combination of information across occlusion bound-
detectors with different preferred orientations and spatiotempo- aries. The experiments described herein may be naturally
ral frequency bands that intersect a common plane. The result is extended to examine the integration of first-stage mechanisms
a family of velocity-tuned detectors that measure the stimulus ener- over these other domains.
gy in fixed, orientationally broadband, planar regions of spa-
METHODS
Planar Stimuli in the experiment subtended 2.2 of visual angle, had a
0.8 Plaid 1 duration of 0.43 s and were displayed in 12-bit precision on a cal-
Plaid 2 ibrated gamma-corrected gray-scale monitor. Observers fixated a
0.7 In plane small black dot 0.2 above the stimulus and were given several
Off plane
hours practice for each stimulus type over the course of several
0.6 days until performance saturated. Observers were given auditory
Threshold SNR
feedback after every trial in all experimental sessions. Only one

0.5 signal type was used in a given 1-h experimental session. Average
signal energy was varied using the method of constant stimuli. The
0.4 average background noise energy was fixed at 0.26 degrees2 . s.
Thresholds were estimated from the maximum-likelihood fits
of two-parameter Weibull functions, for which the threshold
0.3
parameter is a measure of signal-to-noise power at 81.1% cor-
rect for a 2AFC task. Standard errors were computed both from
0.2
0.1
Fig. 5. Threshold SNRs for detecting the five types of pattern stimuli
replotted from experiments 1 and 2, where Plaid 1 in the legend denotes
0 the plaid from the first experiment and Plaid 2 from the second. Plaid 1 dif-
PS ML fers from Plaid 2 in that its energy is more diffusely spread over frequency.
Black bands indicate the predictions of a planar filter, based on subjects
Subject detection thresholds for the component stimulus used in experiment 1.

articles
the Hessian of the likelihood function of the fit and using a Monte bers, Es En is Gaussian distributed, and hence can be charac-
Carlo sampling method. terized by its mean and variance. The mean and variance of
The component filter has an amplitude spectrum given by Es En are simply the sums of the means and variances of the
C(r, , ) = R(r) |cos()|9 |cos( )|9 in spherical fre- 2(2) variables at each frequency. Computing the means and vari-
quency coordinates. The frequency radius r is given by ances of the 2(2) variables that have been scaled by |F(x, y,
0
r = x2 + y2 + (t/2.1)2 (cycles per degree, cycles per degree, t)|2 (s2 |F(x, y, t)|2 + |N(x, y, t)|2) on the signal interval
cycles per s). The elevation shift = 36.9 acts to rotate the filter and |F(x, y, t)|2 (|N(x, y, t)|2) on the noise interval, and
letting FmSn = |F(x, y, t)|m |N(x, y, t)|n, then with a lit-
0
out of the spatial frequency plane into a plane corresponding to a
1.93 degrees per s translation downward. The radial frequency func- tle algebra, the mean and variance E s E n are given by
tion R is a smooth box function whose transitions from 0 to 1 are id = 2s2F2F2 and 2id = 8(s4F4F4 + 2 s2F4F2 N + 2 F4N2), respec-
given by 1/2 cycle of a cosine function. This function has low/high tively. Then p(Es En > 0) = 1 (0, id, 2id), where is the
cutoffs of (0.49, 7.6) cycles per degree and the cosine transition cumulative Gaussian distribution.
regions have widths of 1.45 cycles per degree. The plaid signal was To compute summation predictions, we modeled component
constructed by adding two component filters, rotated by +70 and performance as ideal but with the variance given by 2id + 2internal,
70 within the common plane. The planar signal was construct- so that p(E s E n > 0) = 1 (0, id, 2id + 2internal). Setting
ed using a sum of 10 component filters, rotated by multiples of 18 p(Es En > 0) = 81.1 and plugging in the subjects threshold into id
and constrained to lie in a common plane. For the second experi- and 2id, we solved for 2internal. To generate predictions, the
ment, the filters were modified to decrease the spectral spread. They 2internal was then added to the ideal variance for each of the other
have the form BP(r, , ) = Wr(r)W()W(), where Wx is signal types, and a predicted threshold for 81.1% was computed.
a smooth box function on the variable x. Wr had a transition region Predictions for the planar model were generated in a similar man-
width of 1.45 and low/high frequency cutoffs of (0.49, 7.6) cycles ner. If we represent the planar filter amplitude spectrum as
per degree visual angle. W and W had transition widths of 8 and P( x, y, t) then the planar model computes the energies
high/low cutoffs that spanned 36. These filters were rotated to lie in Epl = |P(x, y, t)|2 |S(x, y, t)|2 on both intervals. The deci-
the same positions as the component and plaid filters for the plaid sion variable mean and variance are then given by id = 2s2P2F2
and planar triplet. The off-plane non-planar triplet component, and 2id = 8(s4P4F4 + 2s2P4F2 N + 2P4N2). Threshold predictions
however, had an elevation shift corresponding to a slower 0.26 were generated from these expressions using the internal noise
degrees per s translation downward. estimated from the observers component stimulus performance.
Ideal observers for the task are matched filter power detec-
tors17. Let F(x, y, t) denote the spectrum of the signal nor- ACKNOWLEDGEMENTS
malized to one, and N(x, y, t) the (complex) spectrum of the P.S. was supported by an NIH training grant, D.C.K. was supported by a grant
noise. Let |X(x, y, t)|2 = X(x, y, t) . X(x, y, t)* denote from the NIH and E.P.S. was supported by a Sloan Fellowship, an NSF CAREER
the inner product of the complex function with its complex con- grant and the Sloan Program in Theoretical Neurobiology at New York University.
jugate. Then the spectrum of the received signal |S(x, y, t)|2 is
s2 |F(x, y, t)|2 + |N(x, y, t)|2 on the signal plus noise inter- RECEIVED 11 AUGUST; ACCEPTED 11 NOVEMBER 1999
val and |N(x, y, t)|2 = N2 on the noise-alone interval. The best
possible performance can be achieved by computing the energy 1. Nakayama, K. Biological image motion processing: a review. Vision Res. 25,
E = |F(x, y, t)|2 |S(x, y, t)|2 dx dy dt, over both signal 625660 (1985).
plus noise and noise intervals and using the difference of these 2. Adelson, E. H. & Movshon, J. A. Phenomenal coherence of moving visual
patterns. Nature 300, 523525 (1982).
values as a decision variable12. In practice, the set of frequencies 3. Albright, T. D. Direction & orientation selectivity of neurons in visual area MT
are quantized, and the integral is computed as a sum over of the macaque. J. Neurophysiol. 52, 11061130 (1984).
1.3 105 frequency samples. 4. Movshon, J. A., Adelson, E. H., Gizzi, M. S. & Newsome, W. T. in Experimental
Brain Research Supplementum II: Pattern Recognition Mechanisms (eds. Chagas,
Efficiencies are approximately given by the ratio of ideal to C., Gattass, R. & Gross, C.) 117151 (Springer, New York, 1986).
human squared energy thresholds. This approximation was cor- 5. Morgan, M. J. Spatial filtering precedes motion detection. Nature 355, 344346
rected through simulations of the ideal, and the simulation val- (1992).
6. Watson, A. B. & Ahumada, A. J. in Motion: Perception & Representation (ed.
ues are shown in the figures. Performance of the matched filter Tsotsos, J. K.) 110 (Association for Computing Machinery, New York, 1983).
depends on the probability, p(Es En > 0), that the difference in 7. Adelson, E. H. & Bergen, J. R. Spatio-temporal energy models for the
energies measured by the filter on two intervals is greater than perception of motion. J. Opt. Soc. Am. A 2, 284299 (1985).
8. Heeger, D. J. Model for the extraction of image flow. J. Opt. Soc. Am. A 4,
zero. To compute this probability, we derive an approximate 14551471 (1987).
expression for the distribution of Es En. For the stimuli used, the 9. Grzywacz, N. M. & Yuille, A. L. A model for the estimate of local image velocity
amplitude spectrum of the signal S(x, y, t) consists of inde- by cells in the visual cortex. Proc. R. Soc. Lond. B Biol. Sci. 239, 129161 (1990).
10. Simoncelli, E. P. & Heeger, D. A model of neuronal responses in visual area MT.
pendent (real and imaginary) Gaussian-distributed random vari- Vision Res. 38, 743761 (1998).
ables at each frequency with zero mean and variance given by 11. Derrington, A. M. & Suero, M. Motion of complex patterns is computed
either s2 |F(x, y, t)|2 + |N(x, y, t)|2 on the signal interval from the perceived motions of their components. Vision Res. 31, 139149
(1991).
or |N(x, y, t)|2 on the noise-alone interval. The power spec- 12. Kersten, D. Statistical efficiency for the detection of visual noise. Vision Res. 27,
trum, as the product of the signal amplitude with its complex 10291043 (1987).
13. Watson, A. B. & Turano, K. The optimal motion stimulus. Vision Res. 35,
conjugate, consists of Chi-squared random variables with two 325336 (1994).
degrees of freedom 2(2) at each frequency, which are scaled by 14. Levinson, E. & Sekuler, R. The independence of channels in human vision
the squared amplitude spectrum of the signal. Because the ener- selective for direction of movement. J. Physiol. (Lond.) 250, 347366 (1975).
15. Watson, A. B., Thompson, P. G., Murphy, B. J. & Nachmias, J. Summation and
gy is the sum of an extremely large number of 2(2) random vari- discrimination of gratings moving in opposite directions. Vision Res. 20,
ables (one for each frequency), weighted by the product of the 341347 (1980).
filter and the signal power spectrum, we can use the law of large 16. Dobkins, K. R. & Teller, D. Y. Infant contrast detectors are selective for direction
of motion. Vision Res. 36, 281294 (1996).
numbers to generate an extremely good bound on performance 17. van Trees, H. L. Detection, Estimation, and Modulation Theory Part III
(<< 1% error from true performance). By the law of large num- (Wiley, New York, 1971).

articles
Touch can change visual slant

perception
Marc O. Ernst1, Martin S. Banks2 and Heinrich H. Blthoff1
1 Max-Planck-Institute for Biological Cybernetics, Spemannstr. 38, D-72076 Tbingen, Germany

2 Vision Science Program & School of Optometry, University of California, Berkeley, California 94720-2020, USA
Correspondence should be addressed to M.O.E. (marc.ernst@tuebingen.mpg.de)
The visual system uses several signals to deduce the three-dimensional structure of the environment,
including binocular disparity, texture gradients, shading and motion parallax. Although each of
these sources of information is independently insufficient to yield reliable three-dimensional
structure from everyday scenes, the visual system combines them by weighting the available
information; altering the weights would therefore change the perceived structure. We report that
haptic feedback (active touch) increases the weight of a consistent surface-slant signal relative to
inconsistent signals. Thus, appearance of a subsequently viewed surface is changed: the surface
appears slanted in the direction specified by the haptically reinforced signal.
Combining different sources of three-dimensional information These weights affect the appearance of the visual world and
helps the interpretation of ambiguous signals and reduces the the manner in which we interact with it. Here we asked how the
effects of measurement noise. The method of combination has nervous system determines weights applied to different estima-
been successfully examined by using cue-conflict protocols in tors. There are at least three ways in which the weights could be
which signals are manipulated independently15. In one study, determined. First, weights could be fixed for a given viewing sit-
for example, subjects adjusted the three-dimensional shape of uation and individual and not subject to change through feed-
convex surfaces with elliptical cross-sections until they appeared back (although they may have been changeable during infancy
cylindrical (circular cross section)4. The shape was specified by and childhood to compensate for anatomical changes in the sen-
conflicting disparity (three-dimensional information provided sory apparatus, for example). Second, weights could be adjust-
by the differences in images between the two eyes) and texture ed by comparing a given estimators output with those of other
gradients (three-dimensional information given by projection of estimators and with feedback from motor behavior. Third,
a surface with statistically regular markings onto the retina). weights could be determined directly from statistical measures
Affected by both the disparity and texture signals, the shape set- of estimator outputs. For example, if the output of one estimator
tings were well described by a linear weight combination rule: for a given viewing situation fluctuated less over time than that of
S = wtSt + wdSd (1) another estimator, the formers weight could be increased rela-
St and Sd are the outputs of shape estimators with weights wt and tive to the latters. We examined the first two of these possibilities.
wd that use texture and disparity signals, respectively. Each shape Specifically, we asked whether the weights assigned to different
estimator may use a variety of input signals. For example, the estimators can be changed by haptic feedback (sensation of touch
disparity-based estimator uses inputs of horizontal disparity, ver- generated by active, exploratory hand and finger movements)
tical disparity and their gradients1. One cannot distinguish a that is consistent with one estimator and not another.
change in weight from a change in estimator gain, so for our pur- In numerous reported visualhaptic interactions, visual appear-
poses a weight change will refer to both possibilities. Equation 1 ance is unaffected by haptic feedback. For example, subjects grasp-
is a maximum-likelihood estimator if the estimators, St and Sd, ing a square viewed through an optical device that distorts the
are Gaussian distributed and statistically independent and the image to make it appear rectangular see and feel the square as a
weights, wt and wd, are equal to the estimators inverse variances, rectangle13. In this case, perception is determined entirely by the
normalized to add up to one68. visual information; thus it is an example of visual capture1317.
This linear weighting scheme suffices for understanding many Perhaps this finding is not surprising, because the visual stimulus
phenomena in visual perception15,911. A statistically sensible clearly specifies perceived shape. Consistent with this idea, touch
method for estimation would give high weight to more informative can affect appearance of an impoverished visual stimulus18. These
estimators and low weight to uninformative ones5, because such visualhaptic studies sought concurrent perceptual effects, but we
weighting should yield more stable percepts58,12. The weights must were primarily concerned with persistent effects of haptic feed-
depend on viewing conditions because, for example, the informa- back on visual perception. To increase the probability of observ-
tion content of the disparity signal decreases as a function of dis- ing a positive influence of haptic feedback, we presented visual
tance4,12. Furthermore, experiments show that the weights vary stimuli with a range of possible interpretations (Fig. 1).
from one individual to another for a given viewing situation. For Many studies of visuomotor adaptation show that humans can
example, some subjects consistently give more weight to disparity, adapt behaviorally to changes in the mapping between the envi-
whereas others preferentially weight texture gradients4,12. ronment and sensory signals15,1921. Helmholtz first demonstrat-

articles
Fig. 1. Two examples of the cue-conflict stimuli. The

texture signal to surface slant is the distortion of the
regular grid patterns marked on the surfaces. The dis-
a parity signal to surface slant gives the right-eye minus
left-eye differences in the images. The reader can visu-
alize the stimuli approximately as they appeared by
cross-fusing the two panels in each row (directing the
left eye to the right half-image and the right eye to the
left half-image). The black crosses, not present in the
experimental stimuli, were added to help the reader
fixate (one fused cross should appear when viewing
the stimuli binocularly). (a) For a texture-specified
slant of 0 and a disparity-specified slant of about 30
(at a viewing distance of 20 cm), most viewers per-
ceive a slant between 0 and 30, because both signals
contribute to the perceived slant. (b) For a texture-
specified slant of about 30 and a disparity-specified
slant of 0 (viewing distance of 20 cm), most people
again perceive a slant between 0 and 30.
b
test (Fig. 2), using a setup involving an image reflect-

ed from a mirror and a force-feedback device
(Fig. 3). The pre- and post-tests were identical and
were used to determine the weights. We presented
visual planes with texture- and disparity-specified
slants that differed by an angle, . Subjects indicat-
ed the slant of the plane that was perceived as fron-
toparallel. The subjects settings were quantified by
, the texture-specified angle at which subjects per-
ceived the surface to be frontoparallel, and were used
to estimate the weights assigned to texture and dis-
parity. We looked for differences between pre- and
post-test judgments due to haptic feedback provid-
ed during the training phase.
ed this plasticity of the visuomotor system by displacing the entire There were three experimental conditions, differing only in
visual field with prismatic spectacles and studying subsequent the training phase. In the texture-feedback condition, haptic feed-
effects on reaching and other sensorimotor behavior22. Initial reach- back was consistent with the slant specified by the texture gradi-
es are strikingly inaccurate, but they improve rapidly. After remov- ent; the non-reinforced disparity gradient varied randomly from
ing the spectacles, errors are observed in the opposite direction. trial to trial. In the disparity-feedback condition, haptic feedback
Thus, adaptation persists after the mapping is restored to its orig- was consistent with the slant specified by disparity, and the tex-
inal state. The observed adaptation could theoretically occur in ture-specified slant varied randomly. The third condition was a
visual perception, in proprioception of body parts or in the trans- control in which no haptic feedback was given. Subjects viewed a
lation from visual to motor coordinates.
Determining the mechanism for adap-
tation proves difficult15,20.
We used a novel approach to find
that haptic feedback can alter subse-
quent visual percepts by changing the
weights given to particular sources of
visual information. Specifically, we
found that haptic feedback to surface
orientation subsequently affects the Fig. 2. Experimental design. The pre- and post-tests were purely visual tasks. The visual plane had differ-
appearance of the surface (when hap- ent texture- and disparity-specified slants. The texture-specified slant is represented by the gray grids and
tic feedback is no longer available). the disparity-specified slant by the light gray planes. The angle between the two specified slants is the con-
We chose slant estimation because it flict angle . The perceived slant is represented by dark gray planes. is the slant specified by the texture
involves a limited set of well under- gradient when the plane (with texture- and disparity-specified slants generally differing) was perceived as
stood environmental and sensory sig- frontoparallel. The decrease in between pre- and post-test indicates an increase in texture weight.
During the pre- and post-test, the plane was clipped with a circular window with a diameter of 13. The
nals and because it allows the use of projected shape of the window at the cyclopean eye was consistent with the texture-specified slant. Use
a simple psychophysical task. As such, of the circular window reduced the probability of slant reversals25. The haptic training phase occurred
the results are easier to interpret than between the pre- and post-tests and consisted of visual and haptic stimulation. The cube, surface and tar-
previous work on visual adaptation. gets could all be seen and felt. During the training phase, the boundary of the plane was a 13 16 rec-
Experiments were conducted in three tangle when it was normal to the line of sight. The shape of the boundary was consistent with the
phases, pre-test, training phase and post- texture-specified slant. (The rectangular boundary increased the salience of the texture signal.)

articles
Fig. 3. Experimental setup. The visual stimulus was generated on a cath-

ode ray tube (CRT) and viewed in a mirror that obscured the subjects
hand. A dot on the screen indicated finger position. The haptic stimulus
was created by a force-feedback device that had six degrees of freedom
and could apply force in the three translation directions. The three-
dimensional position of the fingertip was monitored, and an appropriate
force was applied to the tip when it reached the position of the simu-
lated haptic objects, creating a compelling sensation of touching a solid
surface and cube. The haptic stimulus included realistic simulation of
gravitys effect on the cubes motion across the plane. The haptic stimu-
lus also simulated friction: it was low between the cube and plane and
higher between the fingertip and cube. To view the stimulus, subjects
line of sight was pitched 69 downward from earth horizontal. The
planes slant varied about a fixed axis pitched 21 up from horizontal.
The line of sight was thus perpendicular to the rotation axis. A chin and
forehead rest limited head movements.
recorded sequence of visual images seen in a previous session of

the texture-feedback condition. This condition tested whether

the weight changes were attributable to the visual experience
rather than haptic feedback.
Here we report that haptic feedback can change subsequent
visual percepts by changing the weights given to different sources
of visual information. pre- and post-test. The texture-specified slant at the slant setting ()
is plotted as a function of the conflict angle (). If settings were based
RESULTS entirely on the texture gradient, the data would lie on a horizontal
Figure 4 shows the results for the two haptic-feedback conditions line through 0. If they were based entirely on disparity, the data
and the control. The top row shows the average slant settings in the would lie on the diagonal gray line. Settings were actually interme-
diate, indicating that both signals affected performance. If
No feedback the visual system used sensorimotor feedback to adjust the
Texture feedback Disparity feedback
(control) weights assigned to different slant estimators, we should
observe a change in settings due to the haptic training.
Indeed, post-test settings were slightly, but consistently, dif-
Texture-specified slant
ferent from pre-test settings. When haptic feedback was con-

at setting ()
sistent with the texture-specified slant (left), the average

weight assigned to the disparity-based estimator decreased
from 0.70 to 0.58 (middle left); 9 of the 10 subjects showed
a decrease in this weight (lower left). This means that a stim-
ulus that appeared frontoparallel in the pre-test appeared
slanted in the direction specified by texture in the post-test.
Conflict angle ()
Fig. 4. Results for the haptic-feedback and control conditions.

