Nature Neuroscience June 2002

contents
volume 5 no 6 june 2002

© 2002 Nature Publishing Group http://neurosci.nature.com
http://neurosci.nature.com
editorial
The timing of neuronal activity is
proposed to be important for Of apes and men . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 501
binding features of a complex
sensory stimulus. Christensen and
colleagues recorded news and views
simultaneously from pairs of pro-
jection neurons in the Quick-change artist: from excitatory to inhibitory synapse
pheromone-receptive in minutes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 503
macroglomerular complex of Story C. Landis
male sphinx moths and found SEE ARTICLE, PAGE 539
more synchrony of responses to a
specific odor component among Putting odor maps in sync . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 505
neurons that innervated the same Wei R. Chen and Gordon M. Shepherd
rather than separate glomeruli. SEE ARTICLE, PAGE 557
This synchrony was enhanced by
inhibitory influences from neigh- Posterior parietal cortex: not just where, but how . . . . . . . . . . . . . . . . . . . . . . . 506
boring glomeruli responding to a
Joshua I. Gold and Mark E. Mazurek
different, but chemically similar
SEE ARTICLE, PAGE 580
pheromone. Photograph courtesy
of Photo Research. See pages 505 Illusions, perception and Bayes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 508
and 557.
Wilson S. Geisler and Daniel Kersten
Wnt signals lead cells down the caudal path . . . . . . . . . . . . . . . . . . . . . . . . . . . . 510

Brian Fiske
brief communications
Nax channel involved in CNS sodium-level sensing . . . . . . . . . . . . . . . . . . . . . . 511
T Y Hiyama, E Watanabe, K Ono, K Inenaga, M M Tamkun, S Yoshida
and M Noda
Acute versus chronic NMDA receptor blockade and synaptic
Getting to the brain AMPA receptor delivery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513
through the nose. J J Zhu and R Malinow
Page 514
Sniffing neuropeptides: a transnasal approach to
the human brain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 514
J Born, T Lange, W Kern, G P McGregor, U Bicke and H L Fehm
Nature Neuroscience (ISSN 1097-6256) is published monthly by Nature America Inc., 345 Park Avenue South, New York, NY 10010-1707. Editorial Office: 345 Park Avenue South, New York, NY
10010-1707. Tel: (212) 726 9200, Fax: (212) 696 9635. Annual subscription rates: USA/Canada: US$199/US$213 (personal), US$99/US$106 (student), Canada add 7% for GST: 140911595RT001;
U.K./Europe: £185 (personal), £105 (student); Rest of world (excluding China, Japan, Korea): £235 (personal), £110 (student); Japan: Contact Nature Japan K.K., MG Ichigaya Building 5F,
19-1 Haraikatamachi, Shinjuku-ku, Tokyo 162-0841. Tel: 81 (03) 3267 8751, Fax: 81 (03) 3267 8746. Authorization to photocopy for internal or personal use, or internal or personal use of specif-
ic clients, is granted by Nature Neuroscience to libraries and others registered with the Copyright Clearance Center (CCC) Transactional Routing Service, provided the base fee of $9.00 an article (or
$1.00 a page) is paid direct to CCC, 27 Congress Street, Salem, MA 01970, USA. Back issues: US$45, Canada add 7% for GST; Periodicals postage rate paid at New York, NY and additional mailing
offices. CPC PUB AGREEMENT #40032744. POSTMASTER: Send address changes to Nature Neuroscience Subscription Department, P.O. Box 5054, Brentwood, TN 37024-5054. Printed by Publish-
ers Press, Inc., Lebanon Junction, KY, USA. Copyright © 2002 Nature America Inc.
nature neuroscience • volume 5 no 6 • june 2002 i

contents
commentary
Thalamcortical optimization of tactile processing according to
behavioral state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 517
M A L Nicolelis and E E Fanselow
articles
Progressive induction of caudal neural character by

graded Wnt signaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 525
U Nordström, T M Jessell and T Edlund
SEE NEWS AND VIEWS, PAGE 510
Ion channel properties underlying axonal action potential initiation

in pyramidal neurons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 533
A proposed model of C M Colbert and E Pan
thalamocortical processing.
Page 517.
A rapid switch in sympathetic neurotransmitter release properties
mediated by the p75 receptor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 539
B Yang, J D Slonimsky and S J Birren
Ethanol elicits and potentiates nociceptor responses via

the vanilloid receptor-1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 546
M Trevisani, D Smart, M J Gunthorpe, M Tognetto, M Barbieri, B Campi,
S Amadesi, J Gray, J C Jerman, S J Brough, D Owen, G D Smith, A D Randall,
S Harrison, A Bianchi, J B Davis and P Geppetti
A computational role for slow conductances: single-neuron models
Computing sound duration that measure duration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 552
with a slow conductance. S L Hooper, E Buchman and K H Hobbs
Page 552.
Local inhibition modulates odor-evoked synchronization of
glomerulus-specific output neurons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 557
H Lei, T A Christensen and J G Hildebrand
Decreasing hypothalamic insulin receptors causes hyperphagia

and insulin resistance in rats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 566
S Obici, Z Feng, G Karkanias, D G Baskin and L Rossetti
Calcium–calmodulin-dependent protein kinase IV is required
for fear memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 573
F Wei, C Qiu, Liauw, D A Robinson, N Ho, T Chatila and M Zhuo
Non-spatial, motor-specific activation in posterior parietal cortex . . . . . . . . . . . 580
J L Calton, A R Dickinson and L H Snyder
CaMKIV and SEE NEWS AND VIEWS, PAGE 506
fear memory.
Page 573. Neural activity in early visual cortex reflects behavioral experience
and higher-order perceptual saliency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 589
T S Lee, C F Yang, R D Romero and D Mumford
Motion illusions as optimal percepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 598
Y Weiss, E P Simoncelli and E H Adelson
Stable perception of visually ambiguous patterns . . . . . . . . . . . . . . . . . . . . . . . . 605

D A Leopold, M Wilke, A Maier and N K Logothetis
classified advertising . . . . . . . . . . . . . . . . . . . . . . . . . . . . . see back pages
Stabilizing perception
of ambiguous patterns.
Page 605.
nature neuroscience • volume 5 no 6 • june 2002 ii

editorial
Of apes and men

How do our brains differ from those of our closest living relatives, during development. This limitation is unfortunate because early
the great apes (bonobos, chimpanzees, orangutans and gorillas)? development is when gene activity can profoundly influence over-
Despite our persistant curiosity about the issue, even very basic all brain structure, and differences in early gene activity probably
questions remain unresolved, such as whether differences in cog- drive evolutionary changes. However, such experiments are possi-
nitive abilities emerged from the addition of new types of neurons ble in monkeys, and they would help to focus more traditional
and cortical areas or enlargement of otherwise similar components. comparative studies. For example, using slice cultures of embry-
In a new study, Svante Pääbo and colleagues1 offer a fresh onic human, macaque and mouse brains, Letinic and Rakic2 recent-
approach to the problem by comparing gene and protein expression ly reported a migratory pathway from the telencephalon to the
on a large scale between ape and human brains. The study is diencephalon that exists in humans but is not apparent in macaque
provocative and raises many new questions. Nevertheless, given or rodent brain; this pathway seems to contribute to the expansion
the sheer complexity of neural structures, linking these molecules of particular brain regions in humans. (Because great apes were
to changes in circuits and behavior may prove very difficult. More not examined, it is not known whether this pathway is unique to
modest goals, however, seem achievable. humans.) The authors conclude that small changes in migratory
The authors removed gray matter from the left prefrontal lobe guidance cues during evolution could lead to the expansion of
of adult humans, chimpanzees, orangutans and macaques (which human-specific brain structures. Identifying the molecular basis
had died of natural causes), and then used oligonucleotide and of such differences would be an important advance.
cDNA arrays to compare gene expression in the brain, liver and A major challenge will be to understand what underlies the dif-
blood. Although non-human primate gene chips are not available, ferences in gene expression profiles. Differences in promotor
the authors were able to use probes based on human sequences, regions of genes and other cis-acting regulatory regions will sure-
because primate genomic DNA sequences are highly conserved ly contribute, and it may be possible to identify these. However,
(over 98% identity for coding regions). changes in RNA levels may result from other factors, such as trans-
By summing differences in mRNA levels over all genes for each activating proteins. A detailed comparison of the human and chim-
tissue, the authors report that for blood leukocytes and liver, the panzee genome sequences will be of great use here—and although
human expression pattern was more similar to that of chimpanzees progress on chimpanzee sequencing has so far been limited to
than to that of macaques. This result was expected—humans are chromosome 21, it now seems likely that substantial resources will
more closely related to chimpanzees than chimpanzees are to be committed to this project.
macaques. However, in brain tissue, the chimpanzee and macaque Most importantly, conclusions based on informatics
expression patterns were more similar to each other than to the approaches will need to be placed into a broader biological con-
human pattern. The authors conclude that human evolution text. Even within a small area of the brain, the interpretation of
included large and rapid changes in gene expression levels in the averaged mRNA abundance levels can be complicated by large
brain compared with other organs. differences in the expression of given transcripts between differ-
Human and chimpanzee brains differ substantially, yet their ent cell types. Therefore, several groups are using array tech-
DNA sequences do not. Thus the general conclusion that large nologies together with techniques that offer high spatial
changes in gene expression might account for brain differences is resolution. For example, David Anderson and colleagues3 com-
not necessarily surprising. The goal now is to use this information bined microarray experiments with in situ hybridization to iden-
about expression differences to understand what changes at the tify genes that respected anatomically defined subdivisions of the
level of DNA led to a brain with different functional properties. amygdala. Such gene expression domains also defined subdivi-
First, it will be necessary to verify that the reported gene expres- sions that were not readily apparent with anatomical staining
sion patterns are stable and reproducible. This is important techniques. Other groups are using microarrays and in situ
because microarray technology is still rapidly evolving—as are the hybridization with similar goals of identifying transcripts
bioinformatics tools that are necessary to interpret the results. For expressed preferentially in various neocortical areas. This ‘mol-
example, the oligonucleotide chip technique indicated considerable ecular anatomical’ information could be used in comparative
within-species variation: one human brain sample differed as studies to understand the relationship between human and great
much from the other human samples as it did from chimpanzee ape cortical areas, and perhaps to identify regions that have
brain. Furthermore, it will be critical to determine whether the evolved very recently in humans.
observed changes reflect ‘recent’ (and less informative) transcrip-
tion events associated with learning and memory, for example, or 1. Enard, W. et al. Science 296, 340–343 (2002).
more profound interspecies differences. 2. Letinic, K. & Rakic, P. Nat. Neurosci. 4, 931–936 (2001).
Given the serious ethical issues involved in experimenting on 3. Zirlinger, M., Kreiman, G. & Anderson, D. J. Proc. Natl. Acad. Sci. USA 98,
great apes, similar comparative experiments are unlikely to be done 5270–5275 (2001).
nature neuroscience • volume 5 no 6 • june 2002 501

news and views
Quick-change artist: from excitatory

to inhibitory synapse in minutes
Story C. Landis
Cultured sympathetic neurons contain an excitatory transmitter, norepinephrine, and one that
is inhibitory, acetylcholine. A new paper shows that BDNF increases the ratio of acetylcholine to
norepinephrine release, reversing the effect of neural stimulation from excitation to inhibition.
Many, if not most, neurons synthesize, lishment of functional connections dition4. Several lines of evidence suggest
store and release multiple neuroactive between a single sympathetic neuron and that the effect is presynaptic, resulting
substances. In most cases, neurons use a small clusters of cardiac myocytes4. These from increased norepinephrine release.
small-molecule neurotransmitter and connections can be readily monitored by Not surprisingly, the effect is mediated
one or more neuropeptides. These by trkA, the tyrosine kinase recep-
are stored in synaptic and large tor, which transduces other effects
dense-core vesicles, respectively, of NGF on sympathetic neurons,
and release from these two pools neuron including survival and growth.
can be differentially regulated. In In the present studies, Yang and
a growing number of instances, colleagues examined the effects of
however, the same neuron has myocytes treating the sympathetic neuron
been observed to synthesize two and myocyte cocultures with
small-molecule transmitters, both BDNF3. In striking contrast to the
of which are presumably stored in effects of NGF, after treatment of
synaptic vesicles. The first exam- the cultures for 15 minutes or sev-
ple was uncovered in analyses of eral days with BDNF, neuron stim-
neonatal sympathetic neurons ulation significantly decreases the
developing in microcultures with axon varicosity
rate at which the myocytes con-
cardiac myocytes. Electrophysio- Amy Center
tract. The inhibitory effects of neu-
logical and ultrastructural studies Fig. 1. In the culture system used by Yang and colleagues, indi- ron stimulation observed after
revealed that under these condi- vidual sympathetic neurons form an extensive network of BDNF treatment are blocked by
tions, individual neurons produce processes in association with small clusters of myocytes. The the muscarinic antagonist atropine,
both norepinephrine and acetyl- myocytes contract or beat spontaneously. When the neuron is indicating that they are mediated
choline1,2. The two transmitters stimulated with a depolarizing current, neurotransmitter is by acetylcholine. BDNF does not
have opposing actions on heart released from axonal varicosities. Under control conditions, affect the responses of the heart
myocytes: norepinephrine is exci- neuron stimulation leads to an increase in the myocyte beat rate myocytes to exogenous transmitter
tatory, depolarizing the myocytes due to the release of norepinephrine. When the cultures are agonists, consistent with a presy-
treated with BDNF, neuron stimulation leads to a decrease in
and speeding their beat, whereas naptic mechanism. The rapid
the beat rate. The simplest explanation of these results is that
acetylcholine is inihibitory, hyper- the axonal varicosities shown at higher magnification in the
induction of an inhibitory effect by
polarizing the myocytes and slow- inset contain two populations of synaptic vesicles, one shown in BDNF and recovery of excitation
ing their beat. In this issue, Yang blue that contains norepinephrine and a second shown in purple after its withdrawal make it unlike-
and colleagues describe the very that contains acetylcholine. Under control conditions, norepi- ly that the neurotrophin is affect-
surprising discovery that the neu- nephrine-containing vesicles are predominantly released, ing the relative synthesis of the two
rotrophin brain-derived neu- whereas BDNF suppresses norepinephrine release and transmitters. Instead BDNF seems
rotrophic factor (BDNF) alters enhances acetylcholine release. to increase the release of acetyl-
the relative release of the two choline preferentially. Thus, two
transmitters, favoring the secre- members of the neurotrophin fam-
tion of acetylcholine over norepineph- examining the rate at which the myocytes ily, NGF and BDNF, differentially affect
rine and transforming what was an contract. Under control conditions, they release of the two transmitters contained
excitatory synapse into an inhibitory one beat spontaneously. Stimulation of a in these bifunctional neurons.
within minutes3. nearby sympathetic neuron doubles the Neurotrophins bind and activate the
The culture system used by Yang and beat rate. This effect is blocked by the low-affinity neurotrophin receptor
coworkers (Fig. 1) encourages the estab- adrenergic antagonist, propranolol, indi- p75 NTR in addition to tyrosine kinase
cating that it is mediated by norepi- receptors5,6. Yang and colleagues provide
The author is at the National Institute of
nephrine. Treating cocultures with compelling evidence that the effects of
Neurological Disorders and Stroke, National increased levels of the neurotrophin BDNF on transmitter release by sympa-
Institutes of Health, Bldg. 36, Rm. 5A05, nerve growth factor (NGF) for 15 min- thetic neurons are mediated by p75NTR
36 Convent Dr., Bethesda, Maryland utes enhances the excitatory effects of and not by a tyrosine kinase receptor3.
20892-4150, USA. neuron stimulation so that the rate is First, neonatal sympathetic neurons do
e-mail: landiss@ninds.nih.gov fourfold greater than in the control con- not express significant levels of trkB or

news and views
trkC, the tyrosine kinase receptors acti- function sympathetic neurons contain properties and acquisition of cholinergic
vated by BDNF. Consistent with their two populations of small synaptic vesi- properties occurs during the first two
absence, the tyrosine kinase inhibitor cles, one noradrenergic and one cholin- postnatal weeks in the sympathetic neu-
K252a has no effect on the induction of ergic, and that signaling initiated by NGF rons that innervate sweat glands and
inhibition by BDNF. Second and most and BDNF through trkA and p75 NTR, periosteum in vivo13,14. It is possible that
compelling, sympathetic neurons isolat- respectively, increases the probability of BDNF and/or NGF acutely influence
ed from mice that lack the neu- release from one pool or the other. Con- which transmitter is released from these
rotrophin-binding, long form of p75NTR sistent with this notion, neurons releas- neurons when they are bifunctional.
do not respond to BDNF. In addition, ing norepinephrine and acetylcholine Of significantly more interest is the
stimulation of neurons containing contain both small granular vesicles that possibility raised by the authors that the
increased levels of p75 leads to a contain norepinephrine and small clear release of colocalized transmitters in
decreased myocyte beat rate in the pres- vesicles that are believed to contain central neurons could be independent-
ence of BDNF. Somewhat surprisingly, acetylcholine2,8. Whereas sympathetic ly and rapidly modulated. They provide
this effect was also observed when BDNF neurons form morphologically special- several examples of neurons that con-
was absent and when p75NTR lacking the ized synapses with each other in these tain two classical or small-molecule
ligand binding domain was overex- cultures, the connections that they make transmitters. These include glutamate
pressed. Thus, when expressed at high with myocytes lack well defined pre- and and GABA in hippocampal granule cells,
enough levels, p75NTR seems to be able postsynaptic active zones2,8. In contrast dopamine and glutamate in substantia
to signal in a ligand-independent man- to the wealth of knowledge about the nigra neurons, and GABA and glycine
ner. Whereas the signaling pathways molecular architecture of presynaptic in spinal interneurons3. An additional
downstream of the trk receptors are rel- active zones and regulation of transmitter example that lends itself to the kinds of
atively well understood, the mechanisms release, relatively little is known about analyses that Yang and colleagues have
by which binding of neurotrophins to transmitter release from unspecialized performed in the present studies is the
p75NTR activates downstream signaling axonal varicosities. It would therefore be release of glutamate at autapses formed
are incompletely delineated5,6. One rel- of interest to know whether cholinergic by individual serotonergic raphe neu-
atively well established pathway involves transmission is enhanced by BDNF treat- rons grown in microcultures15. It is clear
the activation of sphingomyelinase and ment at neuron–neuron synapses as it is that the ability to alter the relative
the production of ceramide7. Yang and at neuron–myocyte junctions. Although release of the two transmitters would
colleagues found that stimulating neu- differential regulation of release from two provide a previously unanticipated
rons in cultures that had been treated populations of synaptic vesicles is the opportunity for the generation of synap-
with C2 ceramide to mimic sphin- likeliest explanation, there are alterna- tic plasticity. Although members of the
gomyelinase activation elicited a tives. The amount of transmitter con- neurotrophin family elicit these changes
decrease in beat frequency. These data tained within synaptic vesicles depends in the present case, there is no reason to
are consistent with the notion of sphin- on re-uptake of transmitter or its pre- believe that they are the only triggers.
gomyelinase as a primary mediator. cursor by plasma membrane trans-
They do not, however, rule out the pos- porters, synthesis of transmitter in the 1. Furshpan, E. J. et al. Proc. Natl. Acad. Sci. USA
73, 4225–4229 (1976).
sibility that some of the growing list of cytoplasm, vesicular transporters and
recently identified p75 NTR interacting degradation 9 . Any of these processes 2. Furshpan, E. J. et al. J. Neurosci. 6, 1061–1079
(1986).
proteins also contribute5. could be influenced by neurotrophins,
3. Yang, B., Slonimsky, J. D. & Birren, S. Nat.
At first glance, the results described by resulting in changes in the amount of Neurosci. 5, 539–545 (2002).
Yang et al. 3 give the impression that norepinephrine or acetylcholine stored
4. Lockhart, S. T., Turrigiano, G. G. & Birren, S.
BNDF treatment eliminates norepineph- within vesicles and available for release. J. Neurosci. 17, 9573–9582 (1997).
rine release and initiates de novo acetyl- One of the most striking aspects of this 5. Bibel, M. & Barde, Y. A. Genes Dev. 14,
choline release, therefore giving rise to a study is the rapid switch between nora- 2919–2937 (2000).
complete switch in transmitter output. drenergic and cholinergic transmission 6. Kaplan, D. R. & Miller, F. D. Curr. Opin.
This may be an artifact of the myocyte elicited by BDNF treatment. Previous Neurobiol. 10, 381–391 (2000).
beating assay, which sums the adrenergic studies had established that the relative 7. Dobrowsky, R. T. et al. Science 265, 1596–1599
stimulation and cholinergic inhibition. balance between the synthesis and storage (1994).
Consistent with this interpretation, an of norepinephrine and acetylcholine is 8. Landis, S. C. Proc. Natl. Acad. Sci. USA 73,
increase in beat frequency after neuron altered by treating the neurons with neu- 4220–4224 (1976).
stimulation in BDNF-treated cultures is ropoietic cytokines, leukemia inhibitory 9. Fon, E. A. & Edwards, R. H . Muscle Nerve 24,
581–601 (2001).
unmasked by atropine. Intracellular factor (LIF) or ciliary neurotrophic factor
recordings from the myocytes would (CNTF). LIF and CNTF decrease the 10. Fukada, K. Proc. Natl. Acad. Sci. USA 82,
8795–8799 (1985).
resolve this issue because the cholinergic expression of noradrenergic properties
11. Patterson, P. H. & Nawa, H. Cell 72, 123–137
inhibition has a faster time course than and increase the expression of cholinergic (1993).
the adrenergic inhibition1,2. Finding that properties10–12. In contrast to the effect of 12. Saadat, S., Sendtner, M. & Rohrer, H. J. Cell
the effect is a modulation rather than a BDNF, the alterations in transmitter Biol. 108, 1807–1816 (1989).
switch would in no way diminish the metabolism induced by LIF and CNTF 13. Asmus, S. E., Parsons, S. & Landis, S. C.
interest in the finding but could shed light and the consequent change in functional J. Neurosci. 20, 1495–1504 (2000).
on the underlying mechanisms. output occur over the course of days and 14. Guidry, G. & Landis, S. C. Dev. Biol. 199,
As the authors posit, the most likely weeks. A developmental transmitter 175–184 (1998).
explanation for their results is that dual- switch involving the loss of noradrenergic 15. Johnson, M. D. Neuron 12, 433–442 (1994).
504 nature neuroscience • volume 5 no 6 • june 2002

news and views
principal neurons, the investigators 3

Putting odor maps in sync record responses from pairs of output
neurons to test the question: do neuron
Wei R. Chen and Gordon M. Shepherd pairs show greater synchrony when con-
nected to the same glomerulus compared
Dual recordings of projection neurons in the moth antennal to different glomeruli? An advantage of
lobe suggest a key role for synchrony resulting from the insect for this kind of study lies in
interglomerular interactions in coding odor representations. stimulus control; puffs of different dura-
tions, concentrations and repetition rates

can be controlled precisely, simulating
Insects have antennae, and vertebrates have degrees (constituting its ‘molecular recep- the way a moth normally follows an odor
noses, making it seem that they have adopt- tive range5); conversely, a given odor has plume, which is to sense changes in the
ed different strategies for sensing odors in different affinities for different receptors. frequency of the intermittent odor
their environment. Despite this difference, Each glomerulus receives its input plumes emitted.
the sensory receptor neurons in both organs from a subpopulation of sensory recep- The results show that each odor com-
send their axons to converge in the brain in tor cells expressing one receptor type. ponent activates synchronous bursts of
small rounded regions called glomeruli. There are up to several hundred glomeruli action potentials in paired cells connected
These are dense collections of synaptic con- in an insect antennal lobe and several to the same glomerulus. Synchrony in
nections between sensory neuron axon ter- thousand in a vertebrate olfactory bulb. response to a single odor is much reduced
minals and the dendrites of the principal A given glomerulus thus reflects the mol- when the cells are connected to different
output neurons (Fig. 1). Glomeruli thus ecular receptive range of the receptor glomeruli. The similarity of the bursts
hold a key to understanding the neural basis expressed by the population of sensory extends to the timing of the individual
of the sense of smell. Glomeruli are also the neurons that projects to it. This makes the impulses, which is key, because it means
most distinct anatomical modules in the glomerulus one of the most specific func- that the synchronous impulses sum to
brain, and are therefore of interest to neu- tional modules in the nervous system. amplify the strengths of the individual
roscientists with regard to the general prin- Exposure to different odors activates responses. These results add to previous
ciples of functional module organization, different glomeruli to different extents,
such as cortical columns and barrels. producing the spatial patterns of activa- Odors
Until now, there have been two main tion of the glomerular population. This is C15 BAL
views about the functions of glomeruli. true for all the main types of odors: simple
One is that the stimulating molecules are monomolecular compounds, pheromones Antennae Receptor
encoded in the spatial pattern of activity and complex odor blends. As shown by neurons
in the glomerular population1. A second many different imaging methods, these

is that oscillatory temporal firing patterns spatial patterns form an odor map, an
of the output neurons are sufficient to ‘odor image’, of the information contained Antennal lobe
encode the odor stimuli2. In this issue, Lei in the stimulating molecules (summarized
et al.3 now report evidence for a third view, in ref. 6). The overlaps of molecular recep- CUM TOR
PN
that synchronization of output neuron tive ranges of different receptors enable Glomeruli
INT
activity at specific sites within the odor the system to cover the thousands of PN
map is crucial 4,5. This is an important potential types of molecules in odor space.
question not only for olfaction, but also By the same token, they add to the com-
for other systems where synchronous fir- plexity of the odor maps.
ing of cells is regarded as critical for such Lei et al. used a well tested organism, Mushroom bodies
Amy Center
properties as binding of distributed cell the sphynx moth, which has a pair of dis-
responses into coherent representations of tinct glomeruli, called cumulus and toroid, Fig. 1. The insect olfactory pathway. Two sub-
the stimuli. Understanding the significance that are activated selectively by compo- sets of receptor neurons in the antennae (red
of the new data requires an acquaintance nents of the female sex pheromone: the and blue) express different olfactory receptors
with the organization of a glomerulus. cumulus to component C15 and the toroid and respond preferentially to pheromone com-
Odor stimulation begins with the to component BAL (Fig. 1). Each of these ponents C15 and BAL, respectively. Each sub-
action of odor molecules on olfactory glomeruli has a small population of out- set sends its axons to its respective glomerulus,
cumulus (CUM) and toroid (TOR). Each
receptor proteins in the membranes of cilia put (principal) neurons in the antennal glomerulus has its subset of principal output
on the sensory receptor neurons. Each sen- lobe connected to it. The principal neu- neurons (PN), which project their axons to the
sory neuron expresses one of a hundred or rons send their dendrites to receive input mushroom bodies, where higher-level process-
so types of olfactory receptor proteins in from only a single glomerulus, and trans- ing takes place. Interneurons (INT) connect to
the insect and one of a thousand types in mit their responses through their axons to multiple glomeruli, to mediate inhibitory
the mammal. Various odor molecules will higher centers in the insect brain (the actions within and between glomeruli. Lei et al.3
activate a given receptor type to varying mushroom bodies). The relations between recorded from pairs of PNs and found that
principal neurons, their glomeruli, and each odor component activates synchronous
their projections to the mushroom bodies bursts of action potentials in paired cells con-
The authors are in the Department of nected to the same glomerulus. Furthermore,
Neurobiology, Yale University School of have been demonstrated particularly clear-
lateral inhibition between glomeruli mediated
Medicine, 333 Cedar Street, New Haven, ly by enhancer trap techniques7,8. by interneurons increases synchronization in
Connecticut 06510, USA. Using intracellular recordings, com- response to simultaneous stimulation by both
e-mail: gordon.shepherd@yale.edu bined with staining of the recorded odor components.

news and views
evidence in olfactory bulb slices that a blend of both components is used as the idea, all of the odor processing in the insect
glomerulus can function to synchronize stimulus, the initial mutual inhibition antennal lobe is carried out by glomeruli.
burst responses9,10, and show that syn- between principal neurons occurs simul- In contrast, the vertebrate olfactory
chronization occurs in the insect in taneously, enhancing their subsequent syn- bulb has a second level of powerful
response to natural odor stimulation. The chronized bursts. If this were classical GABAergic interactions between the
authors postulate that synchronization can lateral inhibition, activating the glomeruli dendrites of output cells (mitral/tufted
enhance transmission of weak olfactory sig- together should have reduced the respons- cells) and the granule cell interneurons1.
nals and increase the contrast between the es, as happens in the retina when a light Are these mechanisms folded into the
stimulus and background odors. In terms shines on both the center and the sur- mechanisms of the glomeruli in the
of signal detection theory, this means round. Instead, synchronization is antennal lobe for output to the mush-
enhancing the signal-to-noise ratio, a fun- observed in the insect glomerulus response. room bodies, or are they unique to the
damental property of every sensory system, This result extends previous work point- vertebrate? With questions like these on
indeed, every system in the brain. ing to the importance of synchronization the agenda, there will be plenty of action
These results add to the number of in processing odor information2. Compu- for those interested in how neural mod-
mechanisms that have already been iden- tational models have shown how changes ules process information in the inverte-
tified for signal-to-noise enhancement in sychronization alone, independent of brate and vertebrate brain.
within a glomerulus. These include mas- spike frequencies or differences between
sive convergence of the incoming axons firing rates of different cells, can produce 1. Shepherd, G. M. & Greer, C. A. in The Synaptic
Organization of the Brain (ed. Shepherd, G. M.)
(up to 5000:1), voltage-gated dendritic changes in odor learning13. 159–203 (Oxford, New York, 1998).
properties that amplify the synaptic Lei et al. note that the enhanced pre-
2. Laurent, G. et al. Annu. Rev. Neurosci. 24,
responses11, and glutamate autoreceptors cision of the ensemble spiking bursts of 263–297 (2001).
that mediate re-excitation of the receiving the output cells is a “potential means of 3. Lei, H., Christensen, T. A. & Hildebrand, J. G.
dendrites9. All these mechanisms togeth- strengthening the spatial representation Nat. Neurosci. 5, 557–565 (2002).
er make the glomerulus a functional unit of the stimulus” by the odor maps. Their 4. Kauer, J. S. Trends Neurosci. 14, 79–85 (1991).
par excellence for detecting weak signals study sets a standard for future studies 5. Kashiwadani, H., Sasaki, Y. F., Uchida, N. &
within the noisy odor environment. of temporal patterning, which will Mori, K. J. Neurophysiol. 82, 1786–1792 (1999).
The second important finding of Lei et require precise knowledge of the con- 6. Xu, F. Q., Greer, C. A. & Shepherd, G. M.
al.3 is that the excitatory responses also pro- nections of the recorded cells within the J. Comp. Neurol. 422, 489–495 (2000).
duce an inhibitory hyperpolarization in the glomerular map. 7. Wong, A. M., Wang, J. W. & Axel, R. Cell 109,
other glomerulus. A common hypothesis Comparisons with the vertebrate olfac- 229–241 (2002).
has been that inter-glomerular inhibition tory bulb are particularly intriguing with 8. Marin, E. C., Jefferis, G. S. X. E., Komiyama, T.,
represents the kind of lateral inhibition that regard to cross-phyla principles1. At the Zhu, H. & Luo, L. Cell 202, 243–255 (2002).
is well known in the visual system, where glomerular level in the vertebrate olfactory 9. Carlson, G. C., Shipley, M. T. & Keller, A.
J. Neurosci. 20, 2011–2021 (2000).
it is involved in center–surround contrast bulb, most of the inhibitory GABAergic
enhancement. By analogy, it has been sug- interneurons have dendrites restricted to 10. Schoppa, N. E. & Westbrook, G. L. Neuron 31,
639–651 (2001).
gested in the olfactory bulb 4 and the one glomerulus; in contrast, in the insect
11. Chen, W. R., Midtgaard, J. & Shepherd, G. M.
insect12 that lateral inhibition can sharpen the GABAergic interneuronal dendrites Science 278, 463–467 (1997).
feature detection between cells connected penetrate multiple glomeruli (Fig. 1). In 12. Sachse, S. & Galizia, C. G. J. Neurophysiol. 87,
to neighboring glomeruli. this respect, the insect glomerulus may be 1106–1117 (2002).
Lei et al. also introduce a new insight to a more complex integrative unit than the 13. Linster, C. & Cleland, T. A. J. Comput. Neurosci.
the concept of lateral inhibition: when a vertebrate glomerulus. In support of this 10, 187–193 (2001).
salient spatial cues and our responses to

Posterior parietal cortex: them. The parietal lobes, for example,
have long been known to be central to
visually guided behaviors. Indeed, pari-
not just where, but how etal damage can give rise to a host of clin-
ical disorders, including hemispatial
Joshua I. Gold and Mark E. Mazurek neglect (lack of awareness of the half of
space contralateral to the lesion), simult-
Posterior parietal cortex is critical for spatial processing. Calton agnosia (inability to see multiple objects
et al. show that this area also represents the intention to make simultaneously) and optic ataxia
a certain type of movement, independent of its spatial target. (impairment of visually guided reaching).
These findings have cemented a primary
role in spatial cognition for the parietal
Shine a bright spot of light in an other- lobes, and in particular a subregion called
Joshua Gold is in the Department of
wise dark room, and the natural reaction the posterior parietal cortex (PPC)1.
Neuroscience, University of Pennsylvania,
Philadelphia, Pennsylvania 19104, USA. Mark
is to attend to it, to look at it, possibly Based on these clinical findings, the
Mazurek is in the Department of Physiology even to reach toward it. These reactions PPC has become a prime target for study-
and Biophysics and RPRC, University of are so strong that it probably comes as no ing the neural mechanisms that link spa-
Washington, Seattle, Washington 98195, USA. surprise that extensive regions of the brain tial perception and action. Both
e-mail: jig@shadlen.org seem to be dedicated to processing such electrophysiological recordings in mon-

news and views
neurons responded more strongly

for an impending arm movement
than for an impending eye move-
ment. Some were even the same
neurons that responded in a spa-
tially selective manner in the first
task and thus signaled different
aspects of the impending move-
ment in different behavioral con-

texts. These results indicate that,
contrary to conventional wisdom,
information processing in the PPC
is not always inherently spatial but
is more flexible and may also
encode specific motor intentions.
Fig. 1. A task used by Calton et al. to dissociate effector selection from spatial processing. To appreciate the potential dif-
ficulty in interpreting such data,
consider another possible explana-
keys and imaging studies in humans have tion. Perhaps the PPC responses that were
measured PPC activity while the subject is function6. In many tasks used to study the evident before the location of the target
making judgments about the position, PPC, monkeys are trained to make an eye was specified simply reflected the mon-
motion, shape, color or other attribute of or an arm movement to a visual target. key’s confidence that such a target would
sensory cues to guide movements of the These valuable studies have shown that eventually appear at a particular location.
eye or arm2–4. A common theme emerg- PPC neurons encode intended move- In this case, it would mean that the PPC
ing from these studies is that the PPC is ments toward particular places in space, signals during the delay period reflected
involved in identifying and interpreting and that they do so in an effector-specif- a spatial expectation, evident before any
salient stimuli and transforming this infor- ic manner (for instance, neurons can be spatial instruction was given. However,
mation into signals that guide movements specific for eye or arm movements) 2 . such an expectation would depend criti-
to particular locations5. In this issue, Cal- However, Calton and colleagues reasoned cally on the number of potential target
ton and colleagues now report that some that these tasks might confound two sep- locations: more locations imply more
PPC neurons also represent the intention arate processes, one concerned with uncertainty and therefore should produce
to make either an eye movement or an selecting the appropriate effector (typi- weaker anticipatory signals associated
arm movement before the animal knows cally thought to be the realm of motor with each location, as has been observed
the spatial parameters of the movement6, cortex and not the PPC) and another in brain areas that prepare spatially selec-
suggesting a more general role for the PPC with specifying a spatial goal (a tradi- tive eye-movement responses8,9. Instead,
in the transformation of sensory informa- tional view of PPC function). They there- Calton and colleagues demonstrated that
tion into the preparation for action. fore trained monkeys on two tasks although additional target uncertainty did
The ability to relate neural activity to designed to differentiate the neural sig- increase the monkeys’ response times as
behavior is the compelling feature of these nals responsible for these two processes. expected, it did not change the effector-
kind of studies, but also their Achilles’ In the first, a visual cue indicates the spa- specific signals in the PPC. This control
heel. Even the simplest perceptual tasks tial target of the movement, but the effec- buttresses the argument that these PPC
involve multiple stages of processing relat- tor is not specified until after a delay signals were independent of any spatially
ed to sensation, action and the cognitive period. In the second, a visual cue indi- selective processing related to the
functions that link them, such as atten- cates to the monkey whether it will be impending movement.
tion, intention and decision-making. How making an eye or an arm movement, but The discovery of non-spatial, effec-
can attention to a particular location in the spatial target of the movement is not tor-specific signals suggests that spatial
space be dissociated from the intention to revealed until after the delay (Fig. 1). processing, though undoubtedly funda-
look or reach there? How can a perceptu- For the first task, they found that some mental to PPC function, is not the whole
al decision be dissociated from the plan to neurons in the PPC responded during the story. The PPC is firmly established as
respond in a certain way? Carefully delay period in a spatially selective man- part of the ‘where’ pathway for visual
designed experiments are required to dis- ner. That is, after the monkey was processing10. These new results support
tinguish these mental processes and avoid instructed where the movement would go the idea that this region of cortex also
the dreaded question, “How do you know but not yet whether it would involve the reflects a ‘how’ component of motor
the task was being solved the way you arm or the eyes, the neural responses pre- intention11 that is not necessarily spa-
think it was being solved?” Particular care dicted the direction of movement. This tially selective. Thus, spatial processing
is needed when relating a specific mental result is consistent with the PPC’s well- may just be one part of a more general
process under investigation to neural supported role in spatial processing. For role of the PPC in the transformation of
activity in an area such as the PPC that the second task, they found that some sensory information into the prepara-
represents the convergence of many sen- PPC neurons responded during the delay tion for action.
sory, motor and cognitive processes7. period in an effector-specific manner.
Calton and colleagues used a novel That is, after the monkey was instructed 1. Colby, C. L. & Olson, C. R. in Fundamental
Neuroscience (eds. Zigmond, M. J., Bloom,
task design to meet these challenges and whether to move its arm or eyes but not F. E., Landis, S. C., Roberts, J. L. & Squire,
thus expand our understanding of PPC yet where to direct the movement, certain L. R.) 1363–1383 (Academic, New York, 1999).

news and views
2. Snyder, L. H., Batista, A. P. & Andersen, R. A. 5, 10–16 (2001). 9. Dorris, M. C. & Munoz, D. P. J. Neurosci. 18,
Vision Res. 40, 1433–1441 (2000). 7015–7026 (1998).
6. Calton, J. L., Dickinson, A. R. & Snyder, L. H.
3. Colby, C. L. & Goldberg, M. E. Annu. Rev. Nat. Neurosci. 5, 580–588 (2002). 10. Ungerleider, L. G. & Mishkin, M. in Analysis of
Neurosci. 22, 319–349 (1999). Visual Behavior (eds. Ingle, D. J., Goodale,
7. Sparks, D. L. Curr. Opin. Neurobiol. 9, M. A. & Mansfield, R. J. W.) 549–586 (MIT
4. Culham, J. C. & Kanwisher, N. G. Curr. Opin. 698–707 (1999). Press, Cambridge, Massachusetts, 1982).
Neurobiol. 11, 157–163 (2001). 8. Basso, M. A. & Wurtz, R. H. Nature 389, 11. Goodale, M. A. Curr. Opin. Neurobiol. 3,
5. Gold, J. I. & Shadlen, M. N. Trends Cogn. Sci. 66–69 (1997). 578–585 (1993).
observers’. As one might guess intuitive-

Illusions, perception and ly, the primary objective of an ideal
observer is to compute the probability of
Bayes each possible true state of the environ-

ment given the stimulus on the retina
Wilson S. Geisler and Daniel Kersten (the ‘posterior probability distribution’).
According to Bayes’ theorem, the poste-
A new model shows that a range of visual illusions in humans can rior probability is proportional to the
be explained as rational inferences about the odds that a motion product of the probability of each possi-
stimulus on the retina results from a particular real-world source. ble state of the environment before
receiving the stimulus (the prior proba-
Artists can create powerful illusions of This study is an excellent example of bility) and the probability of the stimu-
distance, size, shape and orientation by how Bayesian concepts are transforming lus given each possible state of the
mimicking on canvas the images that perception research by providing a rigor- environment (the likelihood). Thus,
would be formed on the retina by per- ous mathematical framework for repre- when an ideal observer (Fig. 1) receives
spective projection from a three-dimen- senting the physical and statistical a stimulus, it computes the likelihood
sional environment. These perceptual properties of the environment, describing and then multiplies by the prior proba-
errors seem to reveal a rational (but the tasks that perceptual systems are trying bility distribution to obtain the posterior
automatic) perceptual system designed to perform, and deriving appropriate com- probability distribution. In many appli-
to correctly interpret the retinal images putational theories of how to perform those cations, prior probability distributions
evoked by the world. This appealing tasks, given the properties of the environ- over the space of possible objects, events
explanation of visual illusions is often ment and the costs and benefits associated and/or lightings represent the ideal
viewed as insufficient, however, because with different perceptual decisions. observer’s knowledge of the environ-
many idiosyncratic illusions seem unlike The Bayesian framework had its ment, and likelihood distributions rep-
the rational solution to any problem. beginnings in Helmholtz’s notion of resent the ideal observer’s knowledge of
One such class of illusions concerns ‘unconscious inference’—the idea that projective geometry and the space of pos-
the effect of luminance contrast and the visual system incorporates implicit sible viewpoints. Once the posterior
shape on the perception of motion veloc- knowledge of the environment and probability distribution is determined,
ity. Surprisingly, the apparent speed and image formation, and uses this knowl- an ideal observer convolves the posteri-
direction of a moving pattern often edge to infer, automatically and uncon- or distribution with a utility function (or
changes substantially as the contrast and sciously, object properties from the loss function), which specifies the costs
shape of the pattern is varied. Such illu- ambiguous images they form on the and benefits associated with the differ-
sions are typically interpreted as the retina (see also refs. 2–4). As Helmholtz ent possible errors in the perceptual deci-
errors or epiphenomena of some impre- knew, retinal images are ambiguous sion. The result of this operation is the
cise neural mechanism that is attempt- because of natural variations in view- expected utility (or Bayes’ risk) associat-
ing to compute one of the quantities point and lighting: very different objects ed with each possible interpretation of
relevant for motion perception. Howev- can give rise to similar retinal images, the stimulus. Finally, the ideal observer
er, using the tools and logic of Bayesian and the same object can give rise to very picks the interpretation that has the max-
statistical decision theory, in this issue different retinal images. For example, a imum expected utility.
Weiss et al. 1 show that many of these circle and an ellipse can produce exact- Weiss et al. 1 derived such an ideal
seemingly idiosyncratic motion illusions, ly the same retinal image if the circle is observer and showed that it displays
in fact, may be exactly what one would slanted appropriately in depth, and the many of the same ‘illusions’ perceived by
expect from a rational perceptual system. same circle slanted in depth by different human observers. Their analysis was
Their results suggest that the human amounts can produce many different based on the plausible assumptions that
visual system may be closer to optimal images. Thus, the ambiguous stimulus low velocities are more likely than high
than once believed. in Fig. 1 can be seen as an ellipse in the velocities, and that there is more vari-
frontal plane or as a circle on a slanted ability at low contrasts (that is, retinal
plane. Bayesian statistical decision the- images are less reliable). Motion illusions
Wilson Geisler is in the Center for Perceptual
ory prescribes a framework for opti- can be understood as optimal adapta-
Systems, University of Texas Austin, Seay
mally interpreting such ambiguous tions rather than mistakes in other cases
Psychology Building, Austin, Texas 78712-7789,
USA. Daniel Kersten is in the Psychology Dept.,
retinal images. as well; for example, a Bayesian analysis
University of Minnesota, 75 E. River Rd., Theoretical devices that use Bayesian predicts correctly that changing the
Minneapolis, Minnesota 55455, USA. statistical decision theory to make opti- motion of a shadow alone can change the
e-mail: geisler@.psy.utexas.edu mal interpretations are called ‘ideal apparent three-dimensional motion of

news and views
Fig. 1. Bayesian ideal observers for tasks involving the perception

of objects or events that differ along two physical dimensions, such
as aspect ratio and slant, size and distance, or speed and direction
of motion. When a stimulus is received, the ideal observer com-
putes the likelihood of receiving that stimulus for each possible pair
of dimension values (that is, for each possible interpretation). It
then multiplies this likelihood distribution by the prior probability
distribution for each pair of values to obtain the posterior probabil-
ity distribution—the probability of each possible pair of values
given the stimulus. Finally, the posterior probability distribution is

convolved with a utility function, representing the costs and bene-
fits of different levels of perceptual accuracy, to obtain the expected
utility associated with each possible interpretation. The ideal
observer picks the interpretation that maximizes the expected util-
ity. (Black dots and curves indicate the maxima in each of the plots.)
As a tutorial example, the figure was constructed with a specific
task in mind; namely, determining the aspect ratio and slant of a
tilted ellipse from a measurement of the aspect ratio (x) of the
image on the retina. The black curve in the likelihood plot shows
the ridge of maximum likelihood corresponding to the combina-
tions of slant and aspect ratio that are exactly consistent with x; the
other non-zero likelihoods occur because of noise in the image and
in the measurement of x. The prior probability distribution corre-
sponds to the assumption that surface patches tend to be slanted
away at the top and have aspect ratios closer to 1.0. The asymmet-
ric utility function corresponds to the assumption that it is more
important to have an accurate estimate of slant than aspect ratio.
an object 5 . However, the examples of bution on object shapes and materials, a estimate of speed and direction. Several
Weiss et al.1 are particularly interesting prior probability distribution on lighting, other recent applications of Bayesian
because they seemed unlikely to be opti- and a stimulus likelihood distribution analysis show that the human visual sys-
mal adaptations before the Bayesian that incorporates the interactions between tem can dynamically adjust weights on
analysis was done. For example, this objects and lighting, and the effects of different information sources, often in a
analysis explains the odd combination of perspective projection and viewpoint. near-optimal way9–11.
facts that a thin horizontally moving This Bayesian generative model has Although the Bayesian model
rhombus appears to move diagonally at recently motivated a number of novel and proposed by Weiss et al.1 makes the cor-
low contrasts and horizontally at high testable hypotheses. To give just one rect qualitative predictions for many
contrasts, whereas a fat rhombus appears example, Bloj et al.8 demonstrated that motion illusions, the quantitative pre-
to move horizontally at all contrasts. perceived surface color can depend upon dictions are not perfect. This is not sur-
The Bayesian approach has advan- the perceived three-dimensional surface prising for a nearly parameter-free
tages over other approaches in percep- configuration in a perceptually bistable model, but the potential reasons for less-
tion research. To begin with, it prescribes image, even when the retinal images never than-perfect quantitative predictions are
a principled method for determining change. They predicted this effect from a worth considering. First, the predictions
optimal performance in a given percep- Bayesian ideal observer that understands depend on the exact shape of the prior
tual task. This can be a very useful exer- the effects of mutual surface illumination. probability and likelihood distributions.
cise because it forces one to consider Finally, the Bayesian approach allows The assumption that low velocities are
carefully the various constraints that one to understand precisely how the reli- more probable than high velocities is
apply in the perceptual task, and the ability of different sources of information, likely to be correct qualitatively, but the
Bayesian ideal observer provides an including prior knowledge, should be specific prior probability distribution that
appropriate benchmark against which to combined by a perceptual system. Differ- Weiss et al. assumed is still just an edu-
compare human performance. Further- ent sources of information do not always cated guess. Undoubtedly, the prior prob-
more, ideal observers are often easily keep the same relative reliability, and ability and likelihood distributions
modified by incorporating anatomical, hence a rational perceptual system should incorporated implicitly into the visual
physiological and other constraints, mak- adjust the weights that it assigns to dif- system arise through a combination of
ing them an excellent starting point for ferent information sources contingent evolution and perceptual learning, and
developing testable models6,7. upon their current relative reliabilities. thus it would be appropriate to estimate
Another advantage of the Bayesian This sort of weight adjustment is at the these distributions by measuring and
approach is that it divides perceptual tasks heart of the account of motion illusions analyzing natural scene statistics. Prior
into convenient and intuitive pieces that from Weiss et al.1. When contrast is low, probability and likelihood distributions
can be considered singly and then com- retinal image information becomes less measured in the natural environment
bined to understand the whole. For exam- reliable, and so the Bayesian ideal observ- may lead to more accurate predictions of
ple, the Bayesian approach naturally er shifts more weight to the prior proba- perceptual performance. For example, a
partitions the ‘generative model’ for reti- bility distribution on motion velocity; this Bayesian model derived from co-occur-
nal images into a prior probability distri- shift in relative weight alters the optimal rence statistics for the geometrical rela-

news and views
tionships between edges in natural scenes physical limitations, and from the design (Cambridge Univ. Press, 1996).
quantitatively predicts human ability to compromises required because biological 4. Yuille, A. L. & Bülthoff, H. H. in Perception as
detect contours on complex backgrounds, systems must perform many tasks. In sup- Bayesian Inference (eds. Knill, D. C. &
Richards, R. W.) 123–161 (Cambridge Univ.
under a range of conditions12. plementary material, Weiss et al. demon- Press, 1996).
Second, Weiss et al.1 used the standard strate that plausible neural constraints bring 5. Kersten, D. in The New Cognitive Neurosciences.
utility function that corresponds to pick- the predictions of their model into better 2nd Edn. (ed. Gazzaniga, M. S.) 353–363 (MIT
ing the interpretation with the maximum quantitative agreement with the data. Press, Cambridge, Massachusetts, 1999).
posterior probability (Fig. 1, black dot in Finally, real perceptual systems are 6. Geisler, W. Psychol. Rev. 96, 267–314 (1989).
lower middle panel). The choice of utili- designed through natural selection and 7. Liu, Z. & Kersten, D. Vision Res. 38, 2507–2519
ty function should have a minor effect in perceptual learning, which sometimes find (1998).
their situation, but the choice can be local optima in design space, rather than 8. Bloj, M. G., Kersten, D. & Hurlbert, A. C.
Nature 402, 877–879 (1999).
important in some situations 13,14. For the global optimum of the Bayesian ideal
example, if smaller errors have greater observer. However, this fact does not 9. Saunders, J. A. & Knill, D. C. Vision Res. 41,
3163–3183 (2001).
utility than larger errors, and if the pos- diminish the value of the Bayesian
10. Mamassian, P., Landy, M. S. & Maloney, L. T.
terior probability distribution contains a approach; indeed, the concepts of Bayesian in Statistical Theories of the Brain (eds. Rao, R.,
tall narrow peak and a short wide peak, statistical decision theory lend themselves Olshausen, B. & Lewicki, M.) 13–36 (MIT
then the maximum expected utility may elegantly to rigorous formulations of nat- Press, Cambridge, Massachusetts, 2002).
be at the short peak rather than the tall ural selection15. Thus, even in the broader 11. Ernst, M. O. & Banks, M. S. Nature 415,
peak. Also, the most appropriate utility biological context of plasticity, learning 429–433 (2002).
function when considering biological and natural selection, the Bayesian 12. Geisler, W. S., Perry, J. S., Super, B. J. &
Gallogly, D. P. Vision Res. 41, 711–724 (2001).
vision is arguably one based on fitness15. approach may prove to be optimal.
Third, as Weiss et al.1 point out, their 13. Brainard, D. H. & Freeman, W. T. J. Opt. Soc.
1. Weiss, Y., Simoncelli, E. & Adelson, E. H. Nat. Am. A 14, 1393–1411 (1997).
Bayesian model does not consider certain
Neurosci. 5, 598–604 (2002). 14. Schrater, P. R. & Kersten, D. How optimal
fundamental biological constraints, such as depth cue integration depends on the task. Int.
2. Freeman, W. T. Nature 368, 542–545 (1994).
the limited dynamic range and limited J. Comput. Vision 40, 73–91 (2000).
3. Knill, D. C., Kersten, D. & Yuille, A. in
speed of neural responses. Such constraints Perception as Bayesian Inference (eds. 15. Geisler, W. S. & Diehl, R. Phil. Trans. R. Soc.
undoubtedly arise (ultimately) from bio- Knill, D. C. & Richards, R. W.) 1–21 Lond. B Biol. Sci. 357, 419–448 (2002).
Wnt signals lead cells down the caudal path

During development, progenitor cells along the rostrocaudal axis of the neural tube are instructed to become forebrain, midbrain,
hindbrain or spinal cord. Cells of the caudal brain are believed to arise through reprogramming (‘caudalization’) of cells that initially
show characteristics of rostral brain. On page 525 of this issue, Thomas Edlund and
colleagues now report that Wnt signals from the posterior mesoderm are required for
caudalization. FGF and retinoic acid also contribute to the induction of midbrain,
hindbrain and spinal cord, but previous studies showed that these factors were not
sufficient on their own. Despite earlier indications that Wnt signals were involved in
specifying caudal brain character, because they have a variety of other functions in
development, it was unclear whether their role in caudalization was direct or
indirect. Therefore, the present work provides a crucial additional piece in this
developmental puzzle.
To examine the role of Wnt signaling, the authors used explant cultures of chick
neural plate along with immunohistochemical labeling for expression of a
combination of transcription factors which selectivity delineate the various brain
regions (yellow, rostral forebrain; red/green, rostral and caudal midbrain; light/dark
blue, rostral hindbrain). When explants of caudal neural plate were taken from a
stage at which the cells still exhibited primarily rostral characteristics and were co-
cultured with caudal mesoderm, cells expressed markers for caudal brain regions. If
Wnt signaling was inhibited, however, the cells retained their rostral character. If the
authors then cultured neural plate explants from a later stage when cells were
already specified to eventually make rostral, middle and caudal brain regions, they
found that Wnt signaling was still directly necessary for the induction of caudal
character. Finally, when explants from the eventual rostral forebrain region of the
neural plate were cultured in the presence of FGF and varying concentrations of
Wnt conditioned medium, the authors found that increasing concentrations of Wnt
resulted in expression of progressively more caudal brain markers. Therefore, the new results firmly establish a direct role for a
graded Wnt signal in directing the caudalization of neural plate cells during early neural tube development.
Brian Fiske

Nax channel involved in in which the Nax gene was knocked-out by insertion of the lacZ
gene in-frame; we found that the Nax channel was expressed in
CNS sodium-level sensing neurons in the CVOs, like the SFO and organum vasculosum
laminae terminalis (OVLT), which are important regions for the
control of body fluid ionic balance8. Under thirst conditions,
Takeshi Y. Hiyama1, Eiji Watanabe1, Kentaro Ono2, Nax-deficient mice showed hyperactivity of the neurons in these
Kiyotoshi Inenaga2, Michael M. Tamkun3, areas and ingested excessive salt; wild-type mice stopped salt
Shigeru Yoshida4,5 and Masaharu Noda1 ingestion. This led us to propose that Nax is involved in the sodi-
um-level sensing mechanism in the brain8.

1 Division of Molecular Neurobiology and Center for Transgenic Animals and We verified this possibility by imaging analysis9 of changes in
Plants, National Institute for Basic Biology, and Department of Molecular the intracellular sodium-ion concentration [Na+]i when extra-
Biomechanics, Graduate University for Advanced Studies, Myodaiji-cho, cellular sodium-ion concentration [Na+]o was raised stepwise
Okazaki 444-8585, Japan from the normal amount. We used dorsal root ganglion (DRG)
2 Department of Physiology, Kyushu Dental College, Manazuru, Kokurakitaku,
neurons because all DRG neurons are Nax-positive, whereas only
Kitakyushu 803-8580, Japan
a subpopulation of neurons in the SFO or OVLT are Na x -
3 Department of Physiology, Colorado State University, Fort Collins, Colorado
positive8. When [Na+]o was increased from the control amount of
80523, USA
145 mM to 170 mM (high sodium solution) by bath application
4 Laboratory of Intracellular Metabolism, National Institute for Physiological
of NaCl solution, the [Na+]i of dissociated neurons derived from
Sciences, Myodaiji-cho, Okazaki 444-8585, Japan
wild-type mice showed a pronounced increase from ∼10 mM to
5 Department of Physiology, Nagasaki University School of Medicine, Nagasaki
∼30 mM (118/118 samples; Fig. 1a–c). The time taken to reach
852-8523, Japan
the plateau was 19.3 ± 4.2 min (n = 85; Fig. 1b), and tetrodotox-
Correspondence should be addressed to M.N. (madon@nibb.ac.jp) in (TTX; 1 µM)—a potent blocker of TTX-sensitive voltage-gated
Published online: 6 May 2002, DOI: 10.1038/nn856 sodium channels—did not antagonize the response (Fig. 1c). In
contrast, there was no significant increase in the [Na+]i detected
Mammals feel thirsty or an appetite for salt when the correct bal- in DRG neurons (87/87 samples) isolated from Nax-deficient
ance between water and sodium in the body fluid has been dis- mice (Fig. 1a–c). The cytosolic sodium response was detected in
rupted, but little is known about the mechanism in the brain that DRG neurons of various sizes from wild-type mice, consistent
controls salt homeostasis. It has been postulated that the exis- with the ubiquitous expression of Nax in DRG neurons8.
tence of both an osmoreceptor and a specific sodium receptor is To identify the key factor involved in this response, we raised
essential if the experimental data are to be encompassed1,2. Sev- osmolarity or chloride concentration [Cl–]o to that of the ‘high
eral candidate osmoreceptors have been identified3–5, and here sodium solution’ without changing [Na+]o, and none of the cells
we show that the Nax channel in the circumventricular organs (from either wild-type or Nax-deficient mice) responded with a
(CVO) is a probable candidate for the specific sodium receptor. change in [Na+]i. When the sodium concentration was raised
The Na x channel—formerly called NaG/SCL11 (in rats), with sodium methanesulfonate, however, there was a response
Nav2.3 (in mice) and Nav2.1 (in humans)—has been classified (Fig. 1c). Thus we concluded that neurons responded to the rise
as a subfamily of voltage-gated sodium channels6. The primary in [Na+]o but not to the rise in osmolarity or [Cl–]o. [Na+]o at
structure of Nax, however, is markedly different from the other the half-maximal (C1/2) was 159 mM (Fig. 1d). When [Na+]o was
voltage-gated sodium channel family members and includes dif- lowered from the control amount of 145 mM to 140 or 120 mM,
ferences in the key regions for voltage sensing and inactivation7. no [Na+]i response was seen (Fig. 1d).
Until now, the functional properties of the channel have been We confirmed these findings with whole-cell current record-
poorly understood, as attempts at functional expression of Nax ings of DRG neurons using a patch-clamp technique. When ‘high
in heterologous systems have failed. We recently generated mice sodium solution’ was applied to cells derived from wild-type mice,
Fig. 1. Nax is a sodium concentration–sensitive sodium channel.

(a) Pseudocolor images showing [Na+]i of cells in the control solution Control 170 mM NaCl
([Na+]o = 145 mM) and in high sodium solution ([Na+]o = 170 mM) at
a b 40
+/+
the times indicated with vertical broken lines in (b). Neurons were +/+
[Na+] (mM)
30
derived from wild-type (+/+) and Nax-null (–/–) mice. Scale bars, 50 µm.
i
20
(b) Time-course of [Na+]i responses of the cells indicated by an arrow- 30 −/−
[Na+]i (mM)
head and arrow in (a). Time 0 was the time at which the extracellular 10
fluid was changed. (c) [Na+]i response is dependent on [Na+]o, but not −/− 20
170 mM NaCl
on extracellular [Cl–]o or osmotic pressure. Instead of NaCl, 50 mM 10 0
–10 0 10 20 30
mannitol, 25 mM choline chloride or 25 mM sodium methanesulfonate (min)
was added to the control solution. *P < 0.001 with one-tailed Mann- c 40 * * *
d 2.5 e
Whitney tests; n = 85. Data are shown as mean ± s.d. [Na+]i. +/+
+/+ 2.0
R (mM/min)
(d) Relationship between the [Na+]i increase rate (R) and [Na+]o. R was 30 −/−
[Na+] (mM)
1.5
calculated from the slope between 20% and 80% of maximum [Na+]i. −/−
i
20 1.0
The solid line is the fit of the data to the equation R = RMax/(1 + 0.5 10 s
exp((C1/2 - C)/a)), where C = [Na+]o. The values RMax = 2.02 mM/min, 10 –5 pA
C1/2 = 159 mM and a = 5.21 mM were used; n = 21. (e) Whole-cell cur-
0
f 0
+/+ −/−
0 –0.5
rent responses of DRG neurons to an increase of [Na+]o from 145 to 120 140 160 180 200 220
I Na (pA)
0 m ate etha e
l
Ma aCl
sudium chlo l
-
0 m ontro
So line nnito
ne
X
lfo m rid
C (mM)
TT
N
–5
170 mM (bar). (f) The mean amplitudes of the whole-cell currents
Cl+
C
M
Na
observed in wild-type and Nax-null cells. *P < 0.005 by one-tailed Mann-

17
–10
o
*
M
Ch
Whitney tests; n = 10.

17

a b 40 a Control 170 mM Anti-Nax

* Nax+
Nax−
+/+
30
[Na+] (mM)
i
20
30
[Na+] (mM)
20
−/−
10
i
10
0
Control 170 mM
NaCl [Na+] 10 20 30 (mM)
i
Fig. 2. Nax cDNA transfection conferred [Na ]o sensitivity on Nax- +

b 40
c 3.5
*
deficient cells. (a) EGFP fluorescence image. Top, cells indicated by * * 3.0
arrowheads were transfected with both Nax and EGFP expression vec- 30 2.5
R (mM/min)
[Na+] i (mM)
tors. Bottom, pseudocolor image showing [Na+]i increase in the 2.0
170 mM solution. Scale bar, 50 µm. (b) Comparison of the response of 20 1.5
cells that were cotransfected with both Nax and EGFP expression vec- 1.0
tors (Nax+) and cells transfected only with EGFP expression vector 10 0.5
(Nax–). [Na+]i was measured 5 min after [Na+]i leveled off. *P < 0.001 0
0 –0.5
by one-tailed Mann-Whitney tests; n = 30.
M ol
Cl
e-
120 140 160 180 200 220
su ium hlo ol
0 m at eth e
TX
17 fon m rid
0 m ontr
an
Na
So holin nnit
+T
C (mM)
M a
Cl
c
M e
Na
17
C
d
l
inward currents with an average amplitude of 8.4 pA were observed
(n = 10; Fig. 1e, f). The current amplitude was consistent with that Fig. 3. Sodium-concentration sensitivity was lost in SFO neurons in
the Nax-null mutants. (a) Pseudocolor images showing the [Na+]i of
estimated from the ion-imaging studies (6.7 pA, see Supplemen-
the cells in the control and high sodium solutions. Scale bar, 50 µm.
tary Methods online). By contrast, the current was not observed (b) The [Na+]i response is dependent on [Na+]o, but not on extracel-
in cells derived from Nax-deficient mice (n = 10; Fig. 1e, f), and it lular [Cl–]o or osmotic pressure. Instead of NaCl, 50 mM mannitol,
was not inactivated under ‘high sodium solution’ conditions; 25 mM choline chloride or 25 mM sodium methanesulfonate was
instead, it disappeared rapidly when extracellular sodium was set added to the control solution. *P < 0.001 by one-tailed Mann-
back to normal amounts. There was no voltage dependency when Whitney tests; n = 85. (c) Relationship between the [Na+]i increase
the I-V relation was recorded between –90 and –40 mV during the rate (R) and [Na+]o. The values RMax = 3.04 mM/min, C1/2 = 157 mM,
application of ‘high sodium solution’ (data not shown); and the and a = 4.67 mM were used; n = 20.
current amplitude was not affected by 1 µM TTX (data not shown).
For further confirmation, we constructed an Nax expression
vector using mouse Nax cDNA and introduced it into the disso- such responses (70 samples; Fig. 3a). These results indicate that
ciated DRG neurons from Nax-deficient mice. To identify those the Nax channel is essential for central sodium reception.
that had been successfully transfected with Na x , cells were Thus, Nax is a newly identified type of sodium channel that
cotransfected with an enhanced green fluorescent protein (EGFP) is sensitive to an increase in the extracellular sodium concen-
expression vector (Fig. 2a, top). When [Na+]o was increased from tration, and is likely to be the sodium-level sensor of body flu-
145 mM to 170 mM, an [Na+]i response similar to that in wild- ids in the brain.
type neurons was seen: 93.5% (43/46 samples) of the EGFP-pos-
itive neurons (most of which were also transfected with Nax) Note: Supplementary information is available on the Nature Neuroscience website.
acquired the [Na+]i response (Fig. 2a, bottom, and Fig. 2b). In
all control experiments (n = 48)—in which neurons were trans- Acknowledgments
fected only with an EGFP expression vector—no [Na+]i response We thank T. Mohri for technical advice and comments on the manuscript; M.
was detected (Fig. 2b). Yasuda, A. Tozaki and C. Egusa for technical assistance; and A. Kodama for
These findings strongly support a physiological role for the secretarial assistance. This study was supported by grants-in-aid from the
Nax channel in body fluid homeostasis. Sodium concentrations Ministry of Education, Culture, Sports, Science and Technology of Japan, and
and plasma osmolarity increase by 5–10% during thirst condi- from the Japan Science and Technology Corporation (CREST).
tions 10,11. The sensitivity and threshold of Na x channels to
[Na+]o is in this range of physiological change. SFO and OVLT Competing interests statement
are regions where a blood-brain barrier is missing, enabling The authors declare that they have no competing financial interests.
cells to directly monitor body fluid conditions. We detected a
similar [Na+]i response in neurons that were dissociated from RECEIVED 11 JANUARY 2002; ACCEPTED 25 MARCH 2002.
the SFO of wild-type mice (41/70 samples; 58.6%); all the
[Na+]i-responsive cells in this region were also Nax-immunore- 1. Johnson, A. K. & Edwards, G. L. Curr. Top. Neuroendocrinol. 10, 149–190 (1990).
active cells (41/41 samples; Fig. 3a). When ‘high sodium solu- 2. Denton, D. A., McKinley, M. J. & Weisinger, R. S. Proc. Natl. Acad. Sci. USA
93, 7397–7404 (1996).
tion’ was applied to the cells, the [Na + ] i of these neurons 3. Oliet, S. H. R. & Bourque, C. W. Nature 364, 341–343 (1993).
increased over a time-course similar to that in DRG cells. The 4. Liedtke, W. et al. Cell 103, 525–535 (2000).
Nax-immunopositive SFO neurons responded to an increase in 5. Wells, T. Mol. Cell. Endocrinol. 136, 103–107 (1998).
6. Goldin, A. L. et al. Neuron 28, 365–368 (2000).
[Na+]o, but not to increases in osmolarity or [Cl–]o (Fig. 3b). 7. Goldin, A. L. Annu. Rev. Physiol. 63, 871–894 (2001).
The threshold value for the [Na+]o was 157 mM (Fig. 3c), which 8. Watanabe, E. et al. J. Neurosci. 20, 7743–7751 (2000).
9. Minta, A. & Tsien, R. Y. J. Biol. Chem. 264, 19449–19457 (1989).
is approximately equal to the value obtained from DRG. In con- 10. Wakerley, J. B., Poulain, D. A. & Brown, D. Brain Res. 148, 425–440 (1978).
trast, no SFO neurons derived from Nax-deficient mice showed 11. Nose, H. et al. J. Appl. Physiol. 73, 1419–1424 (1992).

Acute versus chronic NMDARs are crucial in the developmental acquisition of

AMPAR-mediated synaptic transmission, and that chronic dis-
NMDA receptor blockade abling of NMDAR function triggers compensatory mechanisms
for NMDAR-independent AMPAfication.
and synaptic AMPA We monitored AMPAfication during development by mea-
suring the ratio of synaptic AMPAR responses to synaptic
receptor delivery NMDAR responses (A/N). As previously shown in hip-
pocampus1,15 and other brain regions2,3, A/N increases rapid-
ly during the first postnatal week (Fig. 1a and b), with little
J. Julius Zhu1,2 and Roberto Malinow1 change during the second week. To determine the role of
1Cold Spring Harbor Laboratory, 1 Bungtown Road, Cold Spring Harbor, NY
NMDAR activity, we prepared hippocampal slices from rats
of different ages—postnatal day 3 (P3), P6, P9 and P12—and
11724, USA
2Department of Pharmacology, University of Virginia School of Medicine, 1300
incubated them for 20–24 hours in culture medium contain-
Jefferson Park Avenue, Charlottesville, VA 22908, USA
ing the NMDAR antagonist AP5 (200 µM DL-2-amino-5-phos-
phonovaleric acid). In slices treated with AP5 (200 µM), A/N
Correspondence should be addressed to R.M. (malinow@cshl.org)
was significantly smaller (P < 0.05) compared to untreated
Published online: 22 April 2002, DOI: 10.1038/nn850 slices (Fig. 1a and b), but only at ages where A/N was increas-
ing. Thus, NMDAR blockade prevented the developmental
Anatomical and electrophysiological experiments1–6 show that increase in A/N, supporting the view that AMPAfication dur-
central excitatory synapses initially display NMDA (N-methyl- ing development requires NMDAR function.
D-aspartate) receptors (NMDARs) and subsequently mature by To examine the effects of chronic NMDAR blockade, we pre-
acquiring AMPA (α-amino-3-hydroxy-5-methyl-4-isoxazole pro- pared slices from P6 rats and maintained them in culture medi-
pionic acid) receptors (AMPARs). NMDAR activation can lead um, either with or without 200 µM AP5. The A/N ratio was
to rapid synaptic delivery of AMPARs (‘AMPAfication’)7,8, but calculated at 1–2 day intervals for 12 days; AP5 was refreshed
the view that AMPAfication during development requires along with media every 48 hours (Fig. 1c and d). As expected,
NMDAR activation has been challenged by studies showing that A/N of neurons in AP5-treated slices was significantly smaller
chronic removal of NMDAR function (either genetically9,10 or
pharmacologically11–14) has no apparent effect on acquisition of a
AMPAR-mediated synaptic transmission. Here we show that
Fig. 1. Effects of acute and chronic blockade of NMDA receptors on

glutamatergic synapses in hippocampus. (a) Evoked AMPAR-mediated
(–60 mV, measured at the peak) and NMDAR-mediated
(+40 mV, measured at 75 ms after stimulation) synaptic responses in
hippocampal CA1 neurons recorded in physiological solution. Slices
were prepared from rats of different ages (P3, P9, P12) and maintained6 b
in culture medium with (AP5, red traces) or without (Ctrl, black traces)
AP5 for one day (1d). Animals were treated in a manner consistent with
Cold Spring Harbor Laboratory Animal Care and Utilization Review
Board. (b) Plot of the ratio of AMPA response to NMDA response
(A/N) versus time (age of animal + culture time). Horizontal red bars
indicate period of AP5 treatment. For P3+1d cells, control (black dia-
mond) values were 0.86 ± 0.09, n = 14; and AP5 (red diamond),
0.50 ± 0.13, n = 16 (P < 0.05). For P6+1d cells, control (black circle) val-
ues were 1.75 ± 0.15, n = 12; and AP5 (red circle) values were
1.12 ± 0.24, n = 12 (P < 0.05). For P9+1d cells, control (black square)
values were 2.01 ± 0.26, n = 16; and AP5 (red square) values were
1.92 ± 0.26, n = 16 (P = 0.80). For P12+1d cells, control (black triangle)
values were 1.92 ± 0.20, n = 16; and AP5 (red triangle) values were c
1.80 ± 0.18, n = 15 (P = 0.67). (c) Evoked AMPAR- and NMDAR-medi-
ated synaptic responses from CA1 neurons. Slices were prepared from
P6 animals and maintained in culture medium (with or without AP5) for
1–12 d. Scale bar as in (a). (d) Plot of A/N against time (age of animal +
culture time). Values for P6+1d cells are the same as in (b). Values for
P6+2d cells: Ctrl, 2.12 ± 0.34, n = 16; AP5, 1.20 ± 0.15, n = 16
(P < 0.05); P6+4d cells: Ctrl, 2.13 ± 0.29, n = 12; AP5, 1.10 ± 0.13, d
n = 15 (P < 0.005); P6+6d cells: Ctrl, 1.90 ± 0.37, n = 16; AP5,
1.43 ± 0.22, n = 16 (P = 0.29); P6+8d cells: Ctrl, 1.94 ± 0.22, n = 16;
AP5, 1.79 ± 0.23, n = 16 (P = 0.45); P6+10d cells: Ctrl, 2.02 ± 0.32,
n = 16; AP5, 1.96 ± 0.26, n = 16 (P = 0.90); P6+12d cells: Ctrl,
1.92 ± 0.19, n = 16; AP5, 1.80 ± 0.17, n = 16 (P = 0.65). AP5 was
refreshed with culture medium every 48 h. Note that inclusion of AP5
(200 µM) in the medium, which has been shown to completely block
NMDA-R-mediated responses11–13, impaired AMPAfication for the first
4 d. Asterisks indicate significant difference between AP5-treated and
untreated cells (P < 0.05, t-test).
(P < 0.05) than in control slices after 1–3 days of treatment, Acknowledgments
indicating that AP5 blocked AMPAfication. However, A/N began We thank N. Dawkins-Pisani for technical assistance, and H. Cline and
to increase after 4 days of treatment with AP5. Slices treated with members of the Malinow laboratory for helpful comments and discussions. This
AP5 for 8 or more days had significantly higher A/N ratios than study was supported by the National Institutes of Health (R.M.), the Alle Davis
did slices treated for 4 days or less (P < 0.05). After 8 days of AP5 and Maxine Harrison Endowment (R.M.), the Alzheimer’s Association (J.J.Z.)
treatment, slices became indistinguishable from untreated slices. and the Fraxa Medical Research Foundation (J.J.Z.). J.J.Z. is a Naples
The appearance of AMPAR responses after chronic NMDAR Investigator of the NARSAD (National Alliance for Research on Schizophrenia
blockade is consistent with previous pharmacological11–14 and and Depression) Foundation.
genetic9,10 experiments. Together, these findings indicate that

additional NMDAR-independent mechanisms, recruited by Competing interests statement
chronic blockade of NMDAR activity, may be capable of driving The authors declare that they have no competing financial interests.
AMPAfication.
We conclude that blockade of NMDAR activity can impair RECEIVED 23 JANUARY; ACCEPTED 26 MARCH 2002
AMPAfication of glutamatergic synapses during development.
However, our finding that AMPAfication of synaptic responses 1. Durand, G. M., Kovalchuk, Y. & Konnerth, A. Nature 381, 71–75 (1996).
resumed after 4 days of chronic blockade of NMDAR function 2. Wu, G., Malinow, R. & Cline, H. T. Science 274, 972–976 (1996).
indicates that other mechanisms are involved. Although we can- 3. Isaac, J. T., Crair, M. C., Nicoll, R. A. & Malenka, R. C. Neuron 18, 269–280
(1997).
not eliminate the possibility of an incomplete blockade by 4. Nusser, Z. et al. Neuron 21, 545–559 (1998).
200 µM AP5, this is unlikely given that similar results were 5. Petralia, R. S. et al. Nat. Neurosci. 2, 31–36 (1999).
6. Zhu, J. J., Esteban, J. A., Hayashi, Y. & Malinow, R. Nat. Neurosci. 3,
obtained in animals completely lacking functional NMDARs9. It 1098–1106 (2000).
will be interesting to determine the nature of these mechanisms 7. Hayashi, Y. et al. Science 287, 2262–2267 (2000).
and whether other NMDAR-dependent processes, such as long- 8. Liao, D., Scannevin, R. H. & Huganir, R. J. Neurosci 21, 6008–6017 (2001).
9. Li, Y., Erzurumlu, R. S., Chen, C., Jhaveri, S. & Tonegawa, S. Cell 76, 427–437
term potentiation and long-term depression, become NMDAR- (1994).
independent after chronic NMDAR blockade. Lastly, this study 10. Kutsuwada, T. et al. Neuron 16, 333–344 (1996).
shows that multiple ‘reserve mechanisms’ may exist for certain 11. Rao, A. & Craig, A. M. Neuron 19, 801–812 (1997).
12. O’Brien, R. J. et al. Neuron 21, 1067–1078 (1998).
cellular processes, and that they may be recruited only under spe- 13. Gomperts, S. N., Carroll, R., Malenka, R. C. & Nicoll, R. A. J. Neurosci. 20,
cial conditions. This could explain why chronically disabling a 2229–2237 (2000).
14. Luthi, A., Schwyzer, L., Mateos, J. M., Gahwiler, B. H. & McKinney, R. A. Nat.
gene or suppressing a protein’s function often shows no obvious Neurosci. 4, 1102–1107 (2001).
effects or an unexpected phenotype. 15. Hsia, A. Y., Malenka, R. C. & Nicoll, R. A. J. Neurophysiol. 79, 2013–2024 (1998).
Sniffing neuropeptides: a ly to healthy humans (9 female, 27 male, 25–41 years of age), and
the concentration of each peptide was measured within 80 min-
transnasal approach to utes after administration in samples of CSF and systemic blood
obtained through intraspinal (between L4 and L5) and intra-
the human brain venous (forearm) catheters. Catheterization was done two hours
before the sampling period began.
Jan Born1, Tanja Lange2, Werner Kern2, Intranasal administration of each peptide resulted in an ele-
vation of its concentration in the CSF (Fig. 1). We saw statisti-
Gerard P. McGregor3, Ulrich Bickel4 and Horst L. Fehm2
cally significant peptide accumulation in the CSF within 80
1Departments of Neuroendocrinology and 2Internal Medicine, University of minutes after administration with the higher dose of
Lübeck, Ratzeburger Allee 160, 23538 Lübeck, Germany MSH/ACTH(4–10) (10 mg), with the higher and lower doses of
3Department of Physiology, University of Marburg, Deutschhausstrasse 2, 35037 vasopressin (80 and 40 IU) and with insulin (40 IU), as com-
Marburg, Germany pared to pre-administration baseline concentrations and to con-
4Department of Pharmaceutical Sciences, Texas Tech University Health Sciences centrations in subjects administered sterile water as a placebo
Center, School of Pharmacy, 1300 Coulter Drive, Amarillo, Texas 79106, USA (Table 1). Also, a marginally significant (P = 0.05) increase in
Correspondence should be addressed to J.B. (born@kfg.mu-luebeck.de) CSF concentration occurred between 60 and 80 minutes after
administration of the lower dose of MSH/ACTH(4–10) (5 mg).
Published online: 29 April 2002, DOI: 10.1038/nn849 Increases in the CSF concentration of each peptide varied con-
siderably among subjects. For all three peptides, however, mean
Neuropeptides act as neuronal messengers in the brain, influenc- CSF concentrations began to rise within 10 minutes of intranasal
ing many neurobehavioral functions1. Their experimental and ther- administration. For MSH/ACTH(4–10) and insulin, peak levels
apeutic use in humans has been hampered because, when were attained within 30 minutes after administration; for vaso-
administered systemically, these compounds do not readily pass the pressin, CSF concentrations continued to increase for up to 80
blood–brain barrier, and they evoke potent hormone-like side effects minutes after administration. For each peptide, concentrations
when circulating in the blood2,3. We administered three peptides, did not return to baseline before the end of the 80-minute sam-
melanocortin(4–10) (MSH/ACTH(4–10)), vasopressin and insulin, pling period. More prolonged sampling in a subgroup of sub-
intranasally and found that they achieved direct access to the cere- jects receiving the higher doses of MSH/ACTH(4–10) and
brospinal fluid (CSF) within 30 minutes, bypassing the bloodstream. vasopressin showed that concentrations of peptides in the CSF
We selected the three peptides for their well-documented levels were still above those in placebo-treated subjects 100–120
effects on brain functions including learning, memory, and body- minutes after administration (P < 0.03 for MSH/ACTH(4–10),
weight regulation1,4,5. We administered the peptides intranasal- P < 0.009 for vasopressin).

Fig. 1. Peptide accumulation in cerebrospinal fluid (CSF) and blood. Concentrations of (a)
MSH/ACTH(4–10) (b) vasopressin and (c) insulin in CSF (left) and blood serum (right) from a
10 min before to 80 min after their intranasal administration in humans. Doses were
MSH/ACTH(4–10), 10 mg (thick solid line, n = 5) and 5 mg (thin solid, n = 4); arginine–
vasopressin, 80 IU (thick solid, n = 5) and 40 IU (thin solid, n = 4); human insulin, 40 IU (thick
solid, n = 8). Placebo (sterile water), thin dashed line (n = 7 for control of MSH/ACTH(4–10),
n = 5 for control of vasopressin and insulin). Substances were administered with a nasal spray
atomizer, with each puff containing defined amounts of MSH/ACTH(4–10) (0.25 or 0.5 mg),
vasopressin (10 IU) or insulin (10 IU). Total doses were achieved by repeated puffs in each
nostril every 30–45 s. Bars, period of peptide administration (for higher dose). Peptide con-
centrations were determined by radioimmunoassay (RIA) (MSH/ACTH(4–10)11; vaso- b
pressin, Mitsubishi Petrochemicals, Tokyo, Japan; insulin, Pharmacia, Uppsala, Sweden). No
extraction procedure was used for CSF samples. Assay sensitivities were MSH/ACTH(4–10),
0.05 ng/ml; vasopressin, 0.2 pg/ml; insulin, 1.8 pmol/l. Cross-reactivity of the RIAs with natu-
rally occurring related molecules was negligible (<0.1% with MSH/ACTH(4–9) or
ACTH(1–24) for MSH/ACTH(4–10) RIA11; <0.04% with oxytocin, lysine–vasopressin or
C-terminal metabolites for vasopressin RIA; <0.2% with C-peptide and insulin-like growth
factor 1 and 2 for insulin RIA). RIAs were combined with reversed-phase high-performance
liquid chromatography, confirming for each peptide that >90% of the immunoreactivity
recovered in CSF represented the intact peptide. Means, s.e.m. and significance compared to
c
placebo concentration (**, P ≤ 0.01, *, P < 0.05, Mann-Whitney test, for baseline-adjusted
values) are shown. Experiments were approved by the Ethics Committee of the University of
Lübeck and informed consent was obtained from each participant.
Concurrent measurement of the concentrations in blood did not

reveal a significant increase in MSH/ACTH(4–10) or insulin fol-
lowing intranasal administration of these peptides. In addition, there concentrations of MSH/ACTH(4–10) and insulin in the CSF after
was no change in plasma glucose concentration after insulin admin- intranasal administration were not paralleled by any increase in the
istration (P > 0.12, for all comparisons). In contrast, the accumu- concentrations in the blood, and intranasal administration of insulin
lation of vasopressin in CSF was accompanied by a distinct increase did not change blood glucose concentration, it is likely that the pep-
in plasma vasopressin levels, in agreement with previous observa- tides entered the CSF directly, bypassing the bloodstream. In each
tions6. The average increase in CSF concentrations of vasopressin case, an undetectable amount of peptide may have also reached the
correlated slightly but non-significantly with that in blood (r = 0.55, circulation. However, MSH/ACTH(4–10) is rapidly degraded in
one-sided P < 0.10, across subjects and both doses). blood (half life <4 min11), and any insulin reaching the bloodstream
Our data validate in humans the idea that intranasal adminis- would be masked by the endogenous hormone. Notably, in mice,
tration allows peptides to penetrate into the CSF. These data cor- the concurrent systemic injection of insulin has been found not to
roborate previous human studies in which recordings of evoked reduce a strong brain uptake of intranasally administered 125I-
brain potentials provided functional evidence for a facilitated access labeled insulin9. Although these data support the hypothesis that
of neuropeptides to the brain after nasal delivery7,8. Animal studies intranasal peptide administration can result in uptake into the CSF
also have shown that peptides (insulin, nerve growth factor) and independent of entry into the blood, other studies12,13 in conjunc-
larger molecules (e.g., horseradish peroxidase, viruses) accumulate tion with our finding that intranasally administered vasopressin
in brain tissue after intranasal administration3,9,10. As the increased accumulated in plasma indicate that blood–brain transport may,
to a certain extent, add to CSF uptake following
Table 1. Accumulation of MSH/ACTH(4–10), vasopressin and insulin in CSF nasal delivery.
and blood serum. Two routes have been proposed for the
CSF Serum
direct passage of peptides from the nose to
Mean s.e.m. P Mean s.e.m. P the brain: an intraneuronal and an extra-
neuronal pathway 3,10,14 . Intraneuronal
MSH/ACTH(4–10) AUC (ng/ml) × min AUC (ng/ml) × min transport involves the internalization of the
Placebo 7.45 8.98 8.80 1.41 peptide into olfactory neurons, followed by
MSH/ACTH(4–10), 5 mg 21.53 7.11 0.30 10.91 0.26 0.48 axonal transport. However, this route pre-
MSH/ACTH(4–10), 10 mg 514.49 195.4 0.004 10.98 2.98 0.62 sents a greater risk of proteolysis (resulting
Vasopressin AUC (pg/ml) × min AUC (pg/ml) × min from lysosomal degradation) than does
Placebo 254.4 65.6 207.4 202.0 extraneuronal transport, and requires hours
for substances to reach the olfactory
Vasopressin, 40 IU 1,319.1 821.8 0.05 1,674.0 931.8 0.14
bulb3,14. It therefore seems more plausible
Vasopressin, 80 IU 2,481.9 732.4 0.009 3,749.8 348.6 0.009
that peptide molecules travel by the extra-
Insulin AUC (pmol/l) × min AUC (pmol/l) × min cellular route, passing through patent inter-
Placebo 603.2 34.6 3,410.5 106.1 cellular clefts in the olfactory epithelium to
Insulin, 40 IU 1,091.1 219.8 0.028 3,414.3 276.8 0.22 diffuse into the subarachnoid space3,14. AUC
calculations (Table 1) are suggestive of an
Concentrations of the peptides expressed as area under curve (AUC; using the trapezoid method)
within 80 min after intranasal administration. P < 0.05 indicates significance in comparison with
inverse relationship between accumulation
concentrations after placebo administration. Significance of accumulations was also confirmed in com- in the CSF and peptide molecular weight
parisons of average post-administration increases with pre-administration baseline concentrations. (MWs: MSH/ACTH(4–10), 962.1; vaso-

pressin, 1,084.2; insulin, 5,808.0), although other factors, such ty of the intranasal route of peptide administration remains to
as lipophilicity and degree of ionization, probably also affect be proven in clinical trials.
peptide access to the brain3. A rapid accumulation of peptides
in cerebral and spinal CSF and in brain tissue (within 10–20 Acknowledgments
minutes of intranasal administration) has been seen in animals We thank A. Otterbein for technical assistance and W.M. Pardridge, University
and is also suggested by observations in patients3,7,9. Although of California at Los Angeles, Department of Medicine, and S. Gizurarson,
the extent of peptide uptake from CSF into human brain tissue University of Iceland, Faculty of Pharmacy, for comments on the manuscript.
is not known, animal studies have shown significant uptake This work was supported by the Deutsche Forschungsgemeinschaft.
even in more interior brain regions, such as the amygdala10.

Our data cannot be taken to establish that intranasal admin- Competing interests statement
istration results in greater CSF uptake of peptides than does intra- The authors declare that they have no competing financial interests.
venous administration. We administered fairly large doses of
peptide, because with intranasal administration, substantial RECEIVED 15 JANUARY; ACCEPTED 20 MARCH 2002
amounts of a compound may simply pass through the nose with-
out being absorbed. In the case of insulin, for example, previous 1. Strand, F.L. Neuropeptides: Regulators of Physiological Process (MIT Press,
studies suggest that brain uptake from plasma is at least compa- Cambridge, 1999).
rable to that resulting from nasal delivery12,13. This conclusion 2. Pardridge, W.M. J. Neurovirol. 5, 556–569 (1999).
3. Illum, L. Eur. J. Pharm. Sci. 11, 1–18 (2000).
does not, however, detract from our finding that intranasal 4. DeWied, D., Van Wimersma Greidanus, T.J.B., Bohus, B., Urban, L. &
administration can deliver neuropeptides to the brain without Gispen, W.H. Prog. Brain Res. 45, 181–194 (1976).
5. Schwartz, M.W., Woods, S.C., Porte, D. Jr., Seeley, R.J. & Baskin, D.G. Nature
uptake into the circulation. The potential usefulness of nasal 404, 661–671 (2000).
administration derives from the fact that biologically effective 6. Riekkinen, P. et al. Peptides 8, 261–265 (1987).
concentrations of neuropeptides can be achieved in the human 7. Fehm, H.L., Perras, B., Smolnik, R., Kern, W. & Born, J. Eur. J. Pharmacol.
405, 43–54 (2000).
brain without strong systemic, hormone-like side effects. Such 8. Kern, W., Born, J., Schreiber, H. & Fehm, H.L. Diabetes 48, 557–563 (1999).
effects limit the systemic administration of peptide to amounts 9. Sigurdsson, P., Thorvaldsson, T., Gizurarson, S. & Gunnarsson, E. Drug Deliv.
too small to have substantial effects in the brain. Nasal delivery 4, 195–200 (1997).
10. Chen, X., Fawcett, J.R., Rahman, Y., Ala, T.A. & Frey, W.H. 2nd. J. Alzheimer´s
may be useful in the treatment of brain diseases, particularly those Dis. 1, 35–44 (1998).
involving dysfunction of neuropeptide signaling, such as 11. Bickel, U., Born, J., Fehm, H.L., Distler, M. & Voigt, K.H. Eur. J. Clin.
Alzheimer’s disease and obesity5,9,10. Although this theory has Pharmacol. 35, 371–377 (1988).
12. Wallum, B.J. et al. J. Clin. Endocrinol. Metab. 64, 190–194 (1987).
received support from human studies showing beneficial effects 13. Schwartz, M.W. et al. J. Clin. Invest. 88, 1272–1281 (1991).
on sleep and body fat mass after prolonged intranasal adminis- 14. Thorne, R.G., Emory, C.R., Ala, T.A. & Frey, W.H. 2nd. Brain Res. 692,
278–283 (1995).
tration of vasopressin and MSH/ACTH(4–10), respectively, it 15. Fehm, H.L., Smolnik, R., Kern, W. & Born, J. J. Clin. Endocrinol. Metab. 86,
remains to be proven in clinical trials7,15. Nevertheless, the utili- 1144–1148 (2001).

commentary
Thalamcortical optimization of tactile

processing according to behavioral
state
Miguel A. L. Nicolelis1,2,3,4 and Erika E. Fanselow1
We propose a conceptual model that describes the operation of the main thalamocortical loop of the rat
somatosensory system. According to this model, the asynchronous convergence of ascending and descending
projections dynamically alters the physiological properties of thalamic neurons in the ventral posterior medial
(VPM) nucleus as rats shift between three behavioral states. Two of these states are characterized by distinct
modes of rhythmic whisker movements. We posit that these simultaneous shifts in exploratory behavioral
strategy and in the physiological properties of VPM neurons allow rats to either (i) optimize the detection of
stimuli that are novel or difficult to sense or (ii) process complex patterns of multi-whisker stimulation.
The rat somatosensory system is widely tem. The central hypothesis of our model their way to VPM, corticothalamic pro-
recognized as a versatile and invaluable is that the thalamocortical loop dynami- jections also send branches to RT neurons.
experimental model in which to investi- cally adjusts its physiological mode of The RT consists exclusively of inhibitory
gate the principles of the development1, operation, at both cellular and circuit lev- neurons, which project either to the VPM
anatomical organization2, physiological els, in accordance with specific behaviors nucleus, where they terminate near the
properties 3–8, coding strategy 9–12 and used by rats to explore their environ- cell bodies and activate GABA (γ-
plastic potential13–15 of sensory systems ments. This way, the somatosensory sys- aminobutyric acid)A and GABAB recep-
in mammals. As a result, this system has tem can favor either the detection of tors, or locally within RT. As the rat VPM
been scrutinized by a variety of powerful minute, novel tactile stimuli or the nucleus does not contain intrinsic
techniques, such as in vivo and in vitro detailed analysis of stimulus attributes. inhibitory neurons (inhibitory interneu-
patch-clamp recording16,17, intra- and rons), RT neurons are therefore the only
extracellular recording18–20 and optical The thalamocortical circuit source of GABAergic inhibition in this
imaging14,21–24. More recently, chronic The circuit that defines the main thalamic relay nucleus30.
multisite, multi-electrode recordings in thalamocortical loop of the rat
freely behaving animals have been used somatosensory system is illustrated in Underlying cellular properties
to characterize, for the first time, the Figure 1. Ascending action potentials Like other thalamic neurons, rat VPM
simultaneous activity of distinct popula- resulting from mechanical stimulation of neurons have two modes of firing: tonic
tions of neurons that define the main the whiskers travel to the trigeminal mode, in which cells fire single action
thalamocortical circuit of the rat brainstem complex (TBC) and the brain- potentials, and bursting mode, in which
somatosensory system25. This new exper- stem reticular formation (RF) (Fig. 1a). cells fire in rhythmic bursts of 2–10 action
imental approach provides a unique Projections from the TBC then terminate potentials at ∼300–500 Hz. These two
opportunity to correlate the physiologi- on neurons in the VPM nucleus. In the rat modes, which have been described by a
cal properties of thalamocortical neural thalamus, the VPM nucleus contains only number of investigators throughout the
ensembles with the main behaviors rats one type of neuron: excitatory cells that last several decades31–37, alternate accord-
use to extract tactile information from project mainly to layer 4 (CTX IV) of the ing to the animal’s behavioral state (for
their surrounding environment. primary somatosensory cortex (SI). On review, see ref. 38). Traditionally it has
Here, by combining recent physiolog- the way to the cortex, the axons of VPM been thought that thalamic bursts occur
ical and behavioral findings from multi- also send a projection to the thalamic only during states such as slow-wave sleep,
electrode recording studies with data reticular nucleus (RT). VPM and RT neu- anesthesia and seizure activity35,39–41 and
describing the main cellular properties of rons also receive dense ascending cholin- that tonic firing is associated solely with
thalamic neurons, we propose a multilevel ergic projections from the RF 26. These waking states34,42–44. Recently, however,
model of operation for the main thalamo- projections can excite VPM cells through several laboratories have shown that thal-
cortical loop of the rat somatosensory sys- nicotinic and M1-type muscarinic recep- amic bursts can occur during waking
tors27,28, but inhibit RT cells through M2- states45–52, although such bursting is more
1 Department of Neurobiology, 2 Department of type receptors29. prevalent during slow-wave sleep.
Descending projections in the Several authors have posited that sen-
Biomedical Engineering, 3 Department of
Psychological and Brain Sciences and 4 Duke
thalamocortical loop originate primarily sory information is transmitted from the
Center for Neuroengineering, Duke University
in layer 6 of SI cortex (CTX VI) (Fig. 1b). thalamus to the cortex only during the
Medical Center, Durham, North Carolina The distal dendrites of VPM neurons are tonic firing mode, and that no informa-
27710, USA densely innervated by these projections, tion is transmitted during the bursting
Correspondence should be addressed to which activate both ionotropic and mode30,35,39,40,53. This idea has recently
M.A.L.N. (nicoleli@neuro.duke.edu) metabotropic glutamate receptors. On been challenged by experiments showing
nature neuroscience • volume 5 no 6 • june 2002

517
commentary
Fig. 1. Schematic diagram of the main rat

a b thalamocortical loop. (a) The main ascending
pathways; (b) the main descending pathways.
Abbreviations as in text.
A shift from bursting mode to tonic

mode requires that cells be depolarized
for 50–100 ms so that I T will be suffi-

ciently inactivated. This means that the
fast postsynaptic potentials, such as those
generated in VPM by ascending trigemi-
no-thalamic pathways, are not likely to
cause a shift from the burst to tonic fir-
ing mode. Application of the neuromod-
ulator acetylcholine (ACh), by contrast,
has been shown to depolarize thalamic
projection neurons in vitro61,62 and is suf-
ficient to cause these cells to switch from
a bursting to a tonic mode of firing. In
addition, stimulation of the brainstem RF,
which supplies cholinergic innervation to
the thalamus, depolarizes thalamic pro-
jection cells for up to several seconds63.
Given that activity in RF substantially
increases during states of arousal (for a
review, see ref. 53), these projections
that neurons in the lateral geniculate inactivated. However, the inactivation could contribute to the shift from burst
nucleus (LGN) can transmit visual infor- can be removed—that is, IT can be de- to tonic mode as animals wake up or shift
mation during both modes of firing, inactivated—if the neuron becomes suf- from one behavioral state to another.
although the nature of the transmitted sig- ficiently hyperpolarized (to around –70
nal differs between the modes46,49. Dur- mV) for at least 50–100 ms. When IT is Linking physiology and behavior
ing bursting, thalamic neurons are ideally de-inactivated, it is then ready to be acti- Our model includes three natural behav-
tuned to detect fast changes in incoming vated by a sufficient depolarization, such ioral states often seen in rats. We refer to
sensory signals (similar to a high-pass fil- as one caused by an incoming stimulus the first of these as ‘quiet’ behavior, in
ter) and to pass this information on to the and leading to a Ca2+ spike. In this way, which rats are standing or sitting still and
cortex37,54,55. In contrast, during tonic fir- the inactivation state of IT is defined by their whiskers are not moving25,64. The
ing, VPM neurons can produce a more a complex function of time and mem- second is known as ‘whisker twitching’
detailed representation of a stimulus. brane voltage. behavior, during which rats are also
The tonic and bursting modes of thal-
amic firing emerge as the interplay of a b
variety of voltage-gated membrane con-
ductances (Fig. 2a). Depending on how
these conductances are selectively acti-
vated by the different inputs (from the
trigemino-thalamic, corticothalamic and
ascending neuromodulatory pathways)
that converge on thalamic neurons, VPM a
neurons can respond with either tonic or
bursting activity. A critical factor in spec-
ifying the firing mode is the activation
state of a voltage-gated Ca 2+ conduc-
tance, which involves T-type Ca2+ chan-
nels 36,54–60 . When these channels are
activated, Ca2+ flows into the cell, creat-
ing a current known as I T, which pro-
duces a slow, long-duration, nearly
all-or-none depolarization known as the
low-threshold Ca 2+ spike. The T-type Fig. 2. Burst activity in VPM cells and period of hyper-sensitivity after a burst. (a) The firing mode
of a VPM cell can shift from tonic to bursting, depending on the membrane potential and activa-
Ca2+ channels have voltage-dependent
tion state of the IT current. (b) A hyper-sensitive period (∼30 ms before the onset of whisker
characteristics that determine when IT is movement; black arrow at top) precedes a burst, and this increases the probability of a response
activated. When a thalamic neuron is rel- to a peripheral stimulus. The solid line shows the probability, at different times before the stimu-
atively depolarized (>60–65 mV), IT is lus, that a thalamic neuron will respond to a stimulus when it occurs.

commentary
Fig. 3. Neural activity in VPM thalamus

during three behavioral states. (a) A con- a
tinuous 50 s trace of the first principal
component of neural ensemble activity in
VPM during quiet, whisker twitching and
whisking behaviors. (b) Responses of sin-
gle VPM neurons to the presentation of
two infra-orbital electrical nerve stimuli25
with an interstimulus interval of 50 ms b
(bold dotted lines, stimulus presentation;

horizontal dotted lines, baseline firing
level). (c) Average peristimulus time his-
tograms (PSTHs) of neural activity in
VPM neurons before and after one stimu-
lation of the infra-orbital nerve (note that
the peaks of the responses to the stimu-
c
lus have been truncated so the lower-
magnitude activity can easily be seen; bold
dotted lines, stimulus presentation; hori-
zontal dotted lines, baseline firing level).
(d) Rasters showing the activity of four d
single units in VPM during each behavior.
Bursting activity is indicated by asterisks.
(e) Amount of partial directed coherence
observed during each of the three behav-
iors. The top panel is for partial directed
coherence from SI to VPM, the bottom e
panel from VPM to SI. Yellow indicates
the highest intensity of the coherence,
green and blue respectively less, and
white indicates none. (f) Cumulative sum
of cortical area activated by a single infra-
orbital nerve stimulus. Values are normal- f
ized to the maximum activated area in the
quiet state.
standing or sitting still but twitch

their whiskers in very rhythmic,
small-amplitude movements at a
rate of 7–12 Hz65,66. The third behavior, twitching and the concurrent oscillato- VPM neurons, which help to define the
referred to as ‘whisking’, occurs when rats ry activity in SI and VPM can last for receptive fields and multi-whisker respons-
move their whiskers back and forth in several seconds, and then terminates es of these thalamic neurons75,76.
large-amplitude sweeps at a rate of spontaneously. Throughout these The tactile responses of both SI and
∼4–6 Hz. Rats use these whisking move- episodes of 7–12 Hz oscillations, rats VPM neurons vary markedly in several
ments to repeatedly put their whiskers in remain extremely responsive to mechan- ways as a rat shifts between these three
contact with surfaces or objects, gather- ical stimulation of their whiskers. This behavioral states (Fig. 3a and b). First,
ing tactile information as they actively and other findings (see below) indicate when a single tactile stimulus (such as a
explore their environments67. that these thalamocortical oscillations brief deflection of a whisker) is present-
During the quiet and whisking resemble a normal, non-pathological ed, the probability of a neuronal response
behaviors, there is no large-scale, coher- oscillatory state sometimes referred to as is largest during the quiet behavior, small-
ent neural activity among the cells in the µ-rhythm 69 , which has been er during whisking and lowest during
either VPM or SI, and the activity is thus observed in other species70–73. whisker twitching (Fig. 3b). Second, after
referred to as ‘desynchronized’. In con- Notably, when these 7–12 Hz thalam- stimulation during the quiet mode, there
trast, the whisker twitching state is ocortical oscillations and the accompany- is a robust inhibitory period lasting
accompanied by a highly synchronous ing whisker twitching movements are ∼75 ms. This post-stimulus inhibitory
7–12 Hz oscillatory neural activity65,66 blocked by permanent SI cortical lesions68 period is shorter during whisker twitch-
(Fig. 3a), which appears first in the rat or by pharmacological inactivation of SI52, ing, and substantially shorter during
SI cortex and later in the VPM thala- the whisking movements remain. This sug- whisking, relative to that during the quiet
mus52,64. Shortly after the onset of this gests that the cortex does not exert strict behavior (Fig. 3c). Finally, when pairs of
oscillatory neural activity in the thalam- control of the cycle-by-cycle generation of stimuli are presented, the ability of a neu-
ocortical loop, rats start producing the whisking movements74, but rather modu- ron to fire in response to the second stim-
rhythmic, small amplitude whisker lates the activity of subcortical circuits that ulus in the pair is dependent on the
twitching movements characteristic of in turn drive the whisker-moving motor inter-stimulus interval and the animal’s
this behavior, which are phase-locked to neurons. It is important to emphasize that behavioral state. During the quiet and
the neural oscillations 66,68 . Whisker the SI cortex exerts other influences on whisker twitching states, if the inter-stim-

commentary
a is likely to be de-inactivated. For this

window of enhanced sensitivity to be
useful, it needs to occur during a period
when the whiskers are likely to be
mechanically stimulated. Indeed, in the
rat VPM the highest probability of burst-
ing activity occurs ∼30 ms after the onset
of whisker twitching protractions 64
(Fig. 2b, arrow). The highest probabili-

ty of a VPM response occurs when the
b stimulus is preceded by a burst occurring
∼120 ms before the stimulation (Fig. 2,
red box), and the period when IT is de-
inactivated begins at roughly the onset
of whisker protraction (Fig. 2). This
means that VPM neurons are most sen-
sitive to incoming stimulation at the
time when the whiskers begin to move
c forward and when these neurons are
capable of bursting. In this scheme, the
occurrence of each VPM burst ‘resets’
the activation state of IT so that the peri-
od when IT is inactivated occurs during
the retraction phase of the whisker
twitching movements. Thus, as I T
becomes de-inactivated again at the
beginning of the next whisker twitching
Fig. 4. Schematic diagram of neuronal firing in VPM neurons during (a) quiet (b) whisker twitch- protraction, the animal is again ready to
ing and (c) whisking behaviors. In each panel, the horizontal colored bars indicate approximate fir- respond to an incoming stimulus.
ing intensity in each of the brain regions considered in our model. Below these in each panel is a Further analysis52,77–79 revealed that
schematic drawing of the firing patterns in response to a single stimulus to the infra-orbital nerve signals directed from SI to VPM are sig-
(see Fig. 3c). Arrow (top), stimulus presentation. Stimulus peaks truncated in (a) and (c). nificantly more coherent during whisker
twitching episodes than during the other
two behavioral states (Fig. 3e). In addi-
tion, inactivation of the SI cortex by local
ulus interval is less than 75 ms, the prob- tion, in awake monkeys the percentage of infusion of the GABAA agonist muscimol
ability that VPM and SI neurons will spikes involved in bursts is 1.6% in the abolished whisker twitching movements,
respond to the second stimulus in a pair LGN and 29.5% in the somatosensory 7–12 Hz oscillations and bursting activi-
is very small (Fig. 3b). During whisking, thalamus 51 . The percentage of spikes ty in the VPM. These results indicate that,
though, the probability of a response to involved in bursts in the VPM thalamus during whisker twitching, the SI cortex
the second stimulus of a pair is only during each of the three rat behaviors is exerts a powerful rhythmic influence on
reduced if the inter-stimulus interval is 11.34% for quiet; 29.9% for whisker VPM neurons and that these descending
25 ms or less. Thus, the ability of VPM twitching and 4.4% for whisking. Notably, cortical signals are required for the emer-
and SI neurons to respond reliably to during the whisker twitching behavior, gence of 7–12 Hz oscillations and associ-
rapidly repeated stimuli is correlated with each neuron does not fire bursts on every ated VPM bursting activity, as well as for
the duration of the post-stimulus oscillation cycle. Thus, during a given the whisker twitching behavior itself.
inhibitory period, which differs accord- episode of whisker twitching, a neuron Although these VPM bursts may result
ing to behavioral state. may fire a burst on some of the oscilla- from the same cellular properties and cir-
Recently, a more detailed analysis of tion cycles and produce just a single cuits that are responsible for sleep-
the firing properties of thalamocortical action potential or not fire at all on other related oscillatory activity (also known as
cells52 in behaving rats has shown that cycles. This means that the amount of sleep spindles) and pathological states
during the 7–12 Hz oscillations observed neuronal bursting varies over a wide such as epilepsy, it should be noted that
in the whisker twitching behavior, VPM dynamic range. More specifically, the the nature of the oscillatory activity that
and SI neurons fire bursts of action poten- amount of thalamic bursting can be grad- these VPM bursts underlie and the behav-
tials much more frequently (average of ed according to the magnitude of the ioral condition during which they occur
once every 7.2 s) than during the quiet afferent stimulus. (whisker twitching) are both distinct from
(average of once every 45.5 s) or whisking During whisker twitching, VPM neu- those observed during sleep or seizure
(average of once every 28.6 s) behaviors rons show a period directly preceding a activity73. First, during whisker twitching,
(Fig. 3d). For comparison, it was shown burst when the probability of a response SI neuronal firing precedes the oscillato-
in the LGN of rats that the average num- to a tactile stimulus is significantly ry activity observed in the VPM thala-
ber of bursts per second during waking enhanced, and is higher than during any mus 32 and the 7–12 Hz oscillations
states is 0.06, although this varied accord- other behavioral states we studied52. This disappear when the cortex is removed66
ing to the size of the stimulus45. In addi- period occurs during the time when IT or inactivated 52. In contrast, the sleep

commentary
spindles that occur after removal of the signals, but allows them to respond profiles of thalamocortical tactile respons-
cortex80,81 are characterized by less coher- robustly to the presence of a single, iso- es to a single stimulus contain elements of
ent oscillations82. Second, cortical and lated stimulus (Fig. 4a). This could be the responses observed during both the
thalamic cells do not fire on every cycle thought of as a stimulus detection state, quiet state and whisking. For example, the
during the 7–12 Hz oscillations 52,64 as specialized for indicating the presence of amount of cortical area in SI that is acti-
they do during seizure activity, suggesting a transient tactile stimulus. vated by a tactile stimulus is similar dur-
that the oscillation-related bursts are not Whisker twitching, a more sensitive ing the whisker twitching and whisking
as intense as seizure-related bursts. Fur- stimulus detection mode, could occur states, but these are both smaller than that
ther, sleep spindles are characterized by when there is a low level of tactile input during the quiet state (Fig. 3f). Therefore,
periods of 7–12 Hz oscillations which from the periphery. This reduction in it is conceivable that during the whisker
have an average duration of 1.5–2 s and depolarizing drive reaching the VPM thal- twitching state, some partial integration
recur at a frequency of 0.1–0.2 Hz (ref. amus through the trigemino-thalamic of multi-whisker information could be
38). In contrast, the 7–12 Hz oscillations pathway, combined with the existence of achieved by the animal, in addition to
that occur during whisker twitching last a cortically driven inhibitory influence pure stimulus detection. This would sup-
from several seconds to more than a from RT to VPM, would produce a net port the contention that oscillations in the
minute and do not typically wax and inhibitory input to VPM neurons. As a thalamocortical loop serve as a ‘rate-
wane 64,65 . Finally, during the whisker consequence, these neurons would tend controlled oscillator’ that detect small
twitching behavior, rats are never uncon- to move to a more hyperpolarized state changes in afferent input9,85.
scious (as happens during both sleep and (around –70 mV; Fig. 2a), causing some Once the whiskers are deflected by a
multiple forms of seizures): they can read- degree of IT de-inactivation. At this point, novel stimulus, a series of concurrent cel-
ily detect tactile stimuli and quickly switch rhythmic 7–12 Hz oscillations would lular, circuit and behavioral events would
into a different behavioral state (quiet or appear in the SI cortex and would be read- allow the rat thalamocortical loop to
whisking). These are all hallmarks of nor- ily transmitted to the VPM thalamus dynamically shift its processing mode for
mal physiological cortical activity. (Fig. 4b) via corticothalamic projections. acquisition of complex, rapidly presented
Rhythmic activation of these corticothal- stimuli, such as might occur while large
A conceptual model amic projections would produce amplitude whisker movements are made
Our model integrates three levels of sequences of fast disynaptic inhibition, across a surface or object. This series of
analysis: (i) the cellular properties of through the RT nucleus, and delayed whisking-related events would occur as
thalamic neurons, (ii) the distinct extra- monosynaptic depolarization of VPM follows (Fig. 4c): first, whisker stimulation
cellular physiological properties of neurons through activation of ionotrop- sends an ascending excitatory volley to the
ensembles of thalamocortical neurons ic and metabotropic glutamate receptors VPM thalamus (via the trigemino-
observed in awake, freely behaving ani- by corticothalamic projections. This thalamic pathways). The signals are then
mals and (iii) the behavioral strategies sequence of rhythmic synaptic influences detected by VPM neurons and transmit-
(for example, different types of whisker would first supply the hyperpolarization ted to the SI cortex, in part as bursts. The
movements) used by rats during differ- needed to fully de-inactivate IT in VPM excitatory signal from the periphery also
ent behavioral situations. neurons, and then provide the depolar- activates the brainstem RF, which sends
According to this model (Fig. 4), the ization required for these neurons to pro- cholinergic input to the thalamus. This
first mode of processing corresponds to duce a Ca2+ spike with bursts of action cholinergic input depolarizes VPM neu-
the quiet state (Fig. 4a), during which potentials riding on it (Fig 2a). This rons but inhibits RT cells, leading to an
VPM neurons are relatively depolarized mechanism has been proposed for the overall reduction in GABAergic inhibition
(to about –55 mV) and thus in the tonic generation of oscillatory activity in the cat in VPM, as has been suggested else-
firing mode (Fig. 2a). This depolariza- thalamus 84 . Further support for this where 84,86 . Finally, further excitatory
tion could be due to ascending neuro- hypothesis comes from our finding that inputs to the cortex (via the basal forebrain
modulatory input and/or descending 7–12 Hz oscillations do not appear in nuclei and the intralaminar thalamic
excitatory input from the cortex83, both VPM if the SI cortex is inactivated52, indi- nuclei) depolarize cortical neurons and
of which are known to cause thalamo- cating that these oscillations may be interrupt the 7–12 Hz oscillations. The
cortical cells to change from bursting to dependent on descending cortical input. combination of these events (VPM depo-
tonic mode. During the quiet behavior, As the oscillatory thalamic bursting larization, reduction in RT-mediated inhi-
VPM neurons respond to tactile stimuli activity provides periods of hypersensi- bition and elimination of 7–12 Hz
with a stereotyped sequence of excita- tivity to stimulation of the whiskers just oscillations) would allow VPM neurons to
tion and inhibition (Fig. 4a). This phasic before the burst onset (Fig. 2, red box), remain sufficiently depolarized to prevent
sequence probably results from a strong we propose that rats are primed during de-inactivation of IT, and hence switch
depolarization by ascending excitatory these periods to detect minute or slowly from a bursting to a tonic mode of firing.
trigemino-thalamic projections that is changing deflections of the whiskers bet- By remaining in tonic mode, VPM
followed by both short- (GABA A - ter than during the quiet behavior when neurons would be able to respond more
mediated) and long-lasting (GABA B - neither 7–12 Hz oscillations nor VPM rapidly and accurately to the multi-
mediated) inhibition from RT projec- bursting are present. Although this hyper- whisker stimuli that would likely be expe-
tions to the VPM nucleus. These RT sensitive period is short, we propose that it rienced as the whiskers move over objects
neurons are activated by collaterals of still provides an advantage because the and surfaces during the whisking behav-
both thalamocortical and corticothala- level of sensitivity during this period is ior. Though sensitivity to a single stimulus
mic projections (Fig. 1). This response substantially higher than during any other may not be as great during the whisking
pattern renders VPM neurons incapable behavioral states we studied. In addition, state relative to other states, the ability of
of following fast sequences of incoming during whisker twitching movements, the VPM and SI neurons to respond accu-

commentary
rately to rapidly repeated whisker deflec- RECEIVED 28 NOVEMBER 2001; 17. Petersen, C. C. & Sakmann, B. The excitatory
tions is much greater. Thus, as rats use ACCEPTED 29 MARCH 2002 neuronal network of rat layer 4 barrel cortex.
J. Neurosci. 20, 7579–7586 (2000).
whisking movements to explore an object,
VPM neurons firing in tonic mode would 18. Nicolelis, M. A. L. et al. Reconstructing the
1. Rice, F. in The Barrel Cortex of Rodents (eds. engram: simultaneous, multisite, many single
be capable of transmitting to SI all the Jones, E. & Diamond, I.) 1–75 (Plenum, New neuron recordings. Neuron 18, 529–537
information provided by the complex York, 1995). (1997).
sequences of multi-whisker deflections 2. Woolsey, T. A. & Van der Loos, H. The 19. Markram, H. A network of tufted layer 5
generated during this behavior. Therefore, structural organization of layer IV in the pyramidal neurons. Cereb. Cortex 7, 523–533
the key purpose of whisking movements somatosensory region (SI) of mouse cerebral (1997).
cortex: description of a cortical field 20. Markram, H., Helm, P. J. & Sakmann, B.
and the accompanying physiological state composed of discrete cytoarchitectonic units. Dendritic calcium transients evoked by single
of the thalamocortical loop is to facilitate Brain Res. 17, 205–242 (1970). back-propagating action potentials in rat
the sampling, integration and analysis of 3. Chapin, J. K. & Woodward, D. J. Modulation neocortical pyramidal neurons. J. Physiol.
multiple, rapid whisker deflections. In this of sensory responsiveness of single 485, 1–20 (1995).
way, rats could best gain detailed somatosensory cortical cells during 21. Polley, D. B., Chen-Bee, C. H. & Frostig, R. D.
movement and arousal behaviors. Exp. Varying the degree of single-whisker
information about the location, shape, Neurol. 72, 164–178 (1981).
stimulation differentially affects phases of
size and texture of objects in their imme- 4. Simons, D. J. Temporal and spatial integration intrinsic signals in rat barrel cortex.
diate surroundings. in the rat SI vibrissa cortex. J. Neurophysiol. J. Neurophysiol. 81, 692–701 (1999).
54, 615–635 (1985).
22. Masino, S. A. & Frostig, R. D. Quantitative
Conclusions 5. Nicolelis, M. A. L. & Chapin, J. K. The long-term imaging of the functional
The conceptual model presented here spatiotemporal structure of somatosensory representation of a whisker in rat barrel
responses of many-neuron ensembles in the cortex. Proc. Natl. Acad. Sci. USA 93,
emphasizes two overlying principles. rat ventral posterior medial nucleus of the 4942–4947 (1996).
First, the active use of whiskers, in mul- thalamus. J. Neurosci. 14, 3511–3532 (1994).
23. Sheth, B. R., Moore, C. I. & Sur, M. Temporal
tiple types of whisker movements, is 6. Connors, B. W. & Gutnick, M. J. Intrinsic modulation of spatial borders in rat barrel
integral to the processing of tactile firing patterns of diverse neocortical neurons. cortex. J. Neurophysiol. 79, 464–470 (1998).
stimuli in rats. Indeed, our model pro- Trends Neurosci. 13, 99–104 (1990).
24. Kleinfeld, D. & Delaney, K. R. Distributed
poses that two distinct types of whisker 7. Silva, L. R., Amitai, Y. & Connors, B. W. representation of vibrissa movement in the
Intrinsic oscillations of neocortex generated upper layers of somatosensory cortex revealed
movements serve as differential dynam- by layer 5 pyramidal neurons. Science 251, with voltage-sensitive dyes [published
ic ‘filters’ to process specific types of 432–435 (1991). erratum appears in J. Comp. Neurol. 378, 594
incoming tactile information from 8. Ghazanfar, A. A. & Nicolelis, M. A. L. (1997)]. J. Comp. Neurol. 375, 89–108 (1996).
whisker stimulation. The second global Spatiotemporal properties of layer V neurons 25. Fanselow, E. E. & Nicolelis, M. A. L.
operating principle is that, as with the in the rat primary somatosensory cortex. Behavioral modulation of tactile responses in
determination of receptive field prop- Cereb. Cortex 9, 348–361 (1999). the rat somatosensory system. J. Neurosci. 19,
7603–7616 (1999).
erties and maps in SI and VPM75,76, the 9. Ahissar, E., Haidarliu, S. & Zacksenhouse, M.
Decoding temporally encoded sensory input 26. Hallanger, A. E. et al. The origins of
asynchronous convergence of ascend- by cortical oscillations and thalamic phase cholinergic and other subcortical afferents to
ing and descending projections in the comparators. Proc. Natl. Acad. Sci. USA 94, the thalamus in the rat. J. Comp. Neurol. 262,
thalamus is critical for generating the 11633–11638 (1997). 105–124 (1987).
animal’s range of sensory processing 10. Fee, M. S., Mitra, P. P. & Kleinfeld, D. Central 27. Plummer, K. L. et al. Muscarinic receptor
strategies. When this circuit arrange- versus peripheral determinants of patterned subtypes in the lateral geniculate nucleus: a
spike activity in rat vibrissa cortex during light and electron microscopic analysis.
ment is combined with the range of whisking. J. Neurophysiol. 78, 1144–1149 J. Comp. Neurol. 404, 408–425 (1999).
intrinsic cellular properties of cells in (1997).
the thalamocortical loop, a complex 28. Zhu, J. J. & Uhlrich, D. J. Cellular mechanisms
11. Simons, D. J. & Carvell, G. E. Thalamocortical underlying two muscarinic receptor-
and dynamic system emerges, which response transformation in the rat mediated depolarizing responses in relay cells
can quickly shift its physiological prop- vibrissa/barrel system. J. Neurophysiol. 61, of the rat lateral geniculate nucleus.
311–330 (1989). Neuroscience 87, 767–781 (1998).
erties to optimize the processing of the
type of tactile information that is sam- 12. Ghazanfar, A. A., Stambaugh, C. R. & 29. Carden, W. B. & Bickford, M. E. Location of
Nicolelis, M. A. L. Encoding of tactile muscarinic type 2 receptors within the
pled by a particular active exploratory stimulus location by somatosensory synaptic circuitry of the cat visual thalamus.
behavior. Together, these principles thalamocortical ensembles. J. Neurosci. 20, J. Comp. Neurol. 410, 431–443 (1999).
show that the rat somatosensory system 3761–3775 (2000).
30. McCormick, D. A. Neurotransmitter actions
does not merely play the role of a ‘pas- 13. Nicolelis, M. A. L., Lin, R. C. S., Woodward,
in the thalamus and cerebral cortex and their
sive observer’ of the environment. D. J. & Chapin, J. K. Induction of immediate
role in neuromodulation of thalamocortical
spatiotemporal changes in thalamic networks
Instead, it shifts between multiple func- activity. Prog. Neurobiol. 39, 337–388 (1992).
by peripheral block of ascending cutaneous
tional modes to actively examine and information. Nature 361, 533–536 (1993). 31. Poggio, G. & Mountcastle, V. The functional
analyze tactile inputs from the world, 14. Polley, D. B., Chen-Bee, C. H. & Frostig, R. D. properties of ventrobasal neurons studied in
Two directions of plasticity in the sensory- unanesthetized monkeys. J. Neurophysiol. 26,
based on expectations built throughout 775–806 (1963).
a life of whisking. deprived adult cortex. Neuron 24, 623–637
(1999). 32. Andersen, P., McBrooks, C., Eccles, H. &
15. Castro-Alamancos, M. A., Donoghue, J. P. & Sears, T. The ventrobasal nucleus of the
Acknowledgments Connors, B. W. Different forms of synaptic thalamus: potential fields, synaptic
This work was supported by National Institute of plasticity in somatosensory and motor areas transmission and excitability of both
Dental Research grants DE11121-01 and DE13810- of the neocortex. J. Neurosci. 15, 5324–5333 presynaptic and post-synaptic components.
(1995). J. Physiol. 174, 348–369 (1964).
01, a Human Frontier Science Program grant and a
16. Zhu, J. J. & Connors, B. W. Intrinsic firing 33. Andersen, P., Eccles, J. & Sears, T. The ventro-
United States/Israel Bi-national Science Foundation basal complex of the thalamus: types of cells,
patterns and whisker-evoked synaptic
award to M.A.L.N. and a predoctoral NRSA grant responses of neurons in the rat barrel cortex. their responses and their functional
(MH-12316-01A1) to E.E.F. J. Neurophysiol. 81, 1171–1183 (1999). organization. J. Physiol. 174, 370–399 (1964).