Disparity weight
Left, results when the texture-specified slant was reinforced

during haptic training. Middle column, results when the dispar-
ity-specified slant was reinforced during training. Right, results
from the control condition in which no haptic feedback was
provided. Top average slant settings in the pre- and post-test.
The texture-specified slant at the slant setting () is plotted as
a function of the conflict angle (). Individual subjects slant set-
tings were corrected for constant baseline shifts. If settings
were based entirely on the texture gradient, the data would lie
on a horizontal line through 0; if based entirely on binocular
disparity, the data would lie on the diagonal gray line. Error
Change in weight
bars represent standard errors across subjects. Middle row,

average disparity weight for the pre-test (black) and post-test
(gray). Weights were calculated from linear regression fits to
the data in the upper row. Error bars represent regression
errors. Bottom, change in weight for each subject (disparity
weight in pre-test minus disparity weight in post-test). Subjects
are ranked by magnitude of the effect, with subjects initials
below the corresponding bar. The order along the horizontal
axismost positive weight change to most negativewas
Individual subjects determined separately for each panel.

articles
When haptic feedback was consistent with the disparity-specified for the particular stimulus properties under examination. When
slant (middle column), the average weight assigned to the disparity- the attended property is size, length, curvature or angular separa-
based estimator increased from 0.60 to 0.66 (middle panel); 7 of the tion, vision dominates the percept, even when touch is in con-
10 subjects showed an increase in this weight and 2 showed a small flict1317, demonstrating visual capture. The visual system is well
decrease (lower middle). designed for making fine discriminations of size, length, curvature
A statistical test on the pre- and post-test weights revealed a and angular separation, so the reliability of visual estimates of those
significant interaction: the increase in texture weight was greater properties is high. The haptic system is not capable of such fine
after texture-reinforced training than after disparity-reinforced discriminations23, so its reliability is lower than that of the visual
training (F1,9 = 10.597, p < 0.01). These results show that 3045 estimates. A sensible estimation method would give greater weight
minutes of haptic feedback cause an upweighting of the reinforced to the more reliable estimate; that is, the final estimate would be
estimator and a corresponding change in visual appearance. The dominated by vision and barely affected by conflicting haptic
weight change was small and variable across subjects. information. However, when the stimulus property under exami-
The average weight assigned to the disparity-based estimator in nation is coarseness of texture, vision and touch influence the per-
the no-feedback control experiment (Fig. 4, right column) revealed cept 18. Visual and haptic discrimination of coarseness
an average disparity-based estimator weight of 0.56 in the pre-test (just-discriminable percentage change) are comparable18, so the
and 0.57 in the post-test (middle right), values that are statistically reliabilities of visual and haptic estimates are roughly equivalent.
indistinguishable (F1,9 = 0.142, p = 0.715). Thus, haptic feedback Both estimates should, therefore, be able to influence the final esti-
is indeed necessary for the weight change to occur. mate20. In the experiment reported here, the visual information
We conducted a second control experiment (data not shown), was by design ambiguous (texture and disparity specified differ-
in which the texture- and disparity-specified slants were congruent ent slants), so it is reasonable to assume that additional haptic
during the training phase. Haptic feedback was consistent with information consistent with one of the slants would be used.
both visual signals, and the weights did not change. Thus, weight In summary, we have shown that haptic feedback provides
changes occur only when haptic feedback is consistent with one one means for adjusting the weights given different sources of
signal and not the other. visual information. An interesting byproduct is that haptic feed-
back consistent with one source of information can cause a
DISCUSSION change in subsequent visual appearance: a surface with conflict-
We showed that haptic feedback can change the subsequent visu- ing disparity and texture signals will look more like the haptical-
al perception of surface slant. Specifically, when subjects are given ly reinforced slant than it did before.
haptic stimulation consistent with the texture-specified slant of a
visual stimulus, their subsequent visual percepts are closer to the
METHODS
texture slant than they were before training. The opposite occurs Nine naive subjects and M.O.E. completed the three conditions of the
when they are given haptic stimulation consistent with the dis- experiment. All had normal or corrected-to-normal vision and good
parity-specified slant. We interpreted the effect as a change in the stereo vision. Informed consent was obtained.
weights given to different slant estimators, in this case, to texture- The visual stimulus was a textured plane viewed binocularly. The tex-
based and disparity-based estimators. ture was a regular grid mapped onto the plane and then displayed on a
The weight changes were small. When the texture-specified CRT using OpenGL routines. The stimulus was viewed stereoscopically
slant was reinforced, the texture weight increased from pre- to with liquid-crystal shutter glasses. The apparent position of the plane
post-test by an average of 40%. When the disparity slant was rein- was below the mirror (Fig. 3). The surroundings were dark. At the begin-
ning of each stimulus presentation, a bright white field was presented to
forced, the texture weight decreased by 15%. There are two obvi-
maintain light adaptation and make the CRT frame less visible.
ous reasons for the small magnitudes. First, although we The stimulus planes slant was specified by binocular disparity and a
separately manipulated the texture- and disparity-specified slants texture gradient (Fig. 1). In the pre- and post-tests, the disparity- and tex-
during training, the values were correlated (r = 0.59). Second, ture-specified slants differed (conflict angle = 0, 10, 20 or 30, tilt = 0).
calibration of a sensory system is probably best served by slow Despite the differing slant specifications, the stimulus always looked like
changes in response to large amounts of data. If brief exposures to a single plane with a well-defined slant. Each stimulus was presented for
new correlations among information sources resulted in large 500 ms, and subjects reported whether the planes left or right side appeared
changes in calibration, a sensory system could become an unsta- closer. The slant was varied (holding constant) according to an adaptive
ble estimator of environmental information because its estimates staircase procedure (seven reversals) to estimate the surface slant that
would be subject to the vagaries of the particular sequence of appeared frontoparallel (). The estimates were determined from an aver-
age of the last four reversals. From these estimates, we calculated the weights
recent events. Thus, the 1540% change we observed is a rea-
assigned to the disparity- and texture-based slant estimators (Fig. 4).
sonable response to 3045 minutes of altered experience. The haptic-training phase occurred between the pre- and post-tests. Again,
The perceptual effect we observed is persistent: remnants of the disparity- and texture-specified slants were manipulated separately. Dur-
the effect last at least 24 hours, as subjects weights had not ing training, haptic stimulation was provided along with the visual displays.
returned to their initial values by the second day of testing. The The haptic stimulus was created by a haptic force-feedback device (PHAN-
evidence for this is the difference in the pre-test weights in the ToM Model 1.5, http://www.sensable.com). The virtual haptic stimulus
texture- and disparity-feedback conditions (Fig. 4). This persis- consisted of the plane and a small cube lying flat on the plane. The right
tence is striking, given that subjects presumably received sub- index finger was placed in a thimble-like holder attached to the device. To
stantial haptic feedback from normal interaction with their create the haptic sensation, the three-dimensional position of the right index
finger was monitored in real time. When the fingertip reached the simulat-
environment during the intervening period.
ed haptic plane or cube, the device applied an appropriate force on the fin-
It is interesting to consider our observations in light of known gertip, creating a compelling sensation of touching solid objects (the
visualhaptic interactions. As stated earlier, the majority of this stationary plane and movable cube)23. By these means of creating stimuli,
work failed to observe an effect of touch on visual appearance. we could independently manipulate visual and haptic stimulation.
Why did we succeed where others had not? A sensible answer During haptic training, a subject used the index finger to move the cube
comes from analyzing information provided by vision and touch along the plane to the target by pressing down on top of the cube and mov-

articles
ing the finger in the desired direction as rapidly and accurately as possi- 3. Buckley, D. & Frisby, J. P. Interaction of stereo, texture and outline cues in
ble. All subjects experienced a convincing sensation of pushing a real cube the shape perception of three-dimensional rigids. Vision Res. 33, 919933
(1993).
along a real slanted surface. We chose the task of moving a small cube along 4. Johnston, E. B., Cumming, B. G. & Parker A. J. Integration of depth
the surface because it is visually and haptically demanding, thus requiring modules: Stereopsis and texture. Vision Res. 33, 813826 (1993).
more attention than simply touching a surface. The fingertips three-dimen- 5. Landy, M. S., Maloney, L. T., Johnston, E. B. & Young, M. Measurement and
sional position was represented by a small dot. The cube, plane, targets modeling of depth cue combination: in defense of weak fusion. Vision Res.
and dot were always visible. A new stimulus with a new cube, slants and 35, 389412 (1995).
6. Clark, J. J. & Yuille, A. L. Data Fusion for Sensory Information Processing
target positions appeared after successful target acquisition. Systems (Kluwer, Boston, 1990).
The three experimental conditions differed only in the relationship 7. Blthoff, H. H. & Yuille, A. L. in Attention Performance XVI: Information
between haptic feedback and visual stimulation during the training phase. In Integration in Perception and Communication (eds. McClelland, J. & Inui,
the texture-feedback condition, subjects received haptic feedback consistent T.) 3 4970 (MIT Press, Cambridge, Massachusetts, 1996).
8. Knill, D. C. Discrimination of planar surface slant from texture: human and
with the planes texture-specified slant. In other words, the haptic plane had ideal observers compared. Vision Res. 38, 16831697 (1998).
the three-dimensional coordinates of the planes texture slant, and the hap- 9. van Ee, R., Banks, M. S. & Backus, B. T. An analysis of binocular slant
tic cube had the coordinates of a cube lying on that plane. In the disparity- contrast. Perception 28, 11211145 (1999).
feedback condition, they received haptic feedback consistent with the 10. Rogers, B. J. & Bradshaw, M. F. Disparity scaling and the perception of
disparity-specified slant. In the no-feedback condition, subjects viewed the frontoparallel surfaces. Perception 24, 155179 (1995).
11. Bradshaw, M. F., Glennerster, A. & Rogers, B. J. The effect of display size on
same sequence of visual displays seen in the texture-feedback condition, but disparity scaling from differential perspective and vergence cues. Vision
received no haptic feedback. To sustain attention on the visually specified Res. 36, 12551264 (1996).
slant, they had to indicate the perceived slant and time of target acquisition. 12. Backus, B. T. & Banks, M. S. Estimator reliability and distance scaling in
In the two feedback conditions, the slant specified by the reinforced stereoscopic slant perception. Perception 28, 217242 (1999).
visual signal and the haptic stimulus was 10 or 20. The slant of the 13. Rock, I. & Victor, J. Vision and touch: An experimentally created conflict
the two senses. Science 143, 595569 (1964).
non-reinforced signal differed randomly from the reinforced one by 10, 14. Gibson, J. J. Adaptation, after-effect, and contrast in the perception of
20 or 30. There were 240 trials in the training phase, which took curved lines. J. Exp. Psychol. 16, 131 (1933).
3045 minutes to complete. The subjects completed all three conditions 15. Rock, I. The Nature of Perceptual Adaptation (Basic Books, New York,
with a minimum of one day between conditions. The order of the two 1966).
16. Festinger, L., Burnham, C. A., Ono, H. & Bamber, D. in Journal of
feedback conditions was counterbalanced, and the no-feedback condi- Experimental Psychology Monograph Vol. 74, No. 4 (American
tion was presented last. No feedback other than the haptic stimulation Psychological Association, Washington., 1967).
in the two reinforced conditions was given in the experiment. 17. Welch, R. B. & Warren, D. H. in Handbook of Perception and Human
Subjects were questioned upon completing the entire experiment. Performance (eds. Boff, K. R., Kaufman, L. & Thomas, J. P.) 25.125.36
None realized that the texture- and disparity-specified slants differed or (Wiley, New York, 1986).
18. Lederman, S. J. & Abbott, S. G. Texture perception: studies of intersensory
that the haptic stimulation was consistent with one and not the other. organization using a discrepancy paradigm, and visual versus tactual
psychophysics. J. Exp. Psychol. Hum. Percept. Perform. 7, 902915 (1981).
ACKNOWLEDGEMENTS 19. Harris, C. S. Perceptual adaptation to inverted, reversed, and displaced
vision. Psychol. Rev. 72, 419444 (1965).
This research was supported by NSF grant (DBS-9309820) and AFOSR grant 20. Welch R. B. Perceptual Modification: Adapting to Altered Sensory
(93NL366) to M.S.B. and by the Max-Planck Society. The authors thank Robert Environments (Academic, New York, 1978).
Cooper and Mike Landy for comments. 21. Stratton, G. M. Vision without inversion of the retinal image. Psychol. Rev.
4, 341360, 463481 (1897).
22. von Helmholtz, H. Handbuch der Physiologischen Optik Vol. III (Leopold
RECEIVED 5 AUGUST; ACCEPTED 18 NOVEMBER 1999 Voss, Leipzig, 1867).
23. Walk, R. D. Intersensory perception and sensory integration. (Plenum,
New York, 1981).
1. Backus, B. T., Banks, M. S., van Ee, R. & Crowell, J. A. Horizontal and vertical 24. Massie T. & Salisbury, J. K. in Dynamic Systems and Control 1994 Vol. 55
disparity, eye position, and stereoscopic slant perception. Vision Res. 39, (ed. Radcliffe, Clark J.) 295301 (American Society of Mechanical
11431170 (1999). Engineers, New York, 1994).
2. Blake, A., Blthoff, H. H. & Sheinberg, D. Shape from texture: ideal observer 25. Gillam, B. Stereoscopic slant reversals: a new kind of induced effect.
and human psychophysics. Vision Res. 33, 17231737 (1993). Perception 22, 10251026 (1993)

articles
Different lateral amygdala outputs

mediate reactions and actions
elicited by a fear-arousing stimulus
P. Amorapanth, J. E. LeDoux and K. Nader
Center for Neural Science, New York University, New York, New York 10003, USA
Correspondence should be addressed to K.M. (karim@cns.nyu.edu)
Fear-arousing stimuli elicit innate reactions and can reinforce acquisition of new responses. We tested
whether mechanisms mediating these conditioned stimulus (CS) properties were isomorphic or disso-
ciable within the amygdala. Rats trained on a fear-conditioning task (CS paired with footshock) were
then trained on an escape-from-fear task (EFF) in which the CS reinforced a locomotor response
terminating the CS. Lateral nucleus (LA) lesions blocked acquisition of both conditioned freezing
responses and the CSs reinforcement of a new response in the EFF task. Central nucleus (CE) lesions
blocked conditioned freezing but not the EFF, whereas basal nucleus (B) lesions blocked the EFF but
not conditioned freezing. Thus, activation of the LA by a CS seems to trigger conditioned reactions
via CE and conditioned aversion via B activation, reduction of which reinforces new actions .
Stimuli in the presence of painful or threatening stimuli acquire tions813. In contrast, several lines of investigation in which LA
aversive properties and, as a result, later elicit fear reactions. This and basal nucleus of the amygdala were lesioned together sug-
form of learning is extensively studied using a Pavlovian fear- gest that the LA or B is involved in the ability of a CS to serve as
conditioning protocol. In a typical experiment, a tone (condi- a conditioned reinforcer14,15, but it is not clear whether both are
tioned stimulus) is paired with an unconditioned stimulus (US), involved. Further, given that LA projects to CE directly and by
a footshock. Subsequently, the CS elicits unlearned species-typ- way of B16,17, it is important to determine whether the Pavlovian
ical defense reactions in the absence of the US1,2. Thus, fear con- conditioning of fear reactions is mediated by the direct projec-
ditioning does not create fear responses, but instead establishes tion from LA to CE or by way of the projection from LA to B and
the environmental conditions under which innate fear respons- from there to CE. Therefore, we examined the effects of lesions of
es will be expressed. LA, B or CE on the acquisition of both a Pavlovian and an instru-
Although innate fear reactions may be adaptive, ability to take mental conditioned response, with the stimulus that served as
novel actions in threatening situations may also be advantageous. the CS in the Pavlovian task also serving as the conditioned rein-
In the present study we used a modified escape-from-fear3 task forcer in the instrumental task.
to demonstrate mediation of these two kinds of responses by dif-
ferent neural pathways. The task involved two phases. In the first, RESULTS
fear reactions were conditioned to a CS by pairing it with shock. Histology
Subsequently, the animals were placed in a new chamber, where The LA was targeted in 20 rats. Fourteen rats were excluded
the CS was presented. They then learned that an arbitrary because of either insufficient tissue damage to the LA or dam-
response, stepping into the adjoining identical chamber, termi- age that grossly infringed on the B and/or CE. The final LA group
nated the CS. Termination of the CS reinforced the novel action, consisted of six rats. These animals had lesions destroying most
presumably because it decreased the conditioned fear elicited by of the dorsal LA and approximately 75% of the ventral LA
the CS. The CS from the conditioned fear task thus functioned as (Fig. 1b and 2a). The lesions infringed slightly on the dorsal
a conditioned negative reinforcer in the EFF task4. endopiriform nucleus laterally, but spared both B and CE. Fif-
The neural basis for the acquisition of Pavlovian fear condi- teen animals received lesions of CE. Nine of these animals were
tioning is well studied57, but the manner in which it fits into a excluded either because CE was spared or because lesions dam-
broader network of mental and behavioral systems is poorly under- aged LA and/or B. The remaining six animals included in the
stood. Here we begin to integrate the anatomy of fear conditioning analysis had lesions destroying most of both the medial and lat-
with other systems involved in more complex aspects of behavior. eral subnuclei of CE but sparing both LA and B (Fig. 1c and 2b).
Fear conditioning is believed to involve the relay of sensory Of the 16 rats that underwent B lesions, 8 were excluded from
information about the CS first to the lateral nucleus of the amyg- behavioral analyses because the lesions spared most of B or
dala and from there to the central nucleus of the amygdala57. infringed on LA and/or CE. Acceptable lesions in eight animals
The conclusion that these circuits are involved in fear condi- damaged much of B and infringed on the ventral portion of LA
tioning is based on anatomical tracing, lesion, pharmacological as well as the accessory basal nucleus (Fig. 1d and 2c). The dam-
and unit-recording studies. Particularly relevant here, lesions of age to the accessory basal was not consistent and was not evident
LA and CE interfere with the Pavlovian conditioning of fear reac- in all the animals included in this experimental group.