commentary
34. Maffei, L., Moruzzi, G. & Rizzolatti, G. unanesthetized, behaving monkeys. Vis. 70. Chatrain, E., Petersen, M. & Lazarte, J. The
Influence of sleep and wakefulness on the Neurosci. 17, 55–62 (2000). blocking of the rolandic wicket rhythm and
response of lateral geniculate units to sinewave 52. Fanselow, E. F., Sameshima, K., Baccala, L. A. some central changes related to movement.
photic stimulation. Arch. Ital. Biol. 103, EEG Clin. Neurophysiol. 11, 497–510 (1959).
& Nicolelis, M. A. L. Thalamic bursting in rats
596–608 (1965). during different awake behavioral states. Proc. 71. Pfurtscheller, G. & Neuper, C. Simultaneous
35. McCarley, R. W., Benoit, O. & Barrionuevo, G. Natl. Acad. Sci. USA 98, 15330–15335 (2001). EEG 10 Hz desynchronization and 40 Hz
Lateral geniculate nucleus unitary discharge in synchronization during finger movements.
53. Steriade, M., Jones, E. & Llinas, R. Thalamic
sleep and waking: state- and rate-specific Neuroreport 3,1057–1060 (1992).
Oscillations and Signaling (Wiley, New York,
aspects. J. Neurophysiol. 50, 798–818 (1983). 1990). 72. Pfurtscheller, G. & Neuper, C. Event-related
36. Deschenes, M., Paradis, M., Roy, J. P. & synchronization of mu rhythm in the EEG
54. McCormick, D. A. & Feeser, H. R. Functional

Steriade, M. Electrophysiology of neurons of over the cortical hand area in man. Neurosci.
implications of burst firing and single spike
lateral thalamic nuclei in cat: resting Lett. 174, 93–96 (1994).
activity in lateral geniculate relay neurons.
properties and burst discharges. Neuroscience 39, 103–113 (1990). 73. Pinault, D., Vergnes, M. & Marescaux, C.
J. Neurophysiol. 51, 1196–1219 (1984). Medium-voltage 5–9 Hz oscillations give rise to
55. Sherman, S. M. Tonic and burst firing: dual
37. Sherman, S. M. Dual response modes in spike-and-wave discharges in a genetic model
modes of thalamocortical relay. Trends
lateral geniculate neurons: mechanisms and of absence epilepsy: in vivo dual extracellular
Neurosci. 24,122–126 (2001).
functions. Vis. Neurosci. 13, 205–213 (1996). recording of thalamic relay and reticular
56. Jahnsen, H. & Llinas, R. Electrophysiological neurons. Neuroscience 105, 181–201 (2001).
38. Steriade, M. & McCarley, R. Brainstem Control properties of guinea-pig thalamic neurons: an
of Wakefulness and Sleep (Plenum, New York, 74. Carvell, G. E., Miller, S. A. & Simons, D. J. The
in vitro study. J. Physiol. 349, 205–226 (1984).
1990). relationship of vibrissal motor cortex unit
57. Llinas, R. & Jahnsen, H. Electrophysiology of activity to whisking in the awake rat. Somat.
39. Fourment, A., Hirsch, J. C. & Marc, M. E. mammalian thalamic neurons in vitro. Nature Motor Res. 13, 115–127 (1996).
Oscillations of the spontaneous slow-wave 297, 406–408 (1982).
sleep rhythm in lateral geniculate nucleus 75. Krupa, D. J., Ghazanfar, A. A. & Nicolelis,
relay neurons of behaving cats. Neuroscience 58. Jahnsen, H. & Llinas, R. Ionic basis for the M. A. L. Immediate thalamic sensory
14, 1061–1075 (1985). electro-responsiveness and oscillatory plasticity depends on corticothalamic
properties of guinea-pig thalamic neurones in feedback. Proc. Natl. Acad. Sci. USA 96,
40. Steriade, M., McCormick, D. A. & Sejnowski, vitro. J. Physiol. 349, 227–247 (1984). 8200–8205 (1999).
T. J. Thalamocortical oscillations in the
sleeping and aroused brain. Science 262, 59. Deschenes, M., Roy, J. P. & Steriade, M. 76. Ghazanfar, A. A., Krupa, D. J. & Nicolelis,
679–685 (1993). Thalamic bursting mechanism: an inward M. A. L. Role of corticothalamic feedback in
slow current revealed by membrane processing of simple and complex tactile
41. Steriade, M. & Llinas, R. R. The functional hyperpolarization. Brain Res. 239, 289–293 signals. Exp. Brain Res. 141, 88–100 (2001).
states of the thalamus and the associated (1982).
neuronal interplay. Physiol. Rev. 68, 649–742 77. Sameshima, K. & Baccala, L. A. Using partial
(1988). 60. Crunelli, V., Lightowler, S. & Pollard, C. E. A directed coherence to describe neuronal
T-type Ca2+ current underlies low-threshold ensemble interactions. J. Neurosci. Meth. 94,
42. Maffei, L. & Rizzolatti, G. Effect of Ca2+ potentials in cells of the cat and rat 93–103 (1999).
synchronized sleep on the response of lateral lateral geniculate nucleus. J. Physiol. 413,
geniculate units to flashes of light. Arch. Ital. 78. Baccala, L. & Sameshima, K. Parital directed
543–561 (1989). coherence: a new concept in neural structure
Biol. 103, 609–622 (1965).
61. McCormick, D. A. Cellular mechanisms determination. Biol. Cybern. 84, 463–474
43. Coenen, A. M. & Vendrik, A. J. Determination underlying cholinergic and noradrenergic (2001).
of the transfer ratio of cat’s geniculate neurons modulation of neuronal firing mode in the cat
through quasi-intracellular recordings and the 79. Baccala, L. & Sameshima, K. Overcoming the
and guinea pig dorsal lateral geniculate limitations of correlation analysis for many
relation with the level of alertness. Exp. Brain nucleus. J. Neurosci. 12, 278–289 (1992).
Res. 14, 227–242 (1972). simultaneously processed neural structures.
62. Sillito, A. M., Kemp, J. A. & Berardi, N. The Prog. Brain Res. 130, 33–47 (2001).
44. Livingstone, M. & Hubel, D. Effects of sleep cholinergic influence on the function of the
and arousal on the processing of visual 80. Morison, R. & Bassett, D. Electrical activity of
cat dorsal lateral geniculate nucleus (dLGN). the thalamus and basal ganglia in decorticate
information in the cat. Nature 291, 554–561 Brain Res. 280, 299–307 (1983).
(1981). cats. J. Neurophysiol. 8, 309–314 (1945).
63. Steriade, M. & Deschenes, M. in Cellular 81. Steriade, M., Domich, L., Oakson, G. &
45. Weyand, T. G., Boudreaux, M. & Guido, W. Thalamic Mechanisms (eds. Bentivoglio, M. &
Burst and tonic response modes in thalamic Deschenes, M. The deafferented reticular
Spreafico, R.) 51–76 (Elsevier, Amsterdam, thalamic nucleus generates spindle rhythmicity.
neurons during sleep and wakefulness. 1988).
J. Neurophysiol. 85, 1107–1118 (2001). J. Neurophysiol. 57, 260–273 (1987).
64. Nicolelis, M. A. L., Baccala, L. A., Lin, R. C. & 82. Contreras, D., Destexhe, A., Sejnowski, T. J. &
46. Reinagel, P., Godwin, D., Sherman, S. M. & Chapin, J. K. Sensorimotor encoding by
Koch, C. Encoding of visual information by Steriade, M. Control of spatiotemporal
synchronous neural ensemble activity at coherence of a thalamic oscillation by
LGN bursts. J. Neurophysiol. 81, 2558–2569 multiple levels of the somatosensory system.
(1999). corticothalamic feedback. Science 274,
Science 268, 1353–1358 (1995). 771–774 (1996).
47. Guido, W. & Weyand, T. Burst responses in 65. Semba, K., Szechtman, H. & Komisaruk, B. R. 83. McCormick, D. A. & von Krosigk, M.
thalamic relay cells of the awake behaving cat. Synchrony among rhythmical facial tremor, Corticothalamic activation modulates
J. Neurophysiol. 74, 1782–1786 (1995). neocortical ‘alpha’ waves and thalamic non- thalamic firing through glutamate
48. Edeline, J. M., Manunta, Y. & Hennevin, E. sensory neuronal bursts in intact awake rats. ‘metabotropic’ receptors. Proc. Natl. Acad. Sci.
Auditory thalamus neurons during sleep: Brain Res. 195, 281–298 (1980). USA 89, 2774–2778 (1992).
changes in frequency selectivity, threshold and 66. Semba, K. & Komisaruk, B. R. Neural 84. Destexhe, A., Contreras, D. & Steriade, M.
receptive field size. J. Neurophysiol. 84, substrates of two different rhythmical vibrissal Mechanisms underlying the synchronizing
934–952 (2000). movements in the rat. Neuroscience 12, action of corticothalamic feedback through
49. Swadlow, H. A. & Gusev, A. G. The impact of 761–774 (1984). inhibition of thalamic relay cells. J.
‘bursting’ thalamic impulses at a neocortical 67. Carvell, G. E. & Simons, D. J. Biometric Neurophysiol. 79, 999–1016 (1998).
synapse. Nat. Neurosci. 4, 402–408 (2001). analyses of vibrissal tactile discrimination in 85. Ahissar, E. & Zacksenhouse, M. in Advances in
50. Ramcharan, E. J. et al. Cellular mechanisms the rat. J. Neurosci. 10, 2638–2648 (1990). Neural Population Coding (ed. Nicolelis, M. A.
underlying activity patterns in the monkey 68. Welker, W. I. Analysis of sniffing of the albino L.) 75–87 (Elsevier, Amsterdam, 2001).
thalamus during visual behavior. rat. Behavior 22, 223–244 (1964). 86. Destexhe, A. Modelling corticothalamic
J. Neurophysiol. 84, 1982–1987 (2000). 69. Gastout, H. Etude électrocorticographique de feedback and the gating of the thalamus by the
51. Ramcharan, E. J., Gnadt, J. W. & Sherman, la réactivite des rythmes rolandiques. Rev. cerebral cortex. J. Physiol. Paris 94, 391–410
S. M. Burst and tonic firing in thalamic cells of Neurol. 87, 176–182 (1952). (2000).

errata
Thalamcortical optimization of tactile processing according to behavioral state

Miguel A.L. Nicolelis & Erika E. Fanselow
Nat. Neurosci. 5, 517–523 (2002)
The title of this article contained a typographical error. It should have read:
Thalamocortical optimization of tactile processing according to behavioral state
nature neuroscience • volume 5 no 7 • july 2002 1

articles
Progressive induction of caudal

neural character by graded Wnt
signaling
Ulrika Nordström1, Thomas M. Jessell2 and Thomas Edlund1
1 Department of Molecular Biology, Umeå University, S-901 87 Umeå, Sweden
2 Howard Hughes Medical Institute, Department of Biochemistry and Molecular Biophysics, Columbia University, New York, New York 10032, USA
Correspondence should be addressed to T.E. (Thomas.Edlund@molbiol.umu.se)
Published online: 13 May 2002, DOI: 10.1038/nn854
Early in differentiation, all neural cells have a rostral character. Only later do posteriorly
positioned neural cells acquire characteristics of caudal forebrain, midbrain and hindbrain cells.
Caudalization of neural tissue in the chick embryo apparently involves the convergent actions of
(i) fibroblast growth factor (FGF) signaling and (ii) signaling from the caudal paraxial mesoderm,
or ‘PMC activity’, which has not yet been defined molecularly. Here we report evidence that Wnt
signaling underlies PMC activity, and show that Wnt signals act directly and in a graded manner
on anterior neural cells to induce their progressive differentiation into caudal forebrain, midbrain
and hindbrain cells.
The early development of the vertebrate nervous system is accom- onic body plan (http://www.stanford.edu/%7Ernusse/wntwin-
panied by the specification of regionally restricted progenitor dow.html). The multiple patterning roles of Wnts at early devel-
cells along the rostrocaudal axis of the neural tube1,2. Studies in opmental stages has made it difficult to determine whether the
various vertebrates have indicated that cells of caudal neural char- later Wnt signals implicated in rostrocaudal neural patterning
acter are generated through the reprogramming of cells with an act directly or indirectly7,13,18–21,24.
initial rostral character2–4. In chick embryos, this happens during In chick embryos, prospective neural tissue can be separated
late gastrulation4,5. The induction of cells of midbrain and rostral from the adjacent mesoderm at stages when neural cells normal-
hindbrain character requires FGF signaling4,6,7. Retinoic acid ly acquire caudal regional characters. This way, the rostrocaudal
(RA) signaling, derived from the paraxial mesoderm that flanks specification of neural cells and their direct responses to putative
the caudal region of the neural plate, suppresses the generation of patterning signals may be examined. Our findings show that Wnt
cells of midbrain and rostral hindbrain character while inducing signaling is required for the specification of cells of caudal neur-
caudal hindbrain and spinal cord character4,8–10. However, FGF al character both in neural plate explants and in chick embryos
and RA signaling are not sufficient (alone or together) to induce grown in New culture. Through in vitro studies and New culture
these caudal characters in neural cells grown in vitro. This process assays, we found that this caudalizing action of Wnts results from
requires an additional paraxial mesoderm caudalizing signal11–13 a direct action on neural cells. Graded Wnt signaling, in combi-
that has been termed PMC activity4,5. The molecular basis of nation with FGFs, specifies cells of caudal forebrain, midbrain
PMC signaling is not known. and rostral hindbrain character. In the absence of Wnt signaling,
Genes of the Wnt family are expressed in the posterior region caudal neural cells grown in vitro revert to a rostral forebrain
of vertebrate embryos during stages of gastrulation when caudal character. Thus we conclude that Wnt signals mediate the PMC
neural cells are generated14–16, and several lines of evidence have activity necessary for the establishment of caudal neural fates.
implicated Wnt signaling in the specification of caudal neural
character7,13,17–27. Indeed, Wnt signaling is required at several RESULTS
different stages, and in several different germ layers, during the Regional expression of Wnts in chick gastrula
early development of vertebrate embryos. During gastrulation, Paraxial mesodermal tissue that underlies the prospective cau-
for example, Wnt signaling is needed to generate caudal non- dal neural plate of Hamburger and Hamilton (HH) stage 4 and 5
axial mesoderm21,28–30: the inactivation of Wnt genes that are chick embryos can induce cells of midbrain and hindbrain char-
expressed at gastrula stages in mouse and zebrafish embryos leads acter at gastrula stages4,5. In considering candidate mediators of
to defects in trunk and tail structures28,29,31. At an even earlier PMC activity, we noted that Wnt8c (ref. 14) and Wnt11 (ref. 32)
stage, Wnt signaling also helps to establish the anteroposterior are expressed in the posterior region of the chick embryo at a
body axis and to initiate gastrulation21,22. Consistent with these time when neural cells are exposed to factors that direct their
findings, mis-expression of Wnts or downstream components of caudal neural character4. Using in situ hybridization, we found
the Wnt signaling pathway before the onset of gastrulation leads that, beginning in early stage 4, Wnt8c is expressed transiently,
to axis duplications and/or other malformations of the embry- and Wnt11 at increasing levels, in the caudal paraxial mesoderm

articles
Fig. 1. Wnt11 and Wnt8c expression in the posterior region of the

chick embryo at developmental stages when neural cell precursors are
exposed to signals that induce caudal neural characters. Expression of
Wnt11 and Wnt8c in HH stage 4 and 5 embryos was monitored by
whole-mount (a, b) and section (c–f) in situ hybridization. (a, b) Black
arrow, Hensen’s node; black line, level of sections in c and d, respec-
tively. (a–d) Wnt11 was expressed in paraxial mesoderm posterior to
Hensen’s node. (e) Wnt8c was expressed in the primitive streak and
transiently in the mesoderm at early stage 4. (f) At early stage 5, Wnt8c
expression was found in lateral plate mesoderm and in the prospective

caudal neural plate. Scale bars, 0.5 mm.
assayed by monitoring the profile of expression of cell-specific

transcription factors. The expression of Sox2 and Sox3 (Sox2/3)
was used to define neural cells, regardless of their rostrocaudal
position33. Otx2 is expressed in the rostral neural tube with a
caudal limit at the isthmus34, and its expression in the absence
of Pax6, En1, Krox20 or Gbx2 was used as an indicator of neur-
al cells characteristic of rostral forebrain (RFB) levels (prospec-
tive telencephalon)35 (Fig. 2c). Co-expression of Otx2 and Pax6
in the absence of En1, Krox20 or Gbx2 was used to define cells
that underlies the prospective caudal neural plate (Fig. 1a–f). in the caudal forebrain (CFB) (prospective diencephalon)35,36
Thus, the combined patterns of expression of Wnt8c and Wnt11 (Fig. 2c). Co-expression of Otx2 and En1 was used to define
in the caudal paraxial mesoderm mimic the known distribution cells of midbrain (MB) character 37 (Fig. 2c). In 12 somite
of tissues that possess PMC activity5. In addition, from late stage embryos, Gbx2 (ref. 38) was expressed in rhombomeres (r) 1–4,
4 onwards, caudal neural plate cells themselves transiently and Krox20 (ref. 39) was co-expressed with Pax6 in r3 of the
expressed Wnt8c (Fig. 1f and ref. 14). hindbrain (data not shown). Thus, expression of Gbx2 and the
co-expression of Krox20 and Pax6 defined cells of rostral hind-
Wnt3A induces Wnt expression brain (RHB) character (Fig. 2c), whereas expression of Krox20
As both Wnt8c and Wnt11 were expressed in the caudal paraxial in the absence of Pax6 and Gbx2 defined cells of caudal hind-
mesoderm underlying prospective caudal neural plate (Fig. 1a–e), brain (r5-like) character.
and Wnt8c was expressed in prospective caudal neural plate cells Stage 4 and 5 caudal paraxial mesoderm tissue was cultured
(Fig. 1f), we reasoned that Wnt signals derived from the paraxi- together with stage 3 prospective caudal (stage 3C) neural plate
al mesoderm may induce Wnt expression in neural cells. To test explants (Fig. 3a,c and f) in the absence or presence of a soluble
this idea, we examined whether Wnt signaling induces Wnt8c fragment of the mouse Frizzled receptor 8 protein (mFrz8CRD-
expression in neural explants. We included FGF8 in these assays IgG), a selective antagonist of Wnt signals40,41. Stage 3C explants
because FGF signaling is required for caudalization of stage 4 grown alone generated Sox2/3+ and Otx2+ neural cells, a molec-
forebrain cells by paraxial mesoderm4. Wnt3A and FGF8 in com- ular profile characteristic of the rostral forebrain4. Similarly, in
bination, but not Wnt3A or FGF8 alone, induced the expression the presence of mFrz8CRD-IgG, stage 3C explants generated
of Wnt8c in stage 4 rostral forebrain (RFB) cells (Fig. 2b). This Sox2/3+ and Otx2+ cells, indicative of their maintained rostral
result supports the view that Wnts derived from the primitive forebrain character (Fig. 3b). Early stage 4 caudal paraxial meso-
streak and caudal paraxial mesoderm are involved in inducing derm induced Otx2+/En1+ cells of midbrain character in stage
the expression of Wnt8c in neural plate cells.
Wnt signaling specifies caudal neural character

a b
We next examined whether Wnt signaling participates in the
induction of midbrain and hindbrain character by caudal parax-
ial mesoderm. The positional character of neural cells was
Fig. 2. Wnt3A and FGF8 in combination induces the expression of

Wnt8c in prospective rostral forebrain cells. (a) Schematic drawing of a
late gastrula, HH stage 4, chick embryo. Dotted line, presumptive neural
plate; boxed region, explant of the prospective neural plate used for
in vitro cultivation and RT-PCR. (b) RT-PCR expression analysis of Wnt8c c
and the ribosome protein gene S17 in stage 4 RFB explants exposed to
FGF8 (10 ng/ml), Wnt3A (3×, 75 µl of Wnt3A conditioned medium per
ml of culture medium, ∼75 ng/ml) or a combination of Wnt3A (3×) and
FGF8 (10 ng/ml). (c) Schematic representation of the regional neural
markers used in this study. In a 12 somite stage chick embryo, cells in the
rostral forebrain (RFB, red) expressed Otx2; cells in the caudal forebrain
(CFB, yellow) co-expressed Otx and Pax6; cells in the midbrain (MB,
green) co-expressed En1 and Otx2. Gbx2 was widely expressed in the
rostral hindbrain (RHB, blue) and was co-expressed with Krox20 and
Pax6 in cells in rhombomere 3 (dark blue).

articles
Fig. 3. The induction of midbrain and a

hindbrain cells by caudal paraxial meso- b
derm requires Wnt signaling. (a, c, f)
Schematic drawings of HH stages 3, 4 and
5 embryos, respectively. (b, d, e, g, h)
Quail mesoderm was identified by expres-
sion of QCPN, and chick neural tissue by c d
expression of Sox2/3 (95 ± 5% cells/sec-
tion). (a) Dashed line, presumptive neural
plate; red box, prospective stage 3C

neural plate tissue used for in vitro explant
studies. (b) Stage 3C explants (n = 14) cul- e
tured for 24 h in the presence of
mFrz8CRD-IgG expressed Sox2/3 (95 ±
5% cells/section, n = 5 sections; mean ±
s.e.m.) and Otx2 (97 ± 3% cells/section, n g
= 8 sections) but not Pax6, En1, Gbx2 or
f
Krox20. (c, f) Caudal paraxial mesoderm
explants used in recombination experi-
ments are marked by gray boxes.
(d) Chick stage 3C explants recombined
with early HH stage 4 quail mesoderm
h
(n = 12) generated Otx2 /En1 cells (80 ±
+ +
20% cells/section, n = 10 sections) and a

few Gbx2+ cells (15 ± 10% cells/section, n
= 10 sections) and Otx2+/Pax6+cells (10 ±
10% cells/section, n = 8 sections), but no
Krox20+ cells. (e) Stage 3C explants recombined with early HH stage 4 mesoderm cultivated in mFrz8CRD-IgG conditioned medium (120 µl/ml cul-
ture medium; n = 14) generated Otx2+ cells (97 ± 2% cells/section, n = 9 sections) but no Pax6+, En1+, Gbx2+ or Krox20+ cells. (g) Stage 3C explants
recombined with HH stage 5 mesoderm (n = 12) generated only a few Otx2+/En1+ cells (10 ± 10% cells/section, n = 10 sections) and Otx2+/Pax6+cells
(5 ± 5%, n = 5 sections) but generated Gbx2+ cells (80 ± 20%, n = 8 sections) and Krox20+/Pax6+cells (60 ± 35%, n = 10 sections). (h) Stage 3C
explants recombined with stage 5 mesoderm in the presence of mFrz8CRD-IgG (120 µl/ml; n = 9) generated Otx2+ cells (95 ± 5% cells/section, n = 12
sections) and Pax6+ cells (50 ± 40% cells/section, n = 12 sections) but no En1+, Gbx2+ or Krox20+ cells. Scale bar, 100 µm.
3C explants (Fig. 3d). Exposing these conjugates to mFrz8CRD- signaling is required for PMC activity to induce cells of caudal
IgG blocked the generation of En1+ cells but not that of Otx2+ regional neural character.
cells (Fig. 3e). Stage 5 paraxial mesoderm induced Krox20+,
Gbx2+ and Pax6+ cells in stage 3C explants, a marker profile Direct caudalizing action of Wnts
indicative of rostral hindbrain character (Fig. 3g). Exposing these We next addressed whether the induction of caudal neural char-
conjugates to mFrz8CRD-IgG blocked the generation of Krox20+ acter requires Wnt action on neural cells themselves. By stage 4,
and Gbx2+cells but not that of Otx2+ cells (Fig. 3h). Thus, Wnt tissue isolated from different regions along the rostrocaudal axis
Fig. 4. Ongoing Wnt signaling in neural plate cells is required for the acquisition of caudal forebrain but not rostral forebrain character.
(a) Schematic drawing of an HH stage 4 chick embryo. Dotted line, presumptive neural plate. Boxed regions, explants of the prospective neural plate
used for in vitro studies: prospective rostral forebrain (RFB, red) and prospective caudal forebrain (CFB, yellow). (b–e) Sox2/3 was used as a general
neural marker (95 ± 5% cells/section).
(b) RFB explants cultured alone for 24 h
a (n = 12) generated Otx2+cells (95 ± 5%,
b cells/section, n = 8 sections) but did not
generate Pax6-, En1-, Gbx2- or Krox20-
expressing cells. (c) RFB explants cul-
tured in mFrz8CRD-IgG conditioned
medium (100 µl/ml culture medium;
c n = 16) generated Otx2+cells (95 ± 5%
cells/section, n = 5 sections) but did not
generate Pax6, En1, Gbx2 or Krox20
cells. (d) CFB explants cultured alone
(n = 25) generated Otx2+/Pax6+cells
d (95 ± 5% cells/section, n = 5 sections)
but did not generate En1-, Gbx2- or
Krox20-expressing cells. (e) CFB ex-
plants cultured in the presence of
e mFrz8CRD-IgG (100 µl/ml; n = 27) gen-
erated Otx2+cells (95 ± 5% cells/sec-
tion, n = 6 sections) but did not generate
Pax6, En1, Gbx2 or Krox20 cells. Scale
bar, 100 µm.
articles
a cells of diencephalic, midbrain and

b rostral hindbrain character.
We used New culture methods42 to
examine whether the attenuation of
Wnt signaling imposes a more rostral
c character in neural cells in intact
embryos. Control beads, or beads con-
taining mFrz8CRD-IgG, were
implanted beneath the neural plate

adjacent to the prospective mid-
d brain/caudal forebrain regions of
stage 4 embryos (Fig. 6a), and these
embryos were permitted to develop to
the 12–14 somite stage (Fig. 6c and d).
e Beads containing mFrz8CRD-IgG
induced morphological changes
indicative of an expansion of the cau-
dal forebrain region, which was con-
sistently more pronounced on the side
Fig. 5. Ongoing Wnt signaling in neural plate cells is required for the acquisition of midbrain and ros- of bead implantation (n = 5; Fig. 6d).
tral hindbrain character. (a) Schematic drawing of a HH stage 4 chick embryo. Dotted line, presumptive Analysis of the profile of transcription
neural plate; boxed regions, explants of the prospective neural plate used for in vitro studies: prospec- factor expression in these embryos
tive midbrain (MB, green) and prospective hindbrain (HB, blue). (b–e) Sox2/3 was used as a general
showed that cells normally located in
neural marker (95 ± 5% cells/section). (b) MB explants cultured alone (n = 22) generated
Otx2 /En1 cells (90 ± 10% cells/section, n = 10 sections) but not Pax6, En1-, Gbx2- or Krox20- the anterior region of the caudal fore-
+ +
expressing cells. (c) MB explants cultured in the presence of mFrz8CRD-IgG (120 µl/ml culture brain expressed Otx2 but not Pax6,
medium; n = 24) generated Otx2+cells (95 ± 5% cells/section, n = 10 sections), a few Pax6+ cells (5 ± indicating that caudal forebrain cells
5% cells/section, n = 10 sections), but no En1, Gbx2 or Krox20 cells. (d) RHB explants cultured alone had acquired rostral forebrain charac-
(n = 19) did not generate any Otx2+ or En1+cells but did generate Gbx2+ cells (90 ± 10% cells/section, ter (Fig. 6c and d). The domain in
n = 11) and Krox20+/Pax6+ cells (60 ± 20% cells/section, n = 14). (e) RHB explants cultured in the which Otx2 +/Pax6+ caudal forebrain
presence of mFrz8CRD-IgG (120 µl/ml; n = 24) generated Otx2+cells (90 ± 10% cells/section, n = 12 cells were present extended caudally
sections) that co-expressed Pax6 in 50 ± 30% of the cells/section (n = 11 sections). No En1-, Gbx2- or into the region normally occupied by
Krox20- expressing cells were generated. Scale bar, 100 µm.
En1+/Otx2+ midbrain cells. Consistent
with this, the number of En1+/Otx2+
midbrain cells was reduced and En1 was
of the prospective neural plate generates cells of rostral forebrain expressed at a much lower level by the remaining cells. In addi-
(RFB), caudal forebrain (CFB), midbrain (MB) and rostral hind- tion, the number of En1+/Gbx2+ cells characteristic of rhom-
brain (RHB) character in a position-dependent manner bomeres 1 and 2 of the rostral hindbrain was reduced (Fig. 6c
(Figs. 4a and 5a) and in the absence of mesodermal signals4. To and d). Collectively, these results provide evidence of a rostral-
examine whether Wnt signaling is required in neural tissue for to-caudal shift in the positional character of neural cells in
the acquisition of caudal neural character, we cultured stage 4 embryos exposed to mFrz8CRD-IgG.
explants isolated from different rostrocaudal levels of the prospec-
tive neural plate in the absence or presence of mFrz8CRD-IgG Distinct caudal fates imposed by graded Wnt signaling
(see Methods). The requirement for Wnt signaling in the generation of neural
Prospective RFB explants grown with or without mFrz8CRD- cells of three different rostrocaudal characters in explant assays,
IgG generated rostral forebrain–like cells that expressed Otx2, combined with the rostral-to-caudal shift in the positional char-
but not Pax6, En1, Krox20 or Gbx2 (Fig. 4b and c). When grown acter of in the New culture assays, led us to examine whether
alone, prospective CFB explants generated Otx2+ cells and Pax6+ Wnts induce different positional identities through actions at
cells, whereas in the presence of mFrz8CRD-IgG, no Pax6+ cells different concentration thresholds. Wnt3A, Wnt8c and Wnt11
were generated (Fig. 4d and e). In the absence of mFrz8CRD- show similar activities in several different assays
IgG, prospective MB explants generated Otx2+ cells and En1+ (http://www.stanford.edu/%7Ernusse/wntwindow.html), and
cells, whereas in the presence of mFrz8CRD-IgG, Otx2+ cells per- the ability of mFrz8CRD-IgG to block Wnt3A signaling was
sisted and no En1+ cells were generated (Fig. 5b and c). When demonstrated by assaying the block in induction of epidermal
grown alone, prospective RHB explants generated Krox20 +, character in blastula-stage chick epiblast cells in response to
Gbx2+ and Pax6+ cells, whereas in the presence of mFrz8CRD- Wnt3A (S. Wilson and T.E., unpublished data). Thus, we exam-
IgG, Otx2+ cells were generated, and a subset of these expressed ined the actions of Wnts on stage 4 RFB explants using Wnt3A
Pax6 (Fig. 5d and e). Thus, under these conditions, prospective conditioned medium43 (Fig. 7 and Table 1).
hindbrain cells acquired either rostral or caudal forebrain char- Stage 4 RFB explants (Fig. 7a and b) were exposed to differ-
acter. Stage 4 CFB, MB and RHB explants exposed to mFrz8CRD- ent concentrations of Wnt3A and to a constant concentration of
IgG were consistently smaller than explants grown alone (data FGF8. Consistent with previous studies4, Wnt3A alone did not
not shown), suggesting that Wnts exert a proliferative as well as induce caudal neural cells in stage 4 RFB explants at any concen-
a patterning effect during the early differentiation of neural cells. tration tested (data not shown). In the presence of Wnt3A (1×)
These results provide evidence that ongoing Wnt signaling in and FGF8 (10 ng/ml), Otx2+ cells and Otx2+/Pax6+ cells of ros-
prospective neural cells in vitro is required for the generation of tral and caudal forebrain character were generated, but no En1+

articles
Fig. 6. Wnt signaling imposes rostrocau-

dal pattern on neural cells in intact chick a b
embryos. (a, b) Schematic drawings of
late HH stage 4 chick embryos. Dotted
line indicates presumptive neural plate.
(a) Blue dot indicates the site at which
beads soaked in either mFrz8CRD-IgG
or control conditioned medium (CM) c d e f
were grafted beneath the neural plate
adjacent to the prospective midbrain

(MB) region. (b) Blue dot indicates the
site at which beads soaked in either
Wnt3A or CM were grafted beneath the
prospective forebrain (FB) region.
(c, d) Dorsal view of 12 somite (12 som)
chick embryos derived from stage 4
embryos grafted with control (c) or
mFrz8CRD-IgG (d) beads adjacent to
the prospective midbrain region, and
maintained in New culture. The num-
bered horizontal bars indicate the posi-
tions of the sections analyzed for marker
expression. (e, f) Dorsal view of 14
somite chick embryos generated in New
culture from stage 4 embryos grafted
with control (e) or Wnt3A (f) beads
beneath the prospective forebrain.
Numbered horizontal bars indicate the
positions of the sections analyzed for
marker expression. (c–f) Sections taken
at an equal distance from the anterior tip
of the respective embryos.
midbrain cells or Krox20 Gbx2+ hind-

brain cells were detected (Fig. 7c). In
these explants, Otx2+/Pax6+ cells were
located at the periphery and
Otx2+/Pax6– cells at the core. This indicates that cells in the We used New culture methods42 to examine whether Wnt3A
peripheral regions of the explants are exposed to higher concen- induced caudal character in anterior neural tissue in intact chick
trations of Wnt3A than are cells at the core. Wnt3A (2×) and FGF8 embryos. Control beads or beads containing Wnt3A were
(10 ng/ml) induced Otx2 +/Pax6 + caudal forebrain cells and implanted beneath the prospective forebrain of stage 4 embryos
Otx2+/En1+ midbrain cells, but no Krox20+ Gbx2+ hindbrain cells (Fig. 6b), and embryos were permitted to develop to the 12–14
(Fig. 7d). Under these conditions, Otx2+/Pax6+ cells were located somite stage (Fig. 6e and f). Embryos with grafted Wnt3A beads
at the core and Otx2+/En1+ cells at the periphery of the explants, typically showed a reduction in rostral forebrain tissue. The
again a likely reflection of the exposure of peripheral cells to high- domain of Otx2+ rostral forebrain cells was reduced and the
er Wnt levels. Wnt3A (4× or 10×) and FGF8 (10 ng/ml) induced domains of Otx2 + /Pax6 + caudal forebrain cells and of
Otx2+/En1+ midbrain cells and Krox20+, Pax6+ and Gbx2+ ros- En1+/Otx2+ midbrain cells were shifted rostrally (Fig. 6e and
tral hindbrain cells (Fig. 7e). Under these conditions, Otx2+/En1+ f). These findings support the idea that elevated Wnt signaling
cells were located at the core and Krox20+ and Gbx2+ cells at the in the anterior region of embryos leads to a loss of anterior
periphery of the explants. Similar results were obtained with Xeno- neural tissue and/or head structures and to a rostral-to-caudal
pus laevis Wnt8 conditioned medium (data not shown). The con- shift in neural pattern7,16,18,25,26,44.
centration of mFrz8CRD-IgG required to block the generation
of cells of rostral hindbrain character in RHB explants was four- Permissive action of FGF signaling
fold higher than that required to block the generation of cau- To examine whether a variation in the level of FGF signaling
dal forebrain character in CFB explants (data not shown). Stage might also influence caudal regional character, we added FGF8
4 RFB explants exposed to Wnt3A were consistently larger than (10 or 40 ng/ml) to stage 4 RFB explants exposed to Wnt3A (1×)
explants grown alone (data not shown), supporting the view or to Wnt3A (4×). Varying the concentration of FGF did not
that Wnts enhance the proliferation of neural progenitor cells, change the proportion or distribution of caudal neural cells
in addition to their role in specifying rostro–caudal positional induced by Wnt signals (data not shown), suggesting that FGFs
identity. Taken together, these results provide in vitro evidence act solely in a permissive manner during the establishment of
that the caudalizing action of Wnts results from a direct action caudal neural character. Taken together, these findings indicate
on neural cells and that graded Wnt signaling, in combination that Wnts act directly, and in a concentration-dependent man-
with FGFs, specifies cells of caudal forebrain, midbrain and ros- ner, to induce cells of caudal forebrain, midbrain and rostral
tral hindbrain character. hindbrain character in RFB explants.

articles
a b
c
Fig. 7. Graded Wnt3A activity, in combination with FGF8, induces caudal regional character in prospective rostral forebrain cells. (a) Schematic
drawing of an HH stage 4 embryo. Dotted line, presumptive neural plate; red box, prospective RFB epiblast explant used for in vitro studies. (b) RFB
explant cultured alone expressed Sox2/3 (95 ± 3% cells/section, n = 8 sections) and Otx2+ cells (95 ± 5% Otx2+ cells/section, n = 8 sections). (c) RFB
explants cultured in the presence of FGF8 (10 ng/ml) and Wnt3A (1×, 25 µl of Wnt3A conditioned medium per milliliter of culture medium,
∼25 ng/ml) had a central domain of Otx2+/Pax6– cells (95 ± 5% Otx2+ cells/section, n = 9 sections), whereas peripheral cells (typically 1–2 cell diam-
eters) co-expressed Otx2 and Pax6 (96 ± 4% Otx2+/Pax6+ cells/section, n = 7 sections). (d) RFB explants cultured in the presence of FGF8
(10 ng/ml) and Wnt3A (2×) had a central domain of Otx2+/Pax6+ cells (95 ± 5% Otx2+/Pax6+ cells/section, n = 8 sections) and a peripheral domain
(typically 2–3 cell diameters) of En1+/Otx2+ cells (98 ± 2% Otx2+/En1+ cells/section, n = 8 sections). (e) RFB explants cultured in the in the presence
of FGF8 (10 ng/ml) and Wnt3A conditioned medium (4×) had a central domain of Otx2+/En1+ cells (95 ± 5% Otx2+/En1+ cells/section, n = 12 sec-
tions) and a peripheral domain (typically 3–5 cell diameters) of Gbx2+ (90 ± 8% Gbx2+cells/section, n = 12 sections), and Krox20+/Pax6+ cells
(60 ± 25% Krox20+/Pax6+ cells/section, n = 12 sections). In all conditions, >95% of the cells expressed Sox2/3. Scale bar, 100 µm.
DISCUSSION neural plate cells and in the head mesendoderm that underlies
This study reports three main findings: (i) Wnt signaling is the prospective rostral forebrain 16,44. This latter domain of
required to reprogram neural cells of initial rostral forebrain char- expression may explain the finding that gastrula-stage head
acter during the acquisition of caudal regional neural characters; mesoendodermal tissue possesses rostralizing activity4,5,46. Thus,
(ii) Wnts, in combination with FGFs, can induce cells of caudal the exclusion of Wnt signaling from the anterior region of the
forebrain, midbrain and rostral hindbrain characters; and (iii) early embryo is probably involved in maintaining the rostral
Wnts act in a graded manner directly on neural plate cells. These (Pax6−) forebrain character of neural progenitor cells.
findings support Wnt involvement in the signaling pathway by The present studies support the view that Wnts have distinct
which prospective neural plate cells acquire diencephalic, mid- roles in the development of the chick nervous system at blastu-
brain and rostral hindbrain identity during the early phases of la and gastrula stages. At the blastula stage, epiblast cells acquire
chick neural tube development. generic neural fates; Wnt signaling at this stage promotes epi-
The developmental patterns of expression of Wnts and Fgfs dermal fate and blocks neural fate, apparently by preventing epi-
are consistent with the notion that that combined Wnt and FGF blast cells from responding to the neuralizing actions of FGFs41.
signaling in neural plate cells induces caudal regional neural char- At late gastrula stages, after neural cells have become commit-
acters. At early gastrula stages, when prospective caudal neural ted to a neural fate, graded Wnt activity instead induces pro-
cells possess a rostral forebrain character, Wnts are preferential- gressively more caudal neural characters, through actions in
ly expressed in the posterior region of the primitive streak, and combination with FGF signaling. RA signaling at these stages is
thus are located at a considerable distance from caudal neural involved in inducing cells of caudal hindbrain and spinal cord
cells (ref. 14 and data not shown). At these developmental stages, character4,10,47, suggesting that the joint actions of RA, Wnts
secreted Wnt antagonists such as crescent and caronte are and FGFs are required to induce cells in the most caudal regions
expressed in prospective neural plate cells or in tissues that under- of the neural axis.
lie the entire prospective neural plate45, and thus may help to The functions of Wnts in neural patterning reported here
decrease the exposure of anterior neural cells to Wnt signals. Over extend previous findings in vertebrate embryos mutant in com-
this early period, Fgf8 is expressed along the entire length of the ponents of the Wnt signaling pathway. In the mouse,
developing primitive streak4. At late gastrula stages, when neur- Wnt3/Wnt3A double mutant embryos lack all mesoderm, and
al cells are exposed to signals that direct their caudal character, Wnt3A mutants generate ectopic neural plate tissue in place of
both FGFs and Wnts are expressed in the posterior regions of the caudal paraxial mesoderm 21,30,31. Moreover, mutant mouse
chick embryo, whereas Wnt inhibitors are expressed in rostral embryos that lack the function of the Wnt inhibitor dickkopf-1

articles
fail to develop head structures rostral to the Table 1. Marker expression in rostral forebrain explants exposed to FGF8 and
midbrain 26 . In zebrafish, Wnt8 mutant Wnt3A.
embryos lack trunk and tail structures and
have ectopic neural tissue29. In addition, in
zebrafish masterblind mutant embryos, an
Otx2 (RFB) Otx2, En1 (MB)
apparent reduction in Axin1-dependent inhi- Otx2, Pax6 (CFB) Gbx2, Krox20, Pax6 (RHB)
bition of Wnt signaling is accompanied by a
loss of telencephalic structures and an expan-
sion of more caudal neural tissue25. Thus, [Wnt3A] Total no.

these genetic studies provide evidence that of explants
Wnt signaling is required for the induction
and patterning of neural tissue. Such genetic      
38 7 45
analyses do not, however, address whether
Wnts act directly or indirectly on neural cells
1× 12 19 22 1 4   58
to regulate their caudal regional character.
2× 
Wnts have been implicated in caudal 2 4 17 21 26 2 72
neural patterning in X. laevis, but again the
earlier involvement of Wnts in induction of 4× 5  2  24 32 15 78
mesoderm and epidermal ectoderm compli-
cates the task of distinguishing direct from Colored discs represent schematic figures of explants and the typical distribution of cells expressing
indirect Wnt action in rostrocaudal neural different region specific markers, characteristic of rostral forebrain (RFB), caudal forebrain (CFB),
patterning17,44. In blastula-stage ectoderm, midbrain (MB) and rostral hindbrain (RHB). For detailed quantification of marker expression, see
overexpression of the neural inducer Noggin Fig. 7 legend.
together with Wnts or the Wnt effector β-
catenin induces the expression of both cau-
dal neural and mesodermal markers. In contrast, in blastula-stage ing signals in the neural tube by similar or identical signals
ectoderm that has been neuralized by dissociation in Ca2+- and derived from extrinsic tissues may represent a common strate-
Mg2+-free medium, X. laevis Wnt8 induces caudal regional neur- gy for axial neural patterning.
al markers in the absence of markers of dorsal mesoderm24. In
gastrula-stage ectoderm, enhanced Wnt signaling leads to the METHODS
induction of caudal neural markers in adjacent cells, in the Isolation and growth of tissue explants. Prospective neural plate explants
absence of induction of mesodermal markers7. Under these con- were isolated from HH stage 3 and 4 chick embryos, and paraxial meso-
ditions, cells expressing caudal neural markers are induced by an derm explants were isolated from stage 4 and 5 quail embryos. Explants
indirect, non-cell-autonomous mechanism that apparently were cultured in vitro in serum-free OPTI-MEM containing N2 supple-
ment (Gibco-BRL/ Invitrogen, Paisley, UK) and fibronectin (Sigma). The
involves FGF signaling. Thus, these studies in X. laevis are con-
use of chick embryos in this study was approved by the ethical commit-
sistent with our findings in chick, indicating that the actions of tee at Umeå University.
FGFs and graded Wnt signaling on neural cells induce cells of
progressively more caudal neural character. Whole-embryo culture. HH stage 4 chick embryos were maintained in
Our results also provide evidence that Wnts mediate the New culture42 to the 12–14 somite stage. Blue-Beads (Bio-Rad, Hercules,
PMC activity described previously in chick assays4,5,12. In prin- California) soaked in mFrz8CRD-IgG, Wnt3A or control conditioned
ciple, the expression of Wnts in prospective caudal neural cells medium were grafted beneath different regions of the prospective neur-
does not exclude the possibility that a distinct signal derived al plate of host embryos. Embryos in which the bead was still in contact
from the paraxial mesoderm induces Wnt expression in neural with the forebrain or midbrain region at stage 7 (1–3 somite stage) were
cells, and that neurally derived Wnts impose caudal neural char- maintained in culture until the 12–14 somite stage. Eight embryos graft-
acters. However, the expression of Wnt8c and Wnt11 in the cau- ed with control beads, six grafted with mFrz8CRD-IgG beads and five
dal paraxial mesoderm precisely mimics the distribution of grafted with Wnt3A beads were cryosectioned (9 µm). Each section was
collected and analyzed by immunohistochemistry for marker expression.
tissues that possess PMC activity 5 (Fig. 1a–d and data not
shown). Moreover, Wnt3A, in combination with FGF8, induces Preparation of inducing factors. Soluble mouse Wnt3A and control con-
Wnt8c in prospective rostral forebrain cells. Although we can- ditioned media (CM) were obtained from stably transfected, or untrans-
not exclude the possible involvement of other Wnts, we show fected, mouse L-cells, respectively43, grown in Dulbecco’s modified Eagle’s
here that Wnt8c and Wnt11 are likely mediators of PMC activ- medium (DMEM) with 10% Knockout Replacement Serum (Gibco-
ity. BRL). Under these conditions, the CM contain ∼100 µg/ml of Wnt3A
The specification of caudal region neural characteristics by protein (R. Nusse, personal communication). The CM were concentrat-
Wnt signaling may, however, be a two-step process in which an ed using Centriprep 10,000-MWCO filters (Amicon/Millipore, Bedford,
initial phase of mesodermally derived Wnt signaling is consol- Massachusetts), divided into aliquots and stored at –80°C. Wnt3A was
idated by the expression of Wnts in prospective caudal neural used at an estimated concentration of 25–250 ng/ml (1–10×) in explant
assays. The mFrz8CRD-IgG40 and control CM were prepared by trans-
cells. This sequential ‘like-inducing-like’ strategy is reminiscent
fecting HEK-293 cells with the mFrz8CRD-IgG expression construct or
of the mechanisms that establish dorsoventral pattern in the with a lacZ reporter construct using Gene-PORTER 2 (GTSINC, San
neural tube. Sonic hedgehog (Shh) protein derived from the Diego, California). Cells were transferred to serum-free OPTI-MEM
axial mesoderm, and bone morphogenetic proteins (BMPs) (Gibco-BRL), and the CM was harvested after 48 h, concentrated with
derived from the flanking epidermal ectoderm, induce the Centriprep filters (Amicon), divided into aliquots and stored at –80°C.
expression of Shh and Bmps, respectively, in ventral and dorsal mFrz8CRD-IgG or lacZ CM were used at 100–160 µl/ml of explant medi-
midline neural cells48. Thus, the induction of secreted pattern- um. FGF8 (Gibco-BRL) was used at 10–40 ng/ml.

articles
Immunohistochemistry, in situ hybridization and RT-PCR. In situ 20. Bang, A.G., Papalopulu, N., Goulding, M.D. & Kintner, C. Expression of Pax-
hybridization and immunohistochemistry was carried out as 3 in the lateral neural plate is dependent on a Wnt-mediated signal from
described49,50. Rabbit anti-Gbx2 antibodies were raised against the pep- posterior nonaxial mesoderm. Dev. Biol. 212, 366–380 (1999).
21. Liu, P. et al. Requirement for Wnt3 in vertebrate axis formation. Nat. Genet.
tides EEAKGREENFSMDSD and QNRRAKWKRVKAGN. Other anti- 22, 361–365 (1999).
bodies used in this study have been described 5,41. Conditions and 22. Huelsken, J. et al. Requirement for β-catenin in anterior-posterior axis
primers used to analyze Wnt8C (35 cycles) and S17 (26 cycles, as semi- formation in mice. J. Cell Biol. 148, 567–578 (2000).
quantitative control) expression by RT-PCR in pooled explants (n = 9) 23. Fekany-Lee, K., Gonzalez, E., Miller-Bertoglio, V. & Solnica-Krezel, L. The
homeobox gene bozozok promotes anterior neuroectoderm formation in
have been described41. zebrafish through negative regulation of BMP2/4 and Wnt pathways.
Development 127, 2333–2345 (2000).
Acknowledgments 24. Kiecker, C. & Niehrs, C. A morphogen gradient of Wnt/β-catenin signaling

We thank Y. Renoncourt for experimental contributions, members of the Edlund regulates anteroposterior neural patterning in Xenopus. Development 128,
4189–4201 (2001).
lab for discussions and H. Alstermark for technical assistance. We are grateful to 25. Heisenberg, C. P. et al. A mutation in the Gsk3-binding domain of zebrafish
R. Nusse for providing Wnt3A-expressing cells, to J. Nathans for the mFrzCRD- Masterblind/Axin1 leads to a fate transformation of telencephalon and eyes
IgG plasmid and Xwnt8 cell line and to C. Tabin for Wnt probes. T.E. is to diencephalon. Genes Dev. 15, 1427–1434 (2001).
26. Mukhopadhyay, M. et al. Dickkopf1 is required for embryonic head
supported by the Swedish Medical Research Council and by the Foundation for induction and limb morphogenesis in the mouse. Dev. Cell 1, 423–434
Strategic Research. T.M.J. is supported by grants from US National Institute of (2001).
27. van de Water, S. et al. Ectopic Wnt signal determines the eyeless phenotype of
Neurological Disorders and Stroke (NIH-NINDS) and is an Investigator of the
zebrafish masterblind mutant. Development 128, 3877–3888 (2001).
Howard Hughes Medical Institute. 28. Greco, T. L. et al. Analysis of the vestigial tail mutation demonstrates that
Wnt-3a gene dosage regulates mouse axial development. Genes Dev. 10,
313–324 (1996).
Competing interests statement 29. Lekven, A. C., Thorpe, C. J., Waxman, J. S. & Moon, R. T. Zebrafish wnt8
The authors declare that they have no competing financial interests. encodes two wnt8 proteins on a bicistronic transcript and is required for
mesoderm and neurectoderm patterning. Dev. Cell 1, 103–114 (2001).
RECEIVED 25 MARCH; ACCEPTED 19 APRIL 2002 30. Yoshikawa, Y., Fujimori, T., McMahon, A. P. & Takada, S. Evidence that
absence of Wnt-3a signaling promotes neuralization instead of paraxial
mesoderm development in the mouse. Dev. Biol. 183, 234–242 (1997).
31. Takada, S. et al. Wnt-3a regulates somite and tailbud formation in the mouse
1. Lumsden, A. & Krumlauf, R. Patterning the vertebrate neuraxis. Science 274, embryo. Genes Dev. 8, 174–189 (1994).
1109–1115 (1996). 32. Eisenberg, C. A., Gourdie, R. G. & Eisenberg, L. M. Wnt-11 is expressed in
2. Stern, C. D. Initial patterning of the central nervous system: how many early avian mesoderm and required for the differentiation of the quail
organizers? Nat. Rev. Neurosci. 2, 92–98 (2001). mesoderm cell line QCE-6. Development 124, 525–536 (1997).
3. Nieuwkoop, P. D. et al. Activation and organisation of the central nervous 33. Rex, M. et al. Dynamic expression of chicken Sox2 and Sox3 genes in
system in amphibians. J. Exp. Zool. 1–108 (1952). ectoderm induced to form neural tissue. Dev. Dyn. 209, 323–332 (1997).
4. Muhr, J., Graziano, E., Wilson, S., Jessell, T. M. & Edlund, T. Convergent 34. Mallamaci, A., Di Blas, E., Briata, P., Boncinelli, E. & Corte, G. OTX2
inductive signals specify midbrain, hindbrain and spinal cord identity in homeoprotein in the developing central nervous system and migratory cells
gastrula stage chick embryos. Neuron 23, 689–702 (1999). of the olfactory area. Mech. Dev. 58, 165–178 (1996).
5. Muhr, J., Jessell, T. M. & Edlund, T. Assignment of early caudal identity to 35. Bell, E., Ensini, M., Gulisano, M. & Lumsden, A. Dynamic domains of gene
neural plate cells by a signal from caudal paraxial mesoderm. Neuron 19, expression in the early avian forebrain. Dev. Biol. 236, 76–88 (2001).
487–502 (1997). 36. Matsunaga, E., Araki, I. & Nakamura, H. Pax6 defines the di-mesencephalic
6. Storey, K. G. et al. Early posterior neural tissue is induced by FGF in the chick boundary by repressing En1 and Pax2. Development 127, 2357–2365 (2000).
embryo. Development 125, 473–484 (1998). 37. Davis, C. A. & Joyner, A. L. Expression patterns of the homeo box-containing
7. Domingos, P. M. et al. The Wnt/β-catenin pathway posteriorizes neural tissue genes En-1 and En-2 and the proto-oncogene int-1 diverge during mouse
in Xenopus by an indirect mechanism requiring FGF signaling. Dev. Biol. 239, development. Genes Dev. 2, 1736–1744 (1988).
148–160 (2001). 38. Shamim, H. & Mason, I. Expression of Gbx-2 during early development of
8. Maden, M. Heads or tails? Retinoic acid will decide. Bioessays 21, 809–812 the chick embryo. Mech. Dev. 76, 157–159 (1998).
(1999). 39. Nieto, M. A., Bradley, L. C. & Wilkinson, D. G. Conserved segmental
9. Niederreither, K., Vermot, J., Schuhbaur, B., Chambon, P. & Dolle, P. Retinoic expression of Krox-20 in the vertebrate hindbrain and its relationship to
acid synthesis and hindbrain patterning in the mouse embryo. Development lineage restriction. Development Suppl. 2, 59–62 (1991).
127, 75–85 (2000). 40. Hsieh, J. C., Rattner, A., Smallwood, P. M. & Nathans, J. Biochemical
10. Liu, J. P., Laufer, E. & Jessell, T. M. Assigning the positional identity of spinal characterization of Wnt–frizzled interactions using a soluble, biologically
motor neurons. Rostrocaudal patterning of Hox-c expression by FGFs, Gdf11 active vertebrate Wnt protein. Proc. Natl. Acad. Sci. USA 96, 3546–3551
and retinoids. Neuron 32, 997–1012 (2001). (1999).
11. Woo, K. & Fraser, S. E. Specification of the zebrafish nervous system by 41. Wilson, S. et al. The status of Wnt signaling regulates neural and epidermal
nonaxial signals. Science 277, 254–257 (1997). fates in the chick embryo. Nature 411, 325–330 (2001).
12. Bang, A. G., Papalopulu, N., Kintner, C. & Goulding, M. D. Expression of 42. New, D. A. A new technique for the cultivation of the chick embryo in vitro.
Pax-3 is initiated in the early neural plate by posteriorizing signals produced J. Embryol. Exp. Morphol. 3, 320–331 (1955).
by the organizer and by posterior non-axial mesoderm. Development 124, 43. Shibamoto, S. et al. Cytoskeletal reorganization by soluble Wnt-3a protein
2075–2085 (1997). signaling. Genes Cells 3, 659–670 (1998).
13. Erter, C. E., Wilm, T.P., Basler, N., Wright, C. V. & Solnica-Krezel, L. Wnt8 is 44. Niehrs, C. Head in the WNT: the molecular nature of Spemann’s head
required in lateral mesendodermal precursors for neural posteriorization in organizer. Trends Genet. 15, 314–319 (1999).
vivo. Development 128, 3571–3583 (2001). 45. Foley, A. C., Skromne, I. & Stern, C. D. Reconciling different models of
14. Hume, C. R. & Dodd, J. Cwnt-8C: a novel Wnt gene with a potential role in forebrain induction and patterning: a dual role for the hypoblast.
primitive streak formation and hindbrain organization. Development 119, Development 127, 3839–3854 (2000).
1147–1160 (1993). 46. Foley, A. C., Storey, K. G. & Stern, C. D. The prechordal region lacks neural
15. Baranski, M., Berdougo, E., Sandler, J. S., Darnell, D. K. & Burrus, L. W. The inducing ability, but can confer anterior character to more posterior
dynamic expression pattern of frzb-1 suggests multiple roles in chick neuroepithelium. Development 124, 2983–2996 (1997).
development. Dev. Biol. 217, 25–41 (2000). 47. Gould, A., Itasaki, N. & Krumlauf, R. Initiation of rhombomeric Hoxb4
16. Yamaguchi, T. P. Heads or tails: Wnts and anterior-posterior patterning. Curr. expression requires induction by somites and a retinoid pathway. Neuron 21,
Biol. 11, R713–724 (2001). 39–51 (1998).
17. McGrew, L. L., Hoppler, S. & Moon, R. T. Wnt and FGF pathways 48. Jessell, T. M. Neuronal specification in the spinal cord: inductive signals and
cooperatively pattern anteroposterior neural ectoderm in Xenopus. Mech. transcriptional codes. Nat. Rev. Genet. 1, 20–29 (2000).
Dev. 69, 105–114 (1997). 49. Schaeren-Wiemers, N. & Gerfin-Moser, A. A single protocol to detect
18. Fredieu, J. R., Cui, Y., Maier, D., Danilchik, M. V. & Christian, J. L. Xwnt-8 and transcripts of various types and expression levels in neural tissue and
lithium can act upon either dorsal mesodermal or neurectodermal cells to cultured cells: in situ hybridization using digoxigenin-labelled cRNA probes.
cause a loss of forebrain in Xenopus embryos. Dev. Biol. 186, 100–114 (1997). Histochemistry 100, 431–440 (1993).
19. Popperl, H. et al. Misexpression of Cwnt8C in the mouse induces an ectopic 50. Yamada, T., Placzek, M., Tanaka, H., Dodd, J. & Jessell, T. M. Control of cell
embryonic axis and causes a truncation of the anterior neuroectoderm. pattern in the developing nervous system: polarizing activity of the floor plate
Development 124, 2997–3005 (1997). and notochord. Cell 64, 635–647 (1991).

corrigenda
Progressive induction of caudal neural character by graded Wnt signaling

Ulrika Nordström, Thomas M. Jessell and Thomas Edlund
Nat. Neurosci. 5, 525-532 (2002)
The authors wish to correct the phrase “rostral-to-caudal shift” on page 528, which should read “rostrocaudal shift”. The error
occurs twice on this page.
nature neuroscience • volume 5 no 7 • july 2002 1

articles
Ion channel properties underlying

axonal action potential initiation in
pyramidal neurons
© 2002 Nature Publishing Group http://www.nature.com/natureneuroscience
Costa M. Colbert and Enhui Pan
Department of Biology and Biochemistry, University of Houston, 4800 Calhoun Road, Houston, Texas 77204-5513, USA
Correspondence should be addressed to C.C. (ccolbert@uh.edu)
A high density of Na+ channels in the axon hillock, or initial segment, is believed to determine the
threshold for action potential initiation in neurons. Here we report evidence for an alternative mech-
anism that lowers the threshold in the axon. We investigated properties and distributions of ion
channels in outside-out patches from axons and somata of layer 5 pyramidal neurons in rat neocorti-
cal slices. Na+ channels in axonal patches (>30 µm from the soma) were activated by 7 mV less depo-
larization than were somatic Na+ channels. A-type K+ channels, which were prominent in somatic
and dendritic patches, were rarely seen in axonal patches. We incorporated these findings into
numerical simulations which indicate that biophysical properties of axonal channels, rather than a
high density of channels in the initial segment, are most likely to determine the lowest threshold for
action potential initiation.
Determining where and when action potentials originate is crit- With these data in mind, we investigated biophysical mecha-
ical to understanding the computations that neurons perform1–3. nisms that might contribute to a low threshold for action poten-
A precise definition of the conditions for action potential initi- tial initiation in the axon. We recorded currents from membrane
ation requires knowledge of the distribution and of the biophys- patches in somata, initial segments and axons of layer 5 pyrami-
ical properties of the underlying ion channels. Because of the dal neurons in rat neocortical slices. There was a difference in the
very small diameters of axons in the brain, direct electrophysio- voltage dependence of activation of Na+ channels across these
logical measurements of axonal ion channels have only recently regions that favors the initiation of action potentials in the axon.
become feasible. Thus, the determinants of threshold and the site We also found that A-type K+ channels, which were prominent in
of initiation of action potentials in cortical neurons have somatic patches, were only rarely present in patches from the
remained a subject of debate4,5. axon or initial segment.
Considerable recent evidence indicates that action potentials
are often initiated near the soma rather than at the site of synap- RESULTS
tic input in the dendrites6–14, suggesting that the axon has a low We measured currents using the outside-out patch configura-
threshold for action potential initiation. The mechanism under- tion, which provided several advantages over other recording
lying this low threshold has long been assumed to be a high den- modes for our specific experiments. First, because we recorded
sity of Na+ channels in the axon hillock or initial segment, which briefly in the whole-cell configuration before the patch was
is typically 25–30 µm in length as defined by ultrastructural char- pulled, we could assess the health of the cell before forming the
acteristics15,16. Although such a mechanism receives support from patch. Second, we could ensure that the electrode was attached
theoretical studies14,17–19 and from labeling studies20–22, electro- to neuronal rather than glial membrane by evoking an action
physiological studies of action potential initiation in large pyra- potential. Third, the voltage clamp directly set the transmem-
midal neurons indicate that the initial segment is not the site of brane potential, thus minimizing errors in estimating the rest-
action potential initiation in these cells. First, simultaneous ing potential of the neuron. We recorded currents in visually
recordings from the soma and initial segment indicate that action identified somata (Fig. 1a), in initial segments (<30 µm from the
potentials are initiated in the axon beyond the initial segment in soma) and in axons (30–47 µm from the soma).
response to synaptic input in layer 1 (ref. 10). Second, local appli- The set of currents evoked by depolarizing steps from a neg-
cation of tetrodotoxin (a Na+ channel blocker) to the initial seg- ative holding potential differed qualitatively between axonal
ment does not alter the threshold for initiating action potentials and somatodendritic patches. In somatic patches (Fig. 1b) and
in response to somatic current injection. Instead, application of in patches from the basal dendrites (not shown, n = 4), a rapid-
tetrodotoxin to the axon beyond the initial segment raises the ly activating and inactivating inward current (INa) was followed
threshold by 7–10 mV (ref. 23). These results indicate that Na+ by an outward current with both transient (IK(A), A-type) and
channels in the axon proper rather than in the initial segment sustained (IK(DR), delayed rectifier) components9,23,24. In patch-
may determine the threshold. es from the initial segment and axon proper, however, A-type

articles
a Fig. 1. Total ensemble currents in outside-out patches from visually

identified somata and axons of neocortical layer 5 pyramidal neurons.
(a) Infrared-illuminated, differential interference contrast image of a
pyramidal neuron. Small arrows indicate the path of an axon. White
arrow indicates the recording electrode. IS, initial segment. Scale bar,
Soma IS Axon 10 µm. (b) Representative currents in outside-out patches from the
soma and axon. Patches were held at a holding potential of –100 mV and
b IK(A) stepped to test potentials between –50 and 0 mV. Somatic patches had
Soma a prominent transient outward current (IK(A)) in addition to inward INa
71
72
70
68
66
67
69
65
and sustained outward IK(DR) currents. Axonal patches did not have the
Axon transient outward current. (c) In axonal patches, applying greater depo-
larizations (–10 to 40 mV) to increase the driving force of K+ currents
did not result in any transient K+ currents. Waveforms are leak-
30 pA
subtracted averages of 15–20 individual sweeps.
20 ms
0 mV
–50 mV Next, we estimated the relative density of Na+ channels by
–100 mV comparing the magnitude of peak currents in somatic and
c axonal patches. Patches from the somata, initial segments and
Axon axons had peak currents of 26.6 ± 7.9 pA (n = 8),
22.4 ± 0.5 pA (n = 7) and 99.8 ± 33.7 pA (n = 9). Excluding a
40 pA
single outlier in the axon group of 308 pA reduced the peak
5 ms
current to 63.9 ± 13.0 pA (n = 8). Plotting the peak current
+40 mV versus distance from the soma (Fig. 2d) indicated that the cur-
–10 mV
–100 mV rent per patch was uniform throughout the initial segment
before increasing two- to threefold in the axon. Steady-state
inactivation, however, was similar in all patches. (Half-inacti-
current was only occasionally seen (2 of 18 patches). This may vation potentials for soma, initial segment and axon were
have resulted from either a low density of A-type channels or a –66 ± 2.4 mV (n = 6), –66 ± 1.8 mV (n = 5) and –69 ± 1.2 mV
restriction of these channels to specific regions25. In most patch- (n = 5) and slopes were 5.3 ± 0.6 (n = 6), 5.9 ± 1.8 (n = 5) and
es, only the inward and the sustained outward currents were 5.3 ± 0.8 mV (n = 5), respectively.)
present (Fig. 1b and c). Tetrodotoxin (1 µM) blocked the inward Previous work addressing action potential initiation in pyra-
current (n = 2) and 10 mM tetraethylammonium blocked the midal neurons indicated that there may be a 7–10-mV differ-
sustained outward current, as expected for INa and IK(DR). ence in the voltage thresholds of the axon and soma 23 . To
To test whether the biophysical properties and distribution of investigate whether the voltage dependence of Na+ channel
Na+ channels varied in the soma, initial segment and axon, we activation could account for such a difference in threshold, we
compared ensemble averages of Na+ currents (Fig. 2a). We con- numerically simulated a pyramidal neuron using a compart-
structed an activation curve from each ensemble and fit it with
a Boltzmann curve (Fig. 2b, see Methods) to yield a half-activa-
tion potential (V1/2) and a slope (k). Patches from the somata, a 0 mV
–60 mV
initial segments and axons had V1/2 values of –31.6 ± 0.5 mV –100 mV
(n = 7), –30.1 ± 1.4 mV (n = 7) and –38.4 ± 1.6 mV (n = 8) and
k values of 6.8 ± 0.46 (n = 7), 8.1 ± 0.62 (n = 7) and 6.0 ± 0.48
35 pA
(n = 8), respectively. A one-way analysis of variance (ANOVA)
1.25 ms
of the half-activation potentials indicated significant differences
between the axon and soma and between the axon and initial seg- b 1.0
ment. A plot of V1/2 as a function of the distance from the soma
conductance
0.8
Normalized
indicated that this may have been due to a shift in the voltage 0.6 Axon
Soma
dependence near the end of the initial segment rather than a 0.4
gradual decrease with distance (Fig. 2c). Thus, Na+ channels in 0.2
the axon proper required less depolarization for activation than –80 –60 –40 –20 0 20
did Na+ channels in the soma or initial segment. c Command potential (mV)
Activation V1/2 (mV)
–20
–30
Fig. 2. Na+ channel properties differ between the soma and axon.
(a) Representative ensemble currents in an outside-out patch from the –40
soma. The patch was held at a holding potential of –100 mV and stepped 0 15 30 45
to test potentials from –60 to 0 mV. (b) Voltage dependence of activa- Distance from soma (µm)
tion in somatic and axonal patches. Each curve is the best-fit Boltzmann
d
Peak current (pA)
for an individual patch. Axonal Na+ channels (solid lines) were activated 120
by less depolarization than somatic channels (dotted lines). (c) V1/2 of
Na+ currents plotted as a function of position along the axon. (d) Peak 60
Na+ current at 0 mV plotted for each patch as a function of position
0
along the axon. The values in (c) and (d) remained relatively constant 0 15 30 45
throughout the initial segment before changing in the axon. Distance from soma (µm)

articles
Fig. 3. Effect of Na+ channel properties on action potential threshold: a

computer simulations. (a) Reconstructed cell showing the position of
current injection (400 pA) into the apical dendrite. (b) Superimposed 400 pA
somatic action potential waveforms. Rightmost action potential in
each panel corresponds to uniform V1/2 and density. Left, the V1/2 of
300 µm
Na+ channel activation was varied from –2 to –8 mV. Right, the density
of Na+ channels in the initial segment was increased 10-, 20-, 50- and
100-fold. (c) Voltage thresholds measured at the soma for each value
of increased density (squares) and shifted V1/2 (circles). Although both b
modifications lowered the threshold, very high increases in density

20 mV
were required.
8 ms
–8 mV 100x
mental model (Fig. 3). We also investigated how a high densi-

ty of Na+ channels in the initial segment might alter the thresh- V1/2 activation Initial segment density
old. First, we varied the V 1/2 of Na + channels in the axon

beyond the initial segment from –2 mV to –8 mV (Fig. 3b, left). c Initial segment density (n-fold)
The action potential threshold measured at the soma became 100 50 10 1
Voltage threshold (mV)

–40
more negative as the shift in V1/2 became greater (Fig. 3c). The Density
observed shifts in V1/2 were sufficient to account for the dif- Activation
ferences in the experimentally observed threshold. To test the –50
effect on the threshold of Na+ channel density in the initial

segment, we varied the density from 10- to 100-fold, as com- –60
pared with the somatic density, while keeping the V1/2 uniform –8 –6 –4 –2 0
Shift in activation V1/2 (mV)
(Fig. 3b, right). Increasing the Na+ channel density in the ini-
tial segment decreased the threshold, but this required very
high channel densities (Fig. 3c).
We next investigated how the site of initiation varied as a todendritic extent (Fig. 4a and b, 0 mV). That is, the action
function of axonal Na+ channel activation or initial segment potential was initiated simultaneously across much of the neu-
density using the compartmental model. We observed the ini- ron rather than propagating through the neuron. Shifting the
tial development of the action potential by plotting membrane V1/2 of Na+ channels in the axon or increasing the density of
potential as a function of position (a space plot) along the axon, Na + channels in the initial segment altered this pattern of
soma and the apical dendrite (Fig. 4a–c) and as a function of action potential initiation; the action potential was initiated
time (Fig. 4b). With Na + channel activation held uniform first in the axon or initial segment, respectively (Fig. 4a, –7 mV
across the neuron, current injection into the dendrite (350 µm and 100× traces). This bias toward action potential initiation
from the soma, 500 pA) produced a regenerative depolariza- in the axon or initial segment was proportional to the magni-
tion across the neuron’s initial segment and much of its soma- tude of the shift in the V1/2 (Fig. 4c, left traces) or the increase
in density (Fig. 4c, right traces), indicating that suf-
ficiently strong dendritic depolarization may over-
come this bias to produce dendritic initiation26,27.
a Inject 500 pA
c Although both models lowered the threshold and
biased the neuron away from initiating an action
potential in the dendrites, the spatial patterns of action
0 mV 1× potential initiation differed considerably. When we
shifted the V1/2 of axonal Na+ channels toward more
0 mV/ negative potentials, the length of axon that participat-
1× ed in producing the regenerative current during action
–2 mV
potential initiation (200 µm) was much greater than
10×
–7 mV
40 mV Fig. 4. Spatial patterns of action potential initiation: com-
puter simulations. (a) Reconstructed cell showing the posi-
500 µm
50 mV tion of current injection (500 pA) into the apical dendrite
250 µm
–4 mV 20× (top). Space plots show membrane potential (Vm) across the
100×
axodendritic axis. Waveforms in each group represent differ-
ent time points beginning just before action potential initia-
tion and ending before propagation begins. 0 mV/1×, uniform
b s
–6 mV 50× Na+ channel properties; –7 mV, Na+ channel activation
0 mV d d shifted –7 mV in the axon; 100×, 100-fold increase in Na+
s channel density in the initial segment. (b) Action potential
waveforms (left) in the soma (s) and dendrites (d) and corre-
40 mV 80 mV
ms
sponding time derivatives dV/dt (right). 0 mV, uniform Na+
–8 mV 100× channel properties; –7 mV, Na+ channel activation shifted
–7 mV 8 ms s 2.5 ms
d –7 mV in the axon. (c) Space plots as in (a) at intermediate
d
Shift axon V1/2 Increase IS density values of shifted activation (left) and increased initial segment
s
(IS) density (right).

articles
a Fig. 5. Threshold increases when a pyramidal neuron is cut at the end of

the initial segment. (a) Action potentials recorded in response to
Control
somatic current injection in a control pyramidal neuron (dotted line) and
Axon cut 30 mV a pyramidal neuron with its axon cut at the end of the initial segment
10 ms (solid line). Cutting the axon raised the threshold. (b) A brief strong cur-
300 pA rent injection evoked a full-amplitude action potential in both the control
and cut cells. (c) Computer simulations of cut (solid lines) and control
neurons (dotted lines) distinguished between models. With a –7-mV shift
b in Na+ channel activation in the axon and either uniform (black lines) or
10-fold higher (gray lines) density of Na+ channels in the initial segment
30 mV cutting the axon raised the threshold, (upper traces). With 50-fold
1.25 ms higher density of Na+ channels in the initial segment, cutting the axon did
not raise the threshold (lower traces).
2.5 nA
c contribute to a low threshold for action potential initiation. They

imply a mechanism favoring initiation of action potentials in the
–7 mV axon that differs in two ways from the widely accepted scheme.
First, the properties of Na+ channels producing the low threshold
for action potential initiation exist in the axon proper rather than
30 mV
in the axon hillock/initial segment. A long history of reports indi-
10 ms
cating that action potential initiation may occur beyond the axon
50 × hillock/initial segment10,23,28–31 substantiates this idea. Second,
a shift in the voltage dependence of activation of Na+ channels
350 pA in the axon, rather than a high density of Na+ channels in the ini-
tial segment, determines the low threshold. In addition to the
present findings, rapid changes in threshold after the induction of
the length of the initial segment. In fact, the initial segment con- long-term potentiation32 and after pharmacological activation
tributed relatively little to the regenerative current, because the of protein kinase C (ref. 33) seem more readily explained by mod-
electrical load presented by the soma and dendrites limited its ulation of channel properties than by rapid insertion of large
depolarization. In contrast, in the high-Na+-channel-density numbers of Na+ channels into the membrane.
model, the initial segment played the chief role in producing the Although it has been commonly believed that a high density
initial regenerative current. The pattern of initiation in the of Na+ channels in the axon hillock and initial segment deter-
shifted-activation model was thus more consistent with previ- mines the threshold for action potential initiation, this idea has
ous experimental results where direct application of tetrodotox- had mixed experimental support. A high density was apparently
in to the initial segment did not change the threshold23. first suggested in a computer study17 motivated by ultrastructur-
The fortuitous finding of a pyramidal neuron that had been al similarities between the node of Ranvier and the initial seg-
cut through the axon at 25 µm from the soma by the slicing pro- ment15,16. Subsequent immunohistochemical studies20,21 indicated
cedure allowed a further experimental test of the models. We com- that there may be as much as a tenfold higher density of Na+ chan-
pared thresholds in this cut neuron and in a neighboring pyramidal nels in the initial segment, although only a fraction of neurons
neuron (in the same field in the slice) that did not have a cut axon. showed increased densities. A freeze-fracture study found the total
Both cells were healthy by visual criteria and had resting poten- number of particles in the initial segment to be only threefold
tials of –63 mV. The shifted-Na+-channel-activation model pre- higher than in the soma34. Patch-clamp studies in subiculum23
dicted that cutting the axon at the end of the initial segment would and in neocortex (ref. 35 and present results) have not detected
raise the threshold because the cut would eliminate the channels any increase in the density of Na+ channels in the initial segment.
with shifted activation. The high-density model predicted that Of course, each of these methods may have underestimated or
threshold would not increase because the cut would preserve the overestimated the true relative density of functional channels.
high density of channels in the initial segment. Recording from Recent computer simulations of pyramidal neurons used very
the neuron with the cut axon showed an increased threshold in high densities of Na+ channels in the initial segment (200–1,000-
response to somatic depolarization (Fig. 5a). The action potential fold higher than the somatic density) to account for the pattern of
was also greatly reduced in amplitude and rate. To test whether the axonal action potential initiation and back-propagation in pyra-
small amplitude of the action potential in the cut neuron was due midal neurons18,19. Together with the present simulations, the
to inactivation of Na+ channels or was an indication of some other results of these studies indicate that the threshold may be relative-
damage, we stimulated each neuron with a strong depolarizing ly insensitive to Na+ channel density in the initial segment and that
current. The resulting action potentials in each neuron had equiv- moderately elevated densities such as those reported by immuno-
alent amplitude and rate of rise (Fig. 5b), indicating that no sub- histochemical studies are unlikely to contribute strongly to a lower
stantial source of Na+ current near the soma was lost because of threshold. The relatively small effects on the threshold of altering
the cut. Numerical simulation confirmed that cutting the axon at the density of Na+ channels in the initial segment results from mor-
the end of the initial segment raised the threshold in the shifted- phological factors. The initial segment is physically and electrical-
activation model but not in the high-density model (Fig. 5c). ly very near the much larger soma and apical dendrite. Thus, the
membrane potential of the initial segment tends to be ‘clamped’
DISCUSSION to the somatic potential by the soma’s relatively large capacitance.
These data are the first direct electrophysiological measurements When action potentials were initiated in the axon because of the
in cortical axons of ion channel properties and distributions that shifted V1/2, a relatively large length of axon, as compared with the

articles
initial segment, participated in the production of the regenerative bath solution to block Na+ currents in some experiments. We purchased
potential. The initial segment contributed little to the depolariza- all other reagents from Sigma.
tion but served to isolate the axon from the electrical load of the We visualized axons of layer 5 pyramidal neurons using infrared-
soma. The exact length of the axon that participated in initiating illuminated differential interference contrast optics (BX50WI micro-
scope, Olympus) and a Newvicon camera (DAGE-MTI, Michigan City,
the action potential depended on the passive membrane parame-
Indiana). In the oldest animals, axons beyond the initial segment looked
ters, which have not been explicitly determined for the small axons much like those in myelinated tracts, having a ‘railroad track’ appear-
that we studied. Using typical ranges of values for passive para- ance. Whole-cell recording always yielded a cell that could fire an action
meters, however, this length was about 200–300 µm. In this case, it
potential. We made recordings using an Axopatch 200 amplifier (Axon

is more correct to think of a region of initiation rather than a dis- Instruments, Union City, California) using patch pipettes of about
crete site. In myelinated axons, this region of initiation may be 15 MΩ made from EN-1 glass and coated with Sylgard (Dow-Corning,
much longer and may involve multiple nodes of Ranvier. Such a Auburn, Michigan). We made all recordings at room temperature
finding was suggested in a study in vivo that compared ortho- (23 ± 0.5°C). We constructed ensemble waveforms from 15–25 indi-
dromic and antidromic action potentials in a motor neuron31. vidual sweeps and digitally leak subtracted them using scaled versions
We have investigated two distinct models that might account of the steps to –70 mV. We filtered the recordings at 2 kHz (8-pole
Bessel), sampled them at 10 kHz at 16-bit resolution (ITC18, Instrutech
for a low threshold for action potential initiation in the axon.
Corporation, Port Washington, New York) and stored them on a com-
The shifted-activation model used uniform densities, whereas puter (Apple Computer, Cupertino, California) for offline analysis.
the increased-density model used uniform activation. The pres- To construct activation curves, we held patches at –100 mV and stepped
ence or absence of these specializations need not, however, be them to –70, –60, –50, –45, –40, –30, –20, –10 and 0 mV. The peak current
mutually exclusive. In fact, modestly increased densities of Na+ at each potential was then converted to a chord conductance assuming a
channels in the initial segment improved the initial rate of rise Na+ reversal potential of +55 mV. We made a least-squares fit to a single
of the action potential in the simulations without much effect on Boltzmann function (Mathematica, Wolfram Research, Champaign, Illi-
the threshold (for example, Fig. 5c). This finding indicated that a nois) for each individual patch, yielding a half activation (V1/2) and a
possible role for increased density of Na+ channels in the initial slope. Somatic patches included K+ channel blockers to block the fast
transient K+ currents. We included axonal and initial segment patches
segment may be to ensure back-propagation of action potentials
without blockers in the analysis because they lacked the fast transient cur-
into the soma after initiation in the axon. Furthermore, if a high rent, and the relatively small sustained current reached a maximum well
density of Na+ channels indeed exists beyond 30 µm from the after the peak of the Na+ current. We included K+ channel blockers in
soma (despite our patch estimates), then the shifted kinetics and some patches (n = 2 in the axon, n = 3 in the initial segment). For inacti-
density would combine to lower the threshold farther, perhaps vation curves, we held patches at potentials between –100 and –30 mV
confining initiation of the action potential to this region. for 3 s and then stepped them to 0 mV to evoke currents. We normalized
A shift in the voltage dependence of activation of Na+ chan- data to the maximum current (from –100 mV) and fit them to a negative
nels is just one of several regional specializations within the cell Boltzmann curve. We determined significance by an ANOVA, with a
that together would promote action potential initiation in the post hoc Tukey’s test between groups. Statistical significance was taken to
axon. A prominent gradient in the density of A-type channels in be P < 0.05. We reported all summary data as mean ± s.e.m.
We made computer simulations using NEURON40. The model was
the dendrites of CA1 pyramidal neurons serves to limit dendritic similar to that described in Hoffman et al.9 with some modifications.
initiation of action potentials9,26. A similar gradient of nonspe- We visualized a layer 5 pyramidal neuron from a 23-day-old rat using
cific cation channels (Ih) may function similarly by decreasing the biocytin as previously described23 and reconstructed it morphologi-
input resistance of the dendrites36. Our finding that A-type K+ cally using Neurolucida (MicroBrightfield, Colchester, Vermont). Sim-
channels were generally absent in axonal patches would contribute ulations assumed a temperature of 23°C. Passive parameters were
to a lower threshold in the axon. Alternatively, this finding may Rm = 50,000 Ωcm2, Ri = 100 Ωcm and Cm = 1 µF/cm2, except in the
dendrites, where Cm = 2 µF/cm2. The baseline channel densities were
simply reflect the clustering of channels to perinodal regions25. gNa = 0.0336 S/cm2, gK(DR) = 0.0154 S/cm2 and gK(A) = 0.0031 +
Taken together, the presence of these specializations indicates 0.00465 × (distance from soma per 200 µm). The Na+ channel density
that at least under some conditions, pyramidal neurons may be in the axon beyond the initial segment was set threefold higher. The
biased to initiate action potentials from an axonal site. Under Na+ channel model was of the form gNa = gNa m3h. Parameters were
such conditions, synaptic input throughout the dendrites would determined by least-squares fit to axonal Na+ current ensemble aver-
be summed and thresholded to produce an output in the form ages. The maximum value of m3h was 0.35, and thus the peak Na+ con-
ductance in the model was about 12 mS/cm2. Steady-state values for
of an action potential in the axon, which would subsequently m and h were of the form ss(v) = α(v)/(α(v) + β(v)), and time con-
back-propagate into the dendrites to signal that firing has stants τ = 1/(α(v) + β(v)), αm(v) = 0.182(v– V)/(1 – Exp[–{v – V}/6])
occurred35,37. Advantages to placing the thresholding function and βm = 0.124(–v + V)/(1 – Exp[{–v + V}/6]), αh(v) = –0.015(v –
in the axon may include minimizing any effects of changes in Vh)/(1 – Exp[{v – Vh}/6]) and βh = –0.015(–v + Vh)/(1 – Exp[{–v +
dendritic or somatic conductance, or synchronizing the output of Vh}/6]). In the soma, dendrites and initial segment, V was –40 mV. To
a population of neurons38,39. test the effect of altering the voltage dependence of activation, we var-
ied V between –40 and –48 mV. We used a V of –40 mV across the cell
in simulations that varied the initial segment density. We set Vh to
METHODS –69 mV in the axon and –66 mV in the remainder of the neuron. To
Neocortical slices (300 µm) were prepared from 14–24-day-old Sprague- allow for errors in estimating the actual length of the initial segment,
Dawley rats23 according to protocols approved by the University of Hous- simulations tested lengths of both 30 and 50 µm. Action potential
ton Institutional Animal and Use Committee. The external artificial threshold was taken as the point where the second time derivative of
cerebrospinal fluid contained 125 mM NaCl, 2.5 mM KCl, 1.25 mM the membrane potential waveform reached a value of 5 mV/ms2.
NaH2PO4, 25 mM NaHCO3, 2.0 mM CaCl2, 1.0 mM MgCl2 and 25 mM
dextrose. The pipette solution used for outside-out patch recordings con-
tained 140 mM potassium gluconate, 10 mM HEPES, 10 mM EGTA, Acknowledgments
4.0 mM NaCl, 4.0 mM Mg2ATP and 0.3 mM Mg2GTP (pH 7.3). To block This work was supported by a National Institute of Neurological Disorders and
K + channels, 10 mM tetraethylammonium chloride and 10 mM 4- Stroke (NINDS) grant (NS36982) to C.M.C. We thank J. Stringer, D. Johnston,
aminopyridine replaced equi-osmolar amounts of potassium gluconate. M. Rea, A. Eskin, G. Cahill and S. Dryer for reading earlier versions of the
We included 1 µM tetrodotoxin (Alomone Labs, Jerusalem, Israel) in the manuscript, and L. Cleary for the use of the Neurolucida system.

articles
Competing interests statement 20. Catterall, W. A. Localization of sodium channels in cultured neural cells.
J. Neurosci. 1, 777–783 (1981).
The authors declare that they have no competing financial interests. 21. Angelides, K. J., Elmer, L. W., Loftus, D. & Elson, E. Distribution and lateral
mobility of voltage-dependent sodium channels in neurons. J. Cell Biol. 106,
RECEIVED 12 NOVEMBER 2001; ACCEPTED 28 MARCH 2002 1911–1924 (1988).
22. Alessandri-Haber, N. et al. Specific distribution of sodium channels in axons
of rat embryo spinal motoneurones. J. Physiol. (Lond.) 518, 203–214 (1999).
23. Colbert, C. M. & Johnston, D. Axonal action-potential initiation and Na+
1. Mel, B. W. Information processing in dendritic trees. Neural Comput. 6,
channel densities in the soma and axon initial segment of subicular
1031–1085 (1994).
pyramidal neurons. J. Neurosci. 16, 6676–6686 (1996).
2. Koch, C. Biophysics of Computation: Information Processing in Single Neurons
24. Stuart, G. J. & Sakmann, B. Active propagation of somatic action potentials

(Oxford Univ. Press, New York, 1999).
into neocortical pyramidal cell dendrites. Nature 367, 69–72 (1994).
3. Shadlen, M. N. & Newsome, W. T. The variable discharge of cortical neurons:
25. Rasband, M. N., Trimmer, J. S., Peles, E., Levinson, S. R. & Shrager, P. K+
implications for connectivity, computation, and information coding.
channel distribution and clustering in developing and hypomyelinated axons
J. Neurosci. 18, 3870–3896 (1998).
of the optic nerve. J. Neurocytol. 28, 319–331 (1999).
4. Regehr, W. G. & Armstrong, C. M. Where does it all begin? Curr. Biol. 4,
26. Golding, N. L. & Spruston, N. Dendritic sodium spikes are variable triggers of
436–439 (1994). axonal action potentials in hippocampal CA1 pyramidal neurons. Neuron 21,
5. Adams, P. R. The platonic neuron gets the hots. Curr. Biol. 2, 625–627 (1992). 1189–1200 (1998).
6. Svoboda, K., Denk, W., Kleinfeld, D. & Tank, D. W. In vivo dendritic calcium 27. Kloosterman, F., Peloquin, P. & Leung, L. S. Apical and basal orthodromic
dynamics in neocortical pyramidal neurons. Nature 385, 161–165 (1997). population spikes in hippocampal CA1 in vivo show different origins and
7. Turner, R. W., Meyers, D. E. R., Richardson, T. L. & Barker, J. L. The site for patterns of propagation. J. Neurophysiol. 86, 2435–2444 (2001).
initiation of action potential discharge over the somatodendritic axis of rat 28. Carras, P. L., Coleman, P. A. & Miller, R. F. Site of action potential initiation in
hippocampal CA1 pyramidal neurons. J. Neurosci. 11, 2270–2280 (1991). amphibian retinal ganglion cells. J. Neurophysiol. 67, 292–304 (1992).
8. Andreasen, M. & Lambert, J. D. C. Regenerative properties of pyramidal cell 29. Zecevic, D. Multiple spike-initiation zones in single neurons revealed by
dendrites in area CA1 of the rat hippocampus. J. Physiol. (Lond.) 483, voltage-sensitive dyes. Nature 381, 322–325 (1996).
421–441 (1995). 30. Edwards, E. & Ottoson, D. The site of impulse initiation in a nerve cell of a
9. Hoffman, D. A., Magee, J. C., Colbert, C. M. & Johnston, D. K+ channel crustacean stretch receptor. J. Physiol. (Lond.) 143, 138–148 (1958).
regulation of signal propagation in dendrites of hippocampal pyramidal 31. Gogan, P., Gueritaud, J. P. & Tyc-Dumont, S. Comparison of antidromic and
neurons. Nature 387, 869–875 (1997). orthodromic action potentials of identified motor axons in the cat’s brain
10. Stuart, G., Schiller, J. & Sakmann, B. Action potential initiation and stem. J. Physiol. (Lond.) 335, 205–220 (1983).
propagation in rat neocortical pyramidal neurons. J. Physiol. (Lond.) 505, 32. Chavez-Noriega, L. E., Halliwell, J. V. & Bliss, T. V. P. A decrease in firing
617–632 (1997). threshold observed after induction of the EPSP-spike (E-S) component of long-
11. Spruston, N., Schiller, Y., Stuart, G. & Sakmann, B. Activity-dependent action term potentiation in rat hippocampal slices. Exp. Brain Res. 79, 633–641 (1990).
potential invasion and calcium influx into hippocampal CA1 dendrites. 33. Astman, N., Gutnick, M. J. & Fleidervish, I. A. Activation of protein kinase C
Science 268, 297–300 (1995). increases neuronal excitability by regulating persistent Na+ current in mouse
12. Moore, J. W. & Westerfield, M. Action potential propagation and threshold neocortical slices. J. Neurophysiol. 80, 1547–1551 (1998).
parameters in inhomogeneous regions of squid axon. J. Physiol. (Lond.) 336, 34. Matsumoto, E. & Rosenbluth, J. Plasma membrane structure at the axon
285–300 (1983). hillock, initial segment and cell body of frog dorsal root ganglion cells.
13. Moore, J. W., Stockbridge, N. & Westerfield, M. On the site of impulse J. Neurocytol. 14, 731–747 (1985).
initiation in a neurone. J. Physiol. (Lond.) 336, 301–311 (1983). 35. Stuart, G., Spruston, N., Sakmann, B. & Hausser, M. Action potential
14. Luscher, H. R. & Larkum, M. E. Modeling action potential initiation and initiation and backpropagation in neurons of the mammalian CNS. Trends
back-propagation in dendrites of cultured rat motoneurons. J. Neurophysiol. Neurosci. 20, 125–131 (1997).
80, 715–729 (1998). 36. Magee, J. C. Dendritic hyperpolarization-activated currents modify the
15. Farinas, I. & DeFelipe, J. Patterns of synaptic input on corticocortical and integrative properties of hippocampal CA1 pyramidal neurons. J. Neurosci.
corticothalamic cells in the cat visual cortex II. The axon initial segment. 18, 7613–7624 (1998).
J. Comp. Neurol. 304, 70–77 (1991). 37. Johnston, D., Magee, J. C., Colbert, C. M. & Christie, B. R. Active properties
16. Palay, S. L., Sotelo, C., Peters, A. & Orkland, P. M. The axon hillock and the of neuronal dendrites. Annu. Rev. Neurosci. 19, 165–186 (1996).
initial segment. J. Cell Biol. 37, 193–201 (1968). 38. Buhl, E. H. et al. Physiological properties of anatomically identified axo-
17. Dodge, F. A. & Cooley, J. W. Action potential of the motoneuron. IBM J. Res. axonic cells in the rat hippocampus. J. Neurophysiol. 71, 1289–1307 (1994).
Dev. 17, 219–229 (1973). 39. Cobb, S. R., Buhl, E. H., Halasy, K., Paulsen, O., & Somogyi, P.
18. Mainen, Z. F., Joerges, J., Huguenard, J. R. & Sejnowski, T. J. A model of spike Synchronization of neuronal activity in hippocampus by individual
initiation in neocortical pyramidal neurons. Neuron 15, 1427–1439 (1995). GABAergic interneurons. Nature 378, 75–78 (1995).
19. Rapp, M., Yarom, Y. & Segev, I. Modeling back propagating action potential 40. Hines, M. NEURON—a program for simulation of nerve equations. in
in weakly excitable dendrites of neocortical pyramidal cells. Proc. Natl. Acad. Neural Systems: Analysis and Modeling (ed. Eeckman, F.) 127–136 (Kluwer,
Sci. USA 93, 11985–11990 (1996). Norwell, Massachusetts, 1993).

errata
Prefontal cortex in long-term memory: an “interference” approach using magnetic

stimulation
Simone Rossi, Stefano F. Cappa, Claudio Babiloni, Patrizio Pasqualetti, Carlo Miniussi, Filippo Carducci, Fabio Babiloni
and Paolo M. Rossini
Nat. Neurosci. 4, 948–952 (2001)
The title of this article contained a typographical error. It should have read as follows:
Prefrontal cortex in long-term memory: an “interference” approach using magnetic

stimulation
Ion channel properties underlying axonal action potential initiation in pyramidal

neurons
Costa M. Colbert and Enhui Pan
Nat. Neurosci. 5, 533–538 (2002)
A printer’s error introduced an extraneous diagonal line into Fig. 2b on page 534. The correct figure is reproduced below.
b 1.0 Fig. 2. Na+ channel properties differ between

conductance
0.8
Normalized
the soma and axon. (b) Voltage dependence of

0.6 Axon activation in somatic and axonal patches. Each
0.4 Soma
curve is the best-fit Boltzmann for an individual
0.2 patch. Axonal Na+ channels (solid lines) were
–80 –60 –40 –20 0 20 activated by less depolarization than somatic
Command potential (mV) channels (dotted lines).
corrigenda
Neurotrophins use the Erk5 pathway to mediate a retrograde survival response
Fiona L. Watson, Heather M. Heerssen, Anita Bhattacharyya, Laura Klesse, Michael Z. Lin and Rosalind A. Segal
Nat. Neurosci. 4, 981–988 (2001)
In Fig. 5e on page 986, the pluses and minuses for lines “PD to DA” and “PD to CB” were incorrect. The conclusions stated in the text
and the experimental description in the figure legend were correct. The corrected figure is reproduced below.
Fig. 5. Activation of Erk5 promotes survival. (e) Neurons in compart-

e mented cultures were treated with PD98059 (PD) to distal axons or cell
bodies, as indicated. Distal axons were stimulated with neurotrophins
and cell body lysates were immunoblotted for P-CREB. PD treatment of
distal axons alone does not prevent CREB phosphorylation. When PD is
applied to the cell bodies, CREB phosphorylation is inhibited.
A-kinase anchoring proteins in amygdala are involved in auditory fear memory

Marta A.P. Moita, Raphael Lamprecht, Karim Nader and Joseph E. LeDoux
Nat. Neurosci. 5, 837–838 (2002)
The authors wish to correct their supplementary methods online, which gave the wrong sources for three antibodies. The mouse
anti-RIIα and anti-RIIβ antibodies were obtained from Transduction Laboratories (San Diego, California), and the rabbit anti-
AKAP150 antibody was obtained from Santa Cruz Biotechnology (Santa Cruz, California).
nature neuroscience • volume 5 no 10 • october 2002 1017

articles
A rapid switch in sympathetic

neurotransmitter release properties
mediated by the p75 receptor
Bo Yang*, John D. Slonimsky* and Susan J. Birren
Department of Biology, Volen Center for Complex Systems, 415 South St., M/S 008, Brandeis University, Waltham, Massachusetts 02454, USA
* The first two authors contributed equally to this work.
Correspondence should be addressed to S.J.B. (birren@brandeis.edu)
Published online: 29 April 2002, DOI: 10.1038/nn853
Cardiac function is modulated by norepinephrine release from innervating sympathetic neurons.

These neurons also form excitatory connections onto cardiac myocytes in culture. Here we report
that brain-derived neurotrophic factor (BDNF) altered the neurotransmitter release properties of
these sympathetic neuron–myocyte connections in rodent cell culture, leading to a rapid shift from
excitatory to inhibitory cholinergic transmission in response to neuronal stimulation. Fifteen
minutes of BDNF perfusion was sufficient to cause this shift to inhibitory transmission, indicating
that BDNF promotes preferential release of acetylcholine in response to neuronal stimulation. We
found that p75–/– neurons did not release acetylcholine in response to BDNF and that neurons
overexpressing p75 showed increased cholinergic transmission, indicating that the actions of BDNF
are mediated through the p75 neurotrophin receptor. Our findings indicate that p75 is involved in
modulating the release of distinct neurotransmitter pools, resulting in a functional switch between
excitatory and inhibitory neurotransmission in individual neurons.
Nerve growth factor (NGF), BDNF, neurotrophin-3 and neu- of a p75 signaling pathway—markedly increased cholinergic
rotrophin 4/5 comprise the neurotrophin family of neu- transmission and resulted in a functional switch to an inhibitory
rotrophic factors. The effects of neurotrophins on cell function postsynaptic response.
are mediated through ligand-specific binding to members of
the Trk family of receptor tyrosine kinases and through p75, a RESULTS
receptor capable of binding all members of the neurotrophin BDNF promotes cholinergic transmission
family 1. Neurotrophins have extensive effects on neuronal In culture, sympathetic neurons formed functional connections
development and function, as they support neuronal survival, to cardiac myocytes (Fig. 1a and d). Functional connections are
trigger cell death, promote process outgrowth and modulate characterized by the stimulus-evoked release of neurotrans-
synaptic plasticity2. mitter13,17, which leads to a postsynaptic response that can be
Neurotrophins act as short-term (acute) and long-term measured as a change in myocyte beat rate during neuronal
(chronic) modulators of synaptic activity in the central and stimulation (Fig. 1a)5,18. Under these culture conditions, there
peripheral nervous systems3. Studies report that neurotrophins was no spontaneous neuronal activity, which allowed us to mea-
act presynaptically at developing motor synapses4, at sympathetic sure the evoked response of the myocytes. We measured the
neuron–cardiac myocyte synapses5, in hippocampal slices6 and magnitude of the functional postsynaptic response for neu-
in cultured cortical neurons7. Neurotrophins regulate features of ron–myocyte pairs cultured for 3 days in 5 ng/ml NGF alone
neurotransmitter release such as the number of synaptic vesicles, (control) and supplemented with additional NGF, BDNF or cil-
the expression of synaptic proteins, the number of docked vesicles iary neurotrophic factor (CNTF) (Fig. 1a and b). In the con-
and the probability of vesicle fusion7–10. trol condition, neuronal stimulation led to an increase in
In the peripheral nervous system, NGF is involved in estab- myocyte beat rate. There was a small (not statistically signifi-
lishing synaptic connections between sympathetic neurons and cant) increase in myocyte response after the 3-day (chronic)
myocytes during development11,12 and acutely potentiates exci- treatment with 50 compared to 5 ng/ml NGF, suggesting that
tatory neurotransmission in vitro5 via the TrkA receptor. In cul- any actions of NGF on the development of synaptic connectiv-
ture, neonatal sympathetic neurons can release norepinephrine, ity were saturated at 5 ng/ml NGF. In cultures treated with 100
acetylcholine or both, depending on environmental signals13–16. ng/ml BDNF, neuronal stimulation led to a marked decrease in
Thus, the acute modulation of synaptic activity by NGF raises myocyte beat rate (Fig. 1a and b). BDNF did not alter the aver-
the question of whether neurotrophins modulate the release of age baseline beat rate of cardiac myocytes (mean baseline beat
specific neurotransmitters. We found that activation of the p75 rate in control, 32 ± 12 beats/min; BDNF, 37 ± 11 beats/min).
receptor—by BDNF treatment, p75 overexpression or activation Thus, in cultures treated for 3 days, BDNF caused a net change

articles
Fig. 1. BDNF promotes cholinergic transmission in sympathetic neu-

a ron–myocyte cocultures. (a) Neonatal sympathetic neurons were
cultured for 3 days with neonatal ventricular myocytes in the pres-
ence of 5 ng/ml NGF either without (Control, left) or with 100 ng/ml
BDNF (right). For neuron–myocyte pairs (as in d), myocyte beat rate
was measured before, during and after neuronal stimulation. Data is
shown as the relative change in myocyte beat rate compared to the
average baseline for the individual myocyte (mean baseline beat rate
for control, 17 beats/min; BDNF, 33 beats/min). Although baseline
myocyte beat rates varied from cell to cell, average beat rate was not
affected by BDNF treatment (see Results). (b) Averaged data for
control (Con) and cultures grown in the presence of additional NGF
(50 ng/ml), BDNF (100 ng/ml) or CNTF (10 ng/ml). To confirm that
b CNTF functioned as a cholinergic differentiation factor, cultures
were maintained for 3 weeks with or without CNTF and then sub-
jected to the beat rate assay. By the 3-week time-point, CNTF had
induced a functional switch to inhibitory neurotransmission (mean ±
s.e.m., n ≥ 7 for each condition). Cultures treated with BDNF and 3
weeks of CNTF were significantly different from controls (*, P <
0.001). (c) Atropine (1 µM) or propranolol (10 µM) was included in
the bath solution to block cholinergic or noradrenergic transmission,
respectively (mean ± s.e.m., n ≥ 4 for each condition). BDNF + pro-
pranolol was significantly different from control (Con) + propranolol
(*, P < 0.02). (d) Neurons positioned near beating myocytes were
stimulated to elicit action potentials at 2.5 Hz (scale bar, 10 µm).
c
d
ergic transmission was seen. BDNF significantly enhanced this
inhibitory cholinergic transmission (Fig. 1c). Sympathetic neu-
rons can package both norepinephrine and acetylcholine, and
are capable of dual release of these transmitters17. We there-
fore suggest that increased BDNF alters the neurotransmitter
profile of these neurons to favor release of acetylcholine, result-
ing in an inhibitory functional output.
Presynaptic action of BDNF

Do these BDNF-induced changes in postsynaptic responses
reflect changes in the presynaptic release of neurotransmit-
from excitatory to inhibitory sympathetic neurotransmission. ters, or changes in the myocyte response to those transmitters?
This result is in contrast to the effects of ciliary neurotrophic We addressed this question by applying neurotransmitter recep-
factor (CNTF), a known cholinergic differentiation factor for tor agonists directly to cocultured myocytes grown in the pres-
sympathetic neurons19. Synaptic transmission was excitatory ence or absence of BDNF. We used a picospritzer (see Methods)
after 3 days of CNTF treatment, then changed to inhibitory to apply either norepinephrine (10 µM) or the cholinergic ago-
transmission after a three-week culture period (Fig. 1b). Thus, nist muscarine chloride (25 µM) for 25, 50 or 75 ms at a pres-
BDNF-induced changes in neurotransmission are rapid com- sure of 10 pounds per square inch (p.s.i.). Measurements of
pared to the actions of CNTF and may be mediated by an inde- myocyte beat rate showed that the responses were within the
pendent mechanism. dynamic range of the agonist concentrations. BDNF did not
The BDNF-dependent, stimulus-evoked decrease in myocyte affect the myocyte response to norepinephrine or muscarine
beat rate was completely abolished by the acetylcholine receptor chloride (Fig. 2), indicating that BDNF does not alter the post-
antagonist atropine (Fig. 1c), implying an effect of BDNF on synaptic response of myocytes to either excitatory or inhibito-
cholinergic neurotransmission. Atropine did not cause a signif- ry neurotransmitters. These results suggest that BDNF alters
icant change in the level of excitatory noradrenergic transmis- presynaptic properties of sympathetic neurons.
sion in control cultures (compare Fig. 1b and 1c), although it
uncovered an excitatory component of neurotransmission in
BDNF-treated cultures. When propranolol was used to block a b
noradrenergic transmission, only a low level of inhibitory cholin-
Fig. 2. BDNF did not alter the postsynaptic response to neurotransmit-

ter. Sympathetic neuron–myocyte cocultures were grown in control
medium (Con), NGF or BDNF. After 3 days (a) 10 µM norepinephrine
or (b) 25 µM muscarine chloride was applied to a beating myocyte for
different pulse durations. Myocyte beat rate was recorded for 3 min
before and after the pulse. The change in myocyte beat rate after drug
application is shown for each pulse duration (mean ± s.e.m., n = 5 for
each condition).
articles
Fig. 3. Acute modulation of neurotransmitter release by BDNF.

(a) Whole-cell recordings were obtained from control cultures and a
the change in myocyte beat rate in response to neuronal stimulation
was measured (Con). BDNF (100 ng/ml) was then perfused into the
dish for 15 min, the same neuron was stimulated and myocyte
response was measured again (BDNF). After a 15-min washout of the
BDNF (Wash), the neuron was again stimulated and the postsynaptic
response measured. Values are given as mean ± s.e.m., n = 7; *, acute
BDNF treatment significantly different from control (P < 0.02). (b) The
switch to inhibitory transmission was blocked by 1 µM atropine

(mean ± s.e.m., n = 7). (c) Dose–response curve for the cholinergic
switch induced by 15-min application of BDNF (mean ± s.e.m., n = 5
at each concentration).
b
Cholinergic effects of BDNF are rapid
We have previously shown that NGF acutely (within 15 min-
utes) promotes noradrenergic transmission between sympa-
thetic neurons and myocytes 5. Here, by exposing excitatory
connections to a short pulse of BDNF and measuring myocyte
response to neuronal stimulation, we tested the hypothesis that
BDNF rapidly modulates cholinergic release. After the cultures
were grown for three days in the control condition, we identified
a neuron in close proximity to a beating myocyte. We stimu-
lated the neuron and recorded the excitatory response of the
myocyte. We then perfused BDNF into the dish for 15 minutes
and stimulated the neuron again. There was a dramatic change c
to an inhibitory postsynaptic response after the 15-minute
BDNF application (Fig. 3a). This switch was reversible, as
shown by the partial recovery of an excitatory postsynaptic
response after a 15-minute washout of the BDNF. The rapidity
and reversibility of the BDNF effect indicate that the release
properties of the cholinergic vesicle pool can be rapidly modu-
lated and that a non-transcriptional mechanism mediates
BDNF-induced changes in neurotransmitter release.
Inhibition after acute BDNF perfusion was blocked by the addi-
tion of atropine to the bath solution (Fig. 3b), confirming that
BDNF caused an acute increase in cholinergic transmission within
15 minutes. The acute effect of BDNF was dose dependent, with a
shift to predominantly inhibitory transmission between 50 and
75 ng/ml BDNF (Fig. 3c). This dose–response curve is decidedly supports sympathetic neuron survival via activation of the TrkA
steep in the 50–75 ng/ml range. Although the reasons for this are receptor20. Moreover, ectopic expression of TrkB in sympathetic
not clear, it suggests that factors, possibly postsynaptic, other than cells also triggers a survival response21. Thus, if BDNF regulates
the total amount of acetylcholine and norepinephrine released con- synaptic function by activation of any Trk receptor, it should
tribute to the overall level of myocyte excitation and inhibition. also provide trophic support for the sympathetic neurons.
Because Trk-mediated neuron survival requires a lower dose of
p75 regulation of cholinergic transmission neurotrophin than Trk-mediated synaptic regulation5, we rea-
To investigate whether BDNF acts through Trk receptors to acute- soned that a survival response to BDNF would indicate activa-
ly promote cholinergic release, we included K252a, an inhibitor tion of a Trk-mediated signaling pathway21. In the absence of
of Trk receptor tyrosine kinases, during acute BDNF treatments. NGF, however, BDNF did not support the survival of cultured
BDNF continued to promote an inhibitory postsynaptic response sympathetic neurons (Fig. 4b). We therefore conclude that the
within 15 minutes in this condition (Fig. 4a). In previous studies5 actions of BDNF on sympathetic neurons in these cultures are
and in parallel experiments (Fig. 4a), K252a blocked NGF-medi- not mediated through a Trk-dependent pathway. Furthermore,
ated potentiation of excitatory synaptic transmission, revealing BDNF did not decrease the survival response of sympathetic
an NGF-dependent, stimulus-induced inhibition of myocyte beat neurons in the presence of NGF, indicating that the synaptic
rate. Thus, TrkA mediates NGF-dependent synaptic potentia- effects of BDNF are not mediated through cell death pathways
tion, and in the absence of Trk signaling a second NGF signaling associated with the p75 receptor22.
pathway seems to promote inhibitory transmission. Together In the absence of a Trk-mediated pathway, these data suggest
with the inability of K252a to block BDNF-dependent inhibito- a role for p75 in modulating cholinergic transmission. We there-
ry transmission, these data support a model for a non-Trk-medi- fore overexpressed human p75 specifically in sympathetic neu-
ated rapid change in cholinergic release and suggest a role for the rons in neuron–myocyte cocultures. Overexpression of p75
p75 receptor in the regulation of cholinergic transmission. protein was confirmed by immunostaining and by co-expression
Further evidence for a non–Trk-mediated pathway for cholin- of a yellow fluorescent protein (YFP) marker (Fig. 5a and b).
ergic regulation comes from BDNF survival experiments. NGF After 3 days, YFP controls showed excitatory neurotransmission

articles
Fig. 4. Trk-independent effects of BDNF. (a) Myocytes showed an

a increase in beat rate during neuronal stimulation in the absence or
presence of K252a (200 nM). After an excitatory neuron–myocyte
pair was found (Stim 1), either NGF (50 ng/ml) or BDNF (100 ng/ml)
was perfused into the bath in the presence of K252a for 15 min, and
the same neuron was stimulated again (Stim 2). With K252a in the
bath, both NGF and BDNF produced a net change to inhibitory
transmission upon repeated stimulation of the neuron (mean ±
s.e.m., n ≥ 6 for each condition; *, P < 0.02 as compared with
matched control). (b) BDNF neither supported the survival (in the
absence of NGF) nor triggered the death (in the presence of NGF)
of cultured sympathetic neurons. Survival is expressed as the per-
centage of neurons counted at plating that were alive at each time
point (mean ± s.e.m., n = 3).
cells continued to show an evoked increase in myocyte beat

b rate after acute BDNF treatment, whereas wild-type controls
showed inhibitory transmission after the same treatment.
These data conclusively identify p75 as a regulator of acetyl-
choline release from sympathetic neurons. Together with evi-
dence implicating TrkA in the regulation of noradrenergic
transmission5, these results suggest that multiple signaling
pathways control the pattern of neurotransmitter release in
developing sympathetic neurons.
A number of signaling pathways downstream of the p75
receptor have been identified, including the activation of
sphingomyelinase, which leads to the generation of
ceramide1. We investigated the p75-mediated ceramide path-
way by acutely treating neuron–myocyte cocultures for 15
minutes with C2-ceramide, a membrane-permeant ceramide
to connected myocytes (Fig. 5e). In contrast, p75-overexpress- analog that mimics the mobilization of ceramide following p75
ing neurons showed inhibitory transmission, even in the absence activation27. A 15-minute exposure to C2-ceramide resulted in a
of exogenous BDNF, indicating that p75 is involved in the regu- stimulus-evoked inhibition in myocyte beat rate (Fig. 6b). In
lation of cholinergic transmission. control experiments, neuronal stimulation elicited an increase
As cultures expressing human p75 contained only 5 ng/ml in myocyte beat rate after a 15-minute treatment with C2-dihy-
NGF, the switch to an inhibitory connection raised the possibil- droceramide, an inactive ceramide compound. The effect of the
ity of ligand-independent signaling through the overexpressed C2-ceramide treatment cannot be explained by TrkA activation
p75 receptor. Such ligand-independent signaling has been pre- of the ceramide pathway because ceramide-induced TrkA acti-
viously suggested for the intracellular domain of p75 (ref. 23). vation has not been observed within the 15-minute time frame
We therefore examined myocyte responses following the expres- used in our experiments28. Furthermore, the effects of the acute
sion of a mutant p75 (p75-105) deficient in ligand binding24,25 ceramide treatment recapitulated the timing of acute BDNF
(Fig. 5c and d). We examined eight connected neuron–myocyte application (Fig. 3a), consistent with a role for p75 in rapid
pairs expressing mutant p75 in the neuron. Four of the eight changes in synaptic release.
showed inhibitory transmission upon neuronal stimulation (aver-
age change in myocyte beat rate, –9.6 ± 1.4 beats/min), whereas DISCUSSION
the other four showed excitatory transmission (average change Sympathetic neurons modulate myocyte beat rate through activ-
in myocyte beat rate, 9.8 ± 2.5 beats/min). This resulted in an ity-evoked release of neurotransmitter. Here we show that cul-
average change in myocyte beat rate midway between excitation tured neonatal sympathetic neurons rapidly switched between
and inhibition (Fig. 5e). Post-hoc examination of p75-105 excitatory and inhibitory neurotransmission in response to neu-
immunostaining in two neurons, one showing an inhibitory rotrophins. This change in the functional output of sympathetic
response and the other an excitatory response, revealed similar- transmission reflected an increase in stimulus-induced cholin-
ly high levels of p75-105 expression. Thus, though p75-105 could ergic transmission that was mediated presynaptically through the
mediate a switch to inhibitory transmission, the effect was not p75 receptor. Activation of the p75 signaling pathway—by BDNF
as complete as that seen for the wild-type construct. This sug- treatment, p75 overexpression or activation of the ceramide sig-
gests that overexpressed p75 may have both ligand-dependent nal transduction pathway—resulted in increased cholinergic
and ligand-independent components. Furthermore, the bimodal transmission and activity-evoked inhibition of myocyte respons-
distribution of excitatory and inhibitory connections is consis- es. BDNF did not mediate a change to synaptic inhibition in cul-
tent with the idea suggested by the dose–response curve (Fig. 3c) tures derived from p75–/– mice, further implicating p75 in the
of a sharp transition between these two states. regulation of cholinergic transmission. These results suggest that
Further evidence for a regulatory role of p75 comes from cotransmission of norepinephrine and acetylcholine can be
analysis of the effect of BDNF in cultures derived from p75–/– dynamically regulated in developing sympathetic neurons by tar-
mice26. In these cultures, p75-deficient sympathetic neurons get-derived factors that modulate the release properties of dif-
formed excitatory connections to myocytes (Fig. 6a), and p75–/– ferent neurotransmitter pools.

articles
Fig. 5. Overexpression of the p75 receptor induced inhibitory neuro-

transmission. (a, b) A neuron cotransfected with YFP and human p75
a b
(white arrows) overexpressed the p75 protein as detected by p75
immunofluorescence. (a) YFP fluorescence of the transfected neuron;
(b) p75 immunoreactivity in the same neuron (white arrow). Non-
transfected neurons in the same field showed lower endogenous p75
expression (white arrowhead). Exposure time, 200 ms; scale bar, 20 µm.
(c, d) A neuron cotransfected with YFP and p75-105 (white arrows).
(c) YFP fluorescence; (d) p75 immunoreactivity. Non-transfected neu-
rons showing lower p75 immunoreactivity are marked with a white

arrowhead (exposure time, 500 ms). (e) Myocyte response to neuronal c d
stimulation was measured for YFP vector–transfected neurons
(Control), YFP-labeled neurons co-expressing wild-type human p75
(p75) or YFP-labeled neurons co-expressing the mutant human p75-105.
Mean ± s.e.m., n = 4 for control, n = 4 for p75, n = 8 for p75-105; *,
P < 0.001 compared with control.
TrkA and p75 activation

Although our data support a role for p75 in promoting cholin- e
ergic release, previous studies have implicated TrkA activation in
the potentiation of noradrenergic transmission5. How might acti-
vation of these two receptors lead to divergent downstream
effects? Trk and p75 signaling pathways are known to interact,
with p75 enhancing TrkA signaling when the two receptors are
co-activated29,30. In the absence of Trk signaling, p75 activates
cell death pathways31, an action which can be blocked by Trk acti-
vation32. Thus, it appears that p75 has the potential to regulate
events in both Trk-dependent and Trk-independent ways. The
recent observation that proneurotrophins (the initial secreted
form of neurotrophins) act as specific, high-affinity ligands for
p75 raises further possibilities of Trk-independent p75 activity
during the modulation of neuronal function33. Our experiments
indicate that p75 mediates specific cellular responses in the pres-
ence of low-level TrkA activation and that the relative ratio of
Trk and p75 signaling can affect the functional output of sym- 100 ng/ml BDNF), only a small decrease in neuronal number is
pathetic neurotransmission. This model is in accordance with seen22. Notably, more cell death occurred in BDNF-treated cul-
the finding that BDNF activation of p75 shifts the functional out- tures in the presence of KCl, a condition in which Trk receptors
put of sympathetic survival pathways even in the presence of low- are not activated22. The lack of such cell death in our cultures
level TrkA activation22. may also reflect differences in culture conditions, differences in
In our coculture system, BDNF treatment did not lead to a degree of p75 expression or a stronger co-activation of TrkA in
cell death response but rather to local changes in synaptic func- our cultures. Thus, within a range of TrkA and p75 signaling that
tion. BDNF signaling through p75 has previously been shown to supports neuronal survival, the balance of signaling through these
trigger the death of sympathetic neurons, but in a condition close- receptors regulates the neurotransmitter profile of developing
ly resembling that used in our experiments (5 ng/ml NGF and synaptic connections.
Fig. 6. Activation of p75 signaling pathways promotes inhibitory transmission. (a) After identifying a connected neuron–myocyte pair, the aver-
age response of a myocyte to neuronal stimulation was measured in cultures obtained from either p75+/+ or p75–/– mice (Con). BDNF was then
added to the bath for 15 min and the stimulus-induced
myocyte response measured again (BDNF). BDNF did not
promote cholinergic transmission in p75–/– mouse sympa-
thetic cocultures (p75+/+, n = 4; p75–/–, n = 6; *, P < 0.03
compared with control). (b) Acute application of C2-
ceramide resulted in a net change to inhibitory trans-
mission. Neuron–myocyte pairs showing an excitatory
connection were perfused with 20 µM C2-ceramide or
C2-dihydroceramide (control) for 15 min, then myocyte
response to neuronal stimulation was measured. In
some experiments, 1 µM atropine was included to
block cholinergic transmission. Mean ± s.e.m., n = 5 for
each condition. Responses of myocytes treated with
C2-ceramide were significantly different from those of
controls and those of myocytes treated with C2-
ceramide + atropine (P < 0.001).