articles
a Sham lesion b LA lesion indicated that the B and sham groups had comparable freezing
scores (p > 0.05). Rats in groups CE and LA had similar scores
(p > 0.05). Freezing scores in both the sham and B-lesioned groups
were significantly different from scores in groups LA and CE
(p < 0.05 for all). These results demonstrate that the LA and CE are
necessary for the acquisition of freezing behavior, but the B is not.
The findings from the instrumental learning task (EFF) over-
lapped and diverged with those from the Pavlovian task. Lesions of
the LA and B blocked acquisition of the EFF task (Fig. 4b), but
c CE lesion d B lesion lesions of the CE had no effect on this task. An ANOVA compar-
ing the groups (sham, B, CE, LA) with the number of escape
responses over 4 blocks of 5 trials revealed a significant interaction
(F9,87 = 4.3, p < 0.05). Furthermore, there was a main effect of both
group (F3,29 = 5.6, p < 0.05) and blocks (F3,87 = 3.1, p < 0.05). New-
man-Keuls post-hoc analyses with group and block as variables
revealed that all groups demonstrated comparable scores on
block 1 (p > 0.05). Furthermore, only rats with either sham or CE
lesions acquired the EFF task. Specifically, the scores for sham and
Fig. 1. Coronal unilateral images of representative lesions of amygdala subdi- CE rats from block 4 were significantly different from their respec-
visions. (a) Sham; (b) lateral amygdala; (c) central amygdala; (d) basal amygdala. tive scores on block 1 (p < 0.05). Conversely, rats with lesions of the
B or LA did not differ between blocks 1 and 4 (or any of the other
blocks; p > 0.05). These data demonstrate that the LA and B are
Behavioral results necessary for the acquisition of the EFF task, but that the CE is not.
Unlesioned animals receiving paired CSUS presentations froze
significantly more than rats presented with unpaired stimuli DISCUSSION
(F1,11 = 19.1, p < 0.05; Fig. 3a). Behavior of the animals also dif- LA is believed to be the sensory interface, the locus where CS
fered with regard to the acquisition of the EFF (Fig. 3b). A one- information enters the amygdala8,1621. Damage to LA should
way analysis of variance (ANOVA) comparing the number of therefore disrupt amygdala-dependent responses elicited by sen-
escape responses over five trial blocks as a repeated measure with sory stimuli. Indeed, here and in other studies, damage to LA pre-
the CSUS relationship (paired versus unpaired) revealed a sig- vents conditioning of fear reactions to a CS, as well as expression
nificant interaction between these two variables (F3,33 = 3.2, of previously conditioned fear reactions9,11. We also found that
p < 0.05). There was no main effect, however, of either block of lesions of LA disrupted the ability of the same CS to serve as a
trials or CSUS relationship (p > 0.05). Post-hoc Newman-Keuls conditioned reinforcer of a novel instrumental response, a con-
analyses with group and block as variables revealed that the escape ditioned fear-motivated action. Furthermore, we have demon-
scores for paired animals were significantly higher on block 4 than strated that these two properties of a CS are differentially affected
on block 1 (p < 0.05). Recipients of unpaired stimuli, however, by lesions placed in different targets of LA within the amygdala.
demonstrated comparable escape responses on the first and last Specifically, lesions of the CE blocked the acquisition of the freez-
training blocks (p > 0.05). Thus, the acquisition of both the freezing response elicited by the CS in the Pavlovian task, but had no
ing and EFF were contingent on association of the CS with the US. effect on the ability of the CS to reinforce acquisition of the instru-
mental response in the EFF task. Conversely, lesions of the B
Effects of amygdala lesions blocked reinforcing effects of the CS in the EFF task, but had no
Lesions of the LA and CE blocked the acquisition of Pavlovian fear effect on acquisition of freezing to the CS in the Pavlovian task.
conditioning, as measured by freezing to the CS (Fig. 4a). Howev- Different outputs of LA thus seem to mediate the ability of the CS
er, lesions confined to the B had no effect on freezing. An ANOVA to elicit fear reactions and to reinforce novel actions.
comparing the freezing scores with groups revealed a significant The effects of CE lesions on the Pavlovian conditioned freezing
effect of group (F3,29 = 10.7, p < 0.05). Newman-Keuls post-hoc tests response is consistent with previous reports that similar lesions
block freezing as well as other
Table 1. Coordinates relative to the skull surface at bregma (mm) and current duration. conditioned fear responses, such
Posterior Medial/lateral Ventral Current duration (s) as fear-potentiated startle, auto-
LA 2.3 5.1 8 9 nomic and endocrine changes
3.2 5.3 8.1 10
and alterations in pain reactivi-
ty913,22. Thus, with regard to fear
4.0 5.5 8.1 11
conditioning, the CE seems to be
the motor output for the expres-
B 2.1 4.9 9.1 12 sion of various hardwired reac-
2.8 4.9 9.3 15 tions elicited by the Pavlovian
3.3 5.3 9.2 15 CS57. The failure of damage to B
4.2 5.3 9.3 15 to affect conditioned freezing
suggests that, although LA pro-
CE 1.8 4.4 8.4 12 jects to CE directly and by way of
2.3 4.4 8.4 12 B16,17, the direct projection is suf-
ficient to mediate Pavlovian fear
2.8 4.4 8.4 12
conditioning.

articles
Rostral Middle Caudal
a
Fig. 2. Camera lucida drawings of the extent of (a) lateral (b) central and (c) basal amygdala lesions over three rostrocaudal planes. The stippled area
represents the area of neuronal loss. LA, lateral nucleus; B, basal nucleus; Bi, intermediate division; Bpc, parvicellular division; AB, accessory basal
nucleus; ABmc, magnocellular division; ABpc, parvicellular division; BAOT, bed nucleus of the accessory olfactory tract; CO, cortical nucleus; M, medial
nucleus; PAC, periamygdaloid cortex; Ce, central nucleus; AHA, amygdalohippocampal area; AHAl, lateral division; I, intercalated nuclei; Pir, piriform
cortex; DEn, dorsal endopiriform nucleus; opt, optic tract; cp, cerebral peduncle.
The ability of a CS paired with a US to reinforce acquisition of age to the CE on acquisition of conditioned reinforcement shows
a new task defines the CS as a conditioned reinforcer23. Lesions of that CE, required for the expression of Pavlovian fear conditioning,
B blocked the conditioned reinforcing properties of an aversively does not mediate all of the effects of the fear-conditioned CS.
conditioned CS. Past studies using appetitive USs found that com- Although intra-amygdala pathways have been anatomically
bined lesions of B and LA interfere with the ability of an appetitive- mapped in great detail16,17,26, the paucity of studies explicitly exam-
ly conditioned CS to support acquisition of a new task14. ining contributions of component nuclei within the amygdala to
Furthermore, combined lesions of LA and B block the acquisition fear conditioning prevents us from elaborating further on the
of appetitive second-order conditioning24. Although B is required intra-amygdala circuitry over which information is relayed. Most
for the establishment of the new response by the conditioned rein- studies focus on lesions of LA, B or CE, so the effects of lesions
forcer, B is not necessarily the locus of motor control nor the locus of of other areas of the amygdala are generally not known. Howev-
plasticity underlying the association of the stimulus and response. er, preliminary studies aimed at addressing this question suggest
B is instead more likely the source of the conditioned reinforcement. that, whereas damage to LA and CE disrupt freezing, damage to
By way of anatomical interactions between the B and striatal response other areas (medial, cortical, B and accessory amygdala nuclei)
control circuits, conditioned reinforcement established in the amyg- have no effect on auditory fear conditioning as measured by freez-
dala may reinforce novel motor responses14,25. Taken together, the ing (P. Majidishad, D. G. Pelli, & J. E. L., personal communica-
various results suggest some overlap in the mechanisms that enable tion). The contribution of these and other amygdala areas to
a CS to reinforce new learning after being conditioned with either contextual fear conditioning and to conditioned reinforcement
an appetitive or an aversive US. In addition, the lack of effect of dam- is not yet known.

articles
Fig. 3. Behavioral results from

animals that received paired ver- a b
sus unpaired CS and US presenta-
tions. (a) Acquisition of freezing.
Number of escape responses

(b) Escape-from-fear protocol.
Bars represent s.e.
Freezing (s)
The behavioral effects of B
and CE lesions are not due to
general sensory, motor or
learning impairments. Thus, as
animals with lesions of the B
showed normal freezing behav-
ior in the presence of the CS, Paired Unpaired
Blocks of five trials
they were competent to per-
ceive the CS, form an association and perform a conditioned ing scores) and also to those of the sham control group that received
response. By the same logic, the ability of rats with lesions of the CE paired training. This demonstrates that the acquisition of the active
to acquire the EFF task demonstrates that they could perceive the response (stepping into the alternate environment) was not the
CS, form an association and perform an appropriate motor response. default behavior of the rats in this situation and was instead gradu-
To damage LA, B and CE, we used electrolytic lesions, which ally acquired over the training session.
affect both cells in the region and fibers of passage. As a result, it is It has been suggested that avoidance responses, such as running
difficult to conclude whether the effect was due to damage to the away in an active-avoidance protocol, are actually Pavlovian in
cells within the lesion or to fibers of cells in other parts of the brain; nature and not instrumental2, and that the particular response of
we chose this method because it was the only one practical for an animal in a given task depends on the situation28. Thus, it is pos-
inducing damage confined to the LA or B, with little or no involve- sible that the EFF task measured a hard-wired reactive response
ment of the other. However, the dissociation we observed between (running) that is only expressed when an exit is available. This is
effects of electrolytic lesions of LA and B, combined with studies unlikely for a number of reasons. First, running is thought to be an
using fiber-sparing excitotoxic lesions covering both the LA and unconditioned reaction to shock, and no shock was delivered in
B11,15,27, allow us to conclude that cells in LA mediate the acquisi- the EFF protocol29. Second, even when animals approach an exit,
tion of conditioned fear reactions to the CS, and cells in B the acqui- presentation of a previously fear-conditioned stimulus elicits freez-
sition of the conditioned reinforcing properties of the CS. That is, ing and not running away30. Third, if performance in the EFF task
given that excitotoxic lesions of LA together with B prevent the reflected Pavlovian rather than instrumental conditioning, then CS
acquisition of both conditioned fear and conditioned reinforce- presentation should have elicited maximal escape responses on the
ment, the dissociation produced by electrolytic lesions of the indi- first block of trials. Instead, they showed a gradual acquisition curve
vidual nuclei shows the necessity of cells within these structures. over trials, typical of what is observed in instrumental training.
An alternative interpretation of our findings is that the acquisition Thus, the EFF is unlikely to be sampling a Pavlovian response.
of the EFF task depends on the absence of freezing. That is, rats can- Our distinction between the B mediating active responses and
not perform the active response as long as they are freezing, so elim- the CE mediating reactive responses is similar to distinctions made
ination of freezing by damaging some area (like CE) allows the rat to by others 15, though important differences also exist. These
step into the other chamber. If freezing were simply competing with researchers found that combined damage of LA and B interferes
stepping into the adjacent chamber, then all groups with minimal with the ability of an aversive CS to reinforce a new response, where-
freezing should have demonstrated high escape responses on the as lesions of CE, but not combined LA/B lesion, interfere with the
very first block. This was not observed. Animals with lesions of the conditioned reactive responses, and suggested that the amygdala
CE or LA both had reduced freezing responses to the CS in the con- has two learning systems, the LA/B (for conditioned reinforcement)
ditioned fear task. At the same time, in the EFF task, escape scores and the CE (for conditioned reactions)15. However, as noted above,
of both the LA- and CE-lesioned groups to the CS on block 1 were our finding that lesions restricted to the LA block Pavlovian con-
comparable to those of rats with B lesions (which had normal freez- ditioned reactions replicates several past studies8,11. At present, the
reason for the discrepant find-
a b Central
ings is not apparent, although a
Sham number of possibilities have
Number of escape responses
Basal been proposed31,32. The most

Lateral
parsimonious explanation of the
various results is that the LA is
Freezing (s)
required for the Pavlovian con-

ditioning of the CSUS associa-
tion, and that projections of LA
Fig. 4. The effects of various

amygdala lesions on the acquisition
of freezing (a) and the escape-from
Sham Basal Central Lateral fear task (b). Group names refer to
Blocks of five trials the specific subdivision destroyed.
Bars represent s.e.

articles
Fig. 5. Models of fear conditioning. (a) The traditional model of informa-

a tion flow within the amygdala mediating Pavlovian fear conditioning. CS
CS information and US information converge in the lateral nucleus (LA). Information is
then relayed to the central nucleus (CE) either directly or via the basal
nucleus (B). The CE mediates motor output of the various conditioned
fear responses. (b) The new model of how fear is organized in the amyg-
dala. CS and US convergence still occurs in the LA. The direct pathway
US information
from the LA to the CE is sufficient to mediate reflexive or reactive
responses, whereas the indirect pathway is not necessary. The broken
arrow from the B to the CE indicates that sufficiency of the indirect path-
way to mediate these behaviors has not been established. In addition,
there is a second, qualitatively different output of the amygdala: projec-
tions from the LA to the B mediate the ability of a CS to reinforce the
acquisition of new or active responses.
actions, to learn the action at the time through trial and error or
Reactive
to devise a plan of action on the spot. By studying how different
responses actions develop in response to a single eliciting stimulus, we will
CS information
be able to further explore aspects of these cognitiveemotional
b interactions in the brain.

In conclusion, the findings of the present study demonstrate
that the neural pathways and mechanisms mediating the ability of
a CS to elicit reactive responses and to reinforce new responses can
US information
be dissociated (Fig. 5). We suggest that the CE is part of a reactive
response output system that responds to stimuli that predict dan-
ger by eliciting hard-wired defense responses, whereas the B is part
of an active fear output system through which new responses are
acquired to minimize exposure to a noxious stimulus, as proposed
by two-process theory4,3638. Furthermore, these two systems are
outputs of a common learning system involving the LA and its
afferent input from sensory systems processing the CS.
Reactive
responses METHODS
Active Surgery. Sprague-Dawley rats weighing 300350 g were injected with
responses 0.15 mg per kg atropine intraperitoneally (ip), anesthetized with Nem-
butal (i.p., 45 mg per kg) and placed in a stereotaxic frame. Electrolytic
to CE mediate conditioned reactions whereas projections to B medi- lesions were made by passing positive current (0.5 mA) at each site
ate conditioned reinforcement of action. Regardless of the differ- through a monopolar electrode insulated with epoxy to within 200 m of
ences, the two studies are concordant in suggesting different outputs the tip. Coordinates for the various lesion sites and current times are
of the amygdala mediate conditioned reinforcement and condi- given in Table 1. Sham animals received the identical treatment with the
following two exceptions. First, the electrodes were placed 0.51 mm
tioned reflexive responses.
dorsal to the target structure. Second, no current was passed through
Our findings also address in two ways the controversy over the electrode. After surgery, animals were allowed to recover undisturbed
whether the amygdala is actually required for fear conditioning or for one week before commencement of behavioral procedures.
whether it simply modulates the formation of a memory in some
other area33,34. First, given that damage to LA disrupts both the Apparatus. The apparatus consisted of three distinct chambers differ-
Pavlovian and instrumental leaning tasks and that damage to B ing in their dimensions, odor, lighting and location. Fear conditioning
and CE disrupt only one task, LA is probably the key site of learn- occurred in a standard fear-conditioning chamber (Chamber A, Model
ing that supports the expression of fear responses (through the E10-10, Coulbourn Instruments, Lehigh Valley, Pennsylvania). Tests for
CE) and conditioned aversion (through the B), the reduction of freezing were performed in a distinct room using standard operant
boxes, housed in sound attenuated shells (Chamber B, ENV-001, Med
which reinforces a new response. Second, the LA/B, but not the
Associates, Georgia, Vermont).
CE, mediates the ability of emotional events to modulate the con- The avoidance chamber used for the modified EFF task consisted of a
solidation of memory35. This is in contrast to traditional models rectangular Plexiglas box containing two identical compartments sepa-
of fear conditioning which posit the CE as the motor output for rated by a sliding guillotine door that was manually controlled by the
fear conditioning. Although these two phenomena may be medi- experimenter. Both compartments had black Plexiglas floors. The avoid-
ated by distinct processes, our finding that the B can act as an inde- ance chamber itself was situated on the floor of the same room in which
pendent output of the fear system somewhat reconciles these two fear conditioning took place. The drop pan beneath the chamber con-
positions. It is possible that the conditioned aversive properties tained animal bedding. The sole source of illumination was a red light
mediated by the B contribute to memory consolidation. For this bulb centered over the top of the apparatus. The avoidance chamber was
to be the case, however, memory consolidation would have to be a made completely of Plexiglas backed by black construction paper. The
neutral retaining box was of the same type as the animals home cage and
conditioned rather than an unconditioned phenomenon.
was lined with animal bedding.
In real time, the reactive and active systems are probably
engaged over different temporal scales. The reactive responses Behavioral procedures. All animals were habituated to all three cham-
are automatic, instantaneous responses to danger. In contrast, to bers in a counterbalanced manner for 15 min over three consecutive
perform an action, it is necessary to call upon previously learned days.