articles
Regulation of co-release of multiple neurotransmitters taining 5 ng/ml 2.5S NGF (Upstate Biotech, Lake Placid, New York) as
A role for p75 has been established in the KCl-induced release of described5. After 1 day in culture, 1 µM cytosine arabinofuranoside
dopamine from rat mesencephalic neurons34, demonstrating that (AraC; Sigma, St. Louis, Missouri) was added to the dishes to stop cell
p75, as well as Trk receptors, can regulate neurotransmitter division. Under these conditions, AraC does not have an effect on the
survival or function of cultured sympathetic neurons49,50. Where indi-
release. In other experiments, NGF-induced release of glutamate
cated, the medium was supplemented either with 100 ng/ml BDNF (gift
from cerebellar granule cells was not blocked by K252a, suggest- from Regeneron Pharmaceuticals, Tarrytown, New York), 10 ng/ml CNTF
ing a possible further role for p75 in non-evoked release of neu- (R&D Systems, Minneapolis, Minnesota) or 50 ng/ml NGF (final con-
rotransmitter35. Here we show that p75 signaling modulated centration), or with 20 µM C2-ceramide or C2-dihydroceramide (Bio-
action potential–mediated synaptic release and, moreover, reg- mol, Plymouth Meeting, Pennsylvania).
ulated the differential release of distinct neurotransmitter pools For survival assays, neurons were plated at 10,000 cells per well in a
within individual sympathetic neurons. The regulation of 24-well dish. Neurotrophins were added at the time of plating. Percent
cotransmitter release by different target-derived factors may survival was calculated by comparing the number of surviving neurons at
increase the modulatory complexity of sympathetic neurons, 24, 48 and 72 h after plating with the number of neurons that were on
the dish 4–6 h after plating. Duplicate wells for each condition were
indicating that TrkA and p75 signaling helps establish local pat-
counted for a minimum of three experiments per condition.
terns of neurotransmitter release. Human p75 was expressed in cocultures using a Helios Gene Gun (Bio-
The observation that neurotrophin signaling through p75 Rad, Hercules, California). Cells were transfected 1 day after plating either
preferentially promoted the release of acetylcholine raises the with pJPA5-CD8-YFP or pNF314-CD8-GFP (control, gift from G. Banker),
question of how cholinergic and noradrenergic transmitter pools or with pJPA5-CD8-YFP and pCMV5A, an expression vector containing a
are segregated. In sympathetic neurons, acetylcholine is pack- human p75 cDNA (gift from M. Chao). A mutant p75 encoding a p75 pro-
aged in small synaptic vesicles, whereas norepinephrine can be tein deficient in ligand binding (p75-50, gift from M. Chao) was subcloned
found in both large dense-core vesicles and small synaptic vesi- into pCMV5 to generate p75-105. Expression of p75 was detected by stain-
cles14,36,37. It has been suggested that some vesicles might package ing with an anti-p75 antibody that recognized the intracellular domain of
both rat and human p75 protein (Promega, Madison, Wisconsin).
acetylcholine and norepinephrine together38, but analyses of the
localization of cholinergic and catecholaminergic transporters Measurement of functional postsynaptic response. Cultures were visu-
suggest that these transmitters are segregated to discrete pools of alized using an inverted Olympus IX70 microscope (Olympus, Melville,
small vesicles39. New York) with differential interference contrast (DIC) optics. Whole-cell
Experiments using latrotoxin to stimulate vesicle release sug- recordings were obtained using 3–4 MΩ patch electrodes as described5
gest that dense-core and small clear vesicles may be compart- from sympathetic neurons that appeared to be connected to a beating
mentalized differently within synaptic sites and have different myocyte. Synaptic transmission was examined by measuring the func-
calcium requirements for release40. Both Trk and p75 activation tional postsynaptic response of a myocyte to neuronal stimulation. This
are associated with influx of intracellular calcium41, although the was achieved by measuring the stimulus-induced change in myocyte beat
calcium pools mobilized differ for the two receptors42. Thus, the rate5. A baseline beat rate was counted for 4 min before the presynaptic
neuron was stimulated. Then the beat rate was counted for the duration
dynamics of calcium entry may be modulated as a consequence
of the 3-min stimulation. Further manipulations and analyses were car-
of differential Trk and p75 activation, leading to differential neu- ried out on neuron–myocyte pairs showing a functional connection5. In
rotransmitter release. Our observation that p75 signaling pro- some cultures, after the initial stimulation, BDNF or ceramide com-
moted the release of cholinergic vesicles is consistent with an pounds were perfused into the dish for 15 min and myocyte beat rate
effect on the mobilization of different vesicle pools. However, was measured during a second stimulation. BDNF was then washed out
further work is required to fully understand the mechanisms for 15 min and the functional assay repeated. Atropine (1 µM) or pro-
underlying this specificity. pranolol (10 µM) was included in the bath solution of some dishes to
These results also have general implications for understanding block cholinergic or noradrenergic transmission respectively.
the regulation of cotransmission in the nervous system. Although In some experiments, cells were continuously perfused with 200 nM
K252a (Kamiya Biomedical, Seattle, Washington) for the length of the
it is widely accepted that some CNS neurons can synthesize more
experiment. After a connected neuron–myocyte pair was found, 50 ng/ml
than one classical neurotransmitter43,44, current dogma holds NGF or 100 ng/ml BDNF was washed into the bath for 15 min, and the
that they secrete only a single neurotransmitter45. Arguing against stimulation protocol was repeated. Cultures used for K252a experiments
this view, recent studies demonstrate co-release of two classical were grown in either 5 ng/ml (acute BDNF experiments) or 50 ng/ml
neurotransmitters in a number of central systems46–48, suggesting NGF (acute NGF experiments). We had previously established that acute
that cotransmission may have a previously unrecognized role in modulation was not affected by the growth concentration of NGF5.
the regulation of complex circuitry in the nervous system. Our Neurotransmitter receptor agonists were applied to beating myocytes
work in the peripheral nervous system provides evidence for a using a Picospritzer II (General Valve Corp., East Hanover, New Jersey) and
mechanism through which postsynaptic factors could regulate standard patch pipettes. Myocytes were exposed to either 10 µM norepi-
the differential release of cotransmitters to increase the modula- nephrine (Research Biochemicals International, Natick, Massachusetts) or
25 µM muscarine chloride (Sigma, St. Louis, Missouri). The agonist solu-
tory complexity of neuronal circuits.
tions were puffed onto the myocytes at 10 p.s.i. for 25, 50 or 75 ms. Myocyte
beat rate was counted for 3 min before and 3 min after the application.
METHODS
Neonatal sympathetic neurons and ventricular myocytes. Cocultures of Statistics. Significance was analyzed by Student’s t-tests or ANOVA fol-
neonatal sympathetic neurons and cardiac ventricular myocytes were lowed by post-hoc tests using StatView (Abacus Concepts, Berkeley,
prepared and cultured essentially as described12 from neonatal Simon- California) software.
sen white rats (Simonsen Laboratories, Gilroy, California) and p75–/– or
wild-type mice (Jackson Laboratories, Bar Harbor, Maine). Animal pro-
tocols were approved by the Brandeis University Institutional Animal Acknowledgments
Care and Use Committee (IACUC). Freshly isolated sympathetic neu- We thank G. Turrigiano, L. Griffith, E. Marder and P. Sengupta for critical
rons (7,500–15,000 neurons) were plated with cardiac myocytes (50,000 reading of the manuscript, J. Hinterneder for helpful discussions, J. Mead and E.
myocytes) obtained from the same animals and cultured for 3–4 days or Nokes for technical assistance and G. Banker, M. Chao and B. Hempstead for
3 weeks before analysis. The cells were cultured in 2× MAH food con- help with reagents. This work was supported by grants from the US National

articles
Institutes of Health (R01 NS40168) and the Whitehall Foundation to S.J.B. The 24. Yan, H. & Chao, M. V. Disruption of cysteine-rich repeats of the p75 nerve
Pew Scholars Program in the Biomedical Sciences supported this work through a growth factor receptor leads to loss of ligand binding. J. Biol. Chem. 266,
12099–12104 (1991).
Pew Scholars Award to S.J.B. 25. Esposito, D. et al. The cytoplasmic and transmembrane domains of the p75
and Trk A receptors regulate high affinity binding to nerve growth factor.
Competing interests statement J. Biol. Chem. 276, 32687–32695 (2001).
26. Lee, F.-F. et al. Targeted mutation of the gene encoding the low affinity NGF
The authors declare that they have no competing financial interests. receptor p75 leads to deficits in the peripheral sensory nervous system. Cell
69, 737–749 (1992).
RECEIVED 11 MARCH; ACCEPTED 10 APRIL 2002 27. Dobrowsky, R. T., Werner, M. H., Castellino, A. M., Chao, M. V. & Hannun, Y.
A. Activation of the sphingomyelin cycle through the low affinity

neurotrophin receptor. Science 265, 1596–1599 (1994).
28. MacPhee, I. & Barker, P. A. Extended ceramide exposure activates the TrkA
1. Kaplan, D. R. & Miller, F. D. Neurotrophin signal transduction in the nervous
receptor by increasing receptor homodimer formation. J. Neurochem. 72,
system. Curr. Opin. Neurobiol. 10, 381–391 (2000).
1423–1430 (1999).
2. Huang, E. J. & Reichardt, L. F. Neurotrophins: roles in neuronal development
29. Verdi, J. M. et al. p75LNGFR regulates Trk signal transduction and NGF-
and function. Annu. Rev. Neurosci. 24, 677–736 (2001).
induced neuronal differentiation in MAH cells. Neuron 12, 733–745
3. McAllister, A. K., Katz, L. C. & Lo, D. C. Neurotrophins and synaptic
(1994).
plasticity. Annu. Rev. Neurosci. 22, 295–318 (1999).
30. Hantzopoulos, P. A., Suri, C., Glass, D. J., Goldfarb, M. P. & Yancopoulos, G.
4. Lohof, A. M., Ip, N. Y. & Poo, M. Potentiation of developing neuromuscular
D. The low affinity NGF receptor, p75, can collaborate with each of the trks to
synapses by the neurotrophins NT-3 and BDNF. Nature 363, 350–353 (1993).
potentiate functional responses to the neurotrophins. Neuron 13, 187–201
5. Lockhart, S. T., Turrigiano, G. G. & Birren, S. J. Nerve growth factor
(1994).
modulates synaptic transmission between sympathetic neurons and cardiac
31. Casaccia-Bonnefil, P., Carter, B. D., Dobrowsky, R. T. & Chao, M. V. Death of
myocytes. J. Neurosci. 17, 9573–9582 (1997).
oligodendrocytes mediated by the interaction of nerve growth factor with its
6. Gottschalk, W., Pozzo-Miller, L. D., Figurov, A. & Lu, B. Presynaptic
receptor p75. Nature 383, 716–719 (1996).
modulation of synaptic transmission and plasticity by brain-derived
32. Yoon, S. O., Casaccia-Bonnefil, P., Carter, B. & Chao, M. Competitive
neurotrophic factor in the developing hippocampus. J. Neurosci. 18,
signaling between TrkA and p75 nerve growth factor receptors determines
6830–6839 (1998).
cell survival. J. Neurosci. 18, 3273–3281 (1998).
7. Takei, N. et al. Brain-derived neurotrophic factor increases the stimulation-
33. Lee, R., Kermani, P., Teng, K. K. & Hempstead, B. L. Regulation of cell
evoked release of glutamate and the levels of exocytosis-associated proteins in
survival by secreted proneurotrophins. Science 294, 1945–1948 (2001).
cultured cortical neurons from embryonic rats. J. Neurochem. 68, 370–375
34. Blöchl, A. & Sirrenberg, C. Neurotrophins stimulate the release of dopamine
(1997).
from rat mesencephalic neurons via Trk and p75Lntr receptors. J. Biol. Chem.
8. Martínez, A. et al. TrkB and TrkC signaling are required for maturation and
271, 21100–21107 (1996).
synaptogenesis of hippocampal connections. J. Neurosci. 18, 7336–7350 (1998).
35. Numakawa, T., Takei, N., Yamagishi, S., Sakai, N. & Hatanaka, H.
9. Tyler, W. J. & Pozzo-Miller, L. D. BDNF enhances quantal neurotransmitter
Neurotrophin-elicited short-term glutamate release from cultured cerebellar
release and increases the number of docked vesicles at the active zones of
granule neurons. Brain Res. 842, 431–438 (1999).
hippocampal excitatory synapses. J. Neurosci. 21, 4249–4258 (2001).
36. De Potter, W. P., Partoens, P. & Strecker, S. Noradrenaline storing vesicles in
10. Collin, C. et al. Neurotrophins act at presynaptic terminals to activate
sympathetic neurons and their role in neurotransmitter release: an historical
synapses among cultured hippocampal neurons. Eur. J. Neurosci. 13,
overview of controversial issues. Neurochem. Res. 22, 911–919 (1997).
1273–1282 (2001).
37. Landis, S. C. Rat sympathetic neurons and cardiac myocytes developing in
11. Korsching, S. & Thoenen, H. Nerve growth factor in sympathetic ganglia and
microcultures: correlation of the fine structure of endings with
corresponding target organs of the rat: correlation with density of
neurotransmitter function in single neurons. Proc. Natl. Acad. Sci. USA 73,
sympathetic innervation. Proc. Natl. Acad. Sci. USA 80, 3513–3516 (1983).
4220–4224 (1976).
12. Lockhart, S. T., Mead, J. N., Pisano, J. M., Slonimsky, J. D. & Birren, S. J. Nerve
38. Johnson, M. I., Paik, K. & Higgins, D. Rapid changes in synaptic vesicle
growth factor collaborates with myocyte-derived factors to promote
cytochemistry after depolarization of cultured cholinergic sympathetic
development of presynaptic sites in cultured sympathetic neurons.
neurons. J. Cell Biol. 101, 217–226. (1985).
J. Neurobiol. 43, 460–476 (2000).
39. Weihe, E., Tao-Cheng, J. H., Schafer, M. K., Erickson, J. D. & Eiden, L. E.
13. Furshpan, E. J., Landis, S. C., Matsumoto, S. G. & Potter, D. D. Synaptic
Visualization of the vesicular acetylcholine transporter in cholinergic nerve
functions in rat sympathetic neurons in microcultures. I. Secretion of
terminals and its targeting to a specific population of small synaptic vesicles.
norepinephrine and acetylcholine. J. Neurosci. 6, 1061–1079 (1986).
Proc. Natl. Acad. Sci. USA 93, 3547–3552 (1996).
14. Potter, D. D., Landis, S. C., Matsumoto, S. G. & Furshpan, E. J. Synaptic
40. Sudhof, T.C. α-Latrotoxin and its receptors: neurexins and
functions in rat sympathetic neurons in microcultures. II.
CIRL/latrophilins. Annu. Rev. Neurosci. 24, 933–962 (2001).
Adrenergic/cholinergic dual status and plasticity. J. Neurosci. 6, 1080–1098
41. Stoop, R. & Poo, M. Synaptic modulation by neurotrophic factors:
(1986).
differential and synergistic effects of brain-derived neurotrophic factor and
15. Landis, S. C. & Keefe, D. Evidence for neurotransmitter plasticity in vivo:
ciliary neurotrophic factor. J. Neurosci. 16, 3256–3264 (1996).
developmental changes in properties of cholinergic sympathetic neurons.
42. Jiang, H. et al. Nerve growth factor (NGF)-induced calcium influx and
Dev. Biol. 98, 349–372 (1983).
16. Brodski, C., Schnürch, H. & Dechant, G. Neurotrophin-3 promotes the intracellular calcium mobilization in 3T3 cells expressing NGF receptors.
cholinergic differentiation of sympathetic neurons. Proc. Natl. Acad. Sci. USA J. Biol. Chem. 274, 26209–26216 (1999).
97, 9683–9688 (2000). 43. Sloviter, R.S. et al. Basal expression and induction of glutamate decarboxylase
17. Furshpan, E. J., MacLeish, P. R., O’Lague, P. H. & Potter, D. D. Chemical and GABA in excitatory granule cells of the rat and monkey hippocampal
transmission between rat sympathetic neurons and cardiac myocytes dentate gyrus. J. Comp. Neurol. 373, 593–618 (1996).
developing in microcultures: evidence for cholinergic, adrenergic, and dual- 44. Gonzalez-Hernandez, T., Barroso-Chinea, P., Acevedo, A., Salido, E. &
function neurons. Proc. Natl. Acad. Sci. USA 73, 4225–4229 (1976). Rodriguez, M. Co-localization of tyrosine hydroxylase and GAD65 mRNA in
18. Conforti, L., Tohse, N. & Sperelakis, N. Influence of sympathetic innervation mesostriatal neurons. Eur. J. Neurosci. 13, 57–67 (2001).
on the membrane electrical properties of neonatal rat cardiomyocytes in 45. Sulzer, D. & Rayport, S. Dale’s principle and glutamate co-release from
culture. J. Dev. Physiol. 15, 237–246 (1991). ventral midbrain dopamine neurons. Amino Acids 19, 45–52 (2000).
19. Saadat, S., Sendtner, M. & Rohrer, H. Ciliary neurotrophic factor induces 46. Jonas, P., Bischofberger, J. & Sandkühler, J. Co-release of two fast
cholinergic differentiation of rat sympathetic neurons in culture. J. Cell Biol. neurotransmitters at a central synapse. Science 281, 419–424 (1998).
108, 1807–1816 (1989). 47. Sulzer, D. et al. Dopamine neurons make glutamatergic synapses in vitro.
20. Fagan, A.M. et al. TrkA, but not TrkC, receptors are essential for survival of J. Neurosci. 18, 4588–4602 (1998).
sympathetic neurons in vivo. J. Neurosci. 16, 6208–6218 (1996). 48. Walker, M. C., Ruiz, A. & Kullmann, D. M. Monosynaptic GABAergic
21. Atwal, J. K., Massie, B., Miller, F. D. & Kaplan, D. R. The TrkB-Shc site signals signaling from dentate to CA3 with a pharmacological and physiological
neuronal survival and local axon growth via MEK and P13-kinase. Neuron profile typical of mossy fiber synapses. Neuron 29, 703–715 (2001).
27, 265–277 (2000). 49. Chun, L. L. Y. & Patterson, P. H. Role of nerve growth factor in the
22. Bamji, S. X. et al. The p75 neurotrophin receptor mediates neuronal development of rat sympathetic neurons in vitro III. Effect on acetylcholine
apoptosis and is essential for naturally occurring sympathetic neuron death. production. J. Cell Biol. 75, 712–718 (1977).
J. Cell Biol. 140, 911–923 (1998). 50. Martin, D. P., Wallace, T. L. & Johnson, E. M. Cytosine arabinoside kills
23. Majdan, M. et al. Transgenic mice expressing the intracellular domain of the postmitotic neurons in a fashion resembling trophic factor deprivation:
p75 neurotrophin receptor undergo neuronal apoptosis. J. Neurosci. 17, evidence that a deoxycytidine-dependent process may be required for nerve
6988–6998 (1997). growth factor signal transduction. J. Neurosci. 10, 184–193 (1990).

articles
Ethanol elicits and potentiates

nociceptor responses via the
vanilloid receptor-1
M. Trevisani1,*, D. Smart2,*, M. J. Gunthorpe2,*, M. Tognetto3, M. Barbieri1, B. Campi3,

S. Amadesi1, J. Gray2, J. C. Jerman4, S. J. Brough4, D. Owen2, G. D. Smith2, A. D. Randall2,
S. Harrison1, A. Bianchi3, J. B. Davis2 and P. Geppetti1
1 Department of Experimental and Clinical Medicine, Headache Center, University of Ferrara, Via Fossato di Mortara 19, 44100 Ferrara, Italy
2 Neurology-CEDD, GlaxoSmithKline, Third Avenue, Harlow CM19 5AW, UK
3 Department of Pharmacology, University of Catania, Viale Andrea Doria 86, 95123 Catania, Italy
4 Discovery Research, GlaxoSmithKline, Third Avenue, Harlow CM19 5AW, UK
* These authors contributed equally to this work.
Correspondence should be addressed to J.B.D. (John_B_Davis@gsk.com)
Published online: 29 April 2002 DOI: 10.1038/nn852
The vanilloid receptor-1 (VR1) is a heat-gated ion channel that is responsible for the burning
sensation elicited by capsaicin. A similar sensation is reported by patients with esophagitis when
they consume alcoholic beverages or are administered alcohol by injection as a medical treatment.
We report here that ethanol activates primary sensory neurons, resulting in neuropeptide release or
plasma extravasation in the esophagus, spinal cord or skin. Sensory neurons from trigeminal or dor-
sal root ganglia as well as VR1-expressing HEK293 cells responded to ethanol in a concentration-
dependent and capsazepine-sensitive fashion. Ethanol potentiated the response of VR1 to capsaicin,
protons and heat and lowered the threshold for heat activation of VR1 from ∼42ºC to ∼34ºC. This
provides a likely mechanistic explanation for the ethanol-induced sensory responses that occur at
body temperature and for the sensitivity of inflamed tissues to ethanol, such as might be found in
esophagitis, neuralgia or wounds.
The symptoms of patients suffering from persistent trigeminal to establish whether a VR1-mediated mechanism may provide an
neuralgia may be alleviated by the induction of peripheral nerve explanation for the painful burning sensations described above.
block with perineural injections of alcohol. This procedure is ini-
tially painful due to an intense burning sensation1. Similarly, RESULTS
patients with esophagitis provide anecdotal evidence that the con- Neuropeptide release and plasma extravasation
sumption of alcoholic beverages causes a burning pain propor- VR1 is expressed on a subset of peptidergic nociceptors that are
tional to the alcoholic strength of the drink. In addition, the able to signal via the release of neuropeptides. Consequently, the
burning pain produced by the application of alcoholic tinctures activation of VR1 leads to release of the neuropeptides substance-
to skin wounds is familiar to all. Common to all these phenome- P (SP) and calcitonin gene–related peptide (CGRP)8. To inves-
na is a “burning” sensation, which raises the possibility that tigate the events underlying the generation of “burning”
ethanol may produce activity in the nociceptors responsible for responses to ethanol, we first measured the release of SP from
sensing noxious heat. Consequently, we have studied the effect of dorsal spinal cord (Fig. 1a and b), esophagus (Fig. 1c and d) and
ethanol upon vanilloid receptor-1 (VR1)2,3, which has been iden- skin (Fig. 1e and f), where central and peripheral endings of pri-
tified previously as a receptor for noxious heat2. VR1—which is mary afferents terminate. Ethanol (0.1–3%, equivalent to
predominantly expressed in afferent Aδ- and C-fibers2—is a poly- 0.017–0.51 M) caused a concentration-dependent release of SP-
modal receptor4 that is activated by noxious heat (above ∼42ºC), like immunoreactivity (SP-LI) from the tissues, including skin
extracellular acidic pH and some lipids—for example, anan- (where exposure to similar concentrations of ethanol can be
damide5,6—or 12- or 15-HPETE7. It is plausible that irritants that expected during various treatments). Pretreatment with cap-
elicit a “burning” sensation may do so by activating VR1 or anoth- saicin or removal of extracellular Ca2+ ions practically abolished
er thermoreceptor. The studies we describe here, which used a the responses.
variety of established methods for studying sensory neurons— The inhibitory effect of capsaicin pretreatment indicated that
including the measurement of neuropeptide release, elevation of ethanol causes neuropeptide release from the central (dorsal
cytoplasmic Ca2+ concentrations in recombinant and native cell spinal cord) and peripheral (esophagus and skin) terminals of
systems, and electrophysiological recording—were done in order capsaicin-sensitive nociceptors in C- and Aδ-fibers, which are

articles
Fig. 1. Ethanol-induced neurotransmitter release and plasma extravasa-

tion. Release of SP-LI from (a) rat dorsal spinal cord, (c) esophagus and
a b (e) skin was induced by increasing concentrations of ethanol. The
effects of capsaicin (CAP) pretreatment (10 µM for 60 min before
ethanol addition), Ca2+-free medium plus 1 mM EGTA (–Ca2+) and 10
µM capsazepine (CZP) on the SP-LI release that was induced by 3%
ethanol from (b) rat dorsal spinal cord, (d) esophagus and (f) skin were
also assessed. Increases in plasma extravasation that were induced by
(g) intra-esophageal instillation of 50% ethanol or (h) intravenous SP
(0.1 nmol/kg) as well as the effect of 1 mg/kg of intravenous tachykinin

NK1 receptor antagonist SR140333 (SR) or intra-esophageal instillation
of 100 µM capsazepine were also analyzed. Data are mean ± s.e.m. of at
c d least five experiments; *, P < 0.05.
endothelial NK1 receptors. In the rat esophagus, a substantial

increase in Evans blue extravasation was caused by both ethanol
(100 µl, 50% ethanol intra-esophageal instillation) and SP (0.1
nmol/kg, intravenously). The NK 1 receptor antagonist
SR140333 significantly inhibited both responses, whereas cap-
sazepine blocked only the effect of ethanol (Fig. 1g and h). Plas-
ma extravasation (6.3 ± 1.1 ng/mg, n = 5) induced by
e f intra-esophageal capsaicin (100 µl, 1 mM) was also inhibited by
capsazepine (2.3 ± 0.4 ng/mg, n = 5, P < 0.01). These data indi-
cate that ethanol stimulated the release of tachykinins, and sub-
sequent plasma extravasation, via VR1 activation.
Responses of primary sensory neurons

The release of neuropeptides and the desensitizing effect of cap-
saicin strongly suggested that the effect of ethanol upon esopha-
gus, skin and dorsal spinal cord was mediated by peptidergic
nociceptors that express VR1. To further dissect these sensory
responses to ethanol, we studied primary neurons isolated from
trigeminal and dorsal root ganglia (DRG), cultured at 37°C, using
g h Ca2+ imaging (Fig. 2). In trigeminal ganglion neurons (TGNs),
ethanol caused a concentration-dependent increase in cytosolic
Ca2+ ion concentration ([Ca2+]i) (Fig. 2a), which was inhibited by
capsazepine (10 µM) (Fig. 2b and c). A similar but less intense
response to ethanol was observed in DRG neurons; this was also
inhibited by capsazepine (Table 1). Ninety-eight per cent of the
TGN and DRG cells grown in culture responded to capsaicin
(0.1–1 µM), which indicated that these were C- or Aδ-fiber noci-
ceptors8. All the cells that responded to ethanol also responded
the major known source of SP in sensory neurons. The role to capsaicin (data not shown). In contrast, a non-VR1-expressing
played by VR1 in this response was supported by the require- cell type, human hepatoma (Hep G2) cells, showed a small
ment for extracellular Ca2+ ions, which suggested the involve- increase in [Ca2+]i in response to ethanol; however, this minor
ment of a Ca2+-permeable ion channel, and by inhibition of the response was not affected by capsazepine (Table 1). A propor-
response by the VR1 antagonist capsazepine9 (10 µM). As well tion of the neuronal response to ethanol was capsazepine-insen-
as antagonizing VR110, capsazepine is reported to inhibit volt- sitive, which indicated that ethanol may exert some effects via
age-gated calcium channels. However, capsazepine had no effect additional, VR1-independent mechanisms. Ethanol also facili-
on KCl (80 mM)–mediated SP-LI release from the dorsal spinal tated VR1 responses to other agonists. Ethanol 1% (0.17 M)
cord (75 ± 7 fmol/g/20 min for capsazepine and 78 ± 8 fmol/g/20 increased the efficacy (Emax relative to control of 1.15 ± 0.03,
min for vehicle; in both cases, n = 5). These data indicated it was n = 4, P < 0.01), but not the potency (pEC50 values of 7.15 ± 0.06
unlikely that inhibition of voltage-gated channels contributed to versus 7.13 ± 0.05), of capsaicin-induced Ca2+ responses in DRG
the inhibitory effect of capsazepine observed in our experiments. cells (Fig. 2d).
In addition, ethanol did not appear to be neurotoxic: capsaicin
(1 µM) caused the release of SP-LI after pretreatment of dorsal Recombinant VR1 responses
spinal cord with 3% (0.51 M) ethanol (124 ± 10 fmol/g/20 min, The responses observed in isolated neurons confirmed that the
n = 5) or vehicle (145 ± 12 fmol/g/20 min, n = 5). Similar results effect of ethanol upon ex vivo tissue preparations was mediated
were obtained for ethanol-induced release of a second sensory via the neurons in those tissues. However, both the native sys-
neuropeptide, CGRP (data not shown). tems express a diverse complement of neuronal receptors that
In various rat tissues, including esophagus11, stimulation might contribute to an ethanol-evoked response. If VR1 plays a
of primary afferents and the subsequent release of SP causes primary role in the ethanol-mediated response, as is suggested
an increase in plasma extravasation via the activation of by inhibition of the response by low extracellular Ca2+ concen-

articles
Fig. 2. Ethanol stimulates and potentiates native and

a
recombinant VR1 responses. (a) Typical traces that
2+ c d
visualize [Ca ]i during exposure to ethanol and cap-
saicin, at 37ºC, in cultured TGNs that were pretreated
with (a) vehicle or (b) capsazepine. (c) Pooled data from
experiments similar to those shown in (a). Eighty neurons
were analyzed; *P < 0.01 versus vehicle. (d) Cytoplasmic
b
2+
[Ca ]i was monitored in DRG cells before and after the
addition of capsaicin (0.1 nM to 10 µM) in the absence
() or presence () of 1% ethanol. Responses were mea-

sured as peak increases in fluorescence minus basal e f g h
amounts; they are expressed relative to a 1 µM capsaicin
control (n = 4). (e) Ethanol induced a concentration-
dependent Ca2+ response. [Ca2+]i was monitored at
37ºC with Fura-2 in hVR1-expressing HEK293 (n = 86)
and wild-type (Wt) HEK293 (n = 148) cells. Responses
were measured relative to the peak response to 5 µM
ionomycin. Ethanol enhancement of Ca2+ response to
(f) capsaicin or (g) protons. [Ca2+]i was monitored with
Fluo-3 in hVR1-expressing HEK293 cells before and after
the addition of capsaicin (0.1 nM to 10 µM) or HCl
(0.4–1.5 mM) in the absence () or presence of 0.1% (), 0.3% () or 1% () ethanol. Responses were measured as peak increases in fluorescence
minus basal amounts and were expressed relative to 100 nM capsaicin (n = 6). (h) The ethanol-induced Ca2+ response is VR1-mediated. Fluo-
3–loaded hVR1-expressing HEK293 cells were preincubated for 30 min with capsazepine (CZP, 1 µM), ruthenium red (RR, 0.3 µM), Rp-cAMP (Ch.Cl,
10 µM), H89 (10 µM), BIM (1 µM), Ro318220 (10 µM) or chelerythrine chloride (10 µM). Rp-cAMP, H89, BIM, Ro318220 and Ch.Cl were added to
block kinase signaling pathways. Changes in [Ca2+]i were monitored before and after the addition of 3% ethanol. Responses were measured as peak
increases in fluorescence minus basal amounts and were expressed relative to a 100 nM capsaicin control (n = 3–8).
tration or capsazepine, then a VR1-dependent response to ethanol—were seen in VR1-expressing cells. The highest con-
ethanol should be obtained in a recombinant system that express- centration tested, 3% ethanol, caused a 165 ± 21% (n = 10)
es VR1. At 37°C, 0.1–3% (0.017–0.51 M) ethanol caused a increase in the capsaicin-gated current (Fig. 3b). Under these
marked concentration-dependent [Ca2+]i increase in human VR1 conditions, in which cells were maintained at room temperature,
(hVR1)-expressing HEK293 cells that was practically absent in ethanol caused no stimulation when applied alone (Fig. 3c). In
wild-type HEK293 cells (Fig. 2e). This showed that VR1 can be contrast, when the fluorimetric imaging plate reader (FLIPR)
activated by ethanol. The effect appeared to be relatively specific platform was used to follow intracellular Ca2+ fluxes, we found
to VR1, as ethanol failed to enhance the endogenous carbachol- that 1% (0.17 M) ethanol increased baseline [Ca2+]i slightly
induced muscarinic Ca2+ response in the same cells (data not (Fig. 2f) and that significant responses to ethanol occurred at
shown). Ethanol also enhanced the Ca2+ response to capsaicin (0.1 37°C (Fig. 2a and e).
nM to 10 µM) in a concentration-dependent manner (Fig. 2f); it These observations may be explained by the suggestion that
increased both the potency and efficacy of capsaicin. Ethanol ethanol is able to potentiate a wide range of VR1 activators,
0.3% (0.051 M) increased the potency of capsaicin from pEC50 including heat. Varying ambient temperature between experi-
8.22 ± 0.03 to 8.76 ± 0.05 (n = 4, P < 0.01) and the efficacy to mental systems might thus explain the changing degree of
1.18 ± 0.02 (P < 0.01). The modulatory effect of ethanol on hVR1 response seen to ethanol alone. Thus, we next established
responses was similarly observed in whole cell patch-clamp whether ethanol had similar effects upon activation of VR1 by
recordings done at 22–24°C (Fig. 3). In control experiments, cap- protons4, the endo-cannabinoid anandamide5,6 and heat (see
saicin alone and capsaicin that was applied with ethanol had no below). Ethanol potentiated the activation of VR1 by protons
effect on wild-type HEK293 cells (n = 7; Fig. 3a), whereas cap- in a concentration-dependent manner analogous to that seen
saicin-gated currents—which were markedly potentiated by with capsaicin and ethanol (Fig. 2g). Ethanol 1% increased the
pEC 50 for protons from 6.26 ± 0.02 to 6.50 ± 0.03
Table 1. TGN and DRG neuron responses to ethanol are inhibited (n = 5, P < 0.01) and the efficacy to 1.18 ± 0.02
by capsazepine. (P < 0.01). The potentiating effect of ethanol was
observed in whole-cell recordings. The application of
Ethanol DRG DRG TGN TGN Hep G2 Hep G2
concentration vehicle CPZ vehicle CPZ vehicle CPZ a pH 6 solution had little effect alone, but did activate
VR1 when 3% (0.51 M) ethanol was coadministered
(Fig. 3e). Similarly, 0.3% ethanol also enhanced the
0.05 M 5±1 2±1 a 8±2 1±1 a 7±2 5±2
(0.3%) anandamide (1 µM)-induced activation of VR1, as
measured by either FLIPR (12,801 ± 1,021 versus 4,625
0.17 M 17 ± 2 7±2 a 23 ± 2 9±2 a 12 ± 2 13 ± 2
(1%) ± 857 fluorescence intensity units (FIU), n = 3,
P < 0.01) or electrophysiology (Fig. 3f). VR1-dependent
0.51 M 32 ± 2 18 ± 2a 38 ± 3 21 ± 3a 17 ± 2 15 ± 3
(3%) effects should be antagonized by recognized VR1 antag-
onists. Both the ethanol-induced Ca 2+ responses
Capsaicin 56 ± 4 9±2 a 71 ± 4 2±2 a ND ND
(Fig. 2e) and current responses to ethanol and capsaicin
Data are shown as a percentage of the control ionomycin response. aP < 0.05 between (Fig. 3a and b) were antagonized by capsazepine (pIC50
cells treated with capsazepine (CPZ, 10 µM) and vehicle (0.01% ethanol) . ND, no data. 6.67 ± 0.02, n = 6) (Figs. 2b, c and 3g) and also by the

articles
a c e g heat-activated current was recruited at a

threshold of ∼42°C; this was accompa-
nied by a large increase in current noise,
which was attributable to the activation
of a channel of high single channel con-
ductance (Fig. 4b). This basal threshold
for VR1 activation at ∼42°C1,3 was con-
sistent with the activity of the endoge-
nous heat-receptor in DRG neurons16,17.

b d f h Ethanol 3% caused a large shift in the
threshold of VR1 activation, so that
channel activity attributable to VR1
was observed at temperatures as low
as 34°C. At higher temperatures,
much greater activation of VR1 was
always seen in the presence of ethanol
(Fig. 4b and c).
Fig. 3. Ethanol modulates capsaicin-, anandamide- or proton-gated inward currents recorded from DISCUSSION
hVR1-expressing HEK293 cells. (a) Sample traces that resulted from the application of 500 nM cap- Our data demonstrate the ethanol-
saicin alone (filled bar) or with 3% ethanol (open bar). (b) Percentage potentiation of the 500 nM cap- mediated modulation of VR1 function.
saicin response induced by 0.3–3% ethanol at 22–24ºC (n = 4–10). (c) At 22–24°C, hVR1-expressing Ethanol (0.1–3%) caused the potentia-
HEK293 cells responded to 1 µM capsaicin but not to 3% ethanol (n = 5). (d) Current-voltage relation of VR1 activity produced by vanil-
tionship for capsaicin-gated hVR1 currents and ethanol-potentiated, capsaicin-gated currents that loids, anandamide, protons or heat.
were assessed with a voltage-ramp protocol (–70 to +70 mV in 1 s). The effects of ethanol on the cur-
rent elicited by endogenous VR1 ligands were also examined. (e) Acid (pH 6) was applied alone or
These concentrations were lower than
with 3% ethanol and then blocked by 10 µM capsazepine (CZP); data from three samples were pooled. those found in the alcoholic beverages
(f) Anandamide (AEA, 1 µM) was applied alone or with 3% ethanol; data from five samples were or medications (up to 30% ethanol) to
pooled. The antagonistic effect of (g) 10 µM capsazepine (hatched bar, n = 4) and (h) 10 µM ruthe- which wounds or damaged tissues may
nium red (RR, shaded bar, n = 4) on 3% ethanol potentiated 500 nM capsaicin (filled bar) currents. be exposed18. Shifts in the heat-medi-
(a–h) The vertical current calibration bar corresponds to 200 pA. ated VR1-gating range were consistent
with the induction of VR1 activity at
normal body temperature. This pro-
non-competitive, but less selective, pore-blocker ruthenium vides further evidence that modulatory factors—for example
red (Figs. 2h and 3h). In addition, current–voltage relation- ethanol or bradykinin19—are capable of sensitizing VR1 so that
ships for currents elicited by capsaicin alone or capsaicin and body temperature is sufficient to activate VR1. The ability of
ethanol (Fig. 3d) showed pronounced outward rectification ethanol to affect C- and Aδ-fiber nociceptors is emphasized by
and reversal potentials that were close to 0 mV (–3.7 ± 1.1 mV, its capacity to release SP from the central and peripheral terminals
n = 5), which are characteristics of VR1-mediated conductance. of these neurons and to cause capsazepine-sensitive excitatory
VR1 accounts for the majority of the effect of ethanol on effects in isolated neurons from sensory ganglia. Such a sensiti-
hVR1-expressing HEK293 cells and at least part of ethanol- zation mechanism may contribute to the persistent pain that
evoked responses in primary neurons. Effects of alcohols on the results from inflammatory or neuropathic injury. A further impli-
activity of other ion channels, including nicotinic, GABA and cation of these results is that interpretations of the effects of
glycine receptor channels, have been described12. However, not ethanol on somatic and visceral tissues now require considera-
all membrane proteins are affected; for example, no effect of tion of the potential contributions of VR1. In addition, because
ethanol upon the activity of endogenous muscarinic receptors of the polymodal nature of VR1, these considerations must also
was found, as might have been expected if the effect were medi- take into account the local temperature, pH and presence of any
ated via non-specific effects upon membrane biophysics. There is endogenous activators of VR1. These factors may vary greatly
evidence for the existence of specific alcohol-binding sites13. between normal and damaged or inflamed tissues, whose
Ethanol has also been linked to protein kinase C (PKC) translo- pathologies engage VR1 function20,21.
cation14 and, as PKC activation leads to an increased probability Our data on the VR1-dependent release of neuropeptides
of VR1 opening15, we tested the effect of PKC inhibitors on the from esophagus—together with reports of the upregulation of
ethanol-induced signal (Fig. 2h). We found that these inhibitors VR1 expression in inflamed bowel22 and enhanced sensitivity to
had no effect, which suggests that PKC is not involved. capsaicin in modeled esophagitis23—highlight a possible role of
VR1 in visceral pain, where tissue injury has increased exposure
Effect upon thermal response threshold VR1 of sensory nerve endings. Our findings provide the following
VR1 is widely believed to be a physiological thermosensor capa- mechanistic explanation for the burning pain, described by
ble of converting noxious heat into the depolarization and firing patients, that ethanol evokes18: exposure to ethanol may suffi-
of sensory neurons. The effect of ambient temperature upon the ciently lower the threshold of VR1 receptors, which are recruited
basal response to ethanol alone is noted above. Thus, we deter- by inflammation, so that they become activated at body temper-
mined the effects of ethanol on the temperature responsiveness of ature or via the activators mentioned above19. These parallels
VR1 (Fig. 4). We found that ethanol had no effect on the small, between in vitro experiments and clinical symptoms encourage
heat-induced increases in leak currents in wild-type HEK293 the further study of VR1 for the development of treatments for
(Fig. 4a and c). In hVR1-expressing HEK293 cells, a well defined painful wounds and inflamed tissues.

articles
Fig. 4. Ethanol shifts the threshold for VR1 heat activation. Whole cell
inward currents were induced by a shift (gray bar) in temperature from
a
24ºC to the indicated temperatures in (a) wild-type HEK293 cells and
(b) hVR1-expressing HEK293 cells in the absence or presence of 3%
(0.51 M) ethanol (open bar). (c) Pooled data from experiments similar b
to those shown in (a) and (b) were used to define the temperature
response profile for heat-evoked currents in wild-type (squares) or
VR1-expressing (circles) cells either in the presence (filled symbols) or
absence (open symbols) of 3% ethanol. Asterisks indicate the lowest
temperature at which the VR1 heat-activated current—in the absence

(*) or presence (**) of ethanol—were significantly different (n = 7,
P < 0.05 by Student’s unpaired t-tests) to control recordings for
HEK293 cells.
METHODS c
Peptide-release assays. Thick slices (∼0.4 mm) of the dorsal horn of lum-
bar enlargements of the dorsal spinal cord, esophagus or skin from the
shaved dorsum of rats were stabilized for 60 min in Kreb’s solution at
37°C. Fractions (4 ml) were collected at 10-min intervals into ethanoic
acid (final concentration 2 N) before, during and after administration
of ethanol. Fractions were freeze-dried, reconstituted with assay buffer
and analyzed for CGRP and SP immunoreactivities as described24.
Plasma extravasation. Male Sprague-Dawley rats were anesthetized

(pentobarbital 60 mg/kg, intraperitoneally). A cannula was inserted
into the oral part of the esophagus and the aboral ending of the esoph-
agus was ligatured. One minute after injection of the dye Evans blue Tyrode buffer. A FLIPR was used to monitor cell fluorescence (λex = 488
(30 mg/kg) into the jugular vein, 50% ethanol (100 µl), its vehicle (100 nm, λem = 540 nm) before and after the addition of agonists. Responses
µl of 0.9% saline) or 1 mM capsaicin (100 µl in 5% ethanol) were given were measured as peak minus basal FIU and, where appropriate, were
through the esophageal cannula; alternatively, SP was administered expressed as a fraction of the maximum capsaicin-induced response.
intravenously. Vehicles did not cause any measurable plasma extrava- Curve-fitting and parameter estimation were done with Graph Pad Prism
sation (data not shown). Capsazepine, administered via the esophageal 3.00 (GraphPad Software, San Diego, California).
cannula, intravenous SR140333 or a combination of their vehicles (1%
intra-esophageal ethanol and 10% intravenous DMSO, respectively) Whole cell patch-clamp electrophysiology. All recordings were made
were administered 15 min before the dye was injected. Rats were killed with standard whole-cell patch-clamp methods, as described3; they
5 min after SP or ethanol were administered and the esophagus was were done at room temperature (22–24ºC) unless otherwise stated. The
removed, weighed and incubated in formamide for 24 h at 37°C. extracellular solution consisted of 130 mM NaCl, 5 mM KCl, 2 mM
Extravasated Evans blue was measured spectrophotometrically at 620 CaCl2, 1 mM MgCl2, 30 mM glucose and 25 mM HEPES-NaOH at pH
nm. Ethical approval for animal work was obtained from the local Uni- 7.3. Patch pipettes (resistance 2–5 MΩ) were filled with 140 mM CsCl,
versity of Ferrara ethics committee. 4 mM MgCl 2, 10 mM EGTA and 10 mM HEPES-CsOH at pH 7.3.
Drugs were applied with an automated perfusion device (time for solu-
Cell culture. DRG and TGNs were taken from 1–3-day-old rats and cul- tion exchange ∼30 ms). Temperature jump experiments were done as
tured as described20,24. The human hepatoma cell line (a gift of F. Bernar- described3. Data are presented as mean ± s.e.m.
di, University of Ferrara) was grown in Dulbecco’s modified Eagle’s Ethanol (0.01%), at a concentration that did not produce any
medium/F12. detectable release of neuropeptides or mobilization of calcium (data
not shown), or DMSO (0.1%) vehicles were used for the delivery of
Measurement of changes in [Ca2+]i. [Ca2+]i was determined with the drugs in in vitro studies. Statistical comparisons were made where
Ca2+ -sensitive fluorescent dye Fura-2, which was loaded into Hep G2 appropriate with Student’s t-tests or the ANOVA and Dunnett’s test.
cells or sensory neurons as its acetoxymethylester form (40 min, 37°C).
The bath solution consisted of 1.4 mM CaCl2, 5.4 mM KCl, 0.4 mM
MgSO4, 135 mM NaCl, 5 mM D-glucose, 10 mM HEPES and 0.1% bovine Acknowledgments
serum albumin at pH 7.4. Alternate excitation at 340 nM and 380 nM was We thank C. Farrant and S. Lomax for preparation of artwork. This work was
supplied and the F340/F380 emission ratio recorded with a dynamic image supported in part by ARCA (Padua) and Cofin (MIUIR, Rome).
analysis system (Laboratory Automation 2.0, RCS, Florence, Italy). Ratio
changes were expressed as a percentage of the peak response to iono- Competing interests statement
mycin (5 µM)24. After 10 min of stabilization, responses to increasing The authors declare that they have no competing financial interests.
concentrations of ethanol (0.1–3% or 0.017 – 0.51 M) and capsaicin (0.1
µM) were recorded in the presence of capsazepine (10 µM) or its vehicle. RECEIVED 26 NOVEMBER 2001; ACCEPTED 25 MARCH 2002.
In some experiments, plated neurons were pre-exposed to capsaicin (10
µM) for 60 min to desensitize them. [Ca2+]i was also monitored with
FLIPR (Molecular Devices, Wokingham, UK), a high-throughput plat- 1. Fardy, M. J. & Patton, D. W. Complications associated with peripheral alcohol
injections in the management of trigeminal neuralgia. Br. J. Oral Maxillofac.
form for the fluorimetric measurement of intracellular Ca2+ concentra- Surg. 32, 387–391 (1994).
tions, as described6. Briefly, hVR1-expressing HEK293 cells loaded with 2. Caterina, M. J. et al. The capsaicin receptor: a heat activated ion channel in
Fluo-3 (4 µM; Teflabs, Austin, USA) were incubated at 25°C for 2 h, while the pain pathway. Nature 389, 816–824 (1997).
dissociated rat DRG cells that had been seeded into 384 plates (10,000 3. Hayes, P. et al. Cloning and functional expression of a human orthologue of
rat vanilloid receptor-1. Pain 88, 205–215 (2000).
cells per well) were incubated with Fluo-4 (2 µM) at 37°C for 1 h. Mod- 4. Tominaga, M. et al. The cloned capsaicin receptor integrates multiple pain-
ulating agents were preincubated with the cells for 30 min at 25°C in producing stimuli. Neuron 21, 531–543 (1998).

articles
5. Zygmunt, P. M. et al. Vanilloid receptors on sensory nerves mediate the 15. Premkumar, L. S. & Ahern, G. P. Induction of vanilloid receptor channel
vasodillator action of anandamide. Nature 400, 452–457 (1999). activity by protein kinase C. Nature 408, 985–990 (2000).
6. Smart, D. et al. The endogenous lipid anandamide is a full agonist at the 16. Cesare, P. & McNaughton, P. A novel heat-activated current in nociceptive
human vanilloid receptor (hVR1). Br. J. Pharmacol. 129, 227–230 (2000). neurons and its sensitization by bradykinin. Proc. Natl. Acad. Sci. USA 93,
7. Hwang S. W. et al. Direct activation of capsaicin receptors by products of 15435–15439 (1996).
lipoxygenases: endogenous capsaicin-like substances. Proc. Natl. Acad. Sci. 17. Nagy, I. & Rang, H. P. Similarities and differences between the responses of
USA 97, 6155–6160 (2000). rat sensory neurons to noxious heat and capsaicin. J. Neurosci. 19,
8. Holzer, P. Capsaicin: cellular targets, mechanisms of action, and selectivity 10647–10655 (1999).
for thin sensory neurons. Pharmacol. Rev. 43, 143–201 (1991). 18. Bolanowski, S. J., Gescheider, G. A. & Sutton S. V. Relationship between oral
9. Bevan, S. et al. Capsazepine: a competitive antagonist of the sensory neurone pain and ethanol concentration in mouthrinses. J. Periodontal. Res. 30,
192–197 (1995).
excitant capsaicin. Br. J. Pharmacol. 107, 544–552 (1992).

10. Docherty, R. J., Yates, J. C. & Piper, A. S. Capsazepine block of voltage- 19. Liang, Y.-F., Haake, B. & Reeh, P. W. Sustained sensitization and recruitment
of rat cutaneous nociceptors by bradykinin and a novel theory of its
activated calcium channels in adult rat dorsal root ganglion neurones in
excitatory action. J. Physiol. 532, 229–239 (2001).
culture. Br. J. Pharmacol. 121, 1461–1467 (1997). 20. Caterina, M. J. et al. Impaired nociception and pain sensation in mice lacking
11. Saria, A. et al. Vascular protein leakage in various tissues induced by the capsaicin receptor. Science 288, 306–313 (2000).
substance P, capsaicin, bradykinin, serotonin, histamine and antigen 21. Davis, J. B. et al. Vanilloid receptor-1 is essential for inflammatory thermal
challenge. Naunyn Schmiedebergs Arch. Pharmacol. 324, 212–218 (1983). hyperalgesia. Nature 405, 183–187 (2000).
12. Forman, S. A. & Zhou, Q. Novel modulation of a nicotinic receptor channel 22. Yiangou, Y. et al. Vanilloid receptor 1 immunoreactivity in inflamed bowel.
mutant reveals that the open state is stabilized by ethanol. Mol. Pharmacol. Lancet 357, 1338–1339 (2001).
55, 102–108 (1999). 23. Smid, S. D. et al. Oesophagitis-induced changes in capsaicin-sensitive
13. Wick, M. J. et al. Mutations of γ-aminobutyric acid and glycine receptors tachykininergic pathways in the ferret lower oesophageal sphincter.
change alcohol cutoff: evidence for an alcohol receptor? Proc. Natl. Acad. Sci. Neurogastroenterol. Motil. 10, 403–411 (1998).
USA 95, 6504–6509 (1998). 24. Tognetto, M. et al. Anandamide excites central terminals of dorsal root
14. Stubbs, C. D. & Slater, S. J. Ethanol and protein kinase C. Alcohol Clin. Exp. ganglion neurons via vanilloid receptor-1 (VR-1) activation. J. Neurosci. 15,
Res. 23, 1552–1560 (1999). 1104–1109 (2001).

articles
A computational role for slow

conductances: single-neuron
models that measure duration
Scott L. Hooper, Einat Buchman and Kevin H. Hobbs
Neuroscience Program, Department of Biological Sciences, Ohio University, Athens, Ohio 45701 USA
Correspondence should be addressed to S.L.H. (hooper@ohio.edu)
Published online: 29 April 2002, DOI: 10.1038/nn838
Humans effortlessly interpret speech and music, whose patterns can contain sound durations up to
thousands of milliseconds. How nervous systems measure such long durations is unclear. We show here
that model neurons containing physiological slow conductances are ‘naturally’ sensitive to duration,
replicate known duration-sensitive neurons and can be ‘tuned’ to respond to a wide range of specific
durations. In addition, these models reproduce several other properties of duration-sensitive neurons
not selected for in model construction. These data, and the widespread presence of slow conductances
in nervous systems, suggest that slow conductances might play a major role in duration measurement.
To understand the problems the nervous system faces in analyzing We present a hypothesis that obviates the need for delayed
speech and music, consider the song ‘Y.M.C.A.’ (Fig. 1). Two mea- synaptic excitation. We propose that duration-sensitive neurons
sures of the song’s chorus are shown both in musical notation possess currents that slowly change during sound-induced inhi-
(line 1) and in an alternative representation in which rectangles bition, and the neurons therefore fire only after certain sound dura-
denote the time and duration of each note (lines 2–4). The song’s tions. Models that reproduce the three types of duration-sensitive
beat, a repeating series of bass drum (~90 Hz) eighth notes, is neurons can be constructed using physiological slow conductances.
shown only in the rectangle representation (line 5); this is the Furthermore, although this was not a goal of our model con-
rhythm people dance to. Almost all the song’s information is construction, these models also reproduce several other properties of
tained in the non-repetitive sequence of long-duration sounds (in duration-sensitive neurons not selected for in model construction.
the original recorded version of ‘Y.M.C.A’, eighth notes last 250,
quarter notes 500 and half notes 1,000 ms) that comprises the RESULTS
song’s melody (lines 2–4). This property—that most information To test the ability of slow conductances to produce duration-sen-
is carried in complex sequences of long-duration sounds—is pre- sitive neurons, we created single-compartment models of neurons
sent in many auditory signals, and especially in speech and music. having INa and IKD and (depending on the model) the transient K
Extracting information from these signals requires measuring long current IA, the hyperpolarization-activated, depolarizing current
durations; in the studies described here we investigated the neur- Ih and the low-threshold Ca current IT. The inhibition duration-
al mechanism underlying this ability. sensitive neurons receive during sound6 was modeled with a
Neurons that fire after sounds of specific durations are presynaptic current open throughout sound duration.
sent in amphibian midbrain1,2 and mammalian inferior collicu- Interaction of IKD and IA can create low-pass models (Fig. 2a).
lus, thalamus and cortex3–5. Three types of duration-sensitive When the model was hyperpolarized for 2, 125, 250, 375, 500,
neuron are known: low-pass (which fire after durations shorter 625, 750, 875 and 1,000 ms, it fired after all hyperpolarizations
than a certain value), high-pass (which fire after durations longer ≤500 ms; in the auditory system it would fire for all sounds ≤500
than a certain value) and band-pass neurons (which fire only ms. Model duration sensitivity arises as follows. At rest, sufficient
after durations within a narrow range)3–5. These neurons respond IKD is open to prevent firing, and IA is almost completely inacti-
to durations of 2–75 ms in bat3 and up to 200 ms in cat5. Whether vated. Hyperpolarization quickly closes IKD but only slowly
similar neurons are present in humans is unknown, but inter- removes IA inactivation (at –80 mV, the time constant of IA inac-
preting speech and music would require neurons sensitive to tivation removal is 500 ms). After short hyperpolarizations, little
durations up to thousands of milliseconds. Sound inhibits dura- IA is available to open, and sufficient IKD closes to induce post-
tion-sensitive neurons6, and duration sensitivity has been pro- inhibitory rebound and firing. As duration increases, more IA
posed to arise because the coincidence of post-inhibitory rebound inactivation is removed (bottom trace) and eventually IA blocks
and delayed excitatory synaptic input drives the neurons above firing. Models tuned to durations from 2 to 1,800 ms, which span
spike threshold3,6–9. For short sounds (<50 ms), axonal and the durations in speech and music, can be constructed using
synaptic delays could produce the delays in excitatory input this physiological IA parameters10–14.
hypothesis requires, but these mechanisms are inadequate to pro- Ih can create high-pass models (Fig. 2b). When this model
duce the delays required to measure longer durations6. was hyperpolarized for the same durations as in Figure 2a, it fired

articles
iological IA and Ih parameters. Models tuned to shorter durations

can be constructed using faster Ih-like currents20 or delay line
mechanisms, which are biologically possible for short durations.
Duration-sensitive neurons can fire short spike bursts, and
slow conductance–based low-, high- and band-pass bursting
models can be constructed (Fig. 2d–f). We obtained a low-pass
burst model (Fig. 2d) by slightly (≤10%) hyperpolarizing the
voltage dependence of the opening and closing rates of IKD (ref.
21). In the absence of IA, any duration of hyperpolarization

Fig. 1. ‘Y.M.C.A’ temporal pattern. Line 1 shows the sound pattern in
musical notation. The pattern consists of five alternations of B and D induces the model to fire a short spike burst at the end. Varying
eighth notes, a quarter note of D, a half note of E, a quarter note of D IA amplitude tunes the model to fire only for durations shorter
and two E and D eighth notes (B above middle C, 494 Hz; D, 587.3 Hz; than a certain value. Bursting in the other models (Fig. 2e, f) was
E, 659.25 Hz). Lines 2–4 are an alternate representation in which the supported by IT, from which hyperpolarization removes inacti-
rectangles represent note duration. Line 5 is the song’s beat (eighth vation. Tuning in the high-pass model was achieved with an IA
notes separated by eighth note rests) in the rectangle representation. whose inactivation removal was faster than that for IT but which,
‘Y.M.C.A’ was composed by J. Morali, H. Belolo and V. Willis and per- when IA and IT were fully activated, could not prevent IT from
formed by the Village People. inducing a burst. IA thus blocks firing after short hyperpolariza-
tions but not long ones. The band-pass model was tuned by
adding a second, slower IA that blocked firing after long hyper-
after all durations ≥500 ms. In this model little IKD is open at polarizations; the two IAs constrained firing to a short duration
rest, and there is therefore no post-inhibitory rebound due to range. Altering conductance amplitude and kinetics changes burst
IKD closing. Instead Ih supports post-inhibitory rebound, and spike number and model tuning; the models can be physiologi-
the model fires only after durations long enough to activate suf- cally tuned through the same ranges as single-spike models.
ficient Ih to drive the model above threshold. Ih is present in many All three models fired fewer spikes as they approach their firing
vertebrate neurons15, including those of the auditory system16,17. limit(s); this decline also occurs in bursting duration–sensitive neu-
Models tuned to durations from 15 to >5,000 ms can be con- rons4,5. In slow-conductance models, this decline is a ‘natural’ con-
structed using physiological Ih parameters10,11,18,19. sequence of the mechanisms underlying model tuning. For instance,
IA and Ih can create band-pass models (Fig. 2c). When this in the bursting low-pass model, the growth of IA eventually blocked
model was hyperpolarized for the same durations as in Figure 2a, firing; this same growth, by partially offsetting the post-inhibitory
and also for 480 and 522 ms, it fired only for durations narrowly rebound, also resulted in the model firing fewer spikes at interme-
centered around 500 ms. The model contains IA and Ih currents diate durations. Similar IA-mediated decreases in post-inhibitory
with the same kinetics as in Figure 2a and b. Ih increases with rebound occur in dorsal cochlear nucleus pyramidal neurons22.
duration, and eventually the model fires. As duration increases Tones are often repeated with brief inter-tone (silent) intervals.
further, enough IA inactivation is removed that IA overcomes Ih To respond correctly to repeated tones, duration-sensitive neurons
and the model no longer reaches threshold. Models tuned to cen- require inter-tone intervals of 30–200 ms23; the models in Figure 2
ter durations from 50 to 3,000 ms can be constructed using phys- required intervals of 60–400 ms. However, the currents in these
models all had the same kinetics (only conductance amplitude
d was varied). When conductance kinetics was also varied (but
a always in the physiological range), the models responded cor-
rectly to repeated tones with inter-tone intervals of 10–60 ms.
In human perception, the accuracy of duration measurement
decreases linearly with duration, and is ±5% of center dura-
tion24. That is, durations between 475 and 525 ms (range 50 ms)
cannot be distinguished from 500 ms (center duration), and
durations between 950 and 1,050 ms (range 100 ms) cannot be
distinguished from 1,000 ms (center duration). This is also seen
b e in band-pass neurons: neurons tuned to short durations fire over
a narrower duration range than neurons tuned to long dura-
tions8. Response range similarly increased with tuned duration
in both our single-spike (Fig. 3a–c; when the model was tuned to
0.25, 0.5 and 1 s, its response ranges were 19, 42 and 81 ms,
respectively) and bursting (Fig. 3d and e) band-pass models.
Fig. 2. Models that reproduce low-, high- and band-pass neurons.

c f Left column, single-spike models; right column, bursting models. In all
panels, top traces are model responses to hyperpolarizations of vari-
ous durations. (a, d) Low-pass models, with firing for all durations
≤500 ms (b, e) High-pass models, with firing for all durations ≥500
ms; (c, f), band-pass models, with firing for a narrow range of dura-
tions centered around 500 ms. Lower traces in (a) and (b) are fraction
of available IA and open Ih in response to sustained hyperpolarization;
gray rectangles are values for which the model can fire. Plots in (d), (e)
and (f) show spike number as a function of duration.

articles
Fig. 3. Band-pass model response range

a d increases with tuned duration. Single-spike
models: (a) tuned to 0.25 s, response range
0.241–0.26 s; (b) tuned to 0.5 s, response
range 0.480–0.522 s; (c) tuned to 1 s, response
range 0.966–1.047 sec. (d, e) Similar data for
bursting models (for clarity, only the ranges
b e over which the models fire two spikes are
shown; one-spike ranges are slightly larger).
Dashed lines are response ranges of 0.25 s sin-

gle-spike (a–c) and 0.5 s burst (d, e) models.
The response ranges become larger (f) because
the curves governing current expression flatten
as duration increases. Curves represent avail-
c f able Ih fraction during a 1.03 s hyperpolariza-
tion for 0.25 s and 1 s single-spike models.
Insets are expansions of the curves near the
fraction where the models can fire; insets have
equal available Ih fraction (∆ = 0.0231) and time
(∆ = 66 ms) ranges. The horizontal gray rectan-
gles show the fraction range in which each
model can fire; the vertical gray rectangles show
the corresponding time range. Although the
0.25 s model can fire over a much wider available Ih range, because of the steepness of its curve in this region, this corresponds to a much shorter
time range. The two curves do not achieve the same fraction at long times because, as a result of the different IA and Ih maximal conductances in
them, their voltages differ in the later portions of the hyperpolarizations.
This increased response range ‘naturally’ arises in slow-con- clamp cannot be obtained in these neurons3,7 (E. Covey, personal
ductance models. The models fire when their currents simulta- communication). This problem can be resolved by hyperpolarizing
neously have values within certain windows. Tuning is achieved duration-sensitive neurons to test whether they show rebound fir-
by adjusting conductance maxima until, at the tuned duration, ing and whether spike number varies with hyperpolarization dura-
all currents have values within these windows. The fraction of tion. To our knowledge, such experiments have not been reported,
available or open current rises as a function of 1 – et/c. As the mod- but about one-third of the neurons in rat inferior colliculus slices
els are tuned to longer durations, their firing occurs further in rebound and spike after square-wave hyperpolarization26.
flatter regions of these curves, and a given current window cor- Third, consistent with the slow-conductance hypothesis,
responds to longer time windows. Thus, during a sustained hyper- which depends on hyperpolarization, inhibitory transmitter
polarization, the 0.25 s single-spike band-pass model fired for antagonists abolish duration sensitivity9. Because the delayed
fractions of available Ih between 0.2602 and 0.2756, whereas the synaptic excitation hypothesis also requires inhibition, these data
1 s single-spike band-pass model fired for fractions between do not distinguish between the hypotheses. An experiment that
0.5291and 0.5298 (Fig. 3f). Although the I h fraction range over would do so would be one blocking excitatory synaptic input.
which firing can occur was 22-fold greater in the 0.25 s model, However, as a multisynaptic pathway mediates sound-evoked
because of the steepness of its curve in this region, the duration responses6, this test may be impossible to perform. In addition,
range to which this Ih fraction range corresponded was much nar- the slow-conductance and synaptic hypotheses are not mutually
rower in the 0.25 s model than in the 1 s model. exclusive; both mechanisms could be present and have comple-
mentary roles in duration sensing, with slow conductance mech-
DISCUSSION anisms becoming increasingly important as duration increases.
It is important to compare our models to experimental data. First, Fourth, auditory neurons have currents similar to those used
sounds occur at different intensities, and the tuning of duration- here. Some inferior colliculus neurons contain IKD currents with
sensitive neurons can vary or not with changes in sound inten- altered activation-voltage dependencies27, as in our single-spike
sity4,5,25. Because slow-conductance activation and inactivation and burst low-pass models. A slow Ih is present in cochlear nucle-
are voltage dependent, changing hyperpolarization amplitude us octopus cells17, and slow IA and IT are present in some inferi-
changes the tuning of slow-conductance models. Slow-conduc- or colliculus neurons; the latter neurons show post-inhibitory
tance models can therefore reproduce, depending on the nature rebound firing27. Unfortunately, in the inferior colliculus work,
of their synaptic input, both sound intensity–independent and duration sensitivity could not be examined, and direct compar-
–dependent neurons. Intensity-independent models have synap- ison with our models is thus impossible. Nonetheless, these data
tic input that does not vary with intensity, or a synapse so strong show that the fundamental building blocks of our models are
that it reaches reversal potential for all inputs. Intensity-depen- present in auditory neurons.
dent models have synaptic input that varies with intensity, and Slow-conductance band-pass neurons can analyze the
a synapse weak enough that postsynaptic hyperpolarization varies ‘Y.M.C.A.’ melody (Fig. 4). For the note B (494 Hz; Fig. 4a), the
with changes in presynaptic activity. 0.25 s neuron fires at 0.25, 0.75 and 1.25 s, and thereby signals
Second, auditory midbrain neurons display duration tuning that immediately before these times a 494 Hz eighth note occurred
under voltage-clamp recording conditions3,7. Because our mod- (because there are no 494 Hz quarter or half notes, the 0.5 s and
els rely on voltage-dependent conductances, these data appear to 1 s neurons do not fire). For the note D (587.3 Hz; Fig. 4b), fir-
contradict the slow-conductance hypothesis. However, good space ing of the 0.25 s or 0.5 s neurons signals when 587.3 Hz eighth or

articles
Fig. 4. Analysis of ‘Y.M.C.A.’ using single-spike slow-conductance band-

a pass neurons. Bottom trace (in each panel), sound pattern of note; next
trace up, firing of a cochlear neuron tuned to that note; three upper
traces, activities of 1, 0.5 and 0.25 s band-pass neurons inhibited when
tones of 494 (a), 587.3 (b), and 659.25 (c) Hz are played. In (a), firing of
0.25 s neuron signals that 494 Hz (B in musical notation) eighth notes
occurred immediately before 0.25, 0.75 and 1.25 s. In (b), firing of 0.25 s
neuron signals that 587.3 Hz (D) eighth notes occurred immediately
before 0.5, 1 and 3.75 s; firing of 0.5 s neuron signals that 587.3 Hz quarter
notes occurred immediately before 1.75 and 3.25 s. In (c), firing of 0.25 s
neuron signals that a 659.25 Hz (E) eighth note occurred immediately
before 3.5 s; firing of 1 s neuron signals that a 659.25 Hz half note occurred
b immediately before 2.75 s. Connections between cochlear and band-pass
neurons are dashed because these connections are multisynaptic6.
were multiplied by 3(37–23.5)/10 = 3.54). When necessary for tuning, con-

ductance kinetics was adjusted by multiplying da/dt and db/dt by con-
stants (all changes were in the physiological range). For INa and IKD, a∞
= α /(α + β) and τa = 1/(α + β), where α and β are membrane voltage
(Vm)–dependent opening and closing rates35. IA, I h and I T a∞, b∞, τa and
c τb were given by equations of the form:
a∞ = 1/(1 + e(Vm –V1/2)/S) and τa = c/(e(Vm–V1/2τ1)/Sτ1 + e(Vm–V1/2τ2)/S τ2)36.
The currents were calculated from I =ga n 1b n 2(V m – E) or I

=Pa n 1bn 2 G(Vm,[Cao],[Cai]), whereg is maximum conductance, E
reversal potential,P maximum permeability, [Cao] and [Cai] external
and internal Ca concentration and G the constant field equation. Vm was
given by dVm/dt = [injected current – Σ(membrane and synaptic cur-
rents)]/Cm, where Cm is membrane capacitance (0.29 nF in all simula-
tions). Models were implemented in ModelMaker (FamilyGenetix,
quarter notes sound. For the note E (659.25 Hz; Fig. 4c), firing of Oxford, UK; fourth-order Runge-Kutta integration, accuracy 0.0001,
the 0.25 s or 1 s neurons signals when 659.25 Hz eighth or half error 0.001) or Simulink/Matlab (Mathworks, Natick, Massachusetts;
notes sound. The model thus transforms the ‘Y.M.C.A.’ melody ODE45 Dormand-Prince integration, relative tolerance 10−4; absolute
tolerance 10−5).
into a duration-based place map in which the firing of specific
Model equations (except for I T, τ is in ms, Vm in mV, I in nA andg
neurons signals the time and duration of each note.
in µS) were as follows:
The brain’s analysis of the beat line (Fig. 1) remains to be
explained. The beat conveys only repeat period (the song’s tempo) Isyn +gsyn (Vm + 85);gsyn = 0.2 during sound
and relative duration (that beat duration is, in this case, 50% of and 0 otherwise
repeat period—eighth notes separated by eighth rests). One pos-
sibility is that higher centers calculate these measures (by Ileak =g leakK . (Vm + 105) +g leakNa. (Vm – 45)
unknown mechanisms) using input from beat-excited duration-
sensitive neurons. We have shown elsewhere, however, that neu- INa = g Na . m3 . h . (Vm – 45)
rons with slow membrane properties vary their firing as a
function of beat parameters and can identify all repeating on/off α m = 0.091 . (Vm + 38)/(1 – e –(Vm + 38)/5);
patterns such as beat lines28, and that slow-conductance models
reproduce these data (E.B., J.B. Thuma, A.L.Weaver, and S.L.H., β m = –0.062 . (Vm + 38)/(1 – e(Vm +38)/5)
Soc. Neurosci. Abstr. 26, 748.9, 2000).
These data and those presented here thus show that, in the- α h = 0.016 . e–(Vm + 55)/15; β h = 2.07/(1 + e–(Vm – 17)/21)
ory, all temporal aspects of music, speech and other, similar pat-
IKD =gKD . n4 . (Vm + 105)
terns can be analyzed by slow-conductance neurons. That
individual cells can ‘measure’ temporal patterns is further sup- α n = 0.01 . (Vm + 45)/(1 – e–(Vm + V 1/2KDα)/5);
ported by work showing pattern-specific variation in second-
messenger concentration, protein and gene expression, β n = 0.17 . e–(Vm + V 1/2KD β )/40
membrane conductance levels, firing delay and peptide secretion
in isolated cells or cell groups driven by varying temporal Single-spike models, V 1/ = –45, V 1/ = –50
input29–34. Taken together, this work suggests that individual neu- 2KDα 2KD β
rons can perform relatively complicated temporal analyses, and Burst models, V 1/ = –47.5, V 1/ = –55
2KD α 2KD β
that slow conductances may be centrally involved in this process.
Ih =gh . ah . (Vm + 30)
METHODS
Conductance activation a (or inactivation, b) was given by da/dt = (a∞ a∞h = 1/(1 + e(Vm + 75)/5.5);
– a)/τa, where a∞ is steady-state activation and τa is the activation time
constant. da/dt and db/dt were temperature compensated using Q10 = 3 τ ah = 270/(e–(Vm + 90)/12.5 + e(Vm +75)/5)
(data were acquired at 23.5°C and models run at 37°C, so da/dt and db/dt

articles
frog’s thalamus: processing of pulse duration and pulse repetition rate.

IA =gA . aA3bA . (Vm + 105) Neurosci. Lett. 63, 215–220 (1996).
3. Casseday, J. H., Ehrlich, D. & Covey E. Neural tuning for sound duration: role of
a∞A = 1/(1 + e–(Vm – V1/2IAa∞)/20); inhibitory mechanisms in the inferior colliculus. Science 264, 847–850 (1994).
4. Galazyuk, A. V. & Feng, A. S. Encoding of sound duration by neurons in the
auditory cortex of the little brown bat, Myotis lucifugus. J. Comp. Physiol. A
τ aA = 0.37 + 1/(e(Vm + 35.8)/19.7 + e–(Vm + 79.7)/12.7) 180, 301–311 (1997).
5. He, J., Hashikawa, T., Ojima, H. & Kinouchi, Y. Temporal integration and
Single-spike models, V1/ = –26 duration tuning in the dorsal zone of cat auditory cortex. J. Neurosci. 17,
2IAa∞ 2615–2625 (1997).
6. Covey, E. & Casseday, J. H. Timing in the auditory system of the bat. Annu.
Burst models, V1/ = –36.
2IAa∞ Rev. Physiol. 61, 457–476 (1999).

7. Covey, E., Kauer, J. A. & Casseday, J. H. Whole-cell patch-clamp recording
b∞A = 1/(1 + e(Vm + 68)/sIAb∞); reveals subthreshold sound–evoked postsynaptic currents in the inferior
colliculus of awake bats. J. Neurosci. 16, 3009–3018 (1996).
8. Ehrlich, D., Casseday, J. H. & Covey, E. Neural tuning to sound duration in
Vm ≥ –80
τ bA =
{ 20
500/(e(Vm + 60)/2 + e–(Vm + 80)/20) Vm < –60

;
the inferior colliculus of the big brown bat, Eptesicus fuscus. J. Neurophysiol.
77, 2360–2372 (1997).
9. Casseday, J. H., Ehrlich, D. & Covey, E. Neural measurement of sound
duration: control by excitatory inhibitory interactions in the inferior
colliculus. J. Neurophysiol. 84, 1475–1487 (2000).
Single-spike models, SIAb∞ = 6; burst models, SIAb∞ = 4. 10. Golowasch, J. & Marder, E. Ionic currents of the lateral pyloric neuron of the
stomatogastric ganglion of the crab. J. Neurophysiol. 67, 318–331 (1992).
11. Huguenard, J. R. & McCormick, D. A. Simulation of the currents involved in
IA1, τa and τb were multiplied by 3 in the burst low- and band-pass rhythmic oscillations in thalamic relay neurons. J. Neurophysiol. 68,
models and by 1.67 in burst high-pass models. IA2, τa and τb were multi- 1373–1383 (1992).
12. Lopez-Lopez, J. R., DeLuis, D. A. & Gonzalez, C. Properties of a transient K+
plied by 10 in the burst band-pass models. current in chemoreceptor cells of rabbit carotid body. J. Physiol. 460, 15–32
(1993).
13. Wang, W. Y., McKenzie, J. S. & Kemm, R. E. Whole-cell K+ currents in identified
 
/ 
–zFVm –zFVm
IT =PCa . aT2 . bT . z F Vm .  [Cai] – [Cao] e
 olfactory bulb output neurones of rats. J. Physiol. 490, 63–77 (1996).
1 – e
2 2
RT RT
  14. Buckingham, S. D. & Spencer, A. N. K+ currents in cultured neurones from a
RT polyclad flatworm. J. Exp. Biol. 203, 3189–3198 (2000).
15. Pape, H. C. Queer current and pacemaker: the hyperpolarization-activated
a∞T = 1/(1 + e–6.2); cation current in neurons. Annu. Rev. Physiol. 58, 299–327 (1996).
16. Chen, C. Hyperpolarization-activated current (Ih) in primary auditory
neurons. Hear. Res. 110, 179–190 (1997).
τ aT = 0.612 + 1/(e (Vm + 13.6)/–16.7+ e(Vm + 16.8)/18.2) 17. Bal, R. & Oertel D. Hyperpolarization-activated, mixed-cation current (Ih) in
octopus cells of the mammalian cochlear nucleus. J. Neurophysiol. 84,
b∞T = 1/(1 + e(Vm + 84.5)/4.03); 806–817 (2000).
18. Hutcheon, B., Miura, R. M. & Puil, E. Models of subthreshold membrane
resonance in neocortical neurons. J. Neurophysiol. 76, 698–714 (1996).
τbT =
{ e(Vm + 467)/66.6
28 + e(Vm + 21.88)/–10.52
Vm < –80
Vm ≥ –80
19. Angstadt, J. D. Persistent inward currents in cultured Retzius cells of the
medicinal leech. J. Comp. Physiol. A 184, 49–61 (1999).
20. Kumamota, E. & Shinnick-Gallagher, P. Slow inward and late slow outward
currents induced by hyperpolarizing pre-pulses in cat bladder
parasympathetic neurones. Eur. J. Physiol. 416, 322–334 (1990).
21. Izhikevich, E. M. Neural excitability, spiking, and bursting. Int. J. Bifurcat.
In the I T equation,P is in nanoliters/s, z is valence (2), F is Faraday’s Chaos 10, 1171–1266 (2000).
constant (9.6 × 104 coulomb (C)/mol), Vm is in volts, R is the gas con- 22. Kanold, P. O. & Manis, P. B. Transient potassium currents regulate the
stant (8.31 J/K mol), T is temperature in K, [Cai] and [Cao] are in mol/liter discharge patterns of dorsal cochlear nucleus pyramidal cells. J. Neurosci. 19,
2195–2208 (1999).
(M), and therefore I is in nA. Vm is in mV in the a∞T, τaT, b∞T and τbT 23. Wu, M. & Jen, P. H.-S. Recovery cycles of neurons in the inferior colliculus,
equations. Both I T τ values were multiplied by 1.2. I T changed [Cai] as the pontine nuclei and the auditory cortex of the big brown bat, Eptesicus
follows: 1 nA = (10−9 C/s) × (10−3 s/ms) × (1 molcharge/9.6 × 104 C) × (1 fuscus. Chin. J. Physiol. 41, 1–8 (1998).
24. Westheimer, G. Discrimination of short time intervals by the human
molCa/2 molcharge) = 5.18 × 10−18 molCa/ms. Neurons had a membrane observer. Exp. Brain Res. 129, 121–126 (1999).
area of 2.9 × 10−4 cm2. Assuming that Ca variation was limited to a 100 25. Brand, A., Urban, A. & Grothe, B. Duration tuning in the mouse auditory
nm shell under the membrane, shell volume was 2.9 × 10−4 cm2 × 100 nm midbrain. J. Neurophysiol. 84, 1790–1799 (2000).
× (10−7 cm/nm) × (10−3 liter/cm3) = 2.9 × 10−12 liter (shell width is small 26. Peruzzi, D., Sivaramakrishnan, S. & Oliver, D. L. Identification of cell types in
brain slices of the inferior colliculus. Neurosci. 101, 403–416 (2000).
enough that the fact that the cell is spherical can be ignored). Therefore, 27. Sivaramakrishnan, S. & Oliver, D. L. Distinct K currents result in
1 nA = (5.18 × 10−18 molCa/ms)/2.9 × 10−12 liter = 1.79 × 10−6 M/ms. [Cai] physiologically distinct cell types in the inferior colliculus of the rat.
was calculated from d[Cai]/dt = IT × (1.79 × 10−6 M/ms nA) + ([Cai∞] – J. Neurosci. 21, 2861–2877 (2001).
28. Hooper, S. L. Transduction of temporal patterns by single neurons. Nature
[Cai])/τ Cai, where [Cai∞] = 3 × 10−8 M and τ Cai = 2 ms. Model maximum Neurosci. 1, 720–726 (1998).
conductances are given in Supplementary Table 1 online. 29. Turrigiano, G., Abbott, L. F. & Marder, E. Activity changes the intrinsic
properties of cultured neurons. Science 264, 974–976 (1994).
30. Itoh, K. Stevens, B., Schachner, M. & Fields, R. D. Regulated expression of the
Note: Supplementary information is available on the Nature Neuroscience website. neural cell adhesion molecule L1 by specific patterns of neural impulses.
Science 270, 1369–1372 (1995).
Acknowledgments 31. Fields, R. D., Eshete, F., Stevens, B. K. & Itoh, K. Action potential-dependent
regulation of gene expression: temporal specificity in Ca+2, cAMP-responsive
We thank E. Covey for helpful discussions. This work was supported by the element binding proteins, and mitogen-activated protein kinase signaling.
Neuroscience Program at Ohio University and the US National Institutes of J. Neurosci. 17, 7252–7266 (1997).
32. Dolmetsch, R. E., Xu, K. & Lewis, R. S. Calcium oscillations increase the
Mental Health. efficiency and specificity of gene expression. Nature 392, 933–936 (1998).
33. Li, W., Llopis, J., Whitney, M., Zlokarnik, G., & Tsien, R. Y. Cell-permeant
Competing interests statement caged InsP3 ester shows that Ca+2 spike frequency can optimise gene
expression. Nature 392, 936–941 (1998).
The authors declare that they have no competing financial interests. 34. Muschol, M. & Salzberg, B. M. Dependence of transient and residual calcium
dynamics on action-potential patterning during neuropeptide secretion.
RECEIVED 12 FEBRUARY; ACCEPTED 14 MARCH 2002 J. Neurosci. 20, 6773–6780 (2000).
35. Hodgkin, A. L. & Huxley, A. F. A quantitative description of membrane
current and its application to conduction and excitation in nerve. J. Physiol.
Lond. 117, 500–544 (1952).
1. Potter, H. D. Patterns of acoustically evoked discharges of neurons in the 36. Huguenard, J. R. & McCormick, D. A. A model of the electrophysiological
mesencephalon of the bullfrog. J. Neurophysiol. 28, 1155–1184 (1965). properties of thalamocortical relay neurons. J. Neurophysiol. 68, 1384–1400
2. Hall, J. & Feng, A. S. Neural analysis of temporally patterned sounds in the (1992).

articles
Local inhibition modulates odor-

evoked synchronization of
glomerulus-specific output neurons
Hong Lei, Thomas A. Christensen and John G. Hildebrand
Arizona Research Laboratories, Division of Neurobiology, University of Arizona, PO Box 210077, Tucson, Arizona 85721-0077, USA
Correspondence should be addressed to T.A.C. (tc@neurobio.arizona.edu)
Published online: 13 May 2002, DOI:10.1038/nn859
At the first stage of olfactory processing in the brain, synchronous firing across glomeruli may help to
temporally bind multiple and spatially distributed input streams activated by a given odor. This
hypothesis, however, has never been tested in an organism in which the odor-tuning properties of
several spatially identifiable glomeruli are known. Using the sphinx moth, an insect that meets these
specific criteria, we recorded odor-evoked responses simultaneously from pairs of projection neurons
(PNs) innervating the same or different glomeruli in the macroglomerular complex (MGC), which is
involved in processing pheromonal information. PNs that branched in the same glomerulus and were
activated by the same pheromone component also showed the strongest coincident responses to
each odor pulse. Glomerulus-specific PN pairs were also inhibited by the pheromone component that
selectively activated PNs in the neighboring glomerulus, and about 70% of all intraglomerular pairs
showed increased synchronization when stimulated with a mixture of the two odorants. Thus, when
two adjacent glomeruli receive their inputs simultaneously, the temporal tuning of output from each
glomerulus is enhanced by reciprocal and inhibitory interglomerular interactions.
Temporal summation of synaptic inputs can greatly increase the acid) contributes to the precision of spike timing in olfactory PNs
probability of activating target neurons in networks that process in both vertebrates and invertebrates20–22, and both fast and slow
sensory information1,2. Studies in the mammalian visual and inhibitory postsynaptic potentials (IPSPs) from GABAergic local
somatosensory systems have shown that synchronized spiking across interneurons are prominent components of PN responses in
a distributed ensemble of neurons may also serve to temporally diverse animal species (reviewed in ref. 13). In the vertebrate
bind the different features of a complex stimulus, thus reinforcing olfactory bulb, for example, dendrodendritic reciprocal synaps-
the population code that is read postsynaptically3,4. Likewise, in the es between mitral/tufted (M/T) cells and granule cells provide a
olfactory systems of both vertebrates and invertebrates, temporal basis for recurrent inhibition of M/T cell activity18,20. Numerous
patterning is centrally involved in encoding chemosensory infor- studies have also suggested that the tuning specificity of a
mation5–9. In some cases, network oscillations modulate the pat- glomerulus (its ‘odor contrast’ in analogy to retinal processing
terns of synchronized spiking activity evoked by odors6–9, whereas for vision) may be enhanced further by lateral inhibition medi-
in others, the timing of odor-evoked synchrony is neither oscilla- ated by reciprocal synaptic connections between glomeruli18,23–26.
tory nor odor specific10–12. Thus the functional importance of pre- This idea, however, remains controversial on the grounds that it
cise timing relationships between the different neural elements that is not yet known which features of an olfactory stimulus provide
encode olfactory information is not known for any organism. the local contrast27.
In the insect antennal lobe, the structural and functional ana- In a recent study, simultaneous extracellular recordings
log of the vertebrate olfactory bulb, the first synapses occur in from pairs of M/T cells presumed to innervate different
anatomically discrete and functionally distinguishable modules glomeruli showed synchronized firing in response to some
called glomeruli (reviewed in ref. 13). A glomerulus is the site of olfactory stimuli9. In that study, pairs of M/T cells innervat-
convergence of numerous olfactory receptor cells (ORCs) expressing the same glomerulus were not examined, and the full range
ing one or a few types of odorant receptors14,15, and each recep- of odorants represented in these glomeruli was not known9. A
tor probably recognizes only a limited number of physical rigorous test for odor-evoked synchrony across olfactory
determinants associated with a given odor molecule16–18. Dif- glomeruli requires that the tuning characteristics of several
ferent odorants thus activate different subsets of glomeruli, pro- readily identifiable glomeruli be well defined. Here we used
ducing a specific spatial representation of each odorant at the intracellular methods to record simultaneously from pairs of
earliest stages of processing in the brain13,18,19. Information about PNs in the MGC of the male sphinx moth—an insect olfacto-
the specific chemical and spatiotemporal features of the odor ry system that meets these criteria. PNs that innervated the
stimulus is then transmitted in multiple, parallel channels to same glomerulus and responded selectively to the same odor-
higher centers through glomerular PNs. GABA (γ-aminobutyric ant (a component of the sex pheromone) showed greater syn-

articles
Table 1. Physiological features of paired MGC-PN recordings

Responses to olfactory stimuli were obtained in a total of 75 experiments involving simultaneous intracellular recordings from
pairs of MGC-PNs. In 34 pairs, both neurons showed primarily excitatory responses to only one odorant; cumulus neurons
responded to C15; toroid neurons responded to BAL. Pheromone-responsive pairs were then classified according to their
response specificity (C15/C15, BAL/BAL or C15/BAL). Each pair of neurons was then scored for four traits associated with
synchronous firing, as measured by the correlation coefficient between PNs. Expression of odor-evoked synchrony (% synchronous
events relative to total spikes) is indicated by the filled triangles (large, 75–100%; medium, 50–74%; small, 15–49%). Expression of
the remaining traits (% increase in correlation) is indicated by the filled circles (large, >100%; medium, 50–100%; small, 5–49%).
None of the pairs showed oscillatory synchrony; six pairs showed all other traits (indicated by asterisks). None of the C15/BAL
pairs showed blend-dependent effects on synchrony. n, not tested.
chronization of spike discharges as compared to PNs that RESULTS

innervated neighboring glomeruli and fired in response to dif- Male Manduca sexta moths are attracted to a blend of two relat-
ferent pheromone components. Furthermore, we found that ed odorants (BAL and C15; see Methods) that represent key com-
odor-evoked synchrony between PNs from one MGC glomeru- ponents of the female sex pheromone 11,22. Each odorant is
lus was augmented by local inhibitory input from the neigh- detected by a different population of ORCs, and the filtered infor-
boring MGC glomerulus encoding the other key component mation is then relayed to two principal glomeruli in the MGC,
of the pheromone. These findings support a crucial role for where the first steps of synaptic processing take place28–30. Our
interglomerular interactions in shaping the glomerular repre- results are based on olfactory responses recorded simultaneous-
sentations for odors in the brain. Although these lateral synap- ly from 34 pairs of MGC-PNs in as many moths (Table 1). In 26
tic interactions serve to enhance the contrast between odor pairs, spiking activity in both PNs was triggered selectively by the
signals, the underlying mechanism differs from ‘classical’ lateral same odorant (Fig. 1): the primary excitatory stimulus for these
inhibition as seen in other sensory systems. PNs was either C15 (n = 11 pairs, Fig. 1a and b, left) or BAL (n =
Fig. 1. PNs innervating the same glomerulus responded

selectively to the same odorant and showed stimulus-
synchronized responses to repetitive stimulation. a
(a) Innervation patterns in the antennal lobe corresponding
to each of the different pairs of PNs analyzed. The most eas-
ily identifiable glomeruli of the MGC are the cloud-shaped
cumulus and the ring-shaped toroid, and MGC-PNs in this
study innervated only one of these two structures. MGC-
PNs in M. sexta do not innervate any of the other, sexually
isomorphic glomeruli in the antennal lobe. (b) Intracellular b
responses to a train of five brief (200-ms) pulses of the single
pheromone components C15 (10 ng), BAL (10 ng) or the
mixture of the two odorants (blend; 10 ng each). Three pairs
of PNs are shown: two cumulus PNs selective for C15 (left),
two toroid PNs selective for BAL (center) and one PN of
each type (right). Note that PNs were depolarized by one
odorant and hyperpolarized by the other. (c) Detail of the
opposing responses to the two odorants. The two traces in
each column show responses recorded from a single neuron c
in each pair. In each case, one odorant depolarized the PN,
whereas the other evoked membrane hyperpolarization
(asterisks) and/or stopped spike activity completely.
d
(d) Cross correlations between three pairs of PNs that
innervated the same or different glomeruli (average of five
responses to the pheromone blend). The correlograms
showed greater odor-evoked synchrony between PNs when
both innervated the cumulus (left; r = 0.60; n = 11) or the
toroid (center; r = 0.51; n = 15), but significantly less
co-activity (r = 0.27) between interglomerular pairs (right;
P = 0.02, Kruskal-Wallis ANOVA).

articles
Fig. 2. Time-series analysis of responses to a

repeated odor pulses recorded simultaneously
d
PN1 5
from pairs of PNs. (a) Raster plots show the 10 ng BAL
Synchronized spike/ bin

PN2 4
responses of two toroid PNs to five consecutive
pulses of BAL. Lower raster series shows only the PN1 x PN2
3
synchronous events between the two neurons
(PN 1 × PN 2). A 5-ms bin was used, reflecting 1s
2
Single-pulse trials (10 ng C15)
instantaneous spike frequencies in the 200 Hz b #1 1
1 ng BAL
range. For some pulses, spontaneous spiking #2

#3
0
resumed in the two PNs before the onset of the
Counts
0 0.1 0 .2 0.3 0 .4
2
subsequent odor pulse (time course of stimulus 1 ng BAL Time (s)
1
train is indicated beneath rasters). The transient 0 PN 1
window of synchrony between PNs was in every 3
Trains of pulses
Counts
case tightly modulated by stimulus dynamics, and 2 PN 2
the duration of each window matched the dura- 1
10 ng BAL
tion of the odor pulse. (b) Time course of 0
PN 1
odorant-evoked synchrony between two cumulus 0 120 160 200 240 280 320
Counts
PNs selective for C15. Raster plots depict the 4
PN 2
2
occurrence of PN synchrony (top) in the
responses to three consecutive odor pulses, and
0 40 80 120 160
post-stimulus time histograms (PSTHs; 5-ms bins) c Response time (ms)
f
Synchronous spikes
summarize the results across trials. The PSTHs 100 * * * Cumulus

illustrate the absence of periodicity in synchro- 80
(% total)
nous events whether odor pulses were separated 60
by 1 min (top) or 1 s (middle and bottom). No 40 Toroid

20 c/c t/t c/t
periodicity was seen whether data were aligned 0
with respect to odor onset (middle) or to the first 1st 2nd 3rd 4th 5th pulse
synchronous event in each pair of responses (bot- e PN-PN synchrony (%) Latency, onset to peak (ms)
tom). (c) The occurrence of synchrony remained 100 e d 30 d d c/c
t/t
bc cd c/t
consistent across repeated odor pulses separated 80 c ILP
ab 20 ab
by 1 s, but in accordance with Fig. 1, the percent- 60
a
b
a
age of synchronous spikes between PNs (relative 40
10 CMB
to total) was greater for intraglomerular PN pairs 20
(c/c or t/t) than for interglomerular pairs (c/t) (c, 0

1 10 1 10 1 10
0
1 10 1 10 1 10
cumulus; t, toroid). All values are mean ± s.d.; Dosage (ng) Dosage (ng)
within each pulse, a significant difference between
two means is indicated by an asterisk (Kruskall-
Wallis test; P < 0.05; n = 11, 15 and 8, for cumulus, toroid and mixed pairs, respectively). (d) Odorant-evoked synchrony is modulated by stimulus
concentration. PSTHs show the averaged responses to five consecutive 150-ms odor pulses (5-ms bins) and illustrate the time course of synchronous
firing between a pair of toroid PNs in response to stimulation with BAL at a dosage of 1 ng (shaded) or 10 ng (black). Stimulation with the higher con-
centration resulted in both an increased frequency and reduced latency of synchronous events. Intracellular records show the responses from both
PNs to a single pulse consisting of 1 ng (top traces) or 10 ng (bottom traces) of BAL. Stimulus intensity and the calculated correlation coefficients
between PNs were positively correlated (r = 0.35 and 0.48 for the 1- and 10-ng stimuli, respectively). Small tick marks in each trace reflect cross-talk
between the two recording electrodes. Stimulus time course is indicated by the solid bar beneath the traces. (e) Summary of dose-dependent effects
on synchrony between PNs in each group. Half of all PN/PN pairs in this study were tested with two concentrations of the pheromone blend (1- and
10-ng dosages). Left, synchrony was expressed as the percentage of synchronous events relative to the total spike counts in each pair of PNs. A sta-
tistically significant effect of concentration was seen in both intraglomerular PN/PN groups (cumulus pairs, n = 7; toroid pairs, n = 8) as well as in the
interglomerular pairs (n = 2) (Kruskal-Wallis test followed by Mann-Whitney U-test; P < 0.05). Right, dose-dependent effect on the latency from stim-
ulus onset to peak of synchronous firing between PNs. Again, the higher concentration shifted this latency to significantly lower values for both intra-
glomerular groups of PNs, as well as for the interglomerular group (P < 0.05). All values are mean ± s.d. Means shown with the same letter are not
significantly different (P > 0.05). (f) Confocal-microscopic montage showing the anatomy of a pair of PNs from the interglomerular group, recorded
simultaneously in the moth antennal lobe. Both PNs were stained with LY (see Methods). The uniglomerular arborizations of each neuron are located
in different glomeruli: one innervates the cumulus (the C15 neuron, larger soma) and the other has branches confined to the toroid (the BAL neu-
ron, smaller soma). As found in previous studies, each PN gave rise to a single axon that projected to the ipsilateral inferior lateral protocerebrum
(ILP). Each axon also extended sparse collateral branches into the ipsilateral calyces of the mushroom bodies (CMB).
15 pairs, Fig. 1a and b, center). In eight additional PN pairs, the inhibitory responses to C15; that is, they were hyperpolarized by
primary stimulus that triggered spiking was different for each the odorant representing the primary excitatory input to the adja-
neuron (Fig. 1a and b, right). cent glomerulus. Similarly, all cumulus neurons were hyperpo-
We identified MGC-PNs through electrical stimulation of the larized by BAL (Fig. 1c)29. We investigated how interglomerular
antennal nerve or iontophoretic injection of Lucifer Yellow (LY) inhibition might be involved in modulating the output from these
stain (Fig. 2f). As found previously28,29, PNs with arborizations two identified glomeruli.
confined to the glomerulus known as the ‘cumulus’ were selec-
tively depolarized by C15, whereas those with branches in the Cross-correlations within and between glomeruli
‘toroid’ glomerulus were depolarized by BAL (Figs. 1 and 2f). In Comparison of two simultaneously impaled PNs showed strik-
this study, all neurons that innervated the toroid also showed ing similarities in their responses to repeated odorant pulses, and

articles
Fig. 3. Time course of synchronous firing a 5 C15 0.35 b

between PNs. (a) PSTHs from two toroid 20 mV
4 PN 1
PNs (black, red) show excitatory responses PN 2
Counts/bin
to stimulation with BAL, but suppression of 0.25
Time (s)
3
background activity with C15 stimulation 2 50 +
2 6
alone (stimulus onset was at time = 0; dura-
PN 2
0.15
tion = 200 ms). (b) Dynamic correlation 0
1 2 100 + +
analysis (color contour plots) show that PN 1 6
0 0.05
spikes in the two PNs showed the greatest
0
0.05 0.15 0.25 0.35 0.05 0.15 0.25 0.35 2 + +
synchrony at the onset of their response win- 200
dows. Color scale represents the normalized 5 0.35 6
BAL
correlation calculated over five consecutive 4
0
responses (range = –1 to +1). (c) Effect of 2 300 + +
Counts/bin
0.25
Time (s)
3 6
stimulus duration on patterns of odor-evoked
synchrony between PNs. Raw data traces 0
2
(top) show responses of a pair of toroid PNs 0.15 2 400 + + +
1 6
to stimulation with a 50-ms pulse of BAL.
Spikes in the PNs became synchronized near 0 0.05 0
response onset. The moment at which the 0.05 0.15 0.25 0.35 0.05 0.15 0.25 0.35 0 0.1 0.2 0.3 0.4
time lag between spikes in the two PNs Blend
Time (s) Time (s) Time (s)
exceeded 5 ms is indicated by the asterisk.
Each panel shows the near-coincidence his-
tograms (top graphs) and the superimposed PSTHs (bottom graphs) for the two PNs stimulated with BAL pulses of 50, 100, 200, 300 and 400 ms,
respectively (stimulus time course indicated by solid bar beneath each panel). All traces were averaged over five trials. The times at which positive
correlations above 0.3 occurred are indicated in each panel (+). Several conditions are readily apparent. The first and largest coincidence peak reli-
ably reflects stimulus onset, regardless of stimulus duration. The distribution of spike activity broadens as a function of stimulus duration, and longer-
duration stimuli may also evoke a greater number of smaller coincidence peaks (bottom panel). This latter effect could be due to the greater
probability that the stimulus time course will be non-uniform as the period of stimulation is prolonged.
this occurred whether the two neurons were tuned to the same ulus pulse (Fig. 2a). Although the two neurons often continued to
or to different odorants (Fig. 1b). Without exception, both PNs fire after the odor pulse had ended, and their spike trains over-
in each pair fired discrete bursts of action potentials in response lapped in time, we saw no synchronous firing except during this
to consecutive pulses of the excitatory odorant, and each burst window defined by the stimulus duration. To test whether the
of spikes was truncated by a distinct membrane hyperpolariza- absence of periodic patterning was due to the effects of adapta-
tion (Fig. 1b and c)22. Time-series analysis showed further that tion, we also tested an inter-stimulus interval of 1 min. Again,
the temporal relationships between PN spike trains differed we found no evidence for an odor-specific pattern of synchrony
markedly depending on the odor-tuning properties of the two that was repeated from trial to trial (Fig. 2b, top). Neither did we
neurons recorded. Olfactory stimulation with the pheromone find any significant change in the strength of PN/PN coherence
blend (C15 + BAL) evoked synchronous firing in pairs of PNs over repeated trials using this pulsatile stimulation protocol
innervating the same glomerulus (correlation coefficients, mean (Fig. 2c; Table 1). Once again, intraglomerular PN pairs showed
± s.d.: 0.60 ± 0.15 for cumulus pairs, n = 11; 0.51 ± 0.26 for toroid a higher percentage of synchronous events per stimulus than did
pairs, n = 15), but significantly less synchrony between PNs inner- the interglomerular pairs (Fig. 2c; Table 1).
vating different glomeruli (0.27 ± 0.10, n = 8, P = 0.02; Fig. 1d).
These differences were evident in spite of the substantial overlap Concentration-dependent effects on PN synchrony
between spike trains in each pair of PNs (Fig. 1b). In addition, Of the 15 toroid pairs studied, eight were tested with two stim-
synchronized spiking did not appear spontaneously but only in ulus concentrations (1- and 10-ng pulses of BAL). Of these eight
response to olfactory stimulation. In all cases, we found more pairs of PNs, all 16 neurons were sensitive to changes in stimulus
synchrony between two PNs innervating the same glomerulus intensity (Fig. 2d). In each case, the 1-ng stimulus evoked fewer
and less synchrony between PNs innervating two neighboring synchronous events than the 10-ng stimulus (mean correlation
glomeruli, even though the two glomeruli encoded information coefficients for the 1- and 10-ng stimuli were 0.24 and 0.64,
about chemically similar odorants. respectively; P < 0.0005; Mann-Whitney U-test; Fig. 2d and e;
Table 1). For cumulus pairs tested with C15 (n = 7), the corre-
Stimulus-locked modulation of PN synchrony sponding values were 0.22 and 0.69 (P < 0.0005). Again, we saw
We also measured the influence of stimulus dynamics on the time no significant change in synchrony over repeated trials with the
course of synchrony between PNs (Fig. 2a and b). In accordance same stimulus, and the temporal pattern of synchrony evoked at
with our earlier studies, a stimulation protocol that produced both odor intensities always reflected stimulus dynamics with
multiple odor pulses separated by 1-s periods of clean air (much great temporal precision (Fig. 2d). In addition to increased syn-
like a natural odor plume12) yielded no evidence of an odor-spe- chrony (Fig. 2e, left), stimulation with the elevated odor con-
cific or periodic pattern of synchrony in the responses of paired centration resulted in significant reductions in the latency of the
cumulus PNs or toroid PNs (Fig. 2b, middle and bottom). response in individual PNs and in the latency to synchronous
Instead, spiking patterns were closely matched to the dynamics events between PNs (Fig. 2e, right). The 10-ng stimulus advanced
of the stimulus itself12. Synchronous firing occurred during a the response onset by as much as 50 ms as compared with the
time window that started at the onset of the response and con- 1-ng stimulus, and this pattern was generally consistent over
tinued for a period that approximated the duration of the stim- repeated stimulus trials (Fig. 2e, right).

articles
Fig. 4. Blend-enhanced synchronization of PNs inner- a

vating the same glomerulus. In about 70% of all cases
examined, synchrony between glomerulus-specific PNs
Counts/bin
Counts/bin
Counts/bin
was enhanced by stimulation with the pheromone blend.
(a) Cross-correlograms (10-ms bins) calculated from the
simultaneous responses of two toroid PNs to a BAL
stimulus. The number of synchronous events between
the two neurons was markedly greater in response to a
mixture of BAL + C15 (Blend, right panel) than to BAL
– – – – – –
alone (middle panel), and there was little or no co- Time (s) Time (s) Time (s)
activity between PNs in the absence of an odor stimulus
(control, left panel). (b) Dynamic correlation analysis b C15
Correlation
showed the time course of synchrony. Again, synchrony
occurred reproducibly at response onset. The blend-
enhanced correlation is evident from the broader distri- BAL
bution of warm colors in the surface plots, and this
correlation was seen across the five consecutive odor
pulses (magnitude of correlation represented by color Blend 30 BAL Blend
20
scale, top right). Mean-rate histograms (inset) show that
PN 2
10
enhanced co-activity in response to the blend is not sim- 0
ply a function of increased spike activity evoked by PN 1 PN #
combining the two odorants. (c) Summary of blend-
enhanced effects on synchrony between PNs in each
group. Left, synchrony was expressed as the percentage Time (s)
of synchronous events relative to the total spike counts
in both PNs. The blend effect occurred in both the c PN-PN synchrony (%)
Latency, response onset
to peak synchrony (ms)
cumulus (7 out of 11) and toroid (11 out of 15) groups abd ab 30
c/c
100 b c b
but was statistically significant only in the former group d
t/t
25
(Kruskal-Wallis test followed by Mann-Whitney U-test; P 80 c/t
20 b ab
< 0.05). Interglomerular synchrony evoked by the blend 60
ab
a
(n = 8) was also significantly weaker than intraglomerular 15
synchrony. Right, differential effects of the individual and 40
10
blended odorants on the latency from stimulus onset to 20 5
synchronous firing between PNs. The odor blend shifted
0 0
this latency to lower values for both the cumulus and C15 Blend BAL Blend Blend C15 Blend BAL Blend Blend
toroid (intraglomerular) groups of PNs, but the effect
was significant only in the latter group (P < 0.05). For the
interglomerular group, the latency was not significantly different from that measured in the other two groups. All values are mean ± s.d. Means
sharing the same letter were not significantly different (P > 0.05).
Dynamics of PN/PN synchronization that defines the sex pheromone in this insect) would evoke pat-
Peristimulus-time histograms (PSTHs) showed that the time terns of PN synchrony that were different from the patterns evoked
course of responses in two PNs selective for the same odorant by the individual odorants. This question is particularly intrigu-
were often nearly indistinguishable (Fig. 3a). Nevertheless, dynam- ing in light of the finding that in the PNs studied here, input from
ic correlation analysis of the underlying pattern of PN firing the neighboring glomerulus was inhibitory (Fig. 1c) and could
showed that the peak period of synchrony between the two neu- thus perform a modulatory function when the two adjacent
rons occurred only during a narrow window near the onset of the glomeruli were activated simultaneously. In 7 of 11 cumulus pairs
response (Fig. 3b). This brief period of co-activity (hereafter and 11 of 15 toroid pairs, the timing of PN/PN synchrony was
referred to as ‘onset synchrony’) was followed first by a period of enhanced by the mixture of odorants (Fig. 4a). The blend evoked
desynchronized spiking, and then by complete suppression of all significantly greater synchrony between PNs innervating the same
spiking until the next stimulus pulse. Onset synchrony between glomerulus (P = 0.05), and it also led to a broadening of the tem-
PNs also occurred before the peak response as measured by mean poral window during which synchrony occurred. This effect was
spike rate (Fig. 3c), indicating that the peaks of synchronous fir- reproducible over repeated trials (Fig. 4b). Comparing the ‘blend
ing did not reflect random coincident firing between the two neu- effect’ across the three groups of PN pairs, it is evident once again
rons. Onset synchrony always accompanied the rising phase of that the degree of synchronous firing between interglomerular PNs
the overlapping PSTHs from the two PNs, and it occurred at all was significantly smaller than that seen in either of the intra-
stimulus durations tested, from 50 to 400 ms (Fig. 3c). Peaks in glomerular groups (Fig. 4c, left; P < 0.05). Furthermore, in about
synchrony occasionally re-appeared during the falling phase of 50% of PN pairs in the latter two groups, we saw no apparent dif-
the odor response (possibly signaling stimulus offset), but the ference in spike rate when the primary excitatory stimulus or the
timing of these peaks was less predictable than it was during the blend containing that odorant was used (Fig. 4b, inset), indicat-
rising phase (Fig. 3c). ing that blend-enhanced synchrony was not a simple consequence
of increased PN spiking.
Blend-enhanced effects on synchrony We next examined the responses evoked by the C15+BAL
In nature, olfactory stimuli are typically mixtures of odorants. We blend to determine whether the blended stimulus had an effect
therefore examined whether the blend of BAL and C15 (a mixture on latency to PN synchrony, as seen with elevated stimulus

articles
Fig. 5. Correlation between PNs in one glomerulus is modulated by

a inhibitory input from the neighboring glomerulus. (a) Simultaneous
PN/PN recordings showed a multiphasic postsynaptic response to a
blend of BAL + C15 (bar shows time course of a 200-ms pulse). In both
neurons the response consisted of a pronounced IPSP (arrow) that
immediately preceded a burst of action potentials22. This example is
from two toroid PNs that were both depolarized by BAL but hyperpo-
larized by C15 as in Fig. 1b. The IPSPs evoked by C15 directly preceded
b the onset of synchronous firing evoked by BAL. The PNs remained syn-
chronized for a period that matched the stimulus duration, and shortly
thereafter became desynchronized (asterisk). Spiking in both PNs then
ceased and the neurons returned to their resting states. (b) Relationship
between spike synchrony in a pair of PNs innervating one glomerulus,
and the strength of inhibitory synaptic input from the neighboring
glomerulus. The correlation between PN pairs is plotted against the
mean amplitude of the IPSP evoked by the odorant that activates the
adjacent glomerulus (y = 0.4 + 0.3x – 0.04x2; P = 0.01). (c) Schematic cir-
cuit diagram relating the two glomerular networks examined in this
olfactory system (round synapses are excitatory; triangular synapses are
inhibitory). The BAL- and C15-selective populations of olfactory recep-
tor cells (ORCs) transmit odor information to the toroid and the cumu-
lus, respectively. ORCs can excite PNs monosynaptically and/or
indirectly through disinhibitory pathways involving a diverse population
of about 300 GABAergic local interneurons (LN1–n). In the presence of
the conspecific pheromone blend, the two MGC glomeruli first act as fil-
c ters that provide specific spatial addresses for inputs from the two
classes of ORCs that are simultaneously activated. The GABAergic LNs
(providing the major inhibitory input to PNs) then serve at least two
important processing functions at this early stage: (i) they organize the
spatial pattern of activity in the activated glomeruli through specific
interglomerular linkages, and (ii) they organize the timing of output sig-
nals simultaneously from each glomerulus through the modulation of
synchrony between PNs (PN1 and 2 in toroid, PN3 and 4 in cumulus).
coefficient for each pair of PNs, we found a strongly positive cor-

relation between these two variables (Fig. 5b).
DISCUSSION
It has been proposed for the mammalian olfactory bulb that
concentrations (Fig. 2e). Stimulation with the blend of odor- neural synchronization across glomerular outputs may enhance
ants also resulted in reductions in the latency of the responses the representation of a complex olfactory stimulus by integrat-
of individual PNs and in the latency to onset synchrony ing the different signal streams activated by the odor into a uni-
between PNs (Fig. 4c, right). This shift in latency occurred in fied olfactory ‘image’ at the level of the sensory cortex9,18. In the
both intraglomerular groups, but was statistically significant sphinx moth, we have shown that synchrony does indeed occur
only in the toroid pairs (P < 0.05). The odor-blend stimulus across glomeruli, but we found that even when a blend of odor-
did not reduce the latency of response when synchrony was ants was used as a stimulus, the firing patterns of PNs from the
measured across glomeruli. same glomerulus always showed the highest correlations
(Fig. 1d). A number of studies have also suggested that lateral
Modulation of synchrony by interglomerular inhibition inhibition (mediated by reciprocal connections between M/T
GABA is centrally involved in the precision of spike timing in and granule cells in the olfactory bulb) may sharpen odor tun-
glomerular PNs of both vertebrates and invertebrates, and fast ing in a glomerulus (in direct analogy to contrast enhancement in
IPSPs from GABAergic local neurons (LNs) are commonly seen a retinal ganglion cell)18,23–25, but this is not universally accept-
in PNs of moths22,30. If PN/PN synchrony is not modulated by ed27. In an attempt to resolve these issues, we used intracellular
oscillations (which could arise from inhibitory feedback loops in recordings from glomerular output neurons to examine the
the glomerular neuropil), we reasoned that feed-forward inhibito- dynamic interactions that occur within and between two adja-
ry connections could serve to synchronize PNs. We therefore test- cent glomeruli with identified odor tuning.
ed whether the activity of one glomerulus was modulated by In the absence of odor, MGC-PNs in moths typically fire
inhibition from its neighbor. Simultaneous intracellular record- action potentials sporadically, and firing between PNs is asyn-
ings from neurons stimulated with the blend of odorants detected chronous (Fig. 1b). When presented with pulses of the correct
multiphasic responses comprising virtually identical, fast-onset olfactory stimulus, however, PNs synchronize their discharges in
IPSPs, followed by depolarizing EPSPs that gave rise to trains of response to each stimulus pulse (Fig. 1d), with PNs from the
action potentials in both neurons (Fig. 5a). In 12 pairs of odor- same glomerulus showing the highest correlations (Fig. 2a). The
ant-matched PNs, we measured the amplitudes of the IPSPs evoked enhanced precision of PN/PN spiking is therefore a potential
by the primary excitatory odorant to the neighboring glomeru- means of strengthening the spatial representation of the stimulus,
lus31. When the average values were plotted against the correlation which according to many recent imaging studies can be defined

articles
by the specific glomerulus (or combination of glomeruli) encod- odorant, and more than half of these PN pairs showed signifi-
ing it19,32–34. In view of this capacity to modulate spike timing cantly greater synchronization in response to the blend of the
on a millisecond time scale, we propose that the MGC glomeruli two odorants (Fig. 4; Table 1). This finding provides new evi-
act as multifunctional coding modules in the brain, participat- dence that the temporal patterning of output from a given
ing in the simultaneous and parallel encoding of the different glomerulus may be further modulated by inhibitory input arising
attributes of the stimulus (quality, quantity and spatiotemporal from neighboring glomeruli18,23–25. Earlier single-unit studies in
features)11,12,22 inherent in a dynamic odor plume35. Our results moths showed that the timing of spike discharges in MGC-PNs is
showing tight correlations in spike timing between PNs with the modulated by bicuculline-sensitive, GABAA-like synapses from
same tuning characteristics provide new support for the long- LNs22,30,43. When these synapses are blocked pharmacological-
held hypothesis that olfactory glomeruli represent the funda- ly, the ability of MGC-PNs to resolve intermittent odor pulses
mental coding modules in early olfactory processing13,16,19,20,25. falls dramatically.
A testable circuit model that can help explain the possible
What coding functions might synchrony perform? organization of these inhibitory interglomerular interactions is
In other sensory systems, there is evidence that the incorporation shown in Fig. 5c. This model incorporates much of our current
of intercellular timing relations into a neural-population model knowledge of the connectivity that constitutes the processing net-
improves the accuracy of ensemble coding and thus facilitates works in MGC glomeruli (reviewed in ref. 22). The model shows
stimulus identification36,37. In motor cortex, the synchronization that when a blend of odorants is being processed, inhibition from
of spike activity between interneurons also facilitates the encoding one glomerulus could reset the timing of spike discharges in PNs
of arm movements by ensembles of neurons38. How might neur- activated in the neighboring glomerulus, thus aiding in their syn-
al synchronization function in encoding olfactory information? chronization. There are both similarities and differences between
Recent results from olfactory-bulb slice preparations show that this organization and that seen in the vertebrate retina. Like reti-
synchronous firing is consistently greater for intra- than for internal ganglion cells, PNs in one glomerulus show enhanced tem-
glomerular pairs of M/T cells, but these studies did not involve poral tuning in the presence of inhibition from the neighboring
odor stimulation39,40. Our results here provide the first evidence glomerulus; unlike ganglion cells, PNs do not show a broaden-
from intracellular recordings that PNs innervating the same ing of their odor tuning in the absence of this inhibition.
glomerulus and tuned to the same odorant are more tightly syn- Although we cannot rule out the possibility that other cellular
chronized than pairs of PNs that process different olfactory inputs. or synaptic mechanisms promote synchronous firing between
This then raises the question: what specific function might PNs26,39,40, we believe that the BAL-evoked IPSP in cumulus PNs
intraglomerular synchrony serve? One suggestion is that it tem- and the C15-evoked IPSP in toroid PNs (Fig. 1c) could serve this
porally integrates information streams from select subsets of the resetting function22,31. Pharmacological experiments to test this
functionally diverse population of output neurons that arise from hypothesis are now in progress.
a single glomerulus41. Neural-ensemble recordings in the moth It must be noted that the principles outlined here for the
MGC11 recently showed that each brief stimulus pulse triggers a glomerular processing of pheromonal information are probably
transient burst of activity across the coding ensemble, but spe- not unique to these specialized odorants or to the sexually dimor-
cific features of the stimulus are encoded in the precise tempo- phic glomeruli dedicated to them. Data from both the MGC44,45
ral relationships superimposed on the ensemble. For example, and non-pheromonal glomeruli46 are consistent with the funda-
different subsets of PNs synchronize at different stimulus con- mental hypothesis that each glomerulus is an identifiable, func-
centrations, suggesting a functional partitioning within the tional unit dedicated to the processing of odorant-specific
glomerulus11. According to this organizational scheme, the com- information. Likewise, both pheromonal and non-pheromonal
plete ensemble could encode the odor signal as well as monitor its odorants activate an odor-specific ‘mosaic’ of antennal-lobe
concentration dynamics because different subsets of PNs would glomeruli in a reproducible pattern (reviewed in ref. 19). Using
synchronize as the ambient concentration of the stimulus multichannel electrode arrays11, we are now probing the hypoth-
changed. Synchronous firing across PNs may therefore help to esis of glomerular chemotopy by recording ensemble responses to
reinforce the spatially organized segregation of odorant-specific a wide array of odorants that are known to trigger activity from
signals encoded in each glomerulus13–20. Recent studies in the both broadly and narrowly tuned ORNs on the moth antennae47.
moth brain indicate that such segregation of odorant-specific We propose that in the context of a blended olfactory stimulus
pathways is maintained even at higher levels of sensory integra- (as typically found in nature), reciprocal inhibitory interactions
tion in the protocerebrum42. between glomeruli provide a temporal mechanism for strength-
Thus, in the moth olfactory system, evidence from both intra- ening the spatial representation of a complex stimulus by syn-
cellular and neural-ensemble recording studies indicates that the chronizing PNs, both within and between the population of
chemical identity of an odor is encoded spatially, according to activated glomeruli. Thus, although a single odorant evokes syn-
which glomeruli are activated (or inhibited) by the stimulus. chronization among outputs from the same glomerulus, the
Other key features of the stimulus (including odor intensity, blending of several odorants increases the probability of lateral
dynamics and the quality of specific odorant blends) are encod- inhibitory interactions between neighboring glomeruli, thereby
ed in specific temporal patterns of activity superimposed on the augmenting the temporal tuning of synchronous output from
spatial ensemble11,12,29,31. It has been suggested that if the target each participating glomerulus.
neurons in higher centers function as coincidence detectors, then Finally, do our results support the suggestion that lateral inhi-
synchronous input to these centers may facilitate the decoding bition operates between glomeruli in the olfactory system18,27?
of these different stimulus features18. Lateral-inhibitory interactions between MGC glomeruli do
indeed help to shape the temporal representations of pheromon-
Interglomerular inhibition shapes odor representations al stimuli, but not in the same sense that lateral inhibition in the
In all experiments involving BAL and C15, PNs that were depo- retina, for example, enhances the local spatial variations in an
larized by one of these odorants were hyperpolarized by the other antagonistic center-surround receptive field. In analogy to ‘on-

articles
center’ retinal ganglion cells that are hyperpolarized by illumi- potentials in LNs is about twice that of PNs43. These criteria were used as
nation of the surround48, MGC-PNs innervating one glomerulus reliable indicators of PNs in all experiments.
(for example, the toroid) are hyperpolarized by the olfactory Olfactory stimuli were delivered to the preparation as reported previ-
input to the neighboring glomerulus (the cumulus). Unlike the ously43. Pulses of air from a constant air stream were diverted through a
glass syringe containing a piece of filter paper to which was applied a sin-
receptive field organization of retinal ganglion cells, however, the
gle odorant (1 or 10 ng) or a blend of two odorants (1 or 10 ng of each).
responses of toroid PNs are not suppressed by uniform stimula- The odor stimulus was pulsed by means of a solenoid-activated valve
tion of the entire receptive field (that is, stimulation with a blend controlled by an electronic stimulator (W-P Instruments, Sarasota, Flori-
of BAL and C15, as in Fig. 4). Rather, the blend of odorants da). In every experiment, the outlet of the stimulus syringe was posi-
specifically leads to an enhancement of synchronized firing tioned about 2 cm from and orthogonal to the center of the antennal
between glomerulus-specific PNs in both the cumulus and the flagellum ipsilateral to the impaled antennal lobe. Stimulus durations
toroid. Thus, reciprocal inhibition between glomeruli at this level varied from 50 ms to 5 s, and multiple odor pulses were separated by
of processing in the olfactory pathway serves to emphasize the intervals of 1 s or 1 min. The odor stimuli used were: (i) E,Z-10,12-hexa-
presence of the individual constituents of the blend, rather than decadienal (bombykal, BAL), the primary 16-carbon aldehyde compo-
to exaggerate the difference (or ‘contrast’) between the two odor- nent of the female’s sex pheromone; (ii) E,Z-11,13-pentadecadienal (C15),
a 15-carbon aldehyde mimic of the second essential component of the
ant molecules. Although this latter function may be reserved for
sex pheromone; and (iii) a mixture of BAL and C15 (blend). Although
higher levels of processing in the protocerebrum, inhibition in we substituted C15 for the natural pheromone component, we refer to
the antennal lobe is important for synchronizing glomerular out- both BAL and C15 as pheromone components.
put, thus enhancing the transmission of weak olfactory signals
and increasing stimulus contrast relative to background odors in Data analysis. Analog signals stored on FM tape were digitized at 20 KHz
the animal’s environment. per channel using Autospike (Syntech, Silversum, the Netherlands) or
Axoscope software (Axon Instruments, Foster City, California). Time
METHODS stamps representing the occurrence of each action potential in an intra-
Preparation. Manduca sexta (L.) (Lepidoptera: Sphingidae) were reared cellular trace were then analyzed with Neuroexplorer (Nex Technologies,
in the laboratory on artificial diet under a long-day photoperiod, and Winston-Salem, North Carolina). Cross-correlograms and correlation
adult male moths, 1–3 days after emergence, were prepared for experi- coefficients were calculated (5-ms bins) to help quantify synchrony
ments as described previously28,29. For electrophysiological recording, between two PNs. These data were corrected by subtracting shift-
the moth was restrained in a plastic tube with its head fully exposed. The predictor values to control for synchrony related solely to the timing of
labial palps, proboscis and cibarial musculature were then removed to the stimulus49. Dynamic correlations (or joint post-stimulus-time his-
allow access to the brain. To eliminate movement, the head was isolated tograms) were used to analyze the time course and temporal patterning
and pinned to a wax-coated glass Petri dish with the antennal lobes fac- of PN synchronization50. The resulting matrix illustrates the magnitude
ing upward. Tracheae and a small part of the sheath overlying one anten- and timing of synchrony, with perfect synchrony between two PNs occur-
nal lobe were then removed with fine forceps. The preparation was ring along the main diagonal. Near-coincidence histograms were extract-
continuously superfused with physiological saline solution containing ed from the main diagonal of each matrix using a bin width of 5–10 ms.
150 mM NaCl, 3 mM CaCl2, 3 mM KCl, 10 mM TES buffer (pH 6.9) and Latency measurements (time between stimulus onset and first synchro-
25 mM sucrose. nous events) were also used as a means to quantify the effects of stimulus
concentration and blending odorants.
Intracellular recording and staining. Intracellular recording and dye-
marking were performed with borosilicate glass microelectrodes filled
with a 4% solution of Lucifer Yellow CH (LY) (Sigma) in 0.2 M LiCl and Acknowledgments
having resistances of 100–400 MΩ (ref. 29). We recorded simultaneous We thank K. Daly, V. Pawlowski and B. Smith for discussions and comments and
activity from two antennal-lobe neurons by independently controlling H. Stein and A.A. Osman for technical assistance. Supported by grants and
two microelectrodes at penetration sites separated by 50–200 µm, contracts from National Institutes of Health (NIDCD).
depending on whether we wished to record from PNs innervating the
same or different glomeruli. The signals from both electrodes were visu- Competing interests statement
alized on an oscilloscope and recorded on FM tape. After carrying out
The authors declare that they have no competing financial interests.
this physiological characterization, we injected cells with LY by passing
hyperpolarizing current (up to 1.5 nA) for 5–15 min. At the completion
of an experiment, the brain was excised and immersed in formaldehyde RECEIVED 4 JANUARY; ACCEPTED 1 APRIL 2002
fixative solution (2.5% formaldehyde in 0.1 M sodium phosphate buffer
with 3% sucrose added). Brains were fixed overnight, dehydrated through 1. Vaadia, E., Ahissar, E., Bergman, H. & Lavner, Y. Correlated activity of
a graded series of ethanol solutions and cleared with methyl salicylate. neurons: a neural code for higher brain functions? in Neuronal Cooperativity
Cleared brains were viewed as whole mounts with a laser-scanning con- (ed. Kruger, J.) 205–279 (Springer-Verlag, Berlin, 1991).
focal microscope (Bio-Rad MRC-600, Cambridge, Massachusetts) 2. Rieke, F., Warland, D., de Ruyter van Steveninck, R. & Bialek, W. Spikes:
equipped with a Nikon Optiphot-2 microscope and both 15-mW kryp- exploring the neural code (MIT Press, Cambridge, Massachusetts, 1997).
3. Murthy, V. N. & Fetz, E. E. Synchronization of neurons during local field
ton/argon and 100-mW argon laser-light sources. Optical sections were potential oscillations in sensorymotor cortex of awake monkeys.
2 or 5 µm thick. Preparations were then returned to 100% ethanol and J. Neurophysiol. 76, 3968–3982 (1996).
embedded in Spurr’s resin. When higher-resolution images were needed, 4. Singer, W. & Gray, C. M. Visual feature integration and the temporal
histological sections of the brain were cut at 48 µm with a sliding micro- correlation hypotheses. Annu. Rev. Neurosci. 18, 555–586 (1995).
tome. 5. Adrian, E. D. Olfactory reactions in the brain of the hedgehog. J. Physiol.
(Lond.) 100, 459–473 (1942).
6. Gelperin, A. Oscillatory dynamics and information processing in olfactory
Sensory stimulation and neuron identification. Orthodromic electrical systems. J. Exp. Biol. 202, 1855–1864 (1999).
stimulation of the antennal nerve and olfactory stimulation of the anten- 7. Laurent, G., Wehr, M. & Davidowitz, H. Temporal representations of odors in
na were used to generate postsynaptic responses in antennal lobe neu- an olfactory network. J. Neurosci. 16, 3837–3847 (1996).
rons. Hook electrodes were placed beneath the antennal nerve, and 8. Mellon, D. & Wheeler, C. J. Coherent oscillations in membrane potential
repetitive 0.1-ms shocks were delivered as a test for monosynaptic input. synchronize impulse bursts in central olfactory neurons of the crayfish.
J. Neurophysiol. 81, 1231–1241 (1999).
Previous studies reported distinct differences in the physiological pro- 9. Kashiwadani, H., Sasaki, Y. F., Uchida, N. & Mori, K. Synchronized oscillatory
files of local and projection neurons in the moth antennal lobe: LNs more discharges of mitral/tufted cells with different molecular receptive ranges in
often receive direct antennal input, and the mean half-width of action the rabbit olfactory bulb. J. Neurophysiol. 82, 1786–1792 (1999).