articles
Phase 1: fear conditioning. Either paired or unpaired fear-conditioning 8. LeDoux, J. E., Cicchetti, P., Xagoraris, A. & Romanski, L. M. The lateral
trials were administered to animals on days 1 and 2. Paired fear-condi- amygdaloid nucleus: Sensory interface of the amygdala in fear conditioning.
J. Neurosci. 10, 10621069 (1990).
tioning consisted of 5 presentations a day for 2 consecutive days of a 9. LeDoux, J. E., Iwata, J., Cicchetti, P. & Reis, D. J. Different projections of the
10-kHz, 75-dB tone for 20 s that coterminated with a 0.5-s, 0.5-mA scram- central amygdaloid nucleus mediate autonomic and behavioral correlates of
bled footshock (Chamber A). The mean intertrial interval was 120 s with conditioned fear. J. Neurosci. 8, 25172529 (1988).
a range from 90 s to180 s. Unpaired procedures entailed the same number 10. Walker, D. L. & Davis, M. Double dissociation between the involvement of
of CS and US presentations as the paired groups; however, the two never the bed nucleus of the stria terminalis and the central nucleus of the amygdala
in startle increases produced by conditioned versus unconditioned fear.
occurred within 60 s of one another. J. Neurosci. 17, 93759383 (1997).
11. Campeau, S. & Davis, M. Involvement of the central nucleus and basolateral
Phase 2: test. On day 3, individual rats were placed in chamber B and complex of the amygdala in fear conditioning measured with fear-
allowed to explore for 2 min. At this time, the CS was presented for 20 s, potentiated startle in rats trained concurrently with auditory and visual
conditioned stimuli. J. Neurosci. 15, 23012311 (1995).
and the amount of time spent freezing was recorded. Freezing was defined 12. Roozendaal, B., Koolhaas, J. M. & Bohus, B. Central amygdala lesions affect
as immobility with the exception of respiratory-related movement. behavioral and autonomic balance during stress in rats. Physiol. Behav. 50,
777781 (1991).
Phase 3. In each trial, the animal was placed in the compartment of the 13. Kapp, B. S., Frysinger, R. C., Gallagher, M. & Haselton, J. Amygdala central
avoidance chamber designated as the start side and allowed to explore. nucleus lesions: effect on heart rate conditioning in the rabbit. Physiol. Behav.
23, 11091117 (1979).
After 10 s, the experimenter opened the door, and the 10-s pre-CS peri- 14. Everitt, B. J. & Robbins, T. W. in The Amygdala: Neurobiological Aspects of
od began, followed by a 30-s presentation of the CS. If the animal stepped Emotion, Memory and Mental Dysfunction (ed. Aggleton, J. P.) 401430
into the alternate environment, the CS was immediately terminated, and (Wiley, New York, 1992).
the rat was allowed to spend 1.5 min in the safe side before being moved 15. Killcross, S., Robbins, T. W. & Everitt, B. J. Different types of fear-conditioned
to a neutral holding cage for the intertrial interval (1.5 min). If an ani- behavior mediated by separate nuclei within amygdala. Nature 388, 377380
(1997).
mal failed to make a response before CS termination, then it was imme- 16. Pitknen, A., Savander, V. & LeDoux, J. L. Organization of intra-amygdaloid
diately moved to the holding cage for the inertrial interval3. The animals circuitries: an emerging framework for understanding functions of the
were given a total of 20 training trials over 2 days. amygdala. Trends Neurosci. 20, 517523 (1997).
17. Amaral, D. G., Price, J. L., Pitknen, A. & Carmichael, S. T. in The Amygdala:
Neurobiological Aspects of Emotion, Memory, and Mental Dysfunction (ed.
Experimental groups. Two groups of rats did not receive any surgery. These Aggleton, J. P.) 166 (Wiley-Liss, New York, 1992).
animals were run to verify that the acquisition of both the EFF task and freez- 18. McDonald, A. J. Cortical pathways to the mammalian amygdala. Prog.
ing were contingent on an associative as opposed to non-associative rela- Neurobiol. 55, 257332 (1998).
tionship between the CS and US. The paired group (n = 7) received paired 19. Turner, B. & Herkenham, M. Thalamoamygdaloid projections in the rat: a
fear conditioning, and the second group, Unpaired (n = 6), received unpaired test of the amygdalas role in sensory processing. J. Comp. Neurol. 313,
295325 (1991).
fear conditioning. Both groups then received phase 2 and 3 as above. 20. Mascagni, F., McDonald, A. J. & Coleman, J. R. Corticoamygdaloid and
All surgical groups received paired presentations of the CS and US corticocortical projections of the rat temporal cortex: a phaseolus vulgaris
during fear conditioning and then received EFF training. The group sizes, leucoagglutinin study. Neuroscience 57, 697715 (1993).
after discarding animals with unacceptable lesions, were sham (n = 10), 21. Romanski, L. M., LeDoux, J. E., Clugnet, M. C. & Bordi, F. Somatosensory
and auditory convergence in the lateral nucleus of the amygdala. Behav.
B (n = 8), CE (n = 6) and LA (n = 6). Neurosci. 107, 444450 (1993).
22. Helmstetter, F. The amygdala is essential for the expression of conditioned
Histology. Upon completion of the behavioral studies, sham and lesioned hypoalgesia. Behav. Neurosci. 106, 518528 (1992).
animals were given an overdose of Nembutal (45 mg per kg), and were 23. Mackintosh, N. J. Conditioning and Associative Learning (eds. Broadbent, D.
perfused with physiological saline followed by 10% buffered formalin. E. et al.) (Oxford Univ. Press, New York, 1983).
24. Hatfield, T., Han, J.-S., Conley, M., Gallagher, M. & Holland, P. Neurotoxic
The brains were stored in 30% sucrose/formalin solution and sectioned lesions of basolateral, but not central, amygdala interfere with Pavlovian
on a cryostat at 60 m. Every other section was collected on a subbed second-order conditioning and reinforcer devaluation effects. J. Neurosci. 16,
slide and stained with cresyl violet. Sections were examined and images 52565265 (1996).
digitally captured under bright-field microscopy using Stereo Investiga- 25. Robbins, T. W., Cador, M., Taylor, J. R. & Everitt, B. J. Limbic-striatal
interactions in reward-related processes. Neurosci. Biobehav. Rev. 13, 155162
tor (v.3.16, MicrobrightField).
(1989).
26. Par, D., Smith, Y. & Pare, J. F. Intra-amygdaloid projections of the basolateral
and basomedial nuclei in the cat: phaseolus vulgaris-leucoagglutinin
ACKNOWLEDGEMENTS anterograde tracing at the light and electron microscopic level. Neuroscience
This research was supported in part by National Institute of Mental Health 69, 567583 (1995).
27. Maren, S. Overtraining does not mitigate contextual fear conditioning
grants MH46516, K02 MH00956 and R37 MH38774 to J.E.L. and a HFS
deficits produced by neurotoxic lesions of the basolateral amygdala.
Fellowship to K.N. The work was also supported by a grant from the W.M. Keck J. Neurosci. 18, 30883097 (1998).
Foundation to New York University. The authors would like to thank T.J. 28. Bolles, R. C. & Fanselow, M. S. A perceptual-defensive-recuperative model of
fear and pain. Behav. Brain Sci. 3, 291323 (1980).
Matthews for discussions. 29. Fanselow, M. S. Conditional and unconditional components of post-shock
freezing. Pavl. J. Biol. Sci. 15, 177182 (1980).
30. Fanselow, M. S. & Lester, L. S. in Evolution and Learning (eds. Bolles, R. C. &
RECEIVED 15 JULY; ACCEPTED 18 NOVEMBER 1999 Beecher, M. D.) 185211 (Erlbaum, Hillsdale, New Jersey, 1987).
31. Nader, K. & LeDoux, J. E. Is it time in invoke multiple fear learning system?
Trends Cog. Sci. 1, 241244 (1997).
1. Blanchard, R. J. & Blanchard, D. C. Defensive reactions in the albino rat. 32. Killcross, A. S., Robbins, T. W. & Everitt, B. J. Response from Killcross,
Learn. Motivat. 2, 351362 (1971). Robbins and Everitt. Trends Cog. Sci. 1, 244246 (1997).
2. Bolles, R. C. Reinforcement, expectancy, and learning. Psychol. Rev. 79, 33. Fanselow, M. S. & LeDoux, J. E. Why we think plasticity underlying Pavlovian
394409 (1972). fear conditioning occurs in the basolateral amygdala. Neuron 23, 229232
3. McAllister, W. R. & McAllister, D. E. in Aversive Conditioning and Learning (1999).
(ed. Brush, F. R.) 105179 (Academic, New York, 1971). 34. Cahill, L., Weinberger, N. M., Roozendaal, B. & McGaugh, J. L. Is the
4. Miller, N. E. Studies of fear as an acquirable drive: I. Fear as motivation and amygdala a locus of conditioned fear? Some questions and caveats. Neuron
fear reduction as reinforcement in the learning of new responses. J. Exp. 23, 227228 (1999).
Psychol. 38, 89101 (1948). 35. Cahill, L. & McGaugh, J. L. Mechanisms of emotional arousal and lasting
5. LeDoux, J. E. Emotion circuits in the brain. Annu. Rev. Neurosci. 23 155184 declarative memory. Trends Neurosci. 21, 294299 (1998).
(2000). 36. Mowrer, O. H. & Lamoreaux, R. R. Fear as an intervening variable in
6. Davis, M. The role of the amygdala in emotional learning. Int. Rev. Neurobiol. avoidance conditioning. J. Comp. Psychol. 39, 2950 (1946).
36, 225266 (1994). 37. Mowrer, O. H. Learning Theory and Behavior (Wiley, New York, 1960).
7. Maren, S. & Fanselow, M. S. The amygdala and fear conditioning: has the nut 38. Miller, N. E. in Handbook of Experimental Psychology (ed. Stevens, S. S.)
been cracked? Neuron 16, 237240 (1996). 435472 (Wiley, New York, 1951).

articles
Distinct representations of eye gaze

and identity in the distributed human
neural system for face perception
Elizabeth A. Hoffman and James V. Haxby
Section on Functional Brain Imaging, Laboratory of Brain and Cognition, National Institute of Mental Health, Building 10, Room 4C104, Bethesda, Maryland
20892-1366, USA
Correspondence should be addressed to E.A.H. (elizabeth_hoffman@nih.gov)
Face perception requires representation of invariant aspects that underlie identity recognition as well as
representation of changeable aspects, such as eye gaze and expression, that facilitate social
communication. Using functional magnetic resonance imaging (fMRI), we investigated the perception of
face identity and eye gaze in the human brain. Perception of face identity was mediated more by regions
in the inferior occipital and fusiform gyri, and perception of eye gaze was mediated more by regions in
the superior temporal sulci. Eye-gaze perception also seemed to recruit the spatial cognition system in
the intraparietal sulcus to encode the direction of anothers gaze and to focus attention in that direction.
Whereas the invariant aspects of a face allow one to recognize ity in the STS, then selective attention to gaze should elicit a
who someone is, the changeable aspects of a face can be used to stronger response in that region than attention to identity. In our
infer information about that persons state of mind. Most face second experiment, we examined the strength of responses elicit-
viewing occurs in the context of social communication after iden- ed by passive viewing of faces in which gazes were averted as com-
tity has been established. One example of information gleaned pared with viewing of faces in which gazes were direct (Fig. 1,
from a face during social interaction is the direction of anothers lower panel). Given that the perception of averted gaze causes a
gaze, which can indicate where that persons attention is directed seemingly reflexive attention shift15, passive viewing of faces in
and can be used to similarly direct ones own attention. which gazes are averted should elicit stronger responses in the
Impaired face recognition (prosopagnosia) is associated with IPS, a brain region associated with covert shifts of spatial atten-
ventral temporal lesions13. Within ventral temporal cortex, neu- tion1618, than viewing of faces in which gazes are direct.
roimaging studies of face perception identify a region in the lat-
eral fusiform gyrus (LFG) that responds more to faces than to RESULTS
other objects47. It is unclear, however, whether the perceptual Accuracy and response times were similar for selective attention to
analysis of all aspects of faces is mediated by this region. Other identity and eye gaze (93 6.0%, mean s.d., versus 92 5.0%,
regions that respond preferentially, but less consistently, to faces n.s.; 686 ms 129 ms versus 722 ms 147 ms, n.s.), indicating that
are identified in the lateral inferior occipital gyri (IOG) and in the tasks were well matched on difficulty and attentional demand.
the posterior superior temporal sulcus (STS)4,6,8. The STS region Four bilateral regions that responded more to faces than to
is also associated with the perception of eye and mouth move- control stimuli (scrambled pictures) were identified, in the LFG,
ment9,10 and may be homologous to a region in the superior bank STS, IOG and IPS (Table 1 and Fig. 2). Of the seven subjects who
of the STS in the monkey in which cells respond preferentially showed significant activations in the regions that were the subject
to faces, eye-gaze direction and face expression1114. We hypoth- of our experimental hypotheses, namely the LFG and the STS,
esized that the human face-responsive region in the STS may also bilateral LFG and IOG regions were identified in all subjects, STS
be more involved in the perception of changeable aspects of faces regions were identified in all subjects on the right and in four sub-
and that the human face-responsive region in the LFG may be jects on the left, and IPS regions were identified in all subjects on
more involved in the perception of face identity. We hypothe- the left and in five subjects on the right. As we had hypothesized,
sized further that perception of the direction of eye gaze would selective attention to face identity and eye gaze had opposite effects
elicit activity in regions associated with spatial perception and in the LFG and STS (region attention interaction, p < 0.001 on
spatially directed attention, namely the intraparietal sulcus (IPS). both the right and left), demonstrating that these regions partici-
We conducted two experiments to test these hypotheses. In pate differentially in the representation of the invariant and change-
our first experiment, we tested whether selective attention to able aspects of a face. In the LFG, attention to identity elicited a
identity and eye-gaze direction modulated activity differently in stronger response than did attention to gaze (1.21% versus 0.90%,
these brain regions (Fig. 1, upper panel). If the representation of n = 7, p < 0.001, on the right; 1.23% versus 0.95%, n = 7, p < 0.001,
identity were more dependent on activity in the LFG, then selec- left). By contrast, in the STS, attention to gaze elicited a stronger
tive attention to identity should elicit a stronger response in that response than did attention to identity on the left (0.74% versus
region than attention to eye-gaze direction. Similarly, if the rep- 0.47%, n = 4, p < 0.001), with no significant difference in the same
resentation of eye-gaze direction were more dependent on activ- direction in the right STS (0.79% versus 0.72%, n = 7, n.s.). The

articles
Selective attention
left IPS than did attention to identity (0.99% versus 0.80%, n = 7,
p < 0.001), with a nonsignificant difference in the same direction
in the right IPS (0.88% versus 0.85%, n = 5, n.s.). Similar to the
findings in the STS, the effects of task on responses in the right
Task cue and left IPS did not differ significantly (p > 0.1).
The enhanced response in the IPS while attending to eye gaze
suggested recruitment of the spatial cognition system. The direc-
tion of anothers eye gaze is a potent cue for directing ones own
spatial attention. Shifts of attention in response to the percep-
tion of averted gaze are observed in monkeys, apes and young
Task cue infants1923, and are elicited in adults while fixating on a face, even
when the direction of gaze is task-irrelevant15, suggesting that
our subjects also made covert, reflexive shifts of attention when
the perceived gaze was averted.
It was also possible that the differential response in the IPS
Passive viewing could be attributed to differences in the aperture of spatial atten-
tion. Presumably, the aperture of attention is narrower when
Averted attending to eye gaze than when attending to identity. To rule out
gaze
this alternative explanation, we conducted a second experiment
using two passive viewing conditions in which subjects were not
instructed to attend to a specific facial feature. Subjects passive-
ly viewed faces that had averted gazes in one condition and pas-
sively viewed faces that had direct gazes in the other (Fig. 1, lower
Direct
gaze
panel). The faces in each condition never repeated, so that the
entire stimulus, not just the eye region in the averted gaze con-
dition, varied from one trial to the next. Passive viewing of faces
with averted gazes elicited significantly stronger responses than
did passive viewing of faces with direct gazes in the IPS bilaterally
Fig. 1. Face-perception tasks. The one-back repetition detection tasks (on the right, 0.35% versus 0.16%, n = 5, p < 0.001; on the left,
in experiment 1 are displayed in the upper panel. Subjects attended 0.17% versus 0.06%, n = 7, p < 0.05) and in the left STS (0.39%
selectively to the direction of eye gaze or the identity of each face. In the versus 0.27%, n = 4, p < 0.01). By contrast, direction of gaze had
passive-viewing conditions in experiment 2 (lower panel), subjects no effect on the response to faces in the right STS or bilaterally
viewed series of faces that either had the eyes all directed away from the in the IOG or LFG. The difference in the right STS, however, was
viewer or all directed at the viewer.
in the same direction as in the left STS, and the sizes of this effect
in the right and left did not differ significantly (p > 0.1).
DISCUSSION
effects of task on responses in the right and left STS did not differ The results of these experiments indicate that face identity and
significantly (p > 0.1). Additionally, we found that, as in the LFG, eye gaze have distinct representations within the distributed
attention to identity elicited a stronger bilateral response in the human neural system for face perception. This distributed sys-
IOG than did attention to gaze (on the right, 0.99% versus 0.85%, tem includes bilateral regions in the IOG, LFG and STS, all of
n = 7, p < 0.005; on the left, 1.10% versus 0.85%, n = 7, p < 0.001). which show a greater response to faces than to other objects4,6,8,24.
As in the STS, attention to gaze elicited a stronger response in the Additionally, a region in the IPS was activated by our tasks,
Table 1. Volumes and stereotaxic brain atlas coordinates45 for the brain regions activated by viewing faces as compared
to viewing scrambled pictures (mean s.d.).
Talairach coordinates (mm)
Region Hemisphere n Volume (cm3) x y z
Fusiform gyrus left 7 2.5 0.7 37 1 60 1 22 4
right 7 2.7 0.6 39 2 55 2 22 3
Superior temporal sulcus left 4 1.8 0.4 45 1 56 3 11 2

right 7 1.9 0.6 50 1 63 4 47
Inferior occipital gyrus left 7 2.5 0.4 31 1 82 1 15 4

right 7 2.2 0.7 41 2 79 3 14 5
Intraparietal sulcus left 7 2.9 0.6 33 2 52 3 45 2

right 5 2.7 0.7 29 3 58 4 50 1

articles
Experiment 1 Experiment 2
Selective attention Passive viewing of faces
to gaze and identity with averted and direct gazes
Lateral view
percent change percent change