articles
10. Kay, L. M. & Laurent, G. Odor- and context-dependent modulation of mitral inhibition of projection neurons in the antennal lobes of the sphinx moth
cell activity in behaving rats. Nat. Neurosci. 2, 1003–1009 (1999). Manduca sexta. J. Comp. Physiol. A 161, 23–32 (1987).
11. Christensen, T. A., Pawlowski, V. M., Lei, H. & Hildebrand, J. G. Multi-unit 31. Christensen, T. A. & Hildebrand, J. G. Coincident stimulation with
recordings reveal context-dependent modulation of synchrony in odor- pheromone components improves temporal pattern resolution in central
specific neural ensembles. Nat. Neurosci. 3, 927–931 (2000). olfactory neurons. J. Neurophysiol. 77, 775–781 (1997).
12. Vickers, N. J., Christensen, T. A., Baker, T. C. & Hildebrand, J. G. Odour- 32. Johnson, B. A., Woo, C. C. & Leon, M. Spatial coding of odorant features in the
plume dynamics influence the brain’s olfactory code. Nature 410, 466–470 glomerular layer of the rat olfactory bulb. J. Comp. Neurol. 393, 457–471 (1998).
(2001). 33. Belluscio, L. & Katz, L. C. Symmetry, stereotypy, and topography of odorant
13. Christensen, T. A. & White, J. Representation of olfactory information in the representations in mouse olfactory bulbs. J. Neurosci. 21, 2113–2122 (2001).
brain. in The Neurobiology of Taste and Smell, Vol. 2 (eds. Finger, T. E., Silver, 34. Fuss, S. H. & Korsching, S. I. Odorant feature detection: activity mapping of
W. L. & Restrepo, D.) 201–232 (Wiley, New York, 2000). structure response relationships in the zebrafish olfactory bulb. J. Neurosci.
14. Ressler, K. J., Sullivan, S. L. & Buck, L. B. Information coding in the olfactory 21, 8396–8407 (2001).
system: evidence for a stereotyped and highly organized epitope map in the 35. Murlis, J., Elkinton, J. S. & Cardé, R. T. Odor plumes and how insects use
olfactory bulb. Cell 79, 1245–1255 (1994). them. Annu. Rev. Entomol. 37, 505–532 (1992).
15. Vassar, R. et al. Topographic organization of sensory projections to the 36. Deadwyler, S. A., Bunn, T. & Hampson, R. E. Hippocampal ensemble activity
olfactory bulb. Cell 74, 309–318 (1994). during spatial delayed-nonmatch-to-sample performance in rats. J. Neurosci.
16. Lancet, D. Vertebrate olfactory reception. Ann. Rev. Neurosci. 9, 329–355 16, 354–372 (1996).
(1986). 37. Nicolelis, M. A. et al. Simultaneous encoding of tactile information by three
17. Singer, M. S. & Shepherd, G. M. Molecular modeling of ligand-receptor primate cortical areas. Nat. Neurosci. 1, 621–630 (1998).
interactions in the OR5 olfactory receptor. Neuroreport 5, 1297–1300 38. Maynard, E. M. et al. Neuronal interactions improve cortical population
(1994). coding of movement direction. J. Neurosci. 19, 8083–8093 (1999).
18. Mori, K., Nagao, H. & Yoshihara, Y. The olfactory bulb: coding and 39. Schoppa, N. E. & Westbrook, G. L. Glomerulus-specific synchronization of
processing of odor molecule information. Science 286, 711–715 (1999). mitral cells in the olfactory bulb. Neuron 31, 639–651 (2001).
19. Galizia, C. G. & Menzel, R. The role of glomeruli in the neural representation 40. Carlson, G. C., Shipley, M. T. & Keller, A. Long-lasting depolarizations in
of odours: results from optical recording studies. J. Insect Physiol. 47, 115–130 mitral cells of the rat olfactory bulb. J. Neurosci. 20, 2011–2021 (2000).
(2001). 41. Christensen, T. A. Anatomical and physiological diversity in the central
20. Shipley, M. T. & Ennis, M. Functional organization of olfactory system. processing of sex-pheromonal information in different moth species. in
J. Neurobiol. 30, 123–176 (1996). Insect Pheromone Research: New Directions (eds. Cardé, R.T. & Minks, A.K.)
21. Stopfer, M., Bhagavan, S., Smith, B. & Laurent, G. Impaired odor 184–193 (Chapman & Hall, New York, 1997).
discrimination on desynchronization of odour-encoding neural assemblies. 42. Lei, H., Anton, S. & Hansson, B. S. Olfactory protocerebral pathways
Nature 390, 70–74 (1997). processing sex pheromone and plant odor information in the male moth
22. Christensen, T. A., Waldrop, B. R. & Hildebrand, J. G. Multitasking in the Agrotis segetum. J. Comp. Neurol. 432, 356–370 (2001).
olfactory system: context-dependent responses to odors reveal dual GABA- 43. Christensen, T. A., Waldrop, B. R., Harrow, I. D. & Hildebrand, J. G. Local
regulated coding mechanisms in single olfactory projection neurons. interneurons and information processing in the olfactory glomeruli of the
J. Neurosci. 18, 5999–6008 (1998). moth Manduca sexta. J. Comp. Physiol. A 173, 385–399 (1993).
23. Rall, W., Shepherd, G. M., Reese, T. S. & Brightman, M. W. Dendrodentritic 44. Rospars, J. P. & Hildebrand, J. G. Anatomical identification of glomeruli in
synaptic pathway for inhibition in the olfactory bulb. J. Exp. Neurol. 14, the antennal lobes of the male sphinx moth Manduca sexta. Cell Tissue Res.
44–56 (1966). 270, 205–227 (1992).
24. Buonviso, N. & Chaput, M. A. Response similarity to odors in olfactory bulb 45. Rospars, J. P. & Hildebrand, J. G. Sexually dimorphic and isomorphic
output cells presumed to be connected to the same glomerulus: glomeruli in the antennal lobes of the sphinx moth Manduca sexta. Chem.
electrophysiological study using simultaneous single-unit recordings. Senses 25, 119–129 (2000).
J. Neurophysiol. 63, 447–454 (1990). 46. King, J. R., Christensen, T. A. & Hildebrand, J. G. Response characteristics of
25. Shepherd, G. M. & Greer, C. A. Olfactory bulb. in The Synaptic Organization an identified, sexually dimorphic olfactory glomerulus. J. Neurosci. 20,
of the Brain (ed. Shepherd, G.M.) 139–169 (Oxford, New York, 1990). 2391–2399 (2000).
26. Desmaisons, D., Vincent, J. D. & Lledo, P. M. Control of action potential 47. Shields, V. D. C. & Hildebrand, J. G. Responses of a population of antennal
timing by intrinsic subthreshold oscillations in olfactory bulb output olfactory receptor cells in the female moth Manduca sexta to plant-associated
neurons. J. Neurosci. 19, 10727–10737 (1999). volatile organic compounds. J. Comp. Physiol. 186, 1135–1151 (2001).
27. Laurent, G. A systems perspective on early olfactory coding. Science 286, 48. Barlow, H. B., Hill, R. M. & Levick, W. R. Retinal ganglion cells responding
723–728 (1999). selectively to direction and speed of image motion in the rabbit. J. Physiol.
28. Hansson, B. S., Christensen, T. A. & Hildebrand, J. G. Functionally distinct (Lond.) 173, 377–407 (1964).
subdivisions of the macroglomerular complex in the antennal lobes of the 49. Nowak, L. G., Munk, M. H. J., James, A. C., Girard, P. & Bullier, J. Cross-
sphinx moth Manduca sexta. J. Comp. Neurol. 312, 264–278 (1991). correlation study of the temporal interactions between areas V1 and V2 of the
29. Heinbockel, T., Christensen, T. A. & Hildebrand, J. G. Temporal tuning of macaque monkey. J. Neurophysiol. 81, 1057–1074 (1999).
odor responses in pheromone-responsive projection neurons in the brain of 50. Aertsen, A. M. H. J., Gerstein, G. L., Habib, M. K. & Palm, G. Dynamics of
the sphinx moth Manduca sexta. J. Comp. Neurol. 409, 1–12 (1999). neuronal firing correlation: modulation of “effective connectivity”.
30. Waldrop, B., Christensen, T. A. & Hildebrand, J. G. GABA-mediated synaptic J. Neurophysiol. 61, 900–917 (1989).

articles
Decreasing hypothalamic insulin

receptors causes hyperphagia and
insulin resistance in rats
Silvana Obici1, Zhaohui Feng1, George Karkanias1, Denis G. Baskin2 and Luciano Rossetti1
1 Albert Einstein College of Medicine, 1300 Morris Park Avenue, Bronx, New York 10461, USA
2 VA Puget Sound Health Care System and University of Washington, 1660 S. Columbian Way, Seattle, Washington 98108, USA
Correspondence should be addressed to L.R. (rossetti@aecom.yu.edu)
We investigated the role of hypothalamic insulin signaling in the regulation of energy balance and
insulin action in rats through selective decreases in insulin receptor expression in discrete hypothala-
mic nuclei. We generated an antisense oligodeoxynucleotide directed against the insulin receptor
precursor protein and administered this directly into the third cerebral ventricle. Immunostaining of
rat brains after 7-day administration of the oligodeoxynucleotide showed a selective decrease of
insulin receptor protein within cells in the medial portion of the arcuate nucleus (decreased by ~80%
as compared to rats treated with a control oligodeoxynucleotide). Insulin receptors in other
hypothalamic and extra-hypothalamic areas were not affected. This selective decrease in hypothala-
mic insulin receptor protein was accompanied by rapid onset of hyperphagia and increased fat mass.
During insulin-clamp studies, physiological hyperinsulinemia decreased glucose production by 55%
in rats treated with control oligodeoxynucleotides but by only 25% in rats treated with insulin
receptor antisense oligodeoxynucleotides. Thus, insulin receptors in discrete areas of the hypothala-
mus have a physiological role in the control of food intake, fat mass and hepatic action of insulin.
Insulin has been postulated to be a signal that provides negative specific disruption of the insulin receptor (IR) gene throughout
feedback to the brain for the long-term regulation of energy bal- the central nervous system leads to increases in body fat and in
ance1–3. According to the ‘lipostatic model’ of the regulation of plasma insulin and leptin levels20. The importance of local pop-
energy balance4, peripheral signals proportional to the size of ulations of neurons expressing insulin receptors in mediating the
energy stores communicate energy status to brain centers involved CNS effects of insulin on energy homeostasis remained to be
in the regulation of food intake and fuel metabolism4–7. Leptin delineated, however. To address this question, we generated a
and insulin are ideal candidates for this lipostatic function, selective, transient defect in hypothalamic insulin signaling by
because their levels are closely related to adiposity7,8 and their infusion of an antisense oligodeoxynucleotide designed to blunt
central administration decreases food intake 1,7,9,10. Indeed, the expression of the IR protein in rat hypothalamic areas sur-
insulin receptors are expressed in several sites within the rounding the third cerebral ventricle (Fig. 1). We found marked
brain11,12, including the medial (containing neuropeptide Y and selective decreases in insulin receptor expression in discrete
(NPY)-expressing neurons) and lateral (containing pro- hypothalamic nuclei. We then examined the effects on brain IR
opiomelanocortin (POMC)-expressing neurons) portions of the protein, food intake, fat mass and insulin action. This rat model
arcuate nucleus of the hypothalamus12. Studies in which insulin allowed us to assess the possible role of hypothalamic insulin sig-
is administered into the third cerebral ventricle (ICV) are among naling in the regulation of energy balance and insulin action.
the strongest evidence to date in support of insulin’s anorectic
role in the central nervous system1,10. These studies showed a RESULTS
decrease in food intake and fat mass after prolonged ICV insulin ICV IR antisense decreased IR protein in hypothalamus
infusions1,10 and a decrease in the hypothalamic expression of We administered IR antisense (IR-AS) and control (scrambled; IR-
NPY, a potent orexogenic peptide, after acute ICV injection of SCR) oligodeoxynucleotides (ODNs) by ICV infusion in the third
insulin or insulin mimetics10,13,14. Notably, ICV injections of cerebral ventricle, and assessed the effects on IR proteins by brain
insulin specifically decrease the expression of NPY in the medial immunostaining and immunoblotting with anti-IR antibodies.
portion of the arcuate nucleus12,13,15. However, insulin receptors After 7 days of infusion (Fig. 1), we saw a selective downregulation
are also present in other brain regions, where insulin may have of IR protein in the medial portion of the arcuate nucleus (∼80%
important roles in neuronal growth16, differentiation17 and func- lower in IR antisense–treated as compared to control mice; Fig. 2a,
tion18,19. Recent loss-of-function studies have provided strong b and Table 1) and in the adjacent habenular nucleus (Fig. 2i, j and
support for a physiological role of brain insulin receptors in the Table 1). This is consistent with the location of these two regions
long-term modulation of energy balance, showing that neuron- close to the third cerebral ventricle and with their high insulin recep-

articles
Fig. 1. Schematic representation of the experimental procedures. a

(a) Surgical implantation of ICV cannulae was performed on day 1 (∼3 Day 1 7 14 21
weeks before the in vivo study). Body weight and food intake recovered
fully by day 7. Intravenous (i.v.) catheters and osmotic minipumps (for ICV infusions
ICV infusions of ODNs) were surgically implanted on day 14; ICV infu- ICV Recovery IV catheters Fat mass
sions were continued for 7 d; and body composition and insulin action cannulae and ICV pump and clamp
were estimated on day 21. (b) A bolus of tritiated water was given i.v.
3 h before the start of tritiated glucose infusion. At t = 0, a primed-con- b
tinuous infusion of labeled glucose was initiated and maintained for the –180 min 0 120 240
remainder of the 4 h study. Pancreatic-insulin clamp study was initiated

at t = 120 min and lasted 120 min. SRIF, somatostatin.
[3 – 3H] glucose
3H O
2
SRIF
bolus Insulin
tor densities. In contrast, other hypothalamic areas, the olfactory Glucose
cortex and the hippocampus were not affected (Table 1 and Fig. 2g,
h, k–p). We confirmed the selective inhibition of IR protein in the
arcuate nucleus by immunoblot analysis of hypothalamic nuclei firmed that copy number was not affected by IR antisense treatment
isolated by the micro-punch technique. Arcuate nuclei of rats treat- (Fig. 3d). To rule out the presence of antisense-generated stable
ed with IR antisense ODN for 3 or 7 days contained ∼46% ± 9 and mRNA fragments, we quantified the IR mRNA with two separate
∼53% ± 13, respectively, of IR protein as compared to controls (Fig. sets of primers that either spanned the sequence of mRNA targeted
3b lanes 5–8 and Fig. 3c). In contrast, and in agreement with the by the antisense ODN (Fig. 3d, ‘upstream primers’) or were located
immunostaining results, the immunoblot of the lateral hypothal- further downstream of the start codon (Fig. 3d, ‘downstream
amic area (LHA) and paraventricular nucleus (PVN) did not show primers’). The results obtained with both sets of primers confirmed
any significant change in IR protein levels (Fig. 3b lanes 1–4). that the IR mRNA levels were not changed by IR antisense treat-
We further characterized the mode of action of the IR antisense ment. These findings support the conclusion that the downregula-
ODN on IR protein through real-time PCR. The IR antisense ODN tion of IR protein in selective arcuate neurons occurs largely by a
was designed to target the translation initiation site of the IR mRNA mechanism of ‘hybrid-arrested’ translation and does not involve
(Fig. 3a). We assessed the IR mRNA copy number in hypothalamic ribonuclease H–dependent degradation of mRNA21.
nuclei isolated by micro-punches through real-time PCR, and con-
AGRP and NPY expression in arcuate
To determine whether an attenuation of insulin
action in the hypothalamus affects the expression
of neuropeptides involved in energy balance, we
examined the expression of NPY, agouti-related
protein (Agrp) and POMC in the arcuate nuclei of
the rats treated with IR antisense and control ODN
by real-time PCR (Fig. 3e). IR antisense treatment
a b c d
Fig. 2. Representative images of brain insulin-receptor
immunostaining. Digitized images were produced by confo-
cal fluorescence microscopy of immunocytochemical stain-
ing for insulin receptors in brains of rats that received third
ventricular infusions of oligonucleotides for 7 d. First and
third columns (a, c, e, g, i, k, m, o; first and third columns),
images from rats that received control (scrambled) oligonu-
cleotides (IR-SCR); second and fourth columns (b, d, f, h, j,
l, n, p), images from rats that received oligonucleotides com-
e f g h plementary to IR mRNA (IR-AS). Neuronal cell bodies are
shown in red pseudocolor (arrows in the first panel), repre-
senting fluorescence of the Cy3-labeled secondary antibod-
ies used for detection of antibody binding. The abundance of
neurons expressing immunocytochemically detectable
insulin receptors was visually lower in the medial region of
the arcuate nucleus (ARCm) adjacent to the third ventricle
(V) in rats that received IR-AS ODN (b). This effect was not
seen in rat that received the control ODN (a) or in the ven-
i j k l trolateral region of the arcuate nucleus (ARCv) (compare c,
d). The abundance of IR-immunoreactive neurons after IR
antisense treatment was also lower in the periventricular
nucleus (PeVN) (compare e, f) and habenula (HAB) (com-
pare i, j). No effect of IR antisense on IR immunostaining was
detected in the ventromedial hypothalamic nucleus (VMN)
(g, h) and dorsomedial hypothalamic nucleus (DMN) (k, l)
or in pyramidal cells of the pyriform cortex (PCX) (m, n)
and region CA1 of the hippocampus (CA1) (o, p). Scale bar,
m n o p 100 µm (applies to all panels).

articles
Table 1. Immunostaining of brain regions from rats infused ICV with IR antisense tracer experiments and post-
(IR-AS) or scrambled (IR-SCR) oligodeoxynucleotides. mortem dissection of fat
Rat ARCmed ARClat PeVN VMH DMH HAB CA1 OCX CPIIIa depots22,23. The selective decrease
IR-AS
in hypothalamic IR protein in rats
receiving IR antisense ODN result-
1 6 30 2 25 9 6 35 41 0.3
ed in rapid onset of hyperphagia
3 7 26 7 28 17 17 38 46 0.3
(average food intake 25.1 ± 1.2 ver-
6 4 21 4 21 5 5 38 37 0.3 sus 16.2 ± 1.3 g/day; P < 0.01) and
7 4 17 5 17 13 2 36 0.0 markedly increased fat mass (Fig. 4

Mean 5.3 23.5 4.5 22.8 11.0 7.5 37.0 40.0 0.2 and Table 2). In contrast, the feed-
s.e.m. 0.6 2.5 0.9 2.1 2.2 2.8 0.7 2.0 0.1 ing behavior of rats treated ICV
IR-SCR with control ODN was quite simi-
2 17 21 9 18 11 36 37 45 2.6 lar to that of rats treated ICV with
4 25 21 12 22 14 43 32 45 3.0 vehicle (ref. 24 and data not
shown). Notably, this occurred
5 24 24 6 26 10 22 32 46 2.3
despite a 65% increase in plasma
8 23 16 11 25 16 34 25 41 2.6
leptin levels (Table 2). Plasma lep-
Mean 22.3 20.5 9.5 22.8 12.8 33.8 31.5 44.3 2.6 tin concentrations were also
s.e.m. 1.6 1.4 1.1 1.6 1.2 3.8 2.1 1.0 0.1 increased by ∼50% in rats treated
t-test ICV with IR antisense ODN for
P= 0.001 0.405 0.027 1.000 0.578 0.004 0.109 0.162 0.0001 only 3 days (data not shown). Lean
aData for choroid plexus represents a staining intensity score, on a 4-level scale from 0 (no staining) to 3
body mass was not affected, and the
(most intense staining seen).
increase in fat mass was largely
ARCmed, arcuate nucleus medial portion, adjacent to third ventricle, region of NPY neurons; ARClat, ventrolateral accounted for by a marked increase
arcuate nucleus, away from ventricle and towards the region of predominately POMC neurons; PeVN, periventricu- in subcutaneous adipose tissue
lar nucleus near the VMH, cells near third ventricle; VMH, main portion of ventromedial hypothalamus; DMH, main
portion of dorsomedial nucleus; HAB, habenular nucleus, adjacent to third ventricle at top of third ventricle cleft; (Fig. 4); intra-abdominal fat was
CA1, region of hippocampus with pyramidal cells; OCX, olfactory (pyriform) cortex; CPIII, choroid plexus in the increased only modestly (4.3 ± 0.5
third ventricle. ICV IR-AS significantly decreased IR protein in ARCmed, PeVN, HAB and CPIII. versus 3.7 ± 0.4 g; P = 0.09).
for 3 days increased the prevalence of NPY (by 53% ± 13 versus Hypothalamic and hepatic insulin action
SCR) and AGRP (by 57% ± 10 versus SCR) transcripts, but did To assess the impact of hypothalamic insulin receptor downreg-
not affect the expression of POMC. Thus, the selective down- ulation on the action of insulin, we performed insulin-clamp
regulation of IR protein in the medial portion of the arcuate (3 mU/kg min) studies in conscious rats23–26. As a result of phys-
nucleus specifically increases the levels of orexogenic peptides iological increases in plasma insulin concentrations (to ∼40
NPY and AGRP. µU/ml), the rate of glucose infusion required to maintain the
plasma glucose at basal levels was lower in IR antisense–treated
Hypothalamic insulin, body fat and food intake (8.7 ± 1.6) than in control (14.9 ± 1.4 mg/kg min) rats (Fig. 5).
Body composition and fat distribution have been assessed by Insulin action on glucose metabolism includes stimulation of
Fig. 3. Effects of IR antisense in discrete nuclei of the

hypothalamus. IR antisense ODN (IR-AS) attenuates
insulin receptor translation only in the arcuate nucleus, +1 +4496
but does not affect its mRNA levels. (a) Schematic illus- a AUG UAA
tration of the IR mRNA. AUG, putative initiation of
translation of the IR precursor protein at position +1; IR mRNA AS
UAA, putative stop codon (as predicted by the IR Upstream primers Downstream primers
receptor sequence deposited with GenBank; see
b c 140
Percentage of control
Methods); AS, sequence of the antisense ODN 1 2 3 4

hybridizing to the area spanning the start of translation. 120
IR 100
Upstream and downstream primers were used to quan- *
GAPDH 80
tify IR mRNA expression by RT-PCR. (b) Immunoblot *
of distinct hypothalamic nuclei after treatment with IR- 5 6 7 8 60
AS. Lanes 1, 3, 5, 7, IR-SCR treatment; lanes 2, 4, 6, 8, IR 40
IR-AS treatment. Lanes 1 and 2 are from LHA and lanes GAPDH 20
3 and 4 from PVN. Lanes 5–8 are from arcuate nuclei 0
after 3 d (5 and 6) or 7 d (7 and 8), respectively, of anti-
d 3 days 7 days
Relative copy number
120
sense treatment. (c) Quantification of immunoblots by
e
Relative copy number
densitometry. IR protein levels of arcuate nuclei after 3

100
80
200 * *
or 7 d of antisense treatment (, IR-AS) or control 60 150
ODN (, IR-SCR); P < 0.04. (d) IR mRNA quantifica- 40
tion by RT-PCR in arcuate nuclei or LHA of rats treated 20 100

0
with control (, IR-SCR) or antisense ODN (, IR- 50
Arcuate LHA Arcuate LHA
AS). (e) Expression of NPY, AGRP and POMC in arcu-
ate nuclei after control (, IR-SCR) or antisense ODN Upstream primers Downstream primers 0
treatment (, IR-AS); P < 0.01. NPY AGRP POMC

articles
a Fig. 4. Selective attenuation of hypothalamic insulin receptors increases

30 food intake and fat mass. The effects of 7 d ICV administration of IR
*
Daily food
* antisense (, IR-AS) and control (scrambled; , IR-SCR) ODNs on (a)
intake (g)
20 *
daily food intake and on (b) subcutaneous and (c) visceral fat mass are
10 shown. Treatment with IR-AS led to increased food intake (P < 0.01)
0 and subcutaneous fat mass (P < 0.01), whereas the modest increase in
2 3 4 5 6 visceral adiposity was not statistically significant.
Days after ODN
administration
b
120
* in the regulation of energy homeostasis. However, the effects of
Subcutaneous
100
insulin on neuronal growth and differentiation16,17 suggest that
80
fat (g)
altered brain insulin signaling during development may account

60
40
for some of insulin’s biological consequences. Furthermore, the
20
role of insulin receptors in specific brain regions has been difficult
0 to examine. We describe here a new approach to this question.
IR-SCR IR-AS Central infusion of antisense ODN directed against the insulin
receptor generated a rapid and site-specific decrease in the expres-
c 5 sion of insulin receptors. By coupling stereotaxic delivery with
brain immunostaining and immunoblotting, we documented
Visceral fat (g)
4
3 that infusion of IR antisense ODN into the third cerebral ventri-
2 cle led to a marked decrease in IR protein in nuclei directly adja-
cent to the third ventricle, while insulin receptor expression in
1
other brain areas was not affected. This is a useful experimental
0
model that may make it possible to explore additional effects of
IR-SCR IR-AS
hypothalamic insulin receptors on reproductive function 20,
counter-regulation22 and insulin secretion23. We confirmed the
results of the immunohistochemistry studies with an indepen-
glucose uptake and inhibition of glucose production. Insulin dent quantification technique that combines micro-punching of
action on peripheral glucose uptake was similar in IR anti- selective hypothalamic nuclei with immunoblotting. Isolated
sense–treated and control rats (19.1 ± 2.0 versus 20.6 ± 1.2 mg/kg arcuate nuclei treated with IR-antisense ODN show ∼50% less
min; Fig. 5 and Table 2). Basal rates of glucose production in the IR protein after either 3 or 7 days of IR antisense treatment. This
two groups were also similar. Physiological increases in plasma decrease in IR protein is consistent with the results we obtained
insulin concentration markedly inhibited (by ∼55%, to 5.6 ± 0.6 by quantification of immunostaining, in which the proportion
mg/kg min) the rate of glucose production in the control rats. In of cells staining positive for IR protein was markedly decreased
contrast, glucose production was decreased by only 25% (to 9.6 (by ∼80%) in the medial portion but not in the lateral aspect of
± 0.6 mg/kg min) in IR antisense–treated rats. Thus, hepatic the arcuate nucleus.
insulin action was markedly impaired after selective attenuation The medial aspect of the arcuate nucleus of the hypothala-
of hypothalamic insulin receptor expression. mus showed ∼80% decrease in the number of neurons that con-
tained immunoreactive insulin receptors, as detected by
DISCUSSION immunostaining. This region is enriched in neurons containing
Insulin receptors are expressed in several regions of the central the orexogenic peptide NPY, whose expression is decreased after
nervous system11,12. Initial analyses have identified pleiotropic systemic or central administration of insulin10,13,14. Furthermore,
functions of brain insulin receptors 10,16–19 , and
experimental evidence supports an important role b
of central insulin receptors in the regulation of ener- a
20 12
Glucose infusion rate
gy homeostasis 1–3,10,13,20. The recent observation that *

Glucose production
10
mice with neuron-specific disruption of the IR gene 15
(mg/kg min)
(mg/kg min)
throughout the brain have increased fat mass and 8

insulin levels as adult20 demonstrates that insulin 10 * 6
receptors within the CNS have a physiological role 4
5
2
0 0
Fig. 5. Selective attenuation of hypothalamic insulin recep- IR-SCR IR-AS IR-SCR IR-AS
tors impairs hepatic insulin action. The effects of 7-day ICV c
administration of IR antisense (, IR-AS) and IR control 70 d 25
Glucose production
(scrambled; , IR-SCR) ODNs on glucose kinetics during

(% inhibition from
60
20
Rate of glucose
the pancreatic-insulin clamp studies are displayed.

disappearance
50
baseline)
(mg/kg min)
Treatment with IR-AS resulted in lower rates of glucose 40 15

infusion (P < 0.01; a) and greater endogenous glucose pro-
duction (P < 0.01; b) as compared to treatment with IR-
30 * 10
SCR. The percentage inhibition of glucose production in 20
5
response to hyperinsulinemia was markedly blunted after 10
ICV IR-AS administration (c), whereas the rate of glucose 0 0
disappearance was not significantly affected (d). IR-SCR IR-AS IR-SCR IR-AS

articles
these NPY-containing neurons co-express a natural antagonist Table 2. Effect of selective decrease of hypothalamic
of the central melanocortin receptors, Agrp, which is also an orex- insulin receptors on food intake, body composition and
ogenic peptide. The expression of both peptides is increased by blood chemistry.
∼50% in the arcuate nuclei of IR antisense–treated rats. This IR-SCR IR-AS P value
moderate increase occurred despite concomitant elevations in Basal
food intake and leptin levels. These findings suggest that insulin
Cumulative food intake
signaling is required to restrain the expression of these orexo- (kcal/5 d) 291 ± 33 443 ± 66* 0.0007
genic peptides in the arcuate nucleus. The rapid hyperphagic BW (g) 297.9 ± 8.5 319.3 ± 19 0.3
effect seen here validates an important physiological role of

∆BW (g) –15.3 ± 5.9 +9.6 ± 2.9* 0.004
insulin receptors within the medial aspect of the arcuate nucleus
FM (g) 40.3 ± 5.5 75.1 ± 15.0* 0.007
in the regulation of food intake and body adiposity. Because the
survival surgery and general anesthesia required for implanta- FFM (g) 257.7 ± 6.1 244.3 ± 8.4 0.3
tion of the indwelling catheters and osmotic minipumps induced Leptin (ng/ml) 0.9 ± 0.1 1.5 ± 0.1* 0.02
a brief period of fasting, in the present study we examined the Insulin (µU/ml) 0.75 ± 0.2 1.72 ± 0.5 0.1
impact of hypothalamic insulin signaling during early re- Glucose (mM) 8.0 ± 0.4 7.8 ± 0.2 0.2
feeding. The pronounced effect of hypothalamic insulin recep- GP (mg/kg min) 12.3 ± 0.5 12. 6 ± 0.8 0.9
tor status in this phase seems to indicate that under normal cir- Insulin clamped
cumstances the increase in plasma insulin levels during re-feeding
Glucose (mM) 7.5 ± 0.3 7.9 ± 0.2 0.7
partially restrains food intake. This observation is also consistent
Insulin (µU/ml) 38.6 ± 4 .6 36 ± 7.2 0.8
with the postulated role of NPY in mediating the hyperphagic
response to fasting3,13,14. *, P < 0.01. BW, body weight; FM, fat mass; FFM, fat-free mass; GP, endoge-
Hyperphagia was associated with a marked increase in sub- nous glucose production.
cutaneous adipose tissue that entirely accounted for the increased
body weight. The preferential increase in fat mass over lean body
mass suggests a role of hypothalamic insulin receptors in regu- ceral adiposity, the effect of hypothalamic insulin receptors seems
lating energy storage. Notably, these effects on food intake and to be independent of this parameter.
body composition occurred in the presence of moderate increas- Understanding the mechanism(s) responsible for suscepti-
es in the plasma leptin concentration. Although leptin and insulin bility to weight gain and insulin resistance in response to
signaling may share common downstream targets (such as phos- increased caloric intake is of vital importance to the fight against
phoinositide-3-kinase)27 in some hypothalamic neurons, it is prevalent metabolic diseases such as obesity and type 2 diabetes
also possible that the increase in food intake and fat mass results mellitus. The observation that a selective decrease in insulin
in enhanced hypothalamic leptin signaling. This in turn may receptor expression in a discrete region of the hypothalamus is
partly antagonize the deficient insulin signaling and may stimu- sufficient to rapidly induce several key features of these meta-
late the melanocortin pathway, perhaps through increased expres- bolic syndromes indicates that selective hypothalamic insulin
sion of POMC. The latter effect may account for the lack of resistance may be involved in the pathophysiology of obesity and
increase in visceral fat mass24 and for the normalization of food insulin resistance. In conclusion, we show that a selective and
intake on day 5. partial decrease in IR protein in arcuate neurons containing NPY
Selective decrease in hypothalamic insulin receptors also and Agrp is sufficient to rapidly and markedly impair energy bal-
results in the rapid onset of hepatic insulin resistance. In fact, ance and insulin action. This observation suggests that hypo-
the rate of hepatic glucose production was markedly increased in thalamic insulin resistance may have a role in the susceptibility
the presence of equal plasma insulin concentrations. This may to obesity and diabetes mellitus.
be partly due to the moderate increase in whole body adiposity
and/or to the transient increase in nutrient intake. In previous METHODS
studies, however, we have not found a significant effect of sim- Design of oligodeoxynucleotide (ODN) antisense against IR mRNA.
ilar changes in caloric intake and fat mass per se on hepatic The antisense ODN (IR-AS) was designed to hybridize to the sequence
insulin action28. We have recently reported that bidirectional spanning the initiation of translation of the IR precursor mRNA
changes in the activity of the melanocortin pathway in the hypo- (Fig. 3a). The sequence from IR-AS ODN (5′-CGGAGCCCATAGCAG-
3′) was scrambled to obtain a control ODN, IR-SCR (5′-CACACGAGC-
thalamus rapidly regulate fat distribution and insulin action24. In CGTAGG-3′). The ODNs were synthesized by Operon Technologies
particular, antagonism of hypothalamic melanocortin receptors (Alameda, California). All ODNs contained a phosphothiorate bond
induces hyperphagia, weight gain, selective increase in visceral between nucleotides 1 and 2 and between nucleotides 14 and 15.
adiposity and insulin resistance. Here, antagonism of hypothal-
amic insulin signaling also resulted in hyperphagia, weight gain Primers and real-time PCR. Total RNA was obtained from individual rats
and insulin resistance. Notably, the induction of hypothalamic by combining left and right homologous nuclei (that is, arcuate nucleus
insulin resistance did not selectively increase visceral adiposity. and PVN). Four rats were analyzed per experimental group. IR mRNA
Taken together, these recent studies support the notion that com- was quantified using two sets of primers (Fig. 3a). The upstream set ampli-
mon hypothalamic pathways are involved in the regulation of fies a fragment of 191 nucleotides beginning at the start codon of IR
energy homeostasis and insulin action. Because glucose pro- mRNA (forward primer (F), 5′-CCTACTGCTATGGGCTCCG-3′; reverse
primer (R), 5′-AGGATCTGCAGATGGCCCTC-3′). The downstream set
duction by the liver is the major source of endogenous fuel, it is
amplifies a fragment of 491 nucleotides beginning at codon 276 of the IR
particularly intriguing that two central neural circuitries involved mRNA (F, 5′-CAGGACTGGCGCTGTGTAAAC-3′; R, 5′-CACAGCTGC-
in the regulation of exogenous fuel (food intake) concomitant- CTCAGGTTCTG-3′). Hypothalamic neuropeptide expression was car-
ly modulate the rate of endogenous glucose production as well. ried out beginning with real-time PCR using the following primers: rat
Although the effect of central melanocortin receptors on glu- NPY (F 5′-GCCATGATGCTAGGTAACAAACG-3′, R 5′-GTTTCATTTC-
cose production are probably due to their potent effects on vis- CCATCACCACATG-3′); rat POMC (F 5′-CCAGGCAACGGAGATGAAC-

articles
3′, R 5′-TCACTGGCCCTTCTTGTGC-3′); rat AGRP (F 5′-GCCATGCT- sion filter setting of 582–625 nm. Each field recorded was scanned as a
GACTGCAATGTT-3′, R 5′-TGGCTAGGTGCGACTACAGA-3′); and β- series of 10 z planes that were separated by 1 µm for a total thickness of
actin (F 5′-TGAGACCTTCAACACCCCAGCC-3′, R 5′-GAGTACTT 10 µm. The final image at each z plane was frame-averaged three times
GCGCTCAGGAGGAG-3′). Total RNA was isolated with Trizol (Invitro- and all 10 z planes were projected into a single image for analysis and
gen, Carlsbad, California) and single-strand cDNA was synthesized with recording as tiff files. All images were captured using the same pinhole,
Superscript (Invitrogen). Real-time PCR reactions were prepared with a laser intensity and photomultiplier tube sensitivity settings.
LightCycler reaction kit (Roche, Indianapolis, Indiana). A real-time PCR Images for quantification (Table 1) were captured with a Hamamatsu
reaction of 20 µl contained 200 nM primers, 1× reaction buffer, 2.3 mM (Bridgewater, New Jersey) cooled CCD camera and the MCID image
MgCl2, 2 µl SYBR Green, 2 µl cDNA and 2 units Taq DNA polymerase. analysis program (Imaging Research, St. Catharines, Ontario, Canada).
The reactions were carried out in capillaries in a LightCycler instrument The data shown in Table 1 for all brain areas (except choroid plexus) rep-
(Roche) and were cycled ∼40 times. resent mean counts of IR-immunopositive neurons on three sections of
each brain region, measured in 400× fields. The data from the three slides
Induction of selective attenuation of hypothalamic insulin receptor were averaged to obtain a mean value per rat (n = 4 per group). For the
expression. We studied 32 10-week male Sprague-Dawley rats (Charles arcuate nucleus, ‘medial’ images and counts were obtained adjacent to
River Breeding Laboratories, Wilmington, Massachusetts). Three weeks the third ventricle, whereas the ‘lateral’ images and counts were made in
before brain harvesting or in vivo studies, we placed a chronic catheter the ventrolateral region of the arcuate nucleus.
in the third cerebral ventricle by stereotaxic surgery25. One week before
all experimental protocols were performed, we placed additional Brain stereotaxic micro-punches of individual hypothalamic nuclei.
catheters in the right internal jugular and left carotid artery24–26,28–32 Brain micro-punches of individual hypothalamic nuclei were prepared
and connected osmotic minipumps (Alzet, Palo Alto, California) to by a modification of a method previously described36. Briefly, rats were
the indwelling catheter placed in the third cerebral ventricle (ICV). killed by decapitation and the brains rapidly removed and frozen in
The minipump delivered a constant infusion of 0.5 µl/h for 7 d. Food isopentane on dry ice at –15°C for 5 min. The brain were then implant-
intake was monitored daily. Rats were randomized into two experi- ed frozen to a pedestal, placed in the cryostat and maintained at –15°C.
mental groups. Each group of rats received 3 or 7 d of infusion of either Brain sections 500 µm thick were made and mounted on glass slides.
IR-AS ODN or IR-SCR (control). All ODNs were purified by HPLC Using anatomical landmarks from a rat brain stereotaxic atlas33, indi-
and dissolved in artificial cerebrospinal fluid at a concentration of vidual hypothalamic nuclei were identified and punched out with a stain-
2 mM. The solutions were infused ICV for 3 or 7 d with osmotic less steel needle under a stereomicroscope. The brain punch was expelled
minipumps at a rate of 1 nmol/h. with a stainless steel insert into an Eppendorf tube on dry ice and then
stored at –80°C until assayed. The reproducibility of micro-punches was
Immunostaining of rat brains. Immunostaining was performed after 7 d established by thawing the sections and inspecting the topography of the
of administration of either IR-AS or IR-SCR ODNs. Rat brains were per- holes by transillumination under a low-power light microscope. Addi-
fused in situ through a cardiac cannula with 100 ml of 4% paraformalde- tionally, the complete removal of arcuate nuclei was validated by mea-
hyde in 0.1 M sodium phosphate buffer (PB), pH. 7.2. After removal suring arcuate-specific genes (POMC and AGRP) in punches immediately
from the cranium, the brains were immersed in PB containing 25% lateral to the arcuate nuclei. In the latter samples, these transcripts were
sucrose for 48 h, then frozen and sectioned at 14 µm with a cryostat. undetectable by highly sensitive real-time PCR.
Slide-mounted coronal sections of the hypothalamus were collected
between stereotaxic levels –2.2 and –4.0 mm caudal to bregma33, which Western blot analysis. Protein analysis was performed on individual rats
included the region targeted by the injector cannula tip. by combining left and right homologous nuclei (that is, arcuate nucleus
and PVN). Each sample was mixed with 100 µl of Laemmli buffer
Immunocytochemistry. Slides containing sections of hypothalamus at the (50 mM Tris-HCl, pH 6.8, 100 mM dithiothreitol, 2% SDS, 0.1% bro-
level of the cannula tip, as well as up to 1 mm anterior and posterior to it, mophenol blue, 10% glycerol) and immediately boiled for 10 min. The
were immunostained by standard procedures34. Briefly, slides were samples were vortexed and centrifuged for 10 min at 4°C and 5,000 g, and
immersed 1 h at 20°C in 0.01 M PB containing 5% normal goat serum and the supernatants were fractionated on 10% SDS-PAGE and blotted onto
1% BSA, followed by anti-IR primary antiserum diluted 1:1,000 in 0.1 M nitrocellulose. Filters were incubated with antibodies against the C terminal
PB with 1% BSA overnight at 6°C. The rabbit polyclonal primary anti- of the IR β-subunit (Transduction, San Diego, California) and against glyc-
serum (provided by J.N. Livingston, Bayer Corp.) was generated against eraldehyde phosphate dehydrogenase (anti-GAPDH, a gift of E.R. Stanley).
the C-terminal 14-amino-acid sequence of the human IR β-
subunit. The specificity of the antiserum for rat IR was verified by immuno- Body composition. On day 8 after the start of ICV infusions, we initiated
in vivo studies in conscious rats fasted for ∼6 h24–26,28–30. To ensure a sim-
precipitation and western immunoblots. The antiserum recognizes bands
ilar postabsorptive state, on the night preceding the clamp study we
of appropriate molecular size for IR β-subunits respectively in immunoblots
assigned a fixed allotment of chow (∼12 g) to all rats.
of rat liver and hypothalamus homogenates separated by SDS-PAGE. The
We administered an intra-arterial bolus injection of 20 µCi of tritiat-
protein immunoprecipitated from brain homogenates and detected by the
ed water (3H2O; New England Nuclear, Boston) to the rats, and obtained
antiserum migrated on the gel slightly more quickly than that from liver, plasma samples at 30 min intervals for 3 h26,28. Steady-state conditions for
consistent with the smaller molecular size of this subunit in brain tissue. plasma 3H2O specific activity were achieved within ∼40 min in all stud-
In addition, hepatic portal vein injection of insulin resulted in the tyro- ies. Five plasma samples obtained between 1 and 3 h after the injection
sine phosphorylation of the protein immunoprecipitated by this antiserum, were used in the calculation of the whole-body distribution space of
consistent with the known behavior of the IR β-subunit tyrosine kinase water. This was obtained by dividing the total radioactivity injected (in
(data not shown). After incubation with primary antiserum, sections were d.p.m.) by the steady-state specific activity of plasma water (in d.p.m./ml).
washed in PBS and immersed 1 h at 20°C in goat anti-rabbit IgG conju- Plasma was assumed to be 93% water. Fat-free mass was calculated as
gated with Cy3 (Jackson ImmunoResearch Laboratories, West Grove, Penn- the whole-body water distribution space divided by 0.73, and fat mass
sylvania) following standard protocols34. Controls included samples with as the difference of body weight and fat-free mass. Epididymal, peri-renal
(i) primary antibody omitted, (ii) secondary antibody omitted and (iii) and omental fat depots were dissected and weighed at the end of each
normal rabbit serum (diluted 1:1,000) substituted for the rabbit anti-insulin experiment. Visceral adiposity was calculated as the sum of epididymal,
antibody35. Immunofluorescence was absent in all controls. peri-renal and omental fat depots and subcutaneous adiposity as the dif-
ference between total fat mass and visceral adiposity.
Microscopy and analysis. Images of immunostained brain sections
(Fig. 2) were captured with a Leica (Solm, Germany) TCS-SP Confocal Measurements of in vivo glucose kinetics. Briefly, a primed-continuous
Laser Scanning Microscope. Images were scanned with a 40× objective infusion of HPLC-purified [γ-3H]glucose (New England Nuclear, Boston;
with a krypton laser at 568 nm, using a 150-µm pinhole and an emis- 40 µCi bolus, 0.4 µCi/min) was administered for the duration of the
articles
study24–26,28–30. Two hours after the basal period, a primed-continuous distributed in the central nervous system of the rat. Nature 272, 827–829
(1978).
infusion of somatostatin (1.8 µg/kg min) and regular insulin (3 mU/kg
12. Baskin, D. G., Figlewicz, D. P., Woods, S. C., Porte, D. Jr. & Dorsa, D. M.
min) were administered, and a variable infusion of a 25% glucose solu- Insulin in the brain. Annu. Rev. Physiol. 49, 335–347 (1987).
tion was started at time zero and periodically adjusted to clamp the plasma 13. Schwartz, M. W. et al. Inhibition of hypothalamic neuropeptide Y gene
glucose concentration at ∼7.5 mM for the rest of the study. Samples for expression by insulin. Endocrinology 130, 3608–3616 (1992).
determination of [3H]glucose specific activity were obtained at 10-min 14. Air, E.L. et al. Small molecule insulin mimetics reduce food intake and body
weight and prevent development of obesity. Nat. Med. 8, 179–183 (2002).
intervals throughout the infusions. 15. Schwartz, M. W. et al. Central insulin administration reduces neuropeptide Y
All values are presented as the mean ± s.e.m. Comparisons among mRNA expression in the arcuate nucleus of food-deprived lean (Fa/Fa) but
groups were made by unpaired Student’s t-test. The study protocol was not obese (fa/fa) Zucker rats. Endocrinology 128, 2645–2647 (1991).
reviewed and approved by the Institutional Animal Care and Use Com- 16. Heidenreich, K. A. Insulin and IGF-I receptor signaling in cultured neurons
mittee of the Albert Einstein College of Medicine. Ann. N. Y. Acad. Sci. 692, 72–88 (1993).
17. Robinson, L. J., Leitner, W., Draznin, B. & Heidenreich, K. A. Evidence that
p21ras mediates the neurotrophic effects of insulin and insulin-like growth
GenBank accession number. Insulin receptor, 204953. factor I in chick forebrain neurons Endocrinology 135, 2568–2573 (1994).
18. Wan, Q. et al. Recruitment of functional GABAA receptors to postsynaptic
domains by insulin. Nature 388, 686–688 (1997).
Acknowledgments 19. Jonas, E. et al. Regulation by insulin of a unique neuronal Ca2+ pool and of
We thank J.N. Livingston for the gift of the IR antiserum and J. Murphy for neuropeptide secretion. Nature 385, 343–347 (1997).
20. Bruning, J. C. et al. Role of brain insulin receptor in control of body weight
performing the immunostaining. This work was supported by grants from the and reproduction. Science 289, 2122–2125 (2000).
National Institutes of Health to L.R. (DK 48321 and DK 45024), from the AECOM 21. Chiang, M. Y., Chan, H., Zounes, M. A., Freier, S. M., Lima, W. F. & Bennett, C.
Diabetes Research & Training Center (DK 20541), the Union of Washington F. Antisense oligonucleotides inhibit intercellular adhesion molecule 1
expression by two distinct mechanisms. J. Biol. Chem. 266, 18162–18171
Diabetes Endocrinology Research Center (DK 17047) and the Career Scientist and (1991).
Merit Review Research Programs of the Department of Veteran Affairs. S.O. 22. Davis, S. N. et al. Evidence that the brain of the conscious dog is insulin
received a postdoctoral fellowship from the American Diabetes Association. sensitive. J. Clin. Invest. 95, 593–602 (1995).
23. Luzi, L. et al. Lack of feedback inhibition of insulin secretion in denervated
human pancreas. Diabetes 41, 1632–1639 (1992).
Competing interests statement 24. Obici, S. et al. Central melanocortin receptors regulate insulin action. J. Clin.
The authors declare that they have no competing financial interests. Invest. 108, 1079–1085 (2001).
25. Liu, L. et al. Intracerebroventricular leptin regulates hepatic but not
peripheral glucose fluxes. J. Biol. Chem. 273, 31160–31167 (1998).
RECEIVED 25 MARCH; ACCEPTED 23 APRIL 2002 26. Barzilai, N., Massilon, D., Vuguin, P., Hawkins, M. & Rossetti, L. Leptin
selectively decreases visceral adiposity and enhances insulin action. J. Clin.
Invest. 100, 3105–3110 (1997).
1. Woods, S. C., Lotter, E. C., McKay, D. L. & Porte, D. Jr. Chronic 27. Niswender, K. D., Morton, G. J., Stearns, W. H., Rhodes, C. J., Myers, M. G. Jr.
intracerobroventricular infusion of insulin reduces food intake and body & Schwartz, M. W. Intracellular signalling. Key enzyme in leptin-induced
weight of baboons. Nature 282, 503–505 (1979). anorexia. Nature 413, 794–795 (2001).
2. Woods, S. C., Stein, L. J., McKay, D. L., & Porte, D. Jr. Suppression of food 28. Barzilai, N. et al. Decreased visceral adiposity accounts for leptin effect on
intake by intravenous nutrients and insulin in the baboon. Am. J. Physiol. hepatic but not peripheral insulin action. Am. J. Physiol. 277, E291–E298
247, R393–R401 (1984). (1999).
3. Schwartz, M. W., Figlewicz, D. P., Baskin, D. G., Woods, S. C. & Porte, D. Jr. 29. Rossetti, L. & Giaccari, A. relative contribution of glycogen synthesis and
Insulin in the brain: a hormonal regulator of energy balance. Endocrin. Rev. glycolysis to insulin-mediated glucose uptake. A dose-response euglycemic
13, 387–414 (1992). clamp study in normal and diabetic rats. J. Clin. Invest. 85, 1785–1792
4. Kennedy, G. C. The role of the depot fat in the hypothalamic control of food (1990).
intake in the rat. Proc. R. Soc. London B Biol. Sci. 140, 578–596 (1953). 30. Rossetti, L., Smith, D., Shulman, G. I., Papachristou, D. & DeFronzo, R. A.
5. Coleman, D. L. Obese and Diabetes: two mutant genes causing Correction of hyperglycemia with phlorizin normalizes tissue sensitivity to
diabetes–obesity syndromes in mice. Diabetologia 14, 141–148 (1978). insulin in diabetic rats. J. Clin. Invest. 79, 1510–1515 (1987).
6. Zhang, Y., Proenca, R., Maffei, M., Barone, M., Leopold, L. & Friedman, J. M. 31. Wang, J., Liu, R., Hawkins, M., Barzilai, N. & Rossetti, L. A nutrient sensing
Positional cloning of the mouse obese gene and its human homologue. pathway regulates leptin gene expression in muscle and fat. Nature 393,
Nature 372, 425–432 (1994). 684–688 (1998).
7. Schwartz, M. W. et al. Central nervous system control of food intake. Nature 32. Wang, J. et al. The effect of leptin on Lep expression is tissue-specific and
404, 661–719 (2000). nutritionally regulated. Nat. Med. 5, 895–899 (1999).
8. Frederich, R. C., Hamman, A., Anderson, S., Lollmann, B., Lowell, B. B. & 33. Paxinos, G. & Watson, C. The Rat Brain in Stereotaxic Coordinates 3rd edn.
Flier, J. S. Leptin levels reflect body lipid content in mice: evidence for diet- (Academic Press, San Francisco, 1997).
induced resistance to leptin action. Nat. Med. 1, 1311–1314 (1995). 34. Baskin, D. G. et al. Leptin receptor long-form splice-variant protein
9. Campfield, L. A., Smith, F. J., Guisez, Y., Devos, R. & Burn, P. Recombinant expression in neuron cell bodies of the brain and co-localization with
mouse OB protein: evidence for a peripheral signal linking adiposity and neuropeptide Y mRNA in the arcuate nucleus. J Histochem. Cytochem. 47,
central neural networks. Science 269, 546–549 (1995). 353–362 (1999).
10. Sipols, A. J., Baskin, D. G. & Schwartz, M. W. Effect of intracerebroventricular 35. Burry, R. W. Specificity controls for immunocytochemical methods.
insulin infusion on diabetic hyperphagia and hypothalamic neuropeptide J. Histochem. Cytochem. 48, 163–166 (2000).
gene expression. Diabetes 44, 147–151 (1995). 36. Palkovits, M. Isolated removal of hypothalamic or other brain nuclei of the
11. Havrankova, J., Roth, J. & Brownstein, M. Insulin receptors are widely rat. Brain Res. 59, 449–450 (1973).