Intraparietal
sulci
Ventral view
Superior
temporal
sulci
percent change
Flattened cortex
Lateral
fusiform
gyri
Intraparietal
sulcus
percent change
Inferior Inferior
occipital occipital
Superior temporal gyri
gyri
sulcus
Fusiform gyrus
Attention Attention Averted Direct
to gaze to identity gaze gaze
Right hemisphere
Left hemisphere
Fig. 2. fMRI results. The left panel shows regions activated by the face-perception tasks (z > 4.0) in the right hemisphere of one subject. Regions are
shown on the folded surface, presented in lateral and ventral views (upper left figures) and on the cortical surface, inflated to show the extent of acti-
vated regions in the sulci. The lower figure shows the cortical surface for the entire right hemisphere presented as a flat, two-dimensional surface.
Sulcal cortex obscured in the folded surface is shown with a darker shade of gray on the inflated and flattened surfaces. On the flattened cortex,
occipital cortex is on the left and frontal cortex is on the right. The right panel shows mean time series in regions of interest, averaged across voxels
in the regions, repetitions of task blocks and subjects. Gray bars indicate the presentation of task blocks. White spaces following task blocks indicate
control task blocks that follow each task. See text for statistical comparisons.
although this region does not typically show selectivity for faces The representation of eye gaze, a changeable aspect of the face,
but, rather, is more typically associated with spatial perception depends more on activity in the STS than on activity in the IOG
and spatial attention1618,25. and LFG. Our results show that selective attention to gaze direc-
The representation of face identity, which is based on aspects of tion elicits a stronger response in the left STS than does attention
facial structure that are invariant across changes in eye gaze or to identity. An earlier neuroimaging study showed that perception
expression, is more dependent on activity in the IOG and LFG than of eye and mouth movement selectively activates the STS bilater-
on activity in the STS. In the monkey, neurons in the convexity of ally9. An event-related potential (ERP) study with scalp electrodes
the inferior temporal (IT) gyrus show greater selectivity for differ- using the same moving stimuli that evoked activity in the STS
ent individual faces than do neurons in the STS11, although some revealed that perception of averted gaze evokes a stronger N170
STS neurons also respond differentially to individual faces1214. The response than does perception of direct gaze30, consistent with our
human ventral temporal regions that are face-responsive may be results. Moreover, N200 responses measured with subdural elec-
homologous to the monkey IT region that is tuned to face identity. trodes placed on ventral face-specific sites do not differ signifi-
This conclusion is consistent with the literature on lesions that cause cantly for perception of averted and direct gaze31, consistent with
prosopagnosia13,26. Neuroimaging research on the role of the LFG our findings in the LFG in experiment 2. A positron emission
has been ambiguous27,28. An early study showed that the fusiform tomography study showed STS activation during perception of
gyrus was activated more when subjects attended to face identity averted and direct gaze10, but no difference between these condi-
than when they attended to gender27. Studies of the effect of face tions, in contrast with our fMRI results and the scalp ERP results30.
inversion on the activity in the LFG, however, generate doubt as to Our results indicate that the STS has a more general role in the
whether this region encodes face identity or simply the generic facial perception of changeable aspects of faces, even when viewing sta-
configuration4,5,29. Inversion impedes recognition of identity but tic images. This conclusion is consistent with electrophysiological
has only a small and nonspecific effect on LFG activity. Our results and lesion studies in monkeys. Monkey STS contains cells that
clearly implicate the LFG in the perception of identity. The effect respond differentially to gaze directions and facial expressions in
of face inversion suggests that LFG activity may reflect the attempt static pictures1114. These findings have led to the proposal that
to perceive identity, not the successful generation of a distinct rep- within face-responsive regions there are independent cell popula-
resentation of an individuals face. tions that perceive social signals from the face1214, and that these

articles
cells are more prevalent in the STS than in IT cortex11. Lesions of the transfer of information about social signals gleaned from the
monkey STS are associated with impaired perception of eye gaze face, particularly those concerning direction of spatial attention38.
with preservation of the perception of face identity32,33. Dissociation Such connections in the human brain could mediate the recruit-
of impairments of face-identity recognition, on the one hand, from ment of the IPS when the STS detects an averted gaze. One fMRI
impairments of the perception of eye-gaze direction or facial study showed that perception of lateral eye movement also acti-
expression, on the other, is reported in human lesion studies, but vates the IPS9, consistent with our contention that the IPS activ-
the anatomical locations of lesions that can cause selective impair- ity is specifically associated with the spatial aspects of perceived
ment of eye-gaze perception are unclear26,32. eye gaze and its role in directing attention.
Neuroimaging studies of other changeable aspects of faces Another fMRI study found that perception of direct gaze, but
also implicate the STS. In one study, perception of facial expres- not averted gaze, elicits activity in the amygdala39, but detected no
sion elicited a response in regions with coordinates close to our face-responsive region in the STS, perhaps because the control task
STS region34. In another study, lip reading also elicited a response (opening and closing of the eyes) also involves perception of facial
in similar regions35. Together with our results, these findings and movement. The amygdala was not included in the volume we
studies of monkey STS suggest that the STS may play a more gen- scanned. An amygdalar response may reflect the emotional signif-
eral role in the representation of changeable aspects of the face. icance of direct gaze, which is perceived as more socially engaging
Studies of the perception of face expression and lip-reading or potentially threatening than is averted gaze 40,41. Monkeys
suggest that these operations also elicit activity in additional sys- respond more emotionally and make more appeasement gestures
tems that process the significance of information gleaned from when gaze is directed at them than when gaze is directed away42,43.
the face. Perception of fearful and disgusted facial expressions Connections between the STS and the amygdala may mediate pro-
elicit further activity in limbic regions associated with process- cessing of the emotional content of direct gaze.
ing emotion34,36,37. Lip-reading elicits further activity in regions Face perception can provide a wealth of information that facil-
associated with auditory processing of speech sounds35. itates social communication. There are two classes of face-percep-
Our results show that the perception of averted eye gaze elic- tion operations that require independent representations. One
its further activity in the spatial cognition system in the IPS. The class involves the perception of the changeable aspects of the face,
IPS is activated during tasks involving spatial perception and such as expression and eye-gaze direction, whereas the other
covert shifts of spatial attention1618,25. Presumably, the IPS was involves the perception of aspects of facial structure that are invari-
recruited in our tasks to encode the spatial direction of anoth- ant across these changes. Perception of changeable aspects pro-
ers gaze and, additionally, perhaps, to mediate covert, reflexive vides information about another persons current state of mind44.
shifts of spatial attention in that direction. The role played by the Eye gaze, in particular, is a powerful social signal that can guide
IPS in mediating covert shifts of spatial attention is presumably our attention and can inform us about the intentions and interest
the same whether that shift is elicited by perceived eye gaze, as of another person. Perception of invariant aspects of facial structure
in our experiment, or by some other spatial cue. underlies the recognition of identity. Our results indicate that face
An alternative account of the differential IPS response is that perception is mediated in humans by a distributed system that
perception of averted gaze elicited eye movements that resulted in comprises multiple regions, and that changeable and invariant
enhanced IPS activity. To rule out this explanation, we recorded aspects of faces have distinct representations within this system.
eye movements using the ISCAN eye tracking system (Burlington,
Massachusetts) while four different subjects performed the selec- METHODS
tive attention and passive viewing tasks outside the scanner. The Tasks. In experiment 1, subjects performed repetition-detection tasks
that directed attention to identity or eye gaze. In each block of trials, nine
mean number of saccadic eye movements was similar for selective
faces were presented sequentially in the center of a screen for 0.5 s with an
attention to identity and eye gaze (18-s stimulus block; 29 5 s, interstimulus interval of 1.5 s. At the beginning of each block of trials, a
mean s.d., versus 28 7 s, n.s.) and for passive viewing of direct cue word (identity or gaze) was displayed for 1 s to inform the subject
and averted gaze (20 7 s versus 18 7 s, n.s.). Moreover, the hor- as to which task to perform. Subjects indicated whether the selectively
izontal and vertical amplitudes of eye movements did not differ attended aspect of each face matched that of the preceding face by press-
for selective attention to identity versus eye gaze (horizontal, ing a button with the right (match) or left (nonmatch) thumb. In a con-
1.6 0.4 versus 1.7 0.3, n.s.; vertical, 0.6 0.3 versus 0.4 0.2, trol task, scrambled, nonsense, color images were presented at the same
n.s.) or for passive viewing of direct versus averted gaze (horizon- rate and in the same format as the stimuli in the repetition-detection
tal, 1.5 0.2 versus 1.3 0.3, n.s.; vertical, 0.6 0.5 versus tasks. In these trials, subjects pressed both the right and left buttons
0.5 0.4, n.s.). Given these results, it is unlikely that the enhanced simultaneously when each stimulus appeared. Blocks of control trials
alternated with repetition-detection blocks. In experiment 2, subjects
IPS activation in experiments 1 and 2 was due to greater eye move-
passively viewed color faces that were blocked by gaze direction (lateral-
ments during selective attention to gaze or passive viewing of avert- ly averted or direct). Subjects were instructed simply to look directly at
ed gaze. These results also suggest that attention to identity, as each picture. Stimuli were presented sequentially at a rate of 2 per s in
compared to attention to eye gaze, did not result in significantly the center of the screen. Each block consisted of 36 stimuli. Face blocks
more scanning of the face or more eye movements with a vertical alternated with control blocks during which nonsense stimuli were pre-
component. Therefore, it is unlikely that differential responses sented at the same rate and in the same format as were stimuli in the face
when attending to gaze and identity can be attributed to differ- blocks. The order of blocks in both experiments was counterbalanced
ences in the parts of the faces that were foveated during these tasks. across subjects and time series. Ten time series, each consisting of eight
Behavioral studies indicate that monkeys and apes use gazes of face blocks and nine control blocks, were obtained for each subject (six in
experiment 1 and four in experiment 2).
others as cues to direct attention19,23. Human infants as young as
three months shift attention in the direction of perceived gaze21,22. Imaging. We scanned 9 healthy volunteers (3 male, 6 female, mean age,
One study with adult subjects suggests that these shifts may be 24 2.5 years). Each subject gave written informed consent and was com-
reflexive, occurring even when the direction of perceived gaze is pensated for participation. Our experimental protocol was approved by
task-irrelevant15. Reciprocal connections between cell popula- the institutional review board of the National Institute of Mental Health.
tions in the superior bank of the STS and the IPS38 could mediate Twenty contiguous, coronal, 5-mm thick slices were obtained in 10 time

articles
series of 102 time points each (TR = 3 s). We used a gradient echo, echo 12. Perrett, D. et al. Neurones responsive to faces in the temporal cortex: studies of
planar imaging sequence (TE = 40 ms, FOV = 20 cm, 64 64 pixels per functional organization, sensitivity to identity and relation to perception.
Hum. Neurobiol. 3, 197208 (1984).
inch matrix, resolution, 3.125 3.125 5 mm) on a GE signa 1.5 Tesla 13. Perrett, D. et al. Visual cells in the temporal cortex sensitive to face view and
magnet (General Electric, Milwaukee, Wisconsin). gaze direction. Proc. R. Soc. Lond. 223, 293317 (1985).
14. Perrett, D. et al. Social signals analyzed at the single cell level: someone is
Statistics. We registered our images using Automated Image Registration looking at me, something touched me, something moved. Int. J. Comp.
(AIR) provided by Roger Woods (Laboratory of Neuroimaging, UCLA). Psychol. 4, 2555 (1990).
15. Friesen, C. & Kingstone, A. The eyes have it! Reflexive orienting is triggered by
Following this, a 1.2-voxel smoothing filter was applied to each scan. We nonpredictive gaze. Psychol. Bull. Rev. 5, 490495 (1998).
identified regions that showed a significant response to faces as compared 16. Corbetta, M., Shulman, G., Miezin, F. & Peterson, S. Superior parietal cortex
to control stimuli in experiment 1; thus, these regions were defined inde- activation during spatial attention shifts and visual feature conjunction.
pendently of response differences between attention to identity and atten- Science 270, 802805 (1995).
17. Corbetta, M. Frontoparietal cortical networks for directing attention and the
tion to gaze or between passively viewed direct gaze and averted gaze. eye to visual locations: Identical, independent, or overlapping neural systems?
Using multiple regression, we selected clusters of 7 or more contiguous Proc. Natl. Acad. Sci. USA 95, 831838 (1998).
voxels that showed a significant difference (z > 4) for the contrast between 18. Nobre, A. et al. Functional localization of the system for visuospatial attention
face and control blocks. The probability of finding a cluster of that size using positron emission tomography. Brain 120, 515533 (1997).
by chance was p < 0.001 in all subjects. Further analysis was restricted to 19. Emery, N., Lorincz, E., Perrett, D., Oram, M. & Baker, C. Gaze following and
joint attention in rhesus monkeys (Macaca mulatta). J. Comp. Psychol. 111,
the seven subjects who showed significant activations in the regions that 286293 (1997).
were the subject of our experimental hypotheses, namely the fusiform 20. Hains, S. & Muir, D. Infant sensitivity to adult eye direction. Child Dev. 67,
gyrus and the STS. Brain atlas coordinates45 for all regions were obtained 19401951 (1996).
using Statistical Parametric Mapping46 (SPM96, Wellcome Department 21. Hood, B., Willen, J. & Driver, J. Adults eyes trigger shifts of visual attention in
of Cognitive Neurology, London). Mean time series were calculated for human infants. Psychol. Sci. 9, 131134 (1998).
22. Vecera, S. & Johnson, M. Gaze detection and the cortical processing of faces:
each region, averaging across voxels and across repetitions of task blocks, Evidence from infants and adults. Vis. Cognit. 2, 5987 (1995).
in each subject, and were analyzed separately for experiments 1 and 2. 23. Tomasello, M., Call, J. & Hare, B. Five primate species follow the visual gaze of
The time series were extracted from regions activated by the face versus conspecifics. Anim. Behav. 55, 10631069 (1998).
control comparison in experiment 1. These regions of activation were 24. Kanwisher, N., Stanley, D. & Harris, A. The fusiform face area is selective for
faces not animals. Neuroreport 10, 183187 (1999).
selected based on their responsiveness to faces generally and were therefore 25. Haxby, J. et al. The functional organization of human extrastriate cortex: A
independent of our specific hypotheses in experiments 1 and 2. Multiple PET-rCBF study of selective attention to faces and locations. J. Neurosci. 14,
regression was used to measure the size of the response to each face task 63366353 (1994).
condition as a percent change above control. Multiple regression used two 26. Tranel, D., Damasio, A. & Damasio, H. Intact recognition of facial expression,
regressors of interest to contrast faces versus control and the two face-task gender, and age in patients with impaired recognition of face identity.
Neurology 38, 690696 (1988).
conditions (experiment 1, identity versus gaze; experiment 2, averted gaze 27. Sergent, J., Ohta, S. & MacDonald, B. Functional neuroanatomy of face and
versus direct gaze). Measures of percent response were then analyzed with object processing. A positron emission tomography study. Brain 115, 1536
a repeated-measures analysis of variance with planned comparisons for (1992).
the interaction between region (LFG versus STS) and face task (identity 28. Sergent, J., Ohta, S., MacDonald, B. & Zuck, E. Segregated processing of facial
identity and emotion in the human brain: a PET study. Vis. Cognit. 1, 349369
versus gaze) in experiment 1 and for the simple differences between face (1994).
task conditions in both experiments in all regions. 29. Aguirre, G., Singh, R. & DEsposito, M. Stimulus inversion and the responses
of face and object-sensitive cortical areas. Neuroreport 10, 189194 (1999).
ACKNOWLEDGEMENTS 30. Puce, A., Smith, A. & Allison, T. ERPs evoked by viewing facial movements.
We wish to thank Leslie Ungerleider and Alex Martin for their comments on an Cognit. Neuropsychol. (in press).
31. McCarthy, G., Puce, A., Belger, A. & Allison, T. Electrophysiological studies of
earlier draft of this manuscript. We also wish to thank Timothy Ellmore and human face perception. II. Response properties of face-specific potentials
Jennifer Schouten for assistance with data collection, analysis and figure generated in occipitotemporal cortex. Cereb. Cortex 9, 431444 (1999).
preparation. Finally, we wish to thank Anders Dale and Bruce Fischl for 32. Campbell, R., Heywood, C., Cowey, A., Regard, M. & Landis, T. Sensitivity to
eye gaze in prosopagnosia patients and monkeys with superior temporal
providing the software for displaying results on inflated and flattened cortical sulcus ablation. Neuropsychologia 28, 11231142 (1990).
surfaces and for their assistance with the use of this software. 33. Heywood, C. & Cowey, A. The role of the face-cell area in the discrimination
and recognition of faces by monkeys. Phil. Trans. R. Soc. Lond. B Biol. Sci. 335,
3138 (1992).
RECEIVED 26 JULY; ACCEPTED 15 NOVEMBER 1999 34. Phillips, M. et al. Neural responses to facial and vocal expressions of fear and
disgust. Proc. R. Soc. Lond. B Biol. Sci. 265, 18091817 (1998).
35. Calvert, G. et al. Activation of auditory cortex during silent lipreading. Science
1. Damasio, A., Damasio, H. & Van Hoesen, G. Prosopagnosia: anatomic basis 276, 593596 (1997).
and behavioral mechanisms. Neurology 32, 331341 (1982). 36. Breiter, H. et al. Response and habituation of the human amygdala during
2. Farah, M. Is face recognition special? Evidence from neuropsychology. Behav. visual processing of facial expression. Neuron 17, 875887 (1996).
Brain Res. 76, 181189 (1996). 37. Morris, J. et al. A differential neural response in the human amygdala to
3. Sergent, J. & Signoret, J. Functional and anatomic decomposition of face fearful and happy facial expressions. Nature 383, 812815 (1996).
processing: Evidence from prosopagnosia and PET study of normal subjects. 38. Harries, M. & Perrett, D. Visual processing of faces in temporal cortex:
Phil. Trans. R. Soc. Lond. B Biol. Sci. 335, 5562 (1992). Physiological evidence for a modular organization and possible anatomical
4. Haxby, J. et al. The effect of face inversion on activity in human neural systems coordinates. J. Cogn. Neurosci. 3, 924 (1991).
for face and object perception. Neuron 22, 189199 (1999). 39. Kawashima, R. et al. The human amygdala plays an important role in gaze
5. Kanwisher, N., Tong, F. & Nakayama, K. The effect of face inversion on the monitoring: A PET study. Brain 122, 779783 (1999).
human fusiform face area. Cognition 68, 111 (1998). 40. Gale, A., Kingsley, E., Brookes, S. & Smith, D. Cortical arousal and social
6. Kanwisher, N., McDermott, J. & Chun, M. The fusiform face area: a module in intimacy in the human female under different conditions of eye contact.
human extrastriate cortex specialized for face perception. J. Neurosci. 17, Behav. Proc. 3, 271275 (1978).
43024311 (1997). 41. Van Hooff, J. A. Facial expressions in higher primates. Symp. Zool. Soc. Lond.
7. McCarthy, G., Puce, A., Gore, J. & Allison, T. Face-specific processing in the 8, 97125 (1962).
human fusiform gyrus. J. Cog. Neurosci. 9, 605610 (1997). 42. Mendelson, M., Haith, M. & Goldman-Rakic, P. Face scanning and
8. Halgren, E. et al. Location of human face-selective cortex with respect to responsiveness to social cues in infant rhesus monkeys. Dev. Psychol. 18,
retinotopic areas. Hum. Brain Mapp. 7, 2937 (1999). 222228 (1982).
9. Puce, A., Allison, T., Bentin, S., Gore, J. & McCarthy, G. Temporal cortex 43. Perrett, D. & Mistlin, A. in Comparative Perception Vol. 2 (eds. Stebbins, W. &
activation in humans viewing eye and mouth movements. J. Neurosci. 18, Berkley, M.) 187215 (Wiley, New York, 1990).
21882199 (1998). 44. Perrett, D., & Emery, N. Understanding the intentions of others from visual
10. Wicker, B., Michel, F., Henaff, M.-A. & Decety, J. Brain regions involved in the signals: neurophysiological evidence. Curr. Psychol. Cogn. 13, 683694 (1994).
perception of gaze: A PET study. Neuroimage 8, 221227 (1998). 45. Talairach, J. & Tournoux, P. Co-Planar Stereotaxic Atlas of the Human Brain
11. Hasselmo, M., Rolls, E. & Baylis, G. The role of expression and identity in the (Thieme, Stuttgart, 1988).
face-selective responses of neurons in the temporal visual cortex of the 46. Friston, K. et al. Statistical parametric mapping in functional imaging: a
monkey. Behav. Brain Res. 32, 203218 (1989). general linear approach. Hum. Brain Mapp. 2, 189210 (1995).

articles
Integration of diverse information

in working memory within the
frontal lobe
V. Prabhakaran1, K. Narayanan2, Z. Zhao2 and J. D. E. Gabrieli1,2
1 Program in Neurosciences and 2Dept. of Psychology, Jordan Hall, Bldg. 420, Stanford University, Stanford, California 94305, USA
Correspondence should be addressed to V.P. (vivekpr@leland.stanford.edu)
Ability to integrate diverse forms of information in current thought, or working memory, is essential
for human reasoning and problem solving. We used functional imaging to identify brain regions
preferentially involved in maintaining integrated versus unintegrated information in working mem-

ory. For equal amounts of verbal and spatial information, activation of prefrontal cortex was greater
for maintaining integrated rather than unintegrated representations. Posterior brain regions
showed the opposite pattern. These results demonstrate frontal-lobe specialization in maintaining
working-memory representations that integrate verbal and spatial information. The role of
prefrontal cortex in integrating multiple forms of information in working memory may underlie its
unique contribution to high-level cognition that demands flexible mental representations.
Working memory involves the short-term maintenance of infor- respectively, had appeared in the previous display. For the ver-
mation relevant to current goals. Neuroimaging studies show bal scans, the letters were shown at study and test in different
that working memory is mediated by frontal and posterior cor- cases, so that subjects coded the verbal identity rather than the
tical regions differing in the types of information maintained (for visual appearance of letters.
instance, verbal, spatial or object) and in the kinds of contribu- In two other scans, subjects were asked to maintain both spa-
tions made to working memory (for instance, directing rehearsal tial and verbal information either in an integrated or in an unin-
versus storing information per se)126. Whereas posterior corti- tegrated fashion. In both scans, subjects saw a target display of
cal regions seem to specialize in the type of information held in four letters and four spatial locations appeared for two seconds
working memory5,10,17,22,23,2729, several findings suggest that pre- and then disappeared (Fig. 2). In the integrated scan, the four
frontal areas have a special role in integrating different types of letters to be remembered were displayed in the four locations to
information in working memory. Electrophysiological studies in be remembered; thus verbal information and spatial informa-
nonhuman primates reveal frontal lobe cells that maintain both tion were bound together. In the unintegrated scan, the four let-
spatial and object information in working memory3032. Neu- ters were presented centrally and separately from the four
roanatomical tracing in nonhuman primates suggests prefrontal indicated locations; thus, verbal information and spatial infor-
cortex as a region of polymodal sensory convergence from pos- mation were separate. Subjects then had to maintain those let-
terior cortical areas (parietal, temporal and visual regions)33,34. ters and locations in working memory over a retention interval.
Neuroimaging studies in humans have shown proximal foci of Subsequently, they saw a single probe letter in a single probe loca-
activation in frontal cortex for maintaining different types of tion and had to judge whether both the letter and the location
information (for example, letters, objects, locations or faces) in had been shown in the previous display (regardless of whether
working memory. The present study aimed to test directly the that particular letter had been in that particular location). Posi-
hypothesis that human frontal cortex has a specialized role in tive-probe trials presented a letter and a location that had been in
maintaining integrated information in working memory35,36. the target display; negative probe trials presented either a differ-
Six subjects (right-handed; mean age, 24.5 years) were ent letter or a different location or both.
scanned via functional magnetic resonance imaging (fMRI). Scans using a 5-second retention interval between the target
Each subject performed four different tasks in four successive and probe displays were compared to a baseline identical in all
scans. Subjects performed two scans in which they were asked to regards except the retention interval between the target and probe
maintain either spatial information or verbal information displays was only 250 ms, so that information did not have to be
(Fig. 1). In the spatial scan, subjects saw a target display of four maintained in working memory. This allowed for isolation of
spatial locations appear for two seconds and then disappear. In areas involved in maintaining different types of information (spa-
the verbal scan, subjects saw a target display of four letters tial, verbal, verbal and spatial/integrated, verbal and spatial/unin-
appear for two seconds and then disappear. Over a retention tegrated) in working memory. As processes involved in encoding
interval, subjects then had to maintain those letters or locations the target information and in responding to the probe were
in working memory. Subsequently, they saw a single lower-case equivalent for the delay (5-second interval) and no-delay (250-ms
probe letter in the verbal scan or a single probe location in the interval) trials, our scans did not allow identification of areas
spatial scan, and decided whether the letter or the location, involved in these processes.

articles
Fig. 1. Spatial and verbal trial types. For the spa-

tial trials, subjects viewed for two seconds and
were asked to remember a target display com-
prising four locations enclosed by parentheses.
Locations were selected randomly on the circum-
ference of an imaginary ellipse. In the spatial con-
dition (left panel), subjects had to maintain
location information for either 5 s (upper halves
of each panel) or 250 ms (lower halves of each
panel). Subsequently, subjects were probed and
had 2.5 s to determine whether the probe loca-
tion was one of the locations in the target display.
Time
For the verbal trials, subjects viewed for two sec- Time
onds and were asked to remember a target dis-
play comprising four upper-case consonants in the
center of the display. They had to maintain letter
information for either 5 s (upper halves of each
panel) or 250 ms (lower halves of each panel).
Subsequently, subjects were probed and had 2.5 s
to determine whether a single lower-case letter No-delay trials No-delay trials
had been in the target display. A 4.75-s delay was

added to the trials with short delays to equalize
the time between target displays.
RESULTS Maintaining only verbal information, in contrast, resulted in

In general, subjects found the spatial and verbal conditions eas- prominent activations in left inferior frontal (Brocas area), left
ier than the bound and separate conditions. Subject reaction inferior parietal and temporal regions (Fig. 3). The different loca-
times (RTs) were significantly shorter in the spatial condition tions of activation reflect the neural separation of short-term mem-
[average RT, 979 ms; F1,5 = 50.5, mean squared error (MSE) = ory for spatial versus verbal information. Both imaging results were
293654, p = 0.0009] and the verbal condition (average RT, 834 consistent with previous imaging studies of maintenance of spa-
ms, F1,5 = 391.8, MSE = 805951, p = 0.0001) than in the bound tial or verbal information in working memory7,10,11,20,22,23,26.
and separate conditions (1200 ms). Subjects were not signifi- When maintaining both spatial and verbal information, for
cantly different in accuracy in the spatial condition (average accu- both positive and negative probe trials, subjects were signifi-
racy, 83.3%; F 1,5 = 0.49, MSE = 52.44, p = 0.51), but were cantly more accurate in the bound (88.5%) than in the separate
significantly more accurate in the verbal condition (average accu- (85.7%) condition (F1,5 = 7.18, MSE = 46.62, p = 0.044) and
racy, 97.6%; F1,5 = 82.35, MSE = 763.04, p = 0.0003) than in the tended to be faster in the bound (1139 ms) than in the separate
bound and separate conditions (87.1%). (1261 ms) condition (F1,5 = 4.59, MSE = 105186.7, p = 0.076).
Maintaining only spatial information resulted in prominent There was no significant effect of delay or condition delay
activations in right frontal and bilateral superior parietal regions. interaction. Further examination of only the positive-probe tri-
Fig. 2. Bound and separate trial types. For all tri-

als, subjects viewed for two seconds and were
asked to remember a target display comprising
four upper-case consonants and four locations
enclosed by parentheses. Locations were selected
randomly on the circumference of an imaginary
ellipse. In the bound condition (left panel), letters
were placed in the locations to be remembered. In
the separate condition (right panel), letters were
located elsewhere. Subjects had to maintain both
letter and location information for either 5 s
Time
(upper halves of each panel) or 250 ms (lower Time
halves of each panel). Subsequently, subjects were
probed and had 2.5 s to determine whether a sin-
gle lower-case letter had been in the target display
and whether its location was one of the locations
in the target display (not necessarily the same let-
ter). A 4.75-s delay was added to the trials with
short delays to equalize the time between target
displays. Examples of two kinds of positive probes
No-delay trials No-delay trials
are shown in the bound condition. In the delay
trial, a positive congruent probe is shown; the probe is congruent with the target because the same letter was presented in the same location. In the
no-delay trial, a positive incongruent probe is shown; the probe is incongruent because the letter had been presented in a different location. For both
kinds of probes, subjects would answer affirmatively because both the letter and the location had been presented in the target.

articles
gruent (1123 ms) than the incongruent (1251 ms) posi-

tive probes (F1,5 = 7.04, MSE = 115585.75, p = 0.038); they
also tended to be more accurate for the congruent (95.8%)
Spatial
than the incongruent positive probes (81.9%; F1,5 = 4.31,
MSE = 1157.59, p = 0.09).
These findings indicate that subjects maintained the
target displays in the bound condition in an integrated
fashion, such that they could quickly compare the con-
gruent probes to the similar, integrated content of work-
ing memory. They were slower, however, when they had
to reorganize the information in working memory to
Verbal compare to the incongruent probes. These results vali-
date the conclusion that verbal and spatial information
in the bound displays were maintained in an integrated
fashion in working memory during the delay period and
before probe presentation.
Fig. 3. Composite activation maps of spatial and verbal conditions. Images show Congruent and incongruent positive-probe trials
regions significantly more active in the delay than in the no-delay conditions coded
occurred equally often in the 5-second delay and 250-mil-
in red/yellow. Right-hemisphere and left-hemisphere views of the brain are shown

for each condition. Regions that were more active when subjects maintained spa- lisecond no-delay conditions. There was no congruency
tial information for 5 s rather than 250 ms are shown in the upper row. Right delay interaction in either reaction time (F1,5 = 0.02,
frontal and bilateral parietal regions show activation. Regions that were more p = 0.90) or accuracy (F1,5 = 0.00, p = 1.0). Therefore, the
active when subjects maintained verbal information for 5 s rather than 250 ms are imaging results do not reflect the behavioral differences
shown in the lower row. Left inferior frontal, left posterior parietal and temporal between congruent and incongruent trials. Rather, they
regions show activation. reflect the mental representation of information before
appearance of the positive (congruent or incongruent)
and negative probe types.
Many brain regions were activated in both the bound
als in the bound condition provided evidence that subjects main- and separate conditions as subjects maintained verbal and
tained verbal and spatial information in an integrated working- spatial information in working memory for 5 seconds versus
memory representation (Fig. 3). The positive-probe trials 250 milliseconds. These regions included frontal, parietal and
consisted of two types of probe trials. Positive congruent-probe temporal areas that were involved in maintaining verbal and
trials presented the same letter in the same location congruent to spatial information (Fig. 3). Scans under bound or separate
what had been shown in the target display. Positive incongru- conditions were directly compared to identify areas preferen-
ent-probe trials presented a previously seen letter in a location tially involved in maintaining integrated versus unintegrated
that had been occupied by a letter different from that shown in information in working memory (Fig. 4). Maintenance of
the target display. Subjects responded affirmatively for both pos- integrated information resulted in greater activity than main-
itive-probe trials, because the letter and location had been in the tenance of unintegrated information in the right frontal cor-
target display. Subjects responded significantly faster for the con- tex, specifically right middle and superior frontal gyri. This
Fig. 4. Difference maps of

bound versus separate scans.
Images show regions that were
significantly more active in the
bound than in the separate
condition coded in red/yellow
and regions that were signifi-
cantly more active in the sepa-
rate than in the bound
condition coded in blue. Four
slices were selected to highlight
the different regions that are
involved in the different condi-
tions. Difference maps were
created by performing paired
subtractions between statistical
parametric maps (SPMs) for
each subject after warping to
the reference template (struc-
tural image averaged across all
subjects) and then averaging
the difference images across
subjects. The resulting func-
tional maps were then overlaid
on the reference template.

articles
p = 0.0005
p = 0.025
Fig. 5. Activation in the right frontal cortex in six subjects. For each subject, a minimum of two slices depicting differential activation levels in the right frontal
cortex were chosen. The right frontal cortex was identified anatomically as a region of interest (ROI) in these slices. Statistical analyses were confined to that
ROI and superimposed on each subjects anatomical MR image. The upper row shows robust right frontal activation for maintaining bound information for
5 s versus 250 ms; the lower row shows that this region was minimally activated while maintaining separate information for 5 s versus 250 ms.
was evident not only in the group average, but also in each mation alone (Fig. 6). Comparison of separate versus spatial and
individual subject (Fig. 5). Maintenance of unintegrated, rel- verbal scans revealed posterior regions that were more involved in
ative to integrated, information resulted in greater activity in maintaining unintegrated verbal and spatial information than in
multiple posterior brain areas, including bilateral parietal, maintaining either spatial or verbal information alone. There was
temporal and cerebellar regions. not, however, any difference in the right prefrontal cortex for
In a second analysis, bound and separate conditions were maintaining unintegrated verbal and spatial information. Thus,
compared to the combination of the purely spatial and verbal both analyses provide convergent evidence for a dissociation of
conditions. Here also, the bound versus spatial and verbal com- the right frontal region preferentially involved in the maintenance
parison revealed substantially greater involvement of the right of integrated working-memory representations, and multiple
prefrontal cortex in maintaining integrated spatial and verbal posterior brain regions preferentially involved in the maintenance
information than in maintaining either spatial or verbal infor- of non-integrated working-memory representations.
Fig. 6. Difference maps of bound or separate condition versus spatial

(SPA) and verbal (VER) conditions. Left column, posterior view of brain;
right column, anterior view. Images show regions that were significantly
more active in the bound or separate condition than in the spatial and
verbal conditions coded in red/yellow. Regions that show greater activa-
tion in the bound condition than in both the spatial and verbal condi-
tions are depicted in the top row. The anterior regions show robust
activation, specifically in the right prefrontal region, whereas posterior
regions are minimally active. Regions that show greater activation in the
separate condition than in both the spatial and verbal conditions are
depicted in the bottom row. The posterior regions show robust activa-
tion, whereas there is no activation in the anterior regions. Difference
maps were created by performing a contrast analysis between the
bound or separate condition versus spatial and verbal conditions.

articles
scan consisted of six cycles, with each cycle comprising blocks of 5-s main-
DISCUSSION
tenance trials alternated with blocks of 250-ms maintenance trials. There
The right prefrontal activation associated with the integration of were 4 trials per block with a total of 24 (12 negative-probe trials and 12
verbal and spatial information is in agreement with meta-analy- positive-probe trials) 5-s maintenance trials and 24 (12 negative-probe
ses showing right prefrontal involvement in both spatial and non- trials and 12 positive-probe trials) 250-ms maintenance trials in each scan.
spatial working memory9,35,36. In contrast, left prefrontal regions In the bound scan, the positive-probe trials were equally divided into con-
show involvement only in nonspatial working memory. Thus, gruent and incongruent trials. The scan order was counterbalanced across
right prefrontal regions seem to have a flexible representational the subjects. Stimuli were generated from a computer and back-project-
architecture that processes both spatial and nonspatial informa- ed onto a screen located above the subjects neck via a magnet-compati-
tion. Posterior activations have been more material specific, with ble projector. Visual images were viewed from a mirror mounted above
specific areas in parietal and temporal regions involved either in the subjects head. The sequence of the presentations of the stimuli were
synchronized with the imaging sequence of the scanner.
verbal or spatial working memory5,10,17,22,23.
Many studies have shown greater activation in prefrontal cor- fMRI methodology. Imaging was performed with a 1.5T whole-body MRI
tex due to increased demands of load, duration or manipulation scanner (General Electric Medical Systems Signa, Rev. 5.5, Waukeshau,
on working memory123. In contrast, greater prefrontal activation Wisconsin). A T2* sensitive gradient echo spiral sequence46 was used
in the present study occurred in the easier condition: subjects were for functional imaging with parameters of TR = 1440 ms,
more accurate and faster in the bound than in the separate condi- TE = 40 ms, flip angle = 83, FOV = 20 cm, inplane resolution = 1.56 mm2,
tion. The prefrontal activation, therefore, reflected the nature of sampling interval = 2.88 s and number of temporal frames or image vol-
the working-memory representation, rather than working-mem- umes = 160. Sixteen 7-mm thick slices with a 0-mm inter-slice interval
ory load or duration, which were equal in the two conditions. and covering the whole brain were acquired in the horizontal plane of the
Talairach and Tournoux atlas47.
Baddeley has proposed separate slave system buffers that allow
for temporary retention of discrete information in working mem- fMRI analysis. Functional images were motion-corrected and normal-
ory, including a visuospatial buffer and a phonological buffer37. ized using SPM96, interpolated to 2 2 4 mm3 voxels and spatially
Logie and others have further decomposed the visuospatial buffer smoothed with a Gaussian filter (full width at half maximum, 8 mm).
into separate visual and spatial buffers38. The present fMRI results Low-frequency noise and differences in global signal were removed. Sin-
provide evidence for another type of buffer, namely, one that gle subject data were analyzed with a fixed-effects model. Group data
allows for temporary retention of integrated information. were analyzed using a random-effects model. For the group analysis,
The capacity to integrate information in working memory images were averaged to create one image of mean activity per condi-
may enhance the efficiency of working memory in several ways. tion and subject. These average images were used to create a series of
For example, a study examining the capacity of short-term mem- SPM{Z} maps depicting differences in brain activity between task con-
ditions. Activation maps for both fixed-effects model and random-effects
ory for visuospatial information found that subjects show equiv-
model analysis were created with SPM96 software
alent accuracy in maintaining the memory of a single feature of (http://www.fil.ion.ucl.ac.uk/spm)48 with an intensity threshold of
objects (for instance, color of an object) as in maintaining four p < 0.025 and spatial-extent threshold of p < 0.05.
features of objects (for instance, color, size, orientation and
shape)39. However, performance decreases as the number of
objects that need to be remembered increases. In other words, ACKNOWLEDGEMENTS
more information could be maintained for visuospatial infor- This work was supported by grants from the National Institute on Aging and the
mation in a bound rather than separate display. In our study, National Center for Research Resources. V.P. is supported by a NRSA training
the finding subjects were significantly more accurate and tend- grant awarded by the National Institutes of Health. The authors thank Mark
ed to be faster in the bound than in the separate condition sug- DEsposito for comments on earlier drafts of this manuscript.
gests more efficient access to integrated relative to unintegrated
information. Further, the mean volume of brain area preferen- RECEIVED 3 SEPTEMBER; ACCEPTED 3 NOVEMBER 1999
tially activated by the unintegrated information (11,483 mm3)
was more than twice that activated preferentially by the inte-
1. Baker, S. C., Frith, C. D., Frackowiak, R. S. J. & Dolan, R. J. Active representa-
grated information (4,368 mm3). The greater activity seen in the tion of shape and spatial location in man. Cereb. Cortex 6, 612619 (1996).
posterior regions suggests that more neural resources are 2. Braver, T. S. et al. A parametric study of prefrontal cortex involvement in
required when maintaining separate information than bound human working memory. Neuroimage 5, 4962 (1997).
3. Cohen, J. D. et al. Activation of prefrontal cortex in a non-spatial working
information. Thus, integration may provide for efficient work- memory task with functional MRI. Hum. Brain Mapp. 1, 293304 (1994).
ing-memory representations. 4. Cohen, J. D. et al. Temporal dynamics of brain activation during a working
Human lesion and neuroimaging studies show that pre- memory task. Nature 386, 604607 (1997).
5. Courtney, S. M., Ungerleider, L. G., Keil, K. & Haxby, J. V. Object and spatial
frontal cortex is essential in reasoning and problem solving4045. visual working memory activate separate neural systems in human cortex.
The capacity to integrate, or relate43, different kinds of infor- Cereb. Cortex 6, 3949 (1996).
mation may endow frontal cortex with a flexible representa- 6. Courtney, S. M., Ungerleider, L. G., Keil, K. & Haxby, J. V. Transient and
sustained activity in a distributed system for human working memory.
tional structure important for the fluid, innovative thinking Nature 386, 608611 (1997).
that underlies successful reasoning and problem solving. More 7. Courtney, S. M. et al. An area specialized for spatial working memory in
generally, the prefrontal activation in the present study may human frontal cortex. Science 279, 13471351 (1998).
8. DEsposito, M. et al. The neural basis of the central executive system of
reflect one solution to the binding problem in the human working memory. Nature 378, 279281 (1995).
brainthe problem of integrating the outputs of multiple, spe- 9. Fiez, J. A. et al. A positron emission tomography study of the short-term
cialized processing pathways. maintenance of verbal information. J. Neurosci. 16, 808822 (1996).
10. Jonides, J. et al. Spatial working memory in humans as revealed by PET.
Nature 363, 623625 (1993).
METHODS 11. Jonides, J. et al. Verbal working memory load affects regional brain activation
as measured by PET. J. Cogn. Neurosci. 9, 462475 (1997).
All studies were done in full compliance with the guidelines of the Insti- 12. Jonides, J., Smith, E. E., Marshuetz, C. & Koeppe, R. A. Inhibition in verbal
tutional Review Board (Assurance #M1272-02) of Stanford University. working memory revealed by brain activation. Proc. Natl. Acad. Sci. USA 95,
Subjects performed four scans: bound, separate, verbal and spatial. Each 84108413 (1998).

articles
13. McCarthy, G. M. et al. Functional magnetic resonance imaging of human 30. Rainer, G., Asaad, W. F. & Miller, E. K. Memory fields of neurons in the
prefrontal cortex activation during a spatial working memory task. Proc. Natl. primate prefrontal cortex. Proc. Natl. Acad. Sci. USA 95, 1500815013
Acad. Sci. USA 91, 86908694 (1994). (1998).
14. McCarthy, G. M. et al. Activation of human prefrontal cortex during spatial 31. Rainer, G., Assad, W. F. & Miller, E. K. Selective representation of relevant
and nonspatial working memory tasks measured by functional MRI. Cereb. information by neurons in the primate prefrontal cortex. Nature 393,
Cortex 6, 600611 (1996). 577579 (1998).
15. Owen, A. M., Evans, A. C. & Petrides, M. Evidence for a two-stage model of 32. Rao, S. C., Rainer, G. & Miller, E. K. Integration of what and where in the
spatial working memory processing within the lateral frontal cortex: a primate prefrontal cortex. Science 276, 821824 (1997).
positron emission tomography study. Cereb. Cortex 6, 3138 (1996). 33. Fuster, J. M. The Prefrontal Cortex: Anatomy, Physiology, and Neuropsychology
16. Owen, A. M. et al. Functional organization of spatial and nonspatial working of the Frontal Lobe (Lippincott-Raven, New York, 1997).
memory processing within the human lateral frontal cortex. Proc. Natl. Acad. 34. Fuster, J. M., Bauer, R. H., & Jervey, J. P. Cellular discharge in the dorsolateral
Sci. USA 95, 77217726 (1998). prefrontal cortex of the monkey in cognitive tasks. Exp. Neurol. 77, 679694
17. Paulesu, E., Frith, C. D. & Frackowiak, R. S. J. The neural correlates of the (1982).
verbal component of working memory. Nature 362, 342345 (1993). 35. DEsposito, M., Aguirre, G. K. & Zarahn, E. Functional MRI studies of spatial
18. Petrides, M., Alivisatos, B., Meyer, E. & Evans, A. C. Functional activation of and non-spatial working memory. Cognit. Brain Res. 7, 113 (1998).
the human frontal cortex during the performance of verbal working memory 36. Owen, A. M. The functional organization of working memory processes
tasks. Proc. Natl. Acad. Sci. USA 90, 878882 (1993). within human lateral cortex: the contribution of functional neuroimaging.
19. Petrides, M., Alivisatos, B., Evans, A. C. & Meyer, E. Dissociation of human Eur. J. Neurosci. 9, 13291339 (1997).
mid-dorsolateral from posterior dorsolateral frontal cortex in memory 37. Baddeley, A. Working Memory (Oxford Univ. Press, New York, 1986).
processing. Proc. Natl. Acad. Sci. USA 90, 873877 (1993). 38. Logie, R. H. & Denis, M. Mental Images in Human Cognition (Elsevier
20. Rypma, B., Prabhakaran, V., Desmond, J. E., Glover, G. H. & Gabrieli, J. D. E. Science, New York, 1991).
Load-dependent roles of frontal brain regions in the maintenance of working 39. Luck, S. J. & Vogel, E. K. The capacity of visual working memory for features
memory. Neuroimage 9, 216226 (1999). and conjunctions. Nature 390, 279281 (1997).
21. Rypma, B. & DEsposito, M. The roles of prefrontal brain regions in 40. Prabhakaran, V., Smith, J. A. L., Desmond, J. E., Glover, G. H. & Gabrieli, J. D.
components of working memory: effects of memory load and individual E. Neural substrates of fluid reasoning: An fMRI study of neocortical
differences. Proc. Natl. Acad. Sci. USA 96, 65586563 (1999). activation during performance of the Ravens Progressive Matrices Test.
22. Smith, E. E. et al. Spatial versus object working memory: PET investigations. Cognit. Psychol. 33, 4363 (1997).
J. Cogn. Neurosci. 7, 337356 (1995). 41. Baker, S. C. et al. Neural systems engaged by planning: a PET study of the
23. Smith, E. E., Jonides, J. & Koeppe, R. A. Dissociating verbal and spatial Tower of London task. Neuropsychologia 34, 515526 (1996).
working memory using PET. Cereb. Cortex 6, 1120 (1996). 42. Duncan, J., Burgess, P. & Emslie, H. Fluid intelligence after frontal lobe
24. Smith, E. E., Jonides, J., Marshuetz, C. & Koeppe, R. A. Components of verbal lesions. Neuropsychologia 33, 261268 (1995).
working memory: Evidence from neuroimaging. Proc. Natl. Acad. Sci. USA 43. Waltz, J. A. et al. A system for relational reasoning in human prefrontal
95, 876882 (1998). cortex. Psychol. Sci. 10, 119125 (1999).
25. Smith, E. E. & Jonides, J. Storage and executive processes in the frontal lobes. 44. Goel, V., Gold, B., Kapur, S. & Houle, S. The seats of reason? An imaging
Science 283, 16571661 (1999). study of deductive and inductive reasoning. Neuroreport 8, 13051310
26. Smith, E. E. & Jonides, J. Neuroimaging analyses of human working memory. (1997).
Proc. Natl. Acad. Sci. USA 95, 1206112068 (1999). 45. Koechlin, E., Basso, G., Pietrini, P., Panzer, S. & Grafman, J. The role of
27. Ungerleider, L. G. & Mishkin, M. in Analysis of Visual Behavior (eds. Ingle, D. anterior prefrontal cortex in human cognition. Nature 399, 148151 (1999).
J., Goodale, M. A. & Mansfield, R. J. W.) 549586 (MIT Press, Cambridge, 46. Glover, G. H. & Lai, S. Self-navigated spiral fMRI: interleaved versus single-
Massachusetts, 1982). shot. Magn. Reson. Med. 39, 361368 (1995).
28. Wilson, F. A. W., Fraser, A. W., O Scalaidhe, S. P. & Goldman-Rakic, P. S. 47. Talairach, J. & Tourneaux, P. Co-Planar Stereotaxic Atlas of the Human Brain
Dissociation of object and spatial processing domains in primate prefrontal (Thieme, Stuttgart, 1988).
cortex. Science 260, 19551957 (1993). 48. Friston, K. J., Frith, C. D., Liddle, P. F. & Frackowiak, R. S. J. Comparing
29. Zeki, S. & Shipp, S. The functional logic of cortical connections. Nature 335, functional (PET) images: The assessment of significant change. J. Cereb.
311317 (1988). Blood Flow Metab. 11, 690699 (1991).

articles
A cultural effect on brain function

E. Paulesu1, E. McCrory2, F. Fazio1, L. Menoncello1, N. Brunswick3, S. F. Cappa4,5, M. Cotelli4,
G. Cossu6, F. Corte6, M. Lorusso7, S. Pesenti7, A. Gallagher2, D. Perani1, C. Price3, C. D. Frith3,
and U. Frith2.
1 Scientific Institute H S. Raffaele, INB-CNR, University of Milan-Bicocca, Milan, Italy

2 Institute of Cognitive Neuroscience, University College London, 7 Queen Square, London WC1N 3AR, UK
3 Wellcome Department of Cognitive Neurology, Institute of Neurology, 12 Queen Square, London WC1N 3GB, UK
4 Neurology Department, University of Brescia, Brescia, Italy
5 Psychology Department, University Vita e Salute H San Raffaele, Milan, Italy
6 Istituto di Fisiologia Umana, University of Parma, Parma, Italy
7 Scientific Institute Eugenio Medea-La Nostra Famiglia, Bosisio Parini, Italy
Correspondence should be addressed to U.F. (u.frith@ucl.ac.uk)
We present behavioral and anatomical evidence for a multi-component reading system in which
different components are differentially weighted depending on culture-specific demands of orthogra-
phy. Italian orthography is consistent, enabling reliable conversion of graphemes to phonemes to
yield correct pronunciation of the word. English orthography is inconsistent, complicating mapping
of letters to word sounds. In behavioral studies, Italian students showed faster word and non-word
reading than English students. In two PET studies, Italians showed greater activation in left superior
temporal regions associated with phoneme processing. In contrast, English readers showed greater
activations, particularly for non-words, in left posterior inferior temporal gyrus and anterior inferior
frontal gyrus, areas associated with word retrieval during both reading and naming tasks.
In English there are 1120 ways of representing 40 sounds (phonemes) We conducted two PET scan studies, again with university stu-
by different letters or letter combinations (graphemes)1. The map- dents. The first study addressed explicit reading. Here participants
pings between graphemes, phonemes and whole word sound are had to read words and non-words aloud. A second study addressed
essentially ambiguous, as illustrated by pairs such as pint/mint, implicit reading, where we assessed physiological responses induced
cough/bough, clove/love. By contrast, in Italian, 33 graphemes are by the mere presence of print in the visual field. Participants were
sufficient to represent the 25 phonemes of the language2, and the not asked to read the stimuli, but to perform a visual feature-detec-
mappings from graphemes to phonemes are unequivocal. Young tion task on words, non-words and false-font stimuli7.
Italian readers can achieve 92% accuracy on word reading tests after Results from both experiments were combined to show only
only 6 months of schooling3, whereas learning to read in English those effects that were reliable in both studies (Fig. 2 and Table 2).
takes much longer. Compared to German, another consistent We identified a common brain system that was active during read-
orthography, accuracy levels in English are lower and reading speed ing, explicitly and implicitly, across the two languages. This sys-
is slower even after three years of schooling4,5. Adult English read- tem included inferior frontal and premotor cortex, superior,
ers are slower at reading non-words than readers of the consistent middle and inferior temporal gyri and fusiform gyrus on the left,
Serbocroat orthography6. We investigated reading in Italian and and superior temporal gyrus on the right. The majority of the left-
English university students. These students read high-frequency reg- hemisphere areas were more active for non-words than for words,
ular words in their respective languages, a set of international words although no single region showed reliably greater activation for
and two sets of non-words derived from both languages. We also words. Interaction effects showed a language-related difference:
investigated the neurophysiological basis of reading in Italian and English readers, particularly when reading non-words, showed
English using positron emission tomography (PET). greater activations in the left posterior inferior temporal region
and in the anteriormost part of the inferior frontal gyrus. When
RESULTS reading words or non-words, Italian readers showed greater acti-
Italian students were faster at both word and non-word reading, vations at the junction between left superior temporal gyrus and
even when the non-words were derived from English words. Con- inferior parietal cortex, a region known as planum temporale.
trol tasks showed that this advantage could not be attributed to
faster reaction times, articulation speed, naming speed or verbal DISCUSSION
fluency (Table 1 and Fig. 1). Both groups were slower at reading In spite of the large amount of empirical data on normal and
non-words as compared with words. This difference was signifi- abnormal reading, it is still a matter of debate as to how word and
cantly greater for English readers. The Italian students read non- non-word reading is achieved, particularly in a deep orthography
words and international words derived from Italian faster than such as English812. Our behavioral results showed that even with
those derived from English. The English students were unaffected simple and regularly spelled stimuli, background effects of the
by the source of the words. complex English orthography incur a cost in terms of reading

articles
Fig. 1. Vocal reaction times in single word and non-word reading.

English subjects
Reading high-frequency words and non-words derived from the words
in each language. Italian students had faster reading latencies than
Italian subjects
English students (F1,70 = 140.04, p < 0.001; Table 1). Both groups read
Reading latency (ms)
words faster than non-words (F2,140 = 114.90, p < 0.0001), but this effect
was more marked for English (group by task interaction; F2,140 = 15,53;
p < 0.0001). In a post-hoc analysis of non-word reading, English students
were equally slow when reading non-words derived from Italian or from
English words. In contrast, Italian students, although significantly faster
than English students for both kinds of non-words (group main effect;
F1,70 = 13.6; p < 0.0005), were faster still with non-words derived from
Italian words (group task interaction effect: F1,70 = 42,51; p < 0.0001).
In reading familiar international words, analysis of variance showed no
overall group effect (F1,70 = 0.97; n.s.), but a group by task interaction
Words
Non-words Non-words International International (F1,70 = 7.9; p < 0.007). This was because Italian subjects were faster at
from Italian from English words words reading international words conforming to their own orthography.
conforming to conforming to
Italian English
speed. According to a dual-route perspective, two processes are spelling patterns. Thus, for English readers, Italian non-words had
necessary in reading, letter-to-sound conversion and access to a the same bigram frequency as English non-words and did not slow
lexicon of orthographic patterns to resolve ambiguities in pro- them down. In contrast, Italians read the English non-words with
nunciation8,9,11. According to a single-route connectionist per- their less-familiar spelling patterns more slowly, as expected from
spective, one process is needed, which involves conversion from their lower bigram frequency in Italian.
orthography to phonology, and in a deep orthography such as Eng- Our PET scan data provide the first cross-cultural anatomi-
lish, an extensive translation from orthography to semantics to cal information about a common reading system for different
phonology10,12. Our data can be interpreted within either of these alphabetic orthographies. Reading in both complex and trans-
models. The complexity of English orthography derives partly from parent orthographies depends on a distributed network of pri-
the historical influence of other orthographies, including Italian marily left-sided language areas. Within the common network,
Table 1. Performance in reading tasks and in control tasks in 36 university students from London and 36 from Milan.
Reading tasks
Words Non-words Non-words International words International
derived from derived from conforming to words conforming
( s.d.) Italian words English words Italian (IW1) to English (IW2)
(ms) (NW1) (ms) (NW2) (ms) (ms) (ms)
English subjects 442.6 47.4 526.4 93.9 528.8 101.5* 450.9 48.2 448.0 57.4**
(n = 36)
Italian subjects 410.7 33.1 437.3 39.0 485.9 56.8+ 424.8 36.5 452.8 53.3++
(n = 36)
Mean difference 31.9 89 42.8 26.1 4.8
Two-tailed t value 3.3 5.2 2.2 2.6 0.37
p value < 0.0015 < 0.0001 < 0.03 < 0.012 0.7
Control tasks
Simple vocal Articulation Semantic verbal Letter verbal Picture-
reaction time speed for pairs of fluency fluency (words naming
words (words in (animal names in starting with m latency
(ms) 15 min) 1 min) in 1 min) (ms)
English subjects 310.5 37.8 40.4 3.5 25.4 6.7 16.5 6.7 546.7 41.8
(n = 36)
Italian subjects 313.1 28.8 41.9 6.1 27.5 6.4 16.6 4.7 561.0 47.5
(n = 36)
Mean difference 2.6 1.5 2.1 0.1 14.4
Two-tailed t value 0.33 1.3 1.3 0.08 1.4
p value 0.73 0.19 0.18 0.93 0.17
The results of the control tasks show that differences in reading speed were not due to differences in the sample population or to more general factors such
as sensorimotor speed, articulation speed or verbal fluency. Other statistical comparisons: *Comparison of NW1 and NW2 for English subjects; mean differ-
ence, 2.4 ms; paired t35 = 0.5; n.s. +Comparison of NW1 and NW2 for Italian subjects; mean difference, 48.6 ms; paired t35 = 9.6; p < 0.0001.
**Comparison of IW1 and IW2 for English subjects; mean difference, 2.9 ms; paired t35 = 0.4; n.s. ++Comparison of IW1 and IW2 for Italian subjects; mean
difference, 28.0 ms; paired t35 = 4.8; p < 0.0001.

articles
vious studies1316. A situation in which Italian readers need to

access word names/orthography is the assignment of stress for
words of more than two syllables with consonant/vowel struc-
ture (for example, tavolo)17,18. In an unpublished PET study (E.P.
et al.), Italian students had to indicate which syllable in a word
was stressed. Strong activation of the posterior basal temporal
area was observed during this task (Table 2). The enhanced acti-
vation shown by English readers in this region may reflect the
fact that the pronunciation of a stimulus in English always
involves such access and hence requires time and resources. One
possibility is that ambiguity is resolved via the activation of mul-
tiple neighboring alternatives in the orthographic lexicon, which
allow the selection of a correct pronunciation. This may explain
the activation of areas related to naming and semantic processing
during a non-word reading task.
The behavioral data showed that reading in a complex and
inconsistent orthography comes at a considerable cost. Reading
in Italian can proceed more efficiently because of the consistent
mapping between individual letter sounds and whole-word

sound. Italian readers showed comparatively stronger activation
of the left planum temporale (at the temporoparietal junction),
a brain region that has been linked to phonological process-
ing1922. Phonological procedures of reading in Italian may involve
a greater proportion of the overall processing effort. Converse-
ly, in the case of English, semantic/orthographic procedures may
predominate. These may be automatically evoked, given the
degree of orthographic complexity. Physiologically, the assump-
tion in both cases is that the enhanced involvement of a particu-
lar component of the reading system is reflected in greater
metabolic demands in the area associated with the component.
As reading seems to involve multiple components, a richer inter-
action among these components must be postulated than pro-
posed by either single- or dual-route theories of reading.
Our evidence adds to the knowledge provided by studies of
neurological patients. The investigation of acquired dyslexia and
dysgraphia has indicated a role for perisylvian lesions in producing
impairments in phonological processing (phonological dyslexia
and dysgraphia)23,24, whereas lexical disorders of written language,
Fig. 2. Functional commonalities and differences between English and
such as surface dyslexia and lexical agraphia, have been associat-
Italian reading systems. Brain areas are rendered on a standard brain con-
forming to stereotactic space. Stereotactic coordinates of the statistically ed with lesions sparing the perisylvian cortex25 or with conditions,
significant areas are given in Table 2. The first row shows the common such as semantic dementia, associated with temporal-lobe atro-
reading system as revealed by conjunction of the main effects of reading phy26. Our findings, although broadly consistent with these func-
minus baseline for the implicit and explicit reading experiments in both tional anatomical assignations, also show that, in normal reading,
English and Italian subjects. The second row shows the main effect of non- these different components have different prominence depending
word reading minus word reading for all groups. No significant difference on the transparency of the orthography. The culture-specific phys-
was found between word reading minus non-word reading. The third row iological differences for normal reading suggest that differences
shows those regions of greater activation in English subjects during non- between Italian and English alexic and agraphic patients may be
word reading. These include the left inferior posterior temporal areas and
due to differences in orthography27. The present findings indicate
the anterior portion of the inferior frontal gyrus. English subjects had
greater activations than Italian subjects when reading non-words. The
that cultural factors, as reflected in orthographic systems, can pow-
fourth row shows the left planum temporale region, which was more erfully shape neurophysiological systems.
active in Italian than English subjects, regardless of word type.
METHODS
Students from London (n = 36; mean age s.d., 20.6 3.3) and Milan
(n = 36; 20.8 3.0), matched on course of study (arts, science and engi-
neering) read bisyllabic words and non-words in their respective languages.
Words were nouns that had stress on the first syllable, and were among
however, some brain areas were more highly activated by one the 7500 most frequent in Italian and in English. Non-words were derived
orthography than the other. from these words by changing one or two phonemes but leaving the syllabic
These differences provide a physiological basis for the differ- structure intact. All participants read all non-words according to their own
ences observed in our behavioral studies. The inferior basal tem- orthography. There were 40 words and 40 non-words, divided into blocks
of 20 and presented on a computer screen in a counterbalanced order.
poral area and the anterior part of the interior frontal gyrus were
English words were as regular as possible, for example, cabin, market, cot-
more strongly activated in English readers, especially for non- tage and apron were matched with cagin, marnet, connage and afton as
word reading. These brain regions have been associated with corresponding non-words. Italian examples were marmo, ponte, moto
object and word naming, and semantic processes in several pre- and carta with margo, ponda, moco and corla as corresponding non-words.

articles
Table 2. Functional commonalities and differences between English and Italian reading systems and meta-analysis of
previous neuroimaging findings during naming/semantic tasks and phonological tasks.
Brain region Left hemisphere Right hemisphere
x y z z score x y z z score
(a) Main effect of reading
Inferior frontal gyrus (BA 44) 44 4 20 4.6
Inferior frontal gyrus / insula 38 22 12 4.4
Precentral gyrus (BA 6/4) 48 8 32 4
Insula 22 2 16 4.4 32 26 6 3.4
Temporo/parietal junction (BA 22/40) 38 36 22 3.6 70 40 24 3.5
Superior temporal gyrus (BA 22) 64 44 12 5.3 54 24 6 5.2
Middle temporal gyrus (BA 21) 60 50 8 5
Inferior temporal gyrus (BA 20) 50 58 22 5.6
Fusiform gyrus (BA 37) 40 52 24 7.4
Caudate nucleus 18 10 22 3.5 26 6 20 3.4
Thalamus 8 30 0 5.2 24 12 14 4
Cerebellum 22 68 26 5
(b) Non-words minus words

Precentral gyrus (BA6) 42 0 44 3.7
Anterior inferior frontal gyrus (BA 45) 42 24 14 3.4
Superior temporal gyrus (BA 22) 70 30 0 2.8
Middle temporal gyrus (BA 21) 48 58 6 3.4
Inferior temporal gyrus (BA 20/37) 52 60 14 3.2
Middle occipital gyrus (BA 19) 48 68 6 2.8
Inferior temporal gyrus (BA 20) 54 52 20 2.7
Inferior temp/fusiform gyri (BA 20/37) 46 68 16 2.3
Fusiform gyrus (BA 37) 48 44 14 2.3
(c) Greater activations for non-words in English readers as compared with Italian readers
Anterior inferior frontal gyrus (BA 45) 46 18 20 2.7
Inferior temporal gyrus (BA 21/37) 58 58 14 2.9
(d) Greater activations for words and non-words in Italian readers as compared with English readers
Superior temporal gyrus (BA 22/42) 48 34 16 2.6
(e) Other tasks activating left basal temporal region for whole-word processing
Naming33 37 46 20
Semantics; words and pictures13 46 46 20
Conjunction word and object naming34 44 62 16
Stress assignment+ 48 58 20
(f) Other tasks activating left anterior inferior frontal gyrus for whole-word processing
Semantic verbal fluency16 36 24 16
Meta-analysis of semantic tasks15* 37 27 14
(g) Other tasks activating left perisylvian temporal region: orthographic translation and sub-lexical phonological processing
Wordspictures35 42 40 20
Wordspictures36 58 46 28
Nonwordword32 50 36 32
Phonological short-term memory22 44 32 24
Localization is based on stereotactic coordinates. These coordinates refer to the location of maximal activation indicated by the highest z score in a particular
anatomical structure. Distances are relative to the intercommissural (ACPC) line in the horizontal (x), antero-posterior (y) and vertical (z) directions. Z
scores indicate the magnitude of the statistical significance. *Average stereotactic coordinates derived by a published meta-analysis (semantic decision)15.
+Stereotactic coordinate of activation associated with stress-assignment task for visually presented trisyllabic Italian words (E.P. et al., unpublished results).

articles
The bigram frequency of the stimuli was analyzed in terms of the number permitting a mixed-effects analysis appropriate for population infer-
of occurrences of a given bigram in a corpus of the 7500 most frequent ence30. The analysis was based on a 2 (English, Italian subjects) 2
words in each language (DeMauro Vocabulario di Base28; CELEX English (implicit, explicit reading) 3 (words, non-words, baseline) factorial
database, http://www.kun.nl/celex). The absolute bigram frequency of the design. We first calculated the main effect of the activation patterns asso-
stimuli across languages was different, as expected by their different ortho- ciated with reading as the conjunction of the four main effects of reading
graphic and phonological structure (English words versus non-words, (reading minus baseline) in each of the four groups (statistical thresh-
266.2 versus 246.4; Italian words versus non-words, 455.2 versus 403.8). old, p < 0.001 corrected for spatial extent)31. We then calculated the main
Within each set of stimuli, there was a nonsignificant trend for a lower effect of non-word minus word reading and vice versa, and the differ-
bigram frequency of the non-words, and there was no significant interac- ences between groups as group task interaction effects. All interaction
tion with language. In a separate task, subjects read 12 familiar interna-
effects were computed on the voxels identified by the linear contrast of the
tional words with the same meaning in both languages: tennis, boiler,
basket, corner, partner, bitter, coma, taxi, panda, bravo, villa and pasta. relevant main effects. For these latter, more subtle comparisons, a thresh-
Only the last six conform to Italian orthography. old of p < 0.01 was adopted32.
We used several control tasks. Simple vocal-reaction time was mea-
sured by asking subjects to say go as quickly as possible every time a ACKNOWLEDGEMENTS
small dot appeared on a computer screen at random intervals. Articula-
The studies were funded by the EEC-BIOMED II grant (contract BMH4-CT96-
tion speed was measured by asking subjects to repeat aloud as quickly as
0274) and by the Gatsby Charitable Foundation. We are grateful to Andrew Holmes
possible, for 15 seconds, pairs of words common to both vocabularies
(gorilla/banana; tennis/polo). Two tasks measured ease of word retrieval. for statistical advice and Caroline Moore for help with preparing the bibliography.
In the verbal-fluency tasks, subjects generated in one minute as many
words as possible starting with the letter m (letter fluency) or belong-
RECEIVED 18 AUGUST; ACCEPTED 16 NOVEMBER 1999

ing to the category animals (semantic fluency). In a picture-naming
task, subjects were instructed to name as quickly as possible pictures
whose names were words common to both vocabularies and were 1. Nyikos, J. in The Fourteenth LACUS Forum 1987 (ed. Empleton, S.) 146163
(Linguistic Associations of Canada and the United States, Lake Bluff, Illinois,
rehearsed in advance: banana, bus, computer, gorilla, hamburger, pizza, 1988).
piano, spaghetti. 2. Lepschy, A. & Lepschy, G. La Lingua Italiana (Bompiani, Milan, 1981).
There were two PET scan experiments. In the explicit reading experi- 3. Cossu, G., Gugliotta, M. & Marshall, J. Acquisition of reading and written
ment, subjects read aloud words and non-words from their respective spelling in a transparent orthography: Two non-parallel processes? Reading
orthographies. The stimuli included those used in the behavioral reading and Writing: An Interdisciplinary Journal 7, 922 (1995).
4. Landerl, K., Wimmer, H. & Frith, U. The impact of orthographic consistency
experiment. The baseline was the resting state. The rate of presentation on on dyslexia: a GermanEnglish comparison. Cognition 63, 315334 (1997).
the computer screen was 2 s on and 1 s off. In the implicit-reading exper- 5. Frith, U., Wimmer, H. & Landerl, K. Differences in phonological recording in
iment7, participants performed a feature-detection task. This involved German- and English-speaking children. Scientific Study of Reading 2, 3154
detecting the presence or absence of ascenders (graphic features that go (1998).
above the midline of the word, as in b, l, or t but not in a, c or o) within 6. Frost, R., Katz, L. & Bentin, S. Strategies for visual word recognition and
orthographical depth: a multilingual comparison. J. Exp. Psychol. Hum.
visually presented words, non-words and false-font strings. The false Percept. Perform. 13, 104115 (1987).
fonts were created by substituting letters in the real words with non-let- 7. Price, C., Wise, R. & Frackowiak, R. Demonstrating the implicit processing of
ters matched for size and presence or absence of ascenders (for example, visually presented words and pseudowords. Cereb. Cortex 6, 6270 (1996).
cannon became and meter became ). The 8. Morton, J. & Patterson, K. in Deep Dyslexia (eds. Coltheart, M., Patterson, K. &
Marshall, J.) 91118 (Routledge and Kegan Paul, London, 1980).
requirements of the task remained constant across stimuli: a subject 9. Coltheart, M., Curtis, B., Atkins, P. & Haller, M. Models of reading aloud: Dual-
pressed one key of a response box with the right index finger if one or route and parallel distributed processing. Psychol. Rev. 100, 589608 (1993).
more ascender was present, and another key with the right middle fin- 10. Plaut, D., McClelland, J., Patterson, K. & Seidenberg, M. Understanding
ger if ascenders were absent. The stimuli included words and non-words normal and impaired word reading: computational principles in quasi-regular
used in the other experiments and were presented at the same rate. domains. Psychol. Rev. 103, 56115 (1996).
11. Zorzi, M., Houghton, G. & Butterworth, B. Two routes or one in reading aloud?
A connectionist dual-process model. J. Exp. Psychol. Hum. Percept. Perform. 24,
Subjects for PET studies. These studies were approved by the Ethics Com- 11311161 (1998).
mittees of the Institute of Neurology (London) and Institute H San Raf- 12. Patterson, K. & Behrmann, M. Frequency and consistency effects in a pure
faele (Milan). Six English and six Italian university students, matched surface dyslexic patient. J. Exp. Psychol. Hum. Percept. Perform. 23, 12171231
for age and IQ, participated in each study. In the explicit reading exper- (1997).
13. Price, C., Moore, C., Humphreys, G., Frackowiak, R. & Friston, K. The neural
iment, Italian subjects mean age and mean IQ s.d. were 27.8 6.6 and regions sustaining object recognition and naming. Proc. R. Soc. Lond. B Biol.
120.2 6.7. English subjects mean age and mean IQ were 23.2 2.9 and Sci. 263, 15011507 (1996).
113.2 5.8. In the implicit reading experiment, Italian subjects mean 14. Vandenberghe, R., Price, C., Wise, R., Josephs, O. & Frackowiak, R. S.
age and mean IQ were 24.7 4.4 and 123 10.3. English subjects mean Functional anatomy of a common semantic system for words and pictures.
age and mean IQ were 24.5 2.9 and 122.2 13.9. Nature 383, 254256 (1996).
15. Poldrack, R. et al. Functional specialization for semantic and phonological
processing in the left inferior prefrontal cortex. Neuroimage 10, 1535 (1999).
Data analysis. A full description of the H215O PET activation scanning 16. Paulesu, E. et al. Functional heterogeneity of left inferior frontal cortex as
technique and data analysis can be found elsewhere29. Regional cerebral revealed by fMRI. Neuroreport 8, 20112017 (1997).
blood flow (rCBF) was measured by recording the distribution of radioac- 17. Nespor, M. Fonologia (il Mulino, Bologna, 1993).
tivity following the intravenous injection of 15O-labeled water (H215O) 18. Cappa, S. F., Nespor, M., Ielasi, W. & Miozzo, A. The representation of stress:
evidence from an aphasic patient. Cognition 65, 113 (1997).
with a CTI Siemens Ecat HR+ PET scanner (CTI, Knoxville, Tennessee) 19. Petersen, S., Fox, P., Posner, M., Mintun, M. & Raichle, M. Positron emission
in London and the GE-Advance scanner (General Electric Medical Sys- tomographic studies of the processing of single words. J. Cogn. Neurosci. 1,
tem, Milwaukee, Wisconsin) in Milan. Twelve consecutive scans were 153170 (1989).
obtained for each subject in each experiment. The three stimulus con- 20. Demonet, J. et al. The anatomy of phonological and semantic processing in
ditions were presented in a counterbalanced order. Task-related differ- normal subjects. Brain 115, 17531768 (1992).
21. Demonet, J., Price, C., Wise, R. & Frackowiak, R. Differential activation of right
ences in regional cerebral blood flow were examined using statistical and left posterior sylvian regions by semantic and phonological tasks: a
parametric mapping (SPM 96) software (Wellcome Department of Cog- positron-emission tomography study in normal human subjects. Neurosci.
nitive Neurology, London, UK) on stereotactically normalized and Lett. 182, 2528 (1994).
smoothed PET images29. For each experiment, data were analyzed accord- 22. Paulesu, E., Frith, C. & Frackowiak, R. The neural correlates of the verbal
component of working memory. Nature 362, 342345 (1993).
ing to a random-effect model. Replications of each task were collapsed
23. Beauvois, M. F. & Derouesne, J. Phonological alexia: three dissociations. J.
into average images, yielding one average scan per reading task per sub- Neurol. Neurosurg. Psychiatry 42, 11151124 (1979).
ject, and the residual variance of subsequent statistical analyses incor- 24. Shallice, T. Phonological agraphia and the lexical route in writing. Brain 104,
porated the appropriate inter- and intra-subject variance components, 413429 (1981).

articles
25. Roeltgen, D. P. & Heilman, K. M. Lexical agraphia. Further support for the two- 31. Friston, K. J., Frackowiak, R. S. J., Mazziotta, J. C. & Evans, A. C. Assessing the
system hypothesis of linguistic agraphia. Brain 107, 811827 (1984). significance of focal activations using their spatial extent. Hum. Brain Mapp.
26. Hodges, J. R., Patterson, K., Oxbury, S. & Funnell, E. Semantic dementia: 1, 214220 (1994).
progressive fluent aphasia with temporal lobe atrophy. Brain 115, 17831806 32. Rumsey, J. et al. Phonological and orthographic components of word
(1992). recognition. Brain 120, 739760 (1997).
27. Jonsdottir, M., Shallice, T. & Wise, R. Phonological mediation and the 33. Nobre, A., Allison, T. & McCarthy, G. Word recognition in the human
graphemic buffer disorder in spelling: cross-language differences? Cognition inferior temporal lobe. Nature 372, 260263 (1994).
59, 169197 (1996). 34. Price, C. & Friston, K. Cognitive conjunction: a new approach to brain
28. De Mauro, T., Mancini, F., Vedovelli, M. & Voghera, M. Lessico di Frequenza activation experiments. Neuroimage 5, 261270 (1997).
dell Italiano Parlato (Etaslibri, Rome, 1993). 35. Bookheimer, S., Zeffiro, T., Blaxton, T., Gaillard, W. & Theodore, W. Regional
29. Friston, K. in Human Brain Function (eds. Frackowiak, R., Friston, K., cerebral blood flow during object naiming and word reading. Hum. Brain
Frith, C., Dolan, R. & Mazziotta, J.) 2541 (Academic, San Diego, 1997). Mapp. 3, 93106 (1995).
30. Frison, L. & Pocock, S. J. Repeated measures in clinical trials: analysis using 36. Menard, M. T., Kosslyn, S. M., Thompson, W. L., Alpert, N. M. & Rauch, S. L.
mean summary statistics and its implications for design Stat. Med. 11, Encoding words and pictures: a positron emission tomography study.
16851704 (1992). Neuropsychologia 34, 185194 (1996).

Nature Neuroscience January 2000

Încărcat de

Informații document

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Nature Neuroscience January 2000

Încărcat de

Drepturi de autor:

Formate disponibile

2000 Nature America Inc. http://neurosci.nature.

volume 3 no 1 january 2000

of multiple classes of ventral and

Sound and meaning: how native language affects reading strategies . . . . . . . . . . . 3

Building inhibitory synapses: exchange factors getting into the act? . . . . . . . . . . . . . . 5

Seeing is relieving: electrical synapses between visualized neurons . . . . . . . . . . . . . . . . 7

Collybistin, a newly identified brain-specific GEF, induces

nature neuroscience volume 3 no 1 january 2000 i

Nuclear Notch1 signaling and the regulation of dendritic development. . . . . . . . 30

The seven-transmembrane receptor Smoothened cell-autonomously

x A model for intradendritic computation of binocular disparity . . . . . . . . . . . . . . . 54

Touch can change visual slant perception . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

Different lateral amygdala outputs mediate reactions and actions

Integration of diverse information in working memory within the

A cultural effect on brain function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

English and Italian speakers

nature neuroscience volume 3 no 1 january 2000 ii

The brain within the genes

nature neuroscience volume 3 no 1 january 2000 1

news and views

Sound and meaning: how native

In the Italian language, spelling is consistently related to pronunciation, whereas English

nature neuroscience volume 3 no 1 january 2000 3

news and views

4 nature neuroscience volume 3 no 1 january 2000

news and views

because the size and shape of postsy-

exchange factors getting ments as well as microtubules. We would

question that is likely to be central to

nature neuroscience volume 3 no 1 january 2000 5

news and views

6 nature neuroscience volume 3 no 1 january 2000

news and views

in response to current application (Fig. 1a

synapses between visualized spaced neurons, which were examined in

nature neuroscience volume 3 no 1 january 2000 7

news and views

tial calculated from this chronization is not so precise as to require

8 nature neuroscience volume 3 no 1 january 2000

news and views

nature neuroscience volume 3 no 1 january 2000 9

news and views

10 nature neuroscience volume 3 no 1 january 2000

news and views

a Component b Plaid c Planar velocity when inconsistent motion compo-

x which purposefully avoids any lasting spa-

nature neuroscience volume 3 no 1 january 2000 11

1. Blakemore, C. & Cooper, G. F. Nature 228, 477478 (1970).

9. Ringach, D. L., Hawken, M. J. & Shapley, R. Nature 387, 281284 (1997).

nature neuroscience volume 3 no 1 january 2000 13

Molecular basis of NMDA receptor-

nature neuroscience volume 3 no 1 january 2000 15

a b on NMDAR responses 8,10, suggest that NO

Fig. 2. Voltage independence of NO-induced inhibition of

16 nature neuroscience volume 3 no 1 january 2000

Fig. 3. Effect of NR2A(C399A) on

receptors. Similar results were obtained in

Normalized response (%)

nature neuroscience volume 3 no 1 january 2000 17

Fig. 4. Physiological effects of endogenous

activity. HEK293 cells stably transfected

points are means s.e. of 3 independent experiments, each

18 nature neuroscience volume 3 no 1 january 2000

[m7G(5)P(5)G] from 1 g of linearized template in