articles
Calcium–calmodulin-dependent
protein kinase IV is required for
fear memory
Feng Wei1, Chang-Shen Qiu1, Jason Liauw1, Daphné A. Robinson1, Nga Ho2,Talal Chatila2 and
Min Zhuo1
1 Washington University Pain Center, Departments of Anesthesiology, Anatomy, and Neurobiology and Psychiatry, and
2 Departments of Pediatrics and Pathology and Immunology and the Center for Immunology, Washington University School of Medicine,
St. Louis, Missouri 63110, USA
Correspondence should be addressed to M.Z. (zhuom@ morpheus.wustl.edu)
The ability to remember potential dangers in an environment is necessary to the survival of animals and
humans. The cyclic AMP–responsive element binding protein (CREB) is a key transcription factor in
synaptic plasticity and memory consolidation. We have found that in CaMKIV–/– mice—which are
deficient in a component of the calcium–calmodulin-dependent protein kinase (CaMK) pathway, a
major pathway of CREB activation—fear memory, but not persistent pain, was significantly reduced.
CREB activation by fear conditioning and synaptic potentiation in the amygdala and cortical areas was
reduced or blocked. We propose that cognitive memory related to a noxious shock can be disassociated
from behavioral responses to tissue injury and inflammation.
Emotional learning and its expression in mammals, such as fear, or prolonged injury, was significantly reduced in CaMKIV–/– mice
require the involvement of higher brain structures, including the as compared with wild-type mice. Consistent with these find-
amygdala, hippocampus and related cortical areas1–5. Cumula- ings, CREB activation by fear conditioning and synaptic poten-
tive evidence consistently shows that within these areas, long- tiation in memory-related areas, including the amygdala and
term changes in synaptic transmission and structure are hippocampus, was blocked or attenuated in CaMKIV–/– mice.
important for the establishment and consolidation of such mem-
ory6–11. CREB is a major transcription factor12,13 that is central- RESULTS
ly involved in the formation of long-term memory in both CaMKIV is crucial for fear memory but not nociception
invertebrates and vertebrates14–22. CREB is activated by phos- We assessed two forms of associative emotional memory in
phorylation of the serine 133 residue, potentially within a short wild-type and CaMKIV–/– mice: contextual and auditory fear
period of time12,13. CREB activation is mediated by two major conditioning. Animals learn to fear a neutral conditioned stim-
pathways, the cAMP signaling pathway and calcium–calmodulin ulus (such as tone) that has been paired with an aversive uncon-
(Ca2+–CaM)-dependent protein kinase pathway23–29. Inhibition ditioned stimulus (such as a foot shock) and with the context
of CREB impairs behavioral performance in various memory in which the animals were conditioned by the pairing of con-
tests across species15–19, whereas overexpression of CREB facili- ditioned and unconditioned stimuli. Early contextual and audi-
tates long-term fear memory20,21. tory fear conditionings are mediated by the hippocampus
Genetic manipulation of signaling pathways upstream of and/or amygdala, whereas late contextual memory may be
CREB has produced varying results. Inhibition of cAMP-protein mediated by areas of the cortex3,5. Contextual and auditory con-
kinase A signal pathways affects both spatial and fear memory in ditionings were measured at 1 hour, 1 day and 7 days after train-
different species30–34. Different results were reported in regard ing (Fig. 1). We found no significant difference in contextual
to the function of CaMKIV, a component of the CaMK pathway, freezing immediately after training or 1 hour later (wild-type,
in behavioral memory. Genetic deletion of CaMKIV does not n = 8 mice; CaMKIV–/–, n = 7). By contrast, at 1 and 7 days, we
affect two common forms of spatial memory in mice35, where- saw significantly less contextual freezing in CaMKIV–/– mice
as transgenic mice expressing a dominant-negative CaMKIV than in wild-type mice (Fig. 1a). Furthermore, when we tested
show deficits in various hippocampus-dependent tasks36. We auditory fear conditioning 1 day or 7 days after training, we saw
hypothesized that different CREB activation pathways preferen- significantly less freezing in response to the tone in CaMKIV–/–
tially encode different forms of memory. We therefore examined mice than in wild-type mice (Fig. 1b). No obvious difference
the function of CaMKIV in behavioral acquisition of contextual in the behavioral response to the foot shock was found between
and auditory fear memory using CaMKIV–/– mice, in which the wild-type and CaMKIV–/– mice.
gene encoding CaMKIV was abolished. We found that fear mem- Next, we asked whether the lack of CaMKIV affected acute
ory, but not behavioral responses to an acute noxious stimulus nociceptive transmission or persistent pain. We saw no signifi-

articles
Fig. 1. CaMKIV is required for fear memory but not behavioral

a b responses to tissue injury. (a, b) Contextual and auditory fear condition-
ing at 1 h, 1 d and 7 d after training (, wild-type mice, n = 8; , CaMKIV–/–
(KO), n = 7). (c) Nociceptive behavioral responses to hind-paw formalin
injection during three different phases. No significant difference was
found between wild-type (hatched) and CaMKIV–/– mice (black). (d) The
responses of animals to a mechanical stimulus that elicited no responses
before dorsal hind-paw CFA injection were recorded 1 and 3 d after CFA
injection. The data are plotted as percentage positive responses to stimu-
lation of the ipsilateral or contralateral hind paw for wild-type and

CaMKIV–/– mice. *, P < 0.05 as compared with wild-type mice.
c
To explore whether CaMKIV-dependent signal pathways con-

tribute to behavioral responses in chronic inflammatory pain,
we tested mechanical allodynia after hind-paw injection of com-
plete Freund’s adjuvant (CFA). Application of a fine von Frey
fiber to the dorsum of a hind paw elicited no response in untreat-
ed mice, but at 1 and 3 days after CFA injection (50%, 10 µl) into
the dorsum of a single hind paw, mice responded to stimulation
of either the same (ipsilateral) or, to a lesser extent, the con-
d tralateral hind paw by hind-paw withdrawal37. Again, we found
that mechanical allodynia was similar in wild-type (n = 6) and
CaMKIV–/– mice (n = 13, Fig. 1d) and that the degree of local
inflammation in wild-type and CaMKIV–/– mice was similar.
Together, these results indicate that CaMKIV is preferentially
involved in fear memory induced by a noxious shock but not in the
behavioral responses to acute noxious stimuli or tissue inflamma-
tion. Thus, behavioral responses to peripheral tissue inflammation
commonly used in pain research do not reflect fear memory38.
CREB activation by fear conditioning

cant difference in tail-flick latency (wild-type, n = 12 mice, mean One major downstream target of CaMKIV is CREB, a key tran-
3.5 ± 0.3 s; CaMKIV–/–, n = 14, 3.9 ± 0.4 s), indicating that spinal scription factor involved in long-term memory, including fear
nociceptive transmission was not significantly altered in memory16,18. Fear conditioning causes activation (phosphoryla-
CaMKIV–/– mice. Likewise, we found no differences in latencies tion) of CREB in the hippocampus and amygdala, as indicated
to a hot-plate test (55°C) (wild-type, n = 17, 13.6 ± 1.1 s; by immunostaining for phosphorylated CREB (pCREB)39. How-
CaMKIV–/–, n = 18, 15.7 ± 0.8 s, P > 0.05). Thus, the behavioral ever, the role of CaMKIV in fear conditioning–induced phos-
responses of CaMKIV–/– and wild-type mice to noxious thermal phorylation of CREB has not been examined. Because two forms
stimuli were indistinguishable. We then examined the response to of fear memory were significantly reduced in CaMKIV–/– mice,
a more prolonged nociceptive stimulus, hind-paw formalin injec- we predicted that the amount of fear conditioning–induced
tion. We had previously showed that formalin injection caused pCREB present should be reduced as well. As the distribution of
three phases of licking responses over 120 min after the injec- CaMKIV immunoreactivity has been described only in the adult
tion37. We found no significant difference between wild-type and rat brain40, we first examined the distribution of CaMKIV in the
CaMKIV–/– mice in any of the three phases of response (wild- hippocampus, amygdala and related cortical areas of wild-type
type, n = 11; CaMKIV–/–, n = 7, Fig. 1c). These results suggest mice (n = 3). Consistent with previous findings in the rat brain40,
that CaMKIV-dependent signal pathways are not required for CaMKIV-labeled neurons were found in the hippocampus, amyg-
behavioral responses to formalin injection. dala, anterior cingulate cortex (ACC), somatosensory cortex and
Fig. 2. Distribution of CaMKIV in selected areas of adult

mouse brain. Coronal section of brain showing the distri-
bution of CaMKIV-labeled neurons in the basolateral
amygdala, the CA1 region of hippocampus, ACC, primary
and secondary somatosensory cortex (S1+2) and insular
cortex of wild-type mice (n = 3 mice, from left to right,
top and middle panels). High-magnification images from
top row show staining in the basolateral amygdala (BL),
pyramidal cell layer of the CA1, area 3 of ACC (Cg3),
layer 4 of somatosensory cortex and agranular insular
cortex (AI), respectively (middle panels). No immunos-
taining was seen in brain sections of the CaMKIV–/– mice
(n = 3 mice, bottom panels). Cg1, area 1 of ACC; pir, pyri-
form cortex; GI, granular insular cortex. Scale bar,
200 µm (top and bottom panels), 50 µm (middle panels).

articles
Fig. 3. CaMKIV contributes to activation of CREB by fear

conditioning. pCREB expression in the amygdala (BM, BL),
hippocampus (CA1, DG), ACC (Cg1, Cg3), S1 and S2 and
insular cortices (GI, AI) from wild-type and CaMKIV–/– mice
in a control situation (Box), after shock–tone unpaired
training or after tone–shock paired training. Notation as in
Fig. 2; BM, basomedial amygdala. Scale bar, 300 µm.
insular cortex (Fig. 2). Higher-magnification confo-

cal images showed strong CaMKIV immunoreactiv-
ity in neuronal nuclei and weak immunoreactivity in
the cytoplasm throughout these areas. No CaMKIV
immunostaining was seen in the brains of
CaMKIV–/– mice (n = 3, Fig. 2).
Next, we measured activation of CREB by fear
conditioning in adult wild-type and CaMKIV–/– mice
through immunocytochemistry. To demonstrate the
associative specificity of pCREB expression in the
various brain areas, we carried out a series of exper-
iments in five different groups of wild-type and
CaMKIV–/– mice: control (no stimulus) and those
conditioned with tone alone, shock alone,
shock–tone (unpaired training) and tone–shock
(paired training). At 45 min after paired tone–shock
presentations, greater pCREB immunoreactivity was
found in the hippocampal CA1 and the basolateral
part of the amygdala of wild-type mice (n = 5 mice)
than in that of wild-type mice (n = 5) receiving unpaired training ever, paired conditioning activated pCREB in the basolateral
(Figs. 3 and 4). Auditory stimulation alone did not produce a nucleus of the amygdala (as compared to wild-type mice receiv-
significant change in pCREB levels in the hippocampus or amyg- ing paired training) was completely abolished in CaMKIV–/– mice
dala (n = 4 mice), whereas shock alone caused modest increases (Fig. 4), suggesting that CaMKIV activity is essential to fear con-
in pCREB in the hippocampus and amygdala (n = 5 mice). In ditioning–induced CREB activation.
both wild-type (n = 5 mice) and CaMKIV–/– mice (n = 5 mice), We also examined three cortical areas involved in the process-
similar basal levels of pCREB were found in different areas of the ing of sensory information and possibly fear-related memory, the
hippocampus, including CA1 (wild-type, 3.7 ± 0.4; CaMKIV–/–, ACC, primary somatosensory cortex and agranular insular cor-
2.9 ± 0.1), CA3 (wild-type, 3.4 ± 0.2; CaMKIV–/–, 3.5 ± 0.2), den- tex. In wild-type mice, paired fear conditioning induced signifi-
tate gyrus (DG) (wild-type, 3.7 ± 0.2, CaMKIV–/–, 3.0 ± 0.3) and cantly larger increases in pCREB level in all three cortical regions
amygdala (wild-type, 3.4 ± 0.1; CaMKIV–/–, 3.2 ± 0.5) (Figs. 3 than did unpaired training (Figs. 3 and 4). Tone or shock alone
and 4). This suggested that deletion of CaMKIV does not affect increased pCREB expression in the somatosensory and insular
basal expression of pCREB in neurons. In the groups of mice cortex of wild-type mice. However, CaMKIV seems to contribute
receiving paired fear conditioning, pCREB levels in the CA1 differently to the activation of CREB by fear conditioning in the
region were significantly lower, but not absent, in CaMKIV–/– different cortical areas examined. In CaMKIV–/– mice, fear con-
mice (n = 6 mice) as compared to wild-type mice (n = 5 mice, ditioning–induced pCREB was not seen in the somatosensory and
Figs. 3 and 4). This finding indicates that CaMKIV contributes insular cortex but was seen, although at a lesser level than in wild-
only partially to the activation of CREB in the CA1 areas. How- type mice, in the ACC (Figs. 3 and 4). Basal expression of pCREB
in these regions was similar in wild-type and CaMKIV–/– mice.
This finding differs from our previous results obtained using pri-
mary cortical cultured cells35, indicating that in vitro–cultured
cell preparations may not completely mimic physiological condi-
tions in vivo. We also examined the expression of pCREB in audi-
tory pathways, including the cochlear nucleus, lateral superior
olivary complex, inferior colliculus and medial nucleus of the
trapezoid body. Similar levels of pCREB after tone alone, unpaired
shock–tone or paired tone–shock were found in wild-type and
CaMKIV–/– mice (data not shown), indicating that the auditory
sensory transmission is normal. This conclusion is supported by
Fig. 4. CaMKIV contributes to activation of CREB by fear conditioning.

Quantification of pCREB staining in basolateral amygdala, hippocampal
CA1, ACC, S1 and AI are illustrated in wild-type and CaMKIV–/– mice
receiving shock–tone unpaired or tone–shock paired training. *, signifi-
cant difference from wild-type and CaMKIV–/– controls; #, significant dif-
ference from unpaired CaMKIV–/– mice.

articles
Fig. 5. Requirement of CaMKIV for amygdala and cortical synaptic

a b potentiation. (a) TBS (arrow) induced synaptic potentiation in the
amygdala in wild-type (, n = 11 slices/9 mice) but not CaMKIV–/–
mice (, n = 6 slices/6 mice). (b) CaMKIV is also required for
potentiation in the ACC by TBS (wild-type, n = 7 slices/5 mice;
CaMKIV–/–, n = 6 slices/6 mice). Insets in (a) and (b), representative
records of the EPSP before (pre) and 40 min after (post) TBS in a
wild-type and CaMKIV–/– slice. (c–d) Synaptic potentiation in the
insular (wild-type, n = 7 slices/6 mice; CaMKIV–/–, n = 8 slices/5
mice) and somatosensory cortices (wild-type, n = 5 slices/5 mice;

CaMKIV–/–, n = 7 slices/6 mice).
to the lateral amygdala by placing a stimulating electrode

in the ventral striatum41. In contrast to the results for the
CA1 region of hippocampus, we found in preliminary
experiments that strong tetanic stimulation did not induce
c d reliable potentiation (data not shown). We therefore used
five trains of theta-burst stimulation (TBS) as a stimulus5.
In slices from wild-type mice, TBS induced significant
synaptic potentiation (n = 11 slices/9 mice; 169.4 ± 8.0%;
Fig. 5a). In slices from CaMKIV –/– mice, however, this
synaptic potentiation was significantly reduced or blocked
(n = 6 slices/6 mice; 124.1 ± 14.1%, P < 0.01 compared to
slices of wild-type mice).
We also examined synaptic potentiation in the ACC,
somatosensory cortex and insular cortex. In the ACC,
strong tetanic stimulation induces only short-term poten-
tiation42. Working with normal slices from wild-type mice,
we found that TBS induced significant synaptic potentia-
behavioral data indicating that immediate responses (such as freez- tion lasting at least 40 min (n = 7 slices/5 mice,
ing response measured after fear conditioning) are normal in 153.7 ± 17.6%). In CaMKIV–/– mice, we saw no potentiation
CaMKIV–/– mice. These results indicate that CaMKIV is crucial after TBS (n = 6 slices/6 mice, 91.3 ± 8.6%), suggesting that
for the activation of CREB in areas related to fear memory in vivo CaMKIV is required for synaptic potentiation in the ACC
and that its contribution is region specific. (Fig. 5b). A similar defect in synaptic potentiation was found
in the insular cortex (wild-type, n = 7 slices/6 mice, 139.3 ±
Synaptic potentiation in the amygdala and cortex 8.9%; CaMKIV–/–, n = 8 slices/5 mice, 118.1 ± 7.6%; P < 0.05
Synaptic plasticity, including long-term potentia-
tion (LTP) and long-term depression (LTD), are
thought to be important for learning and memo- a
ry 3,6–11. In the CA1 region of the hippocampus,
LTP but not LTD is altered in CaMKIV–/– mice35.
Here we wanted to examine synaptic potentiation
in the amygdala, a structure that plays an impor-
tant role in fear memory 1–4,7–11 . We examined
synaptic potentiation at ‘thalamic’ input synapses
Fig. 6. CaMKIV is required for CaM translocation and b

activation of CREB in the hippocampus and amygdala.
(a) Confocal images of CaM and pCREB staining in CA1
pyramidal neuron layer from either wild-type or
CaMKIV–/– mice, with or without high-K+ stimulation.
(b) Summary data of CaM translocation in CA1, CA3 and
DG areas of hippocampus from wild-type (WT; control, n
= 9 sections/3 mice; treated, n = 11 sections/3 mice) and
CaMKIV–/– mice (KO; control, n = 7 sections/2 mice; c
treated, n = 11 sections/3 mice). (c) CaM and pCREB d
staining in neurons of the basolateral amygdala receiving
similar treatment as described in (a). (d) Summary data of
CaM translocation in basolateral amygdala neurons from
CaMKIV–/– (control, n = 7 sections/2 mice; treated, n = 11
sections/3 mice) and wild-type mice (control, n = 9 sec-
tions/3 mice; treated, n = 11 sections/3 mice). Scale bar,
25 µm (a, c). *, P < 0.05 as compared with sections of
wild-type mice.
articles
Fig. 7. CaMKIV contributes to CaM translocation and a

activation of CREB in three cortical areas.
(a–c) Confocal images of CaM and pCREB staining in
(a) Cg1 of ACC, (b) S1 and (c) AI slices from either
wild-type or CaMKIV–/– mice with or without high-K+
stimulation. Scale bar, 25 µm. (d) Summary data of CaM
translocation in these regions after neuronal depolariza-
tion from wild-type (WT; for Cgl and AI, control, n = 15
sections/3 mice, treated, n = 12 sections/3 mice; for S1,
control, n = 9 sections/3 mice, treated, n = 11 sec-

tions/3 mice) and CaMKIV–/– mice (KO; for Cgl and AI, d
control, n = 17 sections/3 mice, treated, n = 15 sec-
tions/3 mice; for S1, control, n = 7 sections/3 mice,
b
treated, n = 11 sections/3 mice). *, P < 0.05 as com-
pared with sections of wild-type mice.
comparing two groups) (Fig. 5c). In the

somatosensory cortex of wild-type mice, TBS
induced a long-lasting enhancement of synap- c
tic responses persisting for at least 40 min after
induction (n = 5 slices/5 mice, 150.2 ± 12.6%).
In CaMKIV–/– mice, however, we saw no synap-
tic potentiation (n = 7 slices/6 mice, 111.9 ±
7.1%; P < 0.05 as compared with potentiation
in wild-type mice) (Fig. 5d). These results pro-
vide the first evidence that CaMKIV contributes
to synaptic potentiation in these cortical areas.
CaM translocation from the cytoplasm to the nucleus in three different areas of the cortex of wild-type mice, including
Neural activity (for example, induced by the application of high- the ACC, somatosensory cortex and insular cortex (Fig. 7a-c).
KCl solution) causes translocation of CaM from the cytoplasm Both CaM translocation and CREB activation were significantly
to the nucleus and subsequent activation of CREB in the nucle- reduced in the ACC and insular cortex and completely absent in
us of central neurons26. CaM-binding proteins in the nucleus the somatosensory cortex of CaMKIV –/– mice (Fig. 7d).
may serve as a sink for Ca2+–CaM, leading to accumulation of
CaM in the nucleus after elevation of intracellular calcium43. If DISCUSSION
this is true, CaM translocation might be affected in mice lacking Two key Ca2+–CaM-dependent protein kinases, CaMKII and
CaMKIV, a CaM-binding protein in the nucleus. To test whether CaMKIV, are important modulators of synaptic plasticity and
KCl depolarization causes CaM translocation from the cytoplasm behavioral memory27,45–48. CaMKII is highly expressed at post-
to the nucleus, we measured the ratio of CaM immunoreactivity synaptic sites, and its activation contributes to the phosphoryla-
between neuronal nuclei and cytoplasm in brain slices from both tion of synaptic proteins, including glutamate receptors, as well as
wild-type and CaMKIV–/– mice treated with either control solu- to the synaptic potentiation known as LTP 27,45. Genetically
tion or 90 mM KCl (Figs. 6 and 7). We stained the same neurons manipulated mice with or without abnormal CaMKII activity
to detect pCREB, because activation of CaMKIV leads to phos- showed defects in the ability to learn and remember during spa-
phorylation of CREB in the nucleus26. In wild-type mice, KCl tial and fear memory tests46–48. Unlike CaMKII, CaMKIV is locat-
application caused significant CaM translocation in neurons ed mainly in the neuronal nuclei and is involved in the regulation
within the CA1 and DG but not the CA3 region of the hip- of activity-triggered gene expression26,35. Our present results pro-
pocampus (Fig. 6a and b). In CaMKIV–/– mice, CaM transloca- vide several new findings regarding the role of CaMKIV in neu-
tion triggered by KCl was significantly lower than that in ronal activation of CREB, synaptic potentiation and behavioral
wild-type mice in the CA1, and completely absent in the DG. memory. Previous in vitro experiments from cultures and slices
Consistently, pCREB was also reduced in the CA1 (Fig. 6a) and have shown that CaMKIV contributes to the activation of CREB
absent in the DG (data not shown). by neuronal activity35. Our in vivo results show that CaMKIV
CaM translocation by neural activity was reported in cultured contributes to the activation of CREB in various memory-relat-
neurons and hippocampus26,44, and it is not known whether sim- ed areas, such as the amygdala and hippocampus. Furthermore,
ilar translocation occurs in neurons in other regions of the brain. the contribution of CaMKIV to CREB activation differs across
Therefore, we carried out similar experiments in amygdala and these areas. In the amygdala, activation of CREB by fear condi-
cortical slices of wild-type and CaMKIV–/– mice. Application of tioning was completely absent in CaMKIV–/– mice. In the hip-
90 mM KCl solution to amygdala slices of wild-type mice caused pocampal CA1 areas, however, activation of CREB was less than
significant CaM translocation and activation of CREB in neurons in wild-type mice, but not absent. Consistent with these
within the basolateral amygdala (Fig. 6c). Notably, just as in CA1 immunostaining results, behavioral changes in spatial (hip-
hippocampal neurons (Fig. 6a), both CaM translocation and pocampus-dependent learning) and fear memory from current
pCREB were significantly lower in slices from CaMKIV–/– mice and previous studies support the conclusion that CaMKIV may
as compared with wild-type mice (Fig. 6c and d). CaM translo- have a more important role in amygdala-related fear memory
cation and activation of CREB by KCl application were also found than in hippocampus-related spatial memory. Our electrophys-

articles
iological studies show that CaMKIV contributes to synaptic obtained with an Olympus (Melville, New York) Fluoview laser-scan-
potentiation in the amygdala, a structure centrally involved in ning confocal microscope. Anatomical terminology is based on the atlas
fear memory. Our data thus provide further evidence in support of Franklin and Paxinos50.
of a connection between synaptic potentiation of glutamatergic To detect nuclear translocation of CaM and CREB phosphorylation,
transmission in the lateral amygdala and fear memory. In addi- we performed experiments using brain slices. Adult male mice were anes-
thetized with halothane and coronal slices of brain were prepared and
tion to the amygdala, we found that CaMKIV also contributes to maintained in interfaced chambers about 28°C, where they were subfused
synaptic potentiation in several cortical areas, including the ACC, with ACSF (124 mM NaCl, 4.4 mM KCl, 2.0 mM CaCl2, 1.0 mM MgSO4,
insular cortex and somatosensory cortex. These findings suggest 25 mM NaHCO3, 1.0 mM Na2HPO4 and 10 mM glucose) bubbled with
that CaMKIV serves as a key intracellular protein kinase con- 95% O2 plus 5% CO2. High K+-containing ACSF (with 4.4 mM KCl
tributing to synaptic potentiation. replaced with 90 mM KCl) was applied through bath solution for 180 s.
Our findings provide further evidence supporting the role of Tetrodotoxin (1 µM) was added to block neuronal activity. Brain slices
CREB in fear memory. Spatial memory, as well as fear memory, were incubated with 1:250 anti-CaM mouse monoclonal antibody
is impaired in CREB–/– mice16. However, spatial memory is not (Chemicon, Temecula, California) and 1:500 anti-pCREB rabbit anti-
affected in CaMKIV–/– mice35. A possible explanation is that serum (Calbiochem, San Diego, California), followed by 1:600 Cy-3 goat
anti-mouse and 1:100 FITC-conjugated AffiniPure donkey anti-rabbit
CaMKIV-independent pathways such as the cAMP/PKA signal-
IgG secondary antibodies (Jackson ImmunoResearch Laboratories, West
ing pathway may have a more dominant role in CREB activation Grove, Pennsylvania). Images were collected sequentially on the Olym-
linked to spatial memory. Consistent with this proposition is the pus Fluoview laser-scanning confocal microscope. Staining for CaM was
observation that the contribution of CaMKIV to CREB activa- quantified by using NIH Image (Scion Image, Scion Corp., Frederick,
tion varies among different brain regions. In some areas, such as Maryland) to obtain measurements of relative fluorescence intensity in
the hippocampal CA1 region, considerable residual pCREB can the whole nuclear area as compared with that in the cytoplasm27,46. All
still be found in CaMKIV–/– mice. data was shown as a ratio between the nuclear and cytosolic area of each
The results presented here, obtained using brain slices, indi- neuron. Only neurons with sharp boundaries and a well-defined nucleus
cate that CaMKIV is required for CaM translocation into the were considered. Thirty neurons were measured from three different sec-
nuclei of central neurons. This translocation, which is triggered tions, for each stimulation, and averaged. For illustration, images were
assembled into montages with Adobe Photoshop software.
by neural activity, is not limited to hippocampal neurons but also
For fear-induced pCREB expression, brain slices were incubated with
occurs in neurons located in the amygdala, ACC, somatosensory a rabbit anti-pCREB antibody at 1:500 dilution and Vectastain ABC kit
cortex and insular cortex; thus, it is probably a common signal- (Vector, Burlingame, California) was used, followed by development
ing mechanism for neurons in the central nervous system. Pre- using nickel-enhanced diaminobenzidine (DAB; Sigma, St. Louis, Mis-
vious studies have shown that CaM translocation reflects the souri). The integrated intensity for the selected regions was normalized
trapping of Ca2+–CaM complexes by nuclear CaM-binding pro- to the corresponding integrated intensity in the adjacent white matter as
teins43. The absence of CaM translocation in CaMKIV –/–mice described previously41. Measurements were made from three randomly
identifies CaMKIV as the crucial sink that traps Ca2+–CaM com- selected non-contiguous sections of each region in each mouse, observed
plexes in neuronal nuclei. This trapping leads to CaMKIV acti- from coded slides and averaged so that each animal had a mean value
vation and subsequent CREB phosphorylation and activation. for regional pCREB immunoreactivity.
Consistently, we found in both in vitro and in vivo conditions
Slice electrophysiology. Transverse slices of amygdala and cortex were
that activation of CREB was significantly reduced or abolished rapidly prepared and maintained in an interface chamber at 30°C, where
in CaMKIV –/– mice. We also found that synaptic potentiation they were subfused with ACSF35,37. In amygdala slices, a bipolar tung-
induced by TBS was reduced or abolished in the same areas. sten stimulating electrode was placed in the ventral striatum, and an
These findings support the existence of an integrated mechanism extracellular recording electrode (3–12 MΩ, filled with ACSF) was placed
involving CaMKIV that links changes in cytosolic Ca2+ concen- in the lateral amygdala41. In the cortical slices, a bipolar tungsten stimu-
trations to nuclear transcriptional activation. lating electrode was placed in layer 5, and extracellular field potentials
Finally, our studies provide strong evidence that cognitive were recorded using a glass microelectrode placed in layer 2/3. Synaptic
memory related to a noxious shock can be disassociated from responses were elicited at 0.02 Hz.
behavioral responses to an acute insult, or prolonged injury, at
the molecular level. Although many signaling pathways may con- Fear conditioning. We used a fear-conditioning shock chamber. Mice
were placed in the chamber for 2 min before fear conditioning. The con-
tribute to normal physiological functions as well as to distorted
ditioned stimulus (CS) used was an 85 dB sound at 2,800 Hz for 30 s,
pathological conditions37,38,49, we believe that it is possible to and the unconditioned stimulus (US) was a continuously scrambled foot
identify signaling molecules that preferentially contribute to one shock at 0.75 mA for 2 s. During training, mice were presented with a
but not the other. Identification of such signaling molecules may 30 s tone (CS) and a shock (US) beginning at 28 s after the onset of US.
have applications useful in the improvement of human health. After CS/US pairing, the mice were allowed to stay in the chamber for
another 30 s for measurement of immediate freezing. Freezing was scored
METHODS every 10 s. During the retention test, each mouse was placed back into
Knockout mice. CaMKIV–/– mice were derived as described35 and bred the shock chamber and the freezing response was recorded for 3 min.
for several generations (F8–F12) in a C57Bl/6 background. Control wild- Subsequently, the mice were put into a novel chamber and monitored
type mice were littermates of mutant mice. The Animal Care and Use for 3 min before the onset of the tone (pre-CS). Immediately after that, a
Committee of Washington University approved the mouse protocols. No tone identical to the CS was delivered for 3 min and freezing responses
visual difference between wild-type and CaMKIV–/– mice is noticeable, were recorded. To test the association specificity of pCREB, in some
and experiments were performed blind. experiments, CS/US unpaired trainings were also performed. The US
preceded the CS by 60 s.
Immunocytochemistry and confocal imaging. For CaMKIV staining,
brain sections were incubated with anti-CaMKIV mouse antibody (1:500; Behavioral experiments. To test acute pain responses, the latency of
Transduction Laboratories, Lexington, Kentucky) and then with FITC- responses to heating of the tail (tail-flick reflex) or to placement on a
conjugated AffiniPure goat anti-mouse IgG at 1:100 dilution (Jackson hot-plate (55°C) was measured as described37. To test inflammatory pain,
ImmunoResearch Laboratories, West Grove, Pennsylvania). Images were formalin (5%, 10 µl) or CFA (50%, 10 µl; Sigma) was injected into the

articles
dorsal side of a hind paw. For formalin, the total time spent licking or 21. Josselyn, S. A., Shi, C., Carlezon, W. A. Jr., Neve, R. L., Nestler, E. J. & Davis,
biting the injected hind paw was recorded during each 5 min interval for M. Long-term memory is facilitated by cAMP response element–binding
protein overexpression in the amygdala. J. Neurosci. 21, 2404–2412 (2001).
2 h. For CFA, mechanical sensitivity was assessed with an innocuous 0.4 22. Silva, A. J., Kogan, J. H., Frankland, P. W. & Kida, S. CREB and memory.
millinewton filament as described37. Hind-paw edema was evaluated Annu. Rev. Neurosci. 21, 127–148 (1998).
with a fine caliper at 3 d after CFA injection. 23. Gonzalez, G. A. & Montminy, M. R. Cyclic AMP stimulates somatostatin
gene transcription by phosphorylation of CREB at serine 133. Cell 59,
Data analysis. Results were expressed as mean ± s.e.m. Statistical compar- 675–680 (1989).
24. Bito, H., Deisseroth, K., & Tsien, R. W. CREB phosphorylation and
isons were made with one- or two-way analysis of variance (ANOVA) with dephosphorylation: a Ca2+- and stimulus duration–dependent switch for
the post-hoc Scheffé F-test in immunocytochemical experiments, or the hippocampal gene expression. Cell 87, 1203–1214 (1996).
Student-Newmann-Keuls test in behavioral experiments, to identify signif- 25. Deisseroth, K., Bito, H. & Tsien, R. W. Signaling from synapse to nucleus:
icant differences. In all cases, P < 0.05 was considered statistically significant. postsynaptic CREB phosphorylation during multiple forms of hippocampal
synaptic plasticity. Neuron 16, 89–101 (1996).
26. Deisseroth, K., Heist, E. K. & Tsien, R.W. Translocation of calmodulin to the
Acknowledgments nucleus supports CREB phosphorylation in hippocampal neurons. Nature
392, 198–202 (1998).
Supported by grants from the National Institute of Neurological Disorders and 27. Soderling, T. R. CaM-kinase: modulator of synaptic plasticity. Curr. Opin.
Stroke 38680, the National Institute on Drug Abuse 10833, the McDonnell Neurobiol. 10, 375–380 (2000).
Center for Higher Brain Function and Alzheimer Disease Research Center at 28. Hardingham, G. E., Arnold, F. J. L. & Bading, H. Nuclear calcium signaling
controls CREB-mediated gene expression triggered by synaptic activity. Nat.
Washington University. Neurosci. 4, 261–267 (2001).
29. West, A.E. et al. Calcium regulation of neuronal gene expression. Proc. Natl.
Acad. Sci. USA 98, 11024–11031 (2001).
Competing interest statement 30. Livingston, M. S., Sziber, P. P. & Quinn, W. G. Loss of calcium calmodulin
The authors declare that they have no competing financial interests. responsiveness in adenylyl cyclase of rutabaga, a Drosophila learning mutant.
Cell 37, 205–215 (1984).
RECEIVED 7 FEBRUARY; ACCEPTED 19 MARCH 2002 31. Foster, J. L., Guttman, J. J., Hall, L. M. & Rosen, O. M. Drosophila cAMP-
dependent protein kinase. J. Biol. Chem. 259, 13049–13055 (1984).
32. Abel, T., Nguyen, P. V., Barad, M., Deuel, T. A. S., Kandel. E. R. &
Bourtchuladze, R. Genetic demonstration of a role for PKA in the late phase
1. Davis, M. The role of the amygdala in fear and anxiety. Annu. Rev. Neurosci. of LTP and the hippocampus-based long-term memory. Cell 88, 615–626
15, 353–375 (1992). (1997).
2. Davis, M., Walker, D. L. & Lee, Y. Amygdala and bed nucleus of the stria 33. Wong, S. T. et al. Calcium-stimulated adenylyl cyclase activity is critical for
terminalis: differential roles in fear and anxiety measured with the acoustic hippocampus-dependent long-term memory and late phase LTP. Neuron 23,
startle reflex. Phil. Trans. R. Soc. Lond. Ser. B 352, 1675–1687 (1997). 787–798 (1999).
3. LeDoux, L. E. Emotion circuits in the brain. Annu. Rev. Neurosci. 23, 155–184 34. Schafe, G. & LeDoux, J. E. Memory consolidation of auditory pavlovian fear
(2000). conditioning requires protein synthesis and protein kinase A in the amygdala.
4. Maren, S. Neurobiology of Pavlovian fear conditioning. Annu. Rev. Neurosci. J. Neurosci. 20 (RC98), 1–5 (2000).
24, 897–931 (2001). 35. Ho, N. et al. Impaired synaptic plasticity and cAMP response
5. Frankland, P. W., O’Brien, C., Ohno, M., Kirkwood, A. & Silva, A. J. element–binding protein activation in Ca2+/calmodulin-dependent protein
α-CaMKII-dependent plasticity in the cortex is required for permanent kinase type IV/Gr-deficient mice. J. Neurosci. 20, 6459–6472 (2000).
memory. Nature 411, 309–313 (2001). 36. Kang, H., Sun, L.D., Atkins, C.M., Soderling, T.R., Wilson, M.A. & Tonegawa,
6. Bliss, T. V. P. & Collingridge, G. L. A synaptic model of memory: long-term S. An important role of neural activity–dependent CaMKIV signaling in the
potentiation in the hippocampus. Nature 361, 31–39 (1993). consolidation of long-term memory. Cell 106, 771–783 (2001).
7. McKernan, M. G. & Shinnick-Gallagher, P. Fear conditioning induces a 37. Wei, F. et al. Genetic enhancement of inflammatory pain by forebrain NR2B
lasting potentiation of synaptic currents in vitro. Nature 390, 607–611 overexpression. Nat. Neurosci. 4, 164–169 (2001).
(1997). 38. Kerchner, G. A. et al. Reply to “Do ‘smart’ mice feel more pain, or are they just
8. Rogan, M. T., Stäubli, U. V. & Ledoux, J. E. Fear conditioning induces better learners?” Nat. Neurosci. 4, 453–454 (2001).
associative long-term potentiation in the amygdala. Nature 390, 604–607
39. Impey, S., Smith, D. M., Obrietan, K., Donahue, R., Wade, C. & Storm, D.R.
(1997).
Stimulation of cAMP response element (CRE)–mediated transcription
9. Maren, S. Long-term potentiation in the amygdala: a mechanism for
during contextual learning. Nat. Neurosci. 1, 595–601 (1998).
emotional learning and memory. Trends Neurosci. 22, 561–567 (1999).
40. Nakamura, Y., Okuno, S., Sato, F. & Fujisawa, H. An immunohistochemical
10. Repa, J. C. et al. Two different lateral amygdala cell populations contribute to
study of Ca2+/calmodulin-dependent protein kinase IV in the rat central
the initiation and storage of memory. Nat. Neurosci. 4, 724–731 (2001).
11. Schafe, G. E., Nader, K., Blair, H. T. & LeDoux, J. E. Memory consolidation of nervous system: light and electron microscopic observation. Neurosci. 68,
Pavlovian fear conditioning: a cellular and molecular perspective. Trends 181–194 (1995).
Neurosci. 24, 540–546 (2001). 41. Schafe, G. E., Atkins, C. M., Swank, M. W., Bauer, E. P., Sweatt, J. D. &
12. Sheng, M., Thompson, M. A. & Greenberg, M. E. CREB: a Ca2+-regulated LeDoux, J. E. Activation of ERK/MAP kinase in the amygdala is required for
transcription factor phosphorylated by calmodulin-dependent kinases. memory consolidation of pavlovian fear conditioning. J. Neurosci 20,
Science 252, 1427–1430 (1991). 8177–8187 (2000).
13. Mayr, B. & Montminy, M. R. Transcriptional regulation by the 42. Sah, P. & Nicoll, R.A. Mechanisms underlying potentiation of synaptic
phosphorylation-dependent factor CREB. Nat. Rew. Mol. Cell Biol. 2, transmission in rat anterior cingulate cortex in vitro. J. Physiol. (Lond.) 433,
599–609 (2001). 615–630 (1991).
14. Dash, P. K., Hochner, B. & Kandel, E. R. Injection of cAMP-responsive 43. Liao, B., Paschal, B. M. & Luby-Phelps, K. Mechanism of Ca2+-dependent
element into the nucleus of aplysia sensory neurons blocks long-term unclear accumulation of calmodulin. Proc. Natl. Acad. Sci. USA 96,
facilitation. Nature 345, 718–721 (1990). 6217–6222 (1999).
15. Yin, J. C. P. et al. Induction of a dominant negative CREB transgene 44. Teruel, M. N., Chen, W., Persechini, A. & Meyer, T. Differential codes for free
specifically blocks long-term memory in Drosophila. Cell 79, 49–58 (1994). Ca2+-calmodulin signals in nucleus and cytosol. Curr. Biol. 10, 86–94 (2000).
16. Bourtchuladze, R., Frenguelli, B., Blendy, J., Cioffi, D., Schutz, G. & Silva, A. J. 45. Lisman, J., Malenka, R. C., Nicoll, R. A. & Malinow, R. Learning mechanisms:
Deficient long-term memory in mice with a targeted mutation of the CaM- the case for CaM-KII. Science 276, 2001–2002 (1997).
responsive element–binding protein. Cell 79, 59–68 (1994). 46. Silva, A. J., Paylor, R., Wehner, J. M. & Tonegawa, S. Impaired spatial learning
17. Bartsch, D. et al. Aplysia CREB2 represses long-term facilitation: relief of in aα-calcium-calmodulin kinase II mutant mice. Science 257, 206–211
repression converts transient facilitation into long-term functional and (1992).
structural change. Cell 83, 979–992 (1995). 47. Bach, M. E., Hawkins, R. D., Osman, M., Kandel, E. R. & Mayford, M.
18. Guzowski, J. F. & McGaugh, J. L. Antisense oligodeoxynucleotide–mediated Impairment of spatial but not contextual memory in CaMKII mutant mice
disruption of hippocampal cAMP response element binding protein levels with a selective loss of hippocampal LTP in the range of the theta frequency.
impairs consolidation of memory for water maze training. Proc. Natl. Acad. Cell 81, 905–915 (1995).
Sci. USA 94, 2693–2698 (1997). 48. Giese, K. P., Fedorov, N. B., Filipkowski, R. K. & Silva, A. J.
19. Falls, W. A., Kogan, J. H., Silva, A. J., Willott, J. F., Carlson, S. & Turner, J. G. Autophosphorylation at Thr286 of the α calcium-calmodulin kinase II in
Fear-potentiation startle, but not prepulse inhibition of startle, impaired in LTP and learning. Science 279, 870–873 (1998).
CREB αδ–/– mutant mice. Behav. Neurosci. 114, 998–1004 (2000). 49. Tang, Y. P. et al. Genetic enhancement of learning and memory in mice.
20. Yin, J. C. P., Del Vecchio, M., Zhou, H. & Tully, T. CREB as a memory Nature 401, 63–69 (1999).
modulator: induced expression of a dCREB2 activator isoform enhances 50. Franklin, K. B. J. & Paxinos, G. The Mouse Brain in Stereotaxic Coordinates
long-term memory in Drosophila. Cell 81, 107–115 (1995). (Academic Press, New York, 1997).

articles
Non-spatial, motor-specific
activation in posterior parietal
cortex
Jeffrey L. Calton1,2, Anthony R. Dickinson1 and Lawrence H. Snyder1
1 Department of Anatomy and Neurobiology, Box 8108, Washington University School of Medicine, 660 S Euclid Ave., St. Louis, Missouri 63110, USA
2 Present address: Department of Psychology, California State University, Sacramento, California 95819, USA
Correspondence should be addressed to L.H.S. (larry@eye-hand.wustl.edu)
A localized cluster of neurons in macaque posterior parietal cortex, termed the parietal reach region
(PRR), is activated when a reach is planned to a visible or remembered target. To explore the role of
PRR in sensorimotor transformations, we tested whether cells would be activated when a reach is
planned to an as-yet unspecified goal. Over one-third of PRR cells increased their firing after an
instruction to prepare a reach, but not after an instruction to prepare a saccade, when the target of
the movement remained unknown. A partially overlapping population (two-thirds of cells) was acti-
vated when the monkey was informed of the target location but not the type of movement to be
made. Thus a subset of PRR neurons separately code spatial and effector-specific information, consis-
tent with a role in specifying potential motor responses to particular targets.
Posterior parietal cortex (PPC) seems to process spatial infor- tion cortex. In contrast, non-spatial, effector-specific intention
mation in a way that is strongly influenced by how that infor- activity might be expected if the dorsal stream is more active in
mation will be used1–3. The response to a visual and auditory sensorimotor transformations, and in particular, if PPC serves
stimulus in a cell’s receptive field is often enhanced when that in part to combine relevant spatial information with specific
stimulus is behaviorally relevant3–9. The enhancement is not motor intentions11–15. Such a role is accepted for the premotor
merely all or none; activity in the lateral intraparietal area (LIP), cortex18, which receives input from PPC19–21, but the appropri-
for example, seems to be continuously modulated by the return ate experiments have not been performed in PPC.
(reward) associated with a given target10. We also wished to know whether spatial information would
Initially, PPC was thought to reflect a supramodal spatial drive activity without effector information. In previous studies,
saliency map whose output can be used by different effectors3,4. animals are often trained to use only one effector. When train-
More recent experiments have shown that there are multiple ing involves multiple effectors, different movements are typical-
maps with outputs associated with particular effectors11–15. This ly segregated in discrete blocks, which provides implicit effector
result does not replace the idea that behavioral enhancement in information22–25. In some sense, effector information is absent
PPC reflects task relevance. Rather, it demonstrates that one in go/no–go tasks, in which animals receive a spatial target before
aspect of task relevance is the choice of effector for the response. the instruction to make or withhold a movement26. But even
We call this an ‘effector-specific intention’. For example, here, only a single effector is involved. Therefore we do not know
memory-related activity for a particular target in LIP is more whether spatial information, without a clear effector-specific
robust when the subject will respond with an eye movement plan, drives activity in PPC.
rather than with an arm movement, whereas the reverse pattern To assess the impact of effector information without spatial
occurs in the parietal reach region (PRR)12. Thus, enhancement information and spatial information without effector informa-
can depend on the choice of which effector to use. tion, we trained monkeys on a task where a delay was interposed
Here we asked whether effector-specific intentions might by between the instruction of what movement to make (reach or
themselves, without spatial information, nonetheless drive sus- saccade) and the target location for the movement. On some tri-
tained activity in PRR. The responses of neurons have been exam- als, we first instructed the type of movement to prepare, and then
ined when both effector and spatial information were known12,15. after a delay we presented the target for the movement. This
Now we ask whether effector-specific intention is only modula- allowed us to test whether PRR neurons distinguished between
tory, or if it can drive a response all on its own. A response from reach and saccade trials, in the interval after receiving the effector-
effector-specific intention alone would be unexpected, given the specific instruction but before receiving the target. On other tri-
prevailing view that the dorsal visual pathway, of which PPC is a als, we revealed the location of the spatial target and after a delay
part, is dedicated to processing spatial information16,17. The pres- gave the effector-specific instruction. This allowed us to assess
ence of non-spatial, effector-specific information in PPC would the impact of knowing the spatial target of the movement, even
also be at odds with its classical role as a purely sensory associa- though the animal did not know the type of movement to be per-

articles
Fig. 1. Localization of the parietal reach region. (a) The standard mem-
ory reach and saccade tasks used to map out PRR12. Task type (reach or
saccade) was indicated by the color of the target. Cells showing signifi-
a cantly higher activity during the delay period following a reach com-
pared to a saccade task cue were identified. (b) PRR (shaded rectangle)
was mapped as the region showing a high proportion of such cells. For
each of two animals (top and bottom rows), left and center maps show
the number of cells at each location (relative to –5 AP, 12 L) with
greater delay activity on reach and saccade trials, respectively. Right,
cells with similar responses on reach and saccade trials (t-test on activity
150–650 ms after target offset; P < 0.05). Coordinates in mm.
b (c) Cue–delay–target task. A cue instructs the type of movement (eye
or arm), and after a variable delay (gray), a target both instructs a spatial
location and triggers the movement. During the delay, the effector to be
used is known, but the spatial goal for the movement is unknown.
(d) Target–delay–cue task. Similar to (c), except the target and cue
appear in reverse order. During the delay, the spatial target is known,
but the effector is unknown.
color cue instructs the type of movement to be performed (reach

or saccade) and then, after a variable delay, a peripheral target
appears, which instructs the goal of the movement. The animal’s
task was to acquire the target using the instructed movement. The
c animal could move only after the movement was fully specified,
that is, after both the cue and target had appeared. Typically, tar-
gets were either inside the receptive field of the recorded cell, or
at a location on the opposite side of the fixation point. We com-
pared delay period responses on reach and saccade trials.
This cue–delay–target task (Fig. 1c) differs in two impor-
tant ways from the standard memory task that was used previ-
d ously to establish effector specificity in LIP and PRR12,15, and
which was used in the present study to delineate PRR (Fig. 1a
and b). First, the cue–delay–target task contains an interval in
which the upcoming movement type (reach or saccade) is
known, but the spatial goal of that movement is not known.
This interval allows us to measure the influence of effector infor-
mation on neuronal activity without spatial information. This
contrasts with conditions in the standard memory task, in
formed. We found that many PRR cells were activated by the which the type of movement and the goal of the movement are
instruction to make a reach without the spatial goal, and that instructed together at the same time.
many cells were activated by the spatial instruction without the A second difference is that in the cue–delay–target task, move-
effector cue. Many cells were responsive to both types of instruc- ments occur ‘on-demand’ with the appearance of the peripheral
tions, but some cells carried only effector information, whereas target. This contrasts sharply with the standard memory task, in
others carried only spatial information. which the movement is fully specified but must be withheld until
a ‘go’ cue is provided. Outside the laboratory, animals often do
RESULTS not wait for a cue, but instead move as soon as a target appears.
Delineation of PRR The cue–delay–target task, unlike a standard memory task,
PRR was defined functionally as the region of PPC containing approximates this type of behavior inside the laboratory.
visually responsive neurons, many of which show sustained firing In a single PRR cell aligned on the presentation of the
during a standard memory reach task (Fig. 1a)12,15. These func- movement-type cue (Fig. 2a, left), a cue instructing a reach result-
tional criteria were applied independent of the properties under ed in an increase in firing rate from baseline (10.5 ± 0.97 spikes/s,
study. Based on these independent functional criteria, we iden- mean ± s.e.m.) to a delay period rate of 27.0 ± 1.2 spikes/s (two-
tified a cube of cortex, of similar anterior–posterior (AP) and tailed t-test, P < 0.05), whereas a cue instructing a saccade had a
medial–lateral (ML) dimensions, in each of two animals significant inhibitory effect (11.3 ± 1.1 to 8.2 ± 1.0 spikes/s;
(Fig. 1b). Changes of 1–2 mm in these borders have minimal P < 0.05). The difference in delay period activity on reach and
effect on our results and do not change our overall conclusions. saccade trials was large and significant (27.0 versus 8.2 spikes/s;
We subsequently obtained high-resolution images of cortex in P < 0.001), despite no spatial information regarding target loca-
monkey 2 (M2) using magnetic resonance imaging and used tion, and despite no more than 1 degree of movement of the arm
those images to localize cells of interest (Fig. 3). or eyes (see below). In previous experiments in parietal, premo-
tor and motor cortex, activity related to an upcoming motor
Effector specificity response has been called motor preparation, motor set or motor
We tested whether neurons in PRR were activated by the instruc- intention27. It is important to recognize, however, that in con-
tion to prepare a reach, without spatial information regarding the trast to these earlier studies, we are describing a signal that occurs
target of that reach. In cue–delay–target trials (Fig. 1c), a central completely without spatial information.

articles
Fig. 2. Neuronal activity in PRR can be evoked by the intention to make

an arm movement even without a spatial target, or by spatial information a
without effector information. (a) A single PRR cell increased its firing
when instructed to prepare a reach (left, dark trace), but not when
instructed to prepare a saccade (light trace). Note the absence of a spa-
tial goal for the movement (cue–delay–target task). The same cell
showed clear spatial tuning (right) on interleaved trials when the move-
ment type had not yet been instructed (target–delay–cue task). Rasters
plot the temporal sequence of individual action potentials occurring in
each of the first eight trials. Traces and rasters are aligned on the presen-
tation of the cue (left) or target (right). The subsequent delay period was
600, 900 or 1200 ms long. Shading indicates the first 600 ms. To avoid
contamination by movement responses, data from trials with 600 ms
delay periods are truncated at 650 ms, causing a slight discontinuity in the
traces. An additional 300 ms of data are included from trials in which the
delay was 900 or 1200 ms. Traces shown for cue–delay–target trials (left)
were calculated using 64 reach and 64 saccade trials (8 trials × 8 direc-
b
tions for each cue). Traces shown for target–delay–cue trials (right) were
calculated using 16 ‘in RF’ (receptive field) and 16 ‘out RF’ trials (8 trials ×
2 effector cues for each direction). In this and subsequent figures, the ori-
gin of the y-axis corresponds to zero spikes/s. (b) Activity in most PRR
neurons, measured in the final 300 ms of the delay period, was greater on
reach than saccade trials (left; cue–delay–target task). Most cells fall to
the right of the diagonal line. Cells with significant effects are plotted as
filled symbols. For comparison, a similar scatter plot is shown for
target–delay–cue responses (right; target–delay–cue task). (c) Across the
population, PRR cells increased their firing when instructed to prepare a
reach (left, dark trace) but not when instructed to prepare a saccade
(light trace) without spatial information. For comparison, responses to
spatial information without effector information are shown on the right.
Trace thickness in (a) and (c) represents mean ± s.e.m.
Spatial responses c
To determine the response to spatial information without effec-
tor-specific intention, a second trial type was randomly inter-
leaved with the cue–delay–target trials. On target–delay–cue trials,
the peripheral target appeared first and then, after a delay period,
the movement cue was presented (Fig. 1d). The same cell that
showed non-spatial, effector-specific responses in the
cue–delay–target task also showed spatial responses in the
target–delay–cue task (Fig. 2a, right). The cell responded to the
peripheral stimulus when it fell inside the receptive field with a
transient increase followed by a sustained elevation in firing (17.5
± 3.0 spikes/s over baseline; P < 0.05). The cell was suppressed
(albeit not significantly) when the stimulus fell outside the recep-
tive field (3.7 ± 2.7 spikes/s below baseline, P > 0.05). These
responses occurred even though the movement type—saccade instruction to prepare a saccade (left, 3.6 ± 0.9 spikes/s increase
or reach—had not yet been specified. Thus, this cell shows clear and 0.9 ± 0.6 spikes/s decrease, respectively). On trials in which
spatial tuning despite uncertainty regarding how that spatial the delay period lasted longer than 600 ms, the differential effect
information will be used. was even larger. In the last 300 ms of the delay interval, firing had
increased on average by 5.6 ± 1.0 spikes/s when a reach was
Population responses instructed, but only by 0.1 ± 0.6 spikes/s when a saccade was
Like the example cell above, many cells in PRR were modulated instructed. Data from a second animal (M2) showed a similar
by effector-specific intention despite the absence of spatial infor- pattern: an increase of 4.8 ± 1.1 spikes/s after a reach instruction,
mation (Fig. 2b, left), and many cells were spatially tuned despite but only 2.1 ± 1.0 spikes/s after a saccade instruction (49 cells).
the absence of effector information (Fig. 2b, right). In this graph, These differences were significant in both animals (P < 0.05).
cells without effector-specific or spatial modulation would fall There was also robust spatial tuning on target–delay–cue trials
on the diagonal lines. Instead, most cells are to the right of this (Fig. 2c, right). Targets inside the receptive field evoked 14.2 ± 1.7
line, indicating a preference for reach cues over saccade cues (left) spikes/s and 8.9 ± 1.2 spikes/s more activity in each of the two ani-
and the presence of spatial tuning (right). mals, respectively, than targets outside the receptive field. Com-
As a measure of overall tendency and time course, responses paring the results of cue–delay–target and target–delay–cue trials
were averaged across every cell recorded in PRR that had a task- shows that over the last 300 ms of the delay period, non-spatial,
related response of any sort and at any time (Methods; Fig. 2c, effector-specific information was 39% (M1, 5.5 versus 14.2 spikes/s)
data from M1). Activity from 82 cells on cue–delay–target trials and 30% (M2, 2.7 versus 8.9 spikes/s) as effective in evoking a neur-
increased after an instruction to prepare a reach but not after an al response as spatial information without an effector instruction.

articles
Fig. 3. Intention cells in the IPS. (a) A roughly coronal MRI section
a b A
* showing a schematic electrode (thin white line) passing through the cen-
ter of the recording chamber (*) and into the IPS (animal M2). The sec-
tion is aligned with the path of the electrode. (b) A roughly horizontal
L R
MRI section, perpendicular to the path of the electrode and 8.0 mm
below the cortical surface (lower red line from a). A, anterior; P, poste-
rior; L, left; R, right. (c) Expanded view of the square in (b), showing the
L R P recording sites of IPS cells with intention-related activity. Cells with
reach-related intention activity (green circles) were found on both sides
c
of the proximal portion of the IPS. Only cells between 3.5 and 8.0 mm in
Midline
depth are shown (red lines in a). Within this range, electrodes traveled
IPS roughly perpendicular to the sulcus. Sites along the same track are jit-
tered by up to 0.25 mm for clarity. IPS, intraparietal sulcus; POS,
parieto-occipital sulcus; STS, superior temporal sulcus. Scale bar, 2 mm.
Alternative explanations: spatial prediction

STS We propose that effector-specific intention activity occurs even
Reach selective
without explicit spatial information. This is a surprising finding,
Saccade selective POS given the prevailing, well-supported notion that the dorsal path-
Nonselective
way is dedicated to the analysis of spatial information16,17. It is
therefore important to consider whether intention activity may
reflect implicit rather than explicit spatial information. Might an
expectation or prediction of target location substitute for explic-
On a cell-by-cell basis, 37% of cells within PRR were sig- it spatial information in our cue–delay–target task?
nificantly more active when the movement-type cue instruct- Cellular activity in LIP6,10, superior colliculus28–30 and the
ed a reach rather than a saccade (Table 1; data from two frontal eye fields (FEF)31 can reflect the probability of a movement
animals; Student’s t-test, P < 0.05). Only 8% of cells showed into the receptive field of the recorded cell. In the superior col-
the reverse preference. Thus, cells favoring the reach cue were liculus, the evoked activity is proportional to the probability that
significantly more common than cells favoring the saccade cue the target will appear in the receptive field28–30, as if on any given
(P < 0.001, chi-square test). Very few cells showed the reverse trial a prior probability or likelihood ratio is associated with each
pattern. Many cells were not activated in either condition, and of a number of potential spatial locations32. Perhaps effector-
some cells were activated equally in the two tasks. This conspecific activity in cue–delay–target trials is a related phenome-
trasts with PPC cells that we recorded outside of PRR. Fewer non. In this case, the activity we observed would still be effector
of these cells were significantly modulated by the movement- specific, but rather than occurring independently of spatial infor-
type cue, and of those that were, cells preferring arm and eye mation, it would be related to a prediction of target location.
movements were present in equal numbers (15% versus 14%, As a test of the spatial probability hypothesis, we examined
respectively; P > 0.05 chi-square test). whether delay period activity depends on the probability of the
target appearing in the receptive field. Such a relationship is cen-
Localization tral to the spatial prediction hypothesis. Although most cells were
Using anatomical MR imaging, we reconstructed the locations tested in blocks of trials where the target might appear at one of
of cells recorded in M2 relative to the intraparietal sulcus (IPS; only two possible locations, other cells were tested in blocks where
Fig. 3). Effector-specific cells preferring arm movements were the target could appear at any one of eight positions, and hence
present in large numbers on both banks of the proximal one- the probability of the target falling inside the receptive field was
third of the IPS. Of 33 reach cells in M2, 13 lay on the medial less for this second group of cells. The spatial anticipation hypoth-
bank of the IPS and 16 lay on the lateral bank. (Two more lay esis predicts that the difference in delay activity for arm com-
on the mesial wall, and two were
indeterminate.) For comparison, Table 1. Summary of neurons.
saccade cells were much more com-
Inside PRR Outside PRR
mon on the lateral bank and gener-
ally lay in the middle third of the M1 M2 Total M1 M2 Total
sulcus. Of 20 saccade intention cells Cue–delay–target task
in M2, 11 lay on the lateral bank of Cells tested 82 49 131 95 142 237
the IPS and only 2 on the medial Arm > eye 36 (44%) 13 (27%) 49 (37%) 15 (16%) 20 (14%) 35 (15%)
bank. (Four were indeterminate, Eye > arm 8 (10) 3 (6) 11 (8) 16 (17) 17 (12) 33 (14)
2 lay on the mesial wall, and 1 lay in Eye = arm > baseline 18 (22) 12 (24) 30 (23) 29 (31) 37 (26) 66 (28)
the STS.) This distribution of reach- Eye = arm = baseline 20 (24) 21 (43) 41 (31) 35 (37) 68 (48) 103 (43)
and saccade-preferring cells is sim-
ilar to that obtained using the stan- Target–delay–cue task
dard memory task that was used to Cells tested 76 48 124 79 142 221
functionally map PRR, both in the Spatial tuned cells 55 (72%) 25 (52%) 80 (65%) 44 (56%) 72 (51%) 116 (52%)
same animals (Fig. 1b) as well as in Data are shown for two animals (M1 and M2). Of the 368 cells tested in the cue–delay–target, all but 25
a previous study in different animals were also tested in the target–delay–cue task. All but 2 cells in the target–delay–cue task were tested in
(Fig. 3 of ref. 12). the cue–delay–target task.

articles
Fig. 4. Effector-specific activity in PRR is not driven by a prediction of

a upcoming target location. (a) Delay period activity showed a similar
preference for reach compared to saccade trials, regardless of whether
the cell was tested under conditions in which the target had a high prob-
ability (left) or low probability (right) of appearing in the receptive field,
in contrast to spatial predictive signals in the superior colliculus28–30.
Only spatially tuned cells are included. Traces show mean activity as a
function of time. (b) Delay period activity in cells without spatial tuning
(top row) was similar to delay activity in cells with strong spatial tuning
b (bottom row). Left column (target–delay–cue task) shows extent of

spatial tuning; right column (cue–delay–target task) demonstrates inten-
tion activity, which, if anything, is more robust among untuned cells.
Untuned cells include all those that failed a test of statistical significance
for spatial tuning (44 cells), excluding those cells in which nonsignificant
spatial tuning exceeded 3 spikes/s (13 cells). Tuned cells are all those
with statistically significant spatial tuning. (c) A wide range of spatial tun-
ing and effector-specific intention activity within individual cells. The
ordinates show spatial tuning, quantified as delay activity on
target–delay–cue trials: inside minus outside receptive field conditions.
The abscissas show intention activity, quantified as delay activity on
cue–delay–target trials: reach minus saccade conditions. Significant spa-
tial tuning is indicated by square versus round; significant effector-
specific intention activity is indicated by solid versus hollow. Many cells
show just one effect, and there is no consistent relationship between
intention activity and spatial tuning.
c
activity. If effector-specific activity reflects predicted spatial infor-
mation, then it should not be present in cells that are not spa-
tially tuned. Of 44 cells without spatial tuning in the
target–delay–cue task, 41% showed effector-specific activity in
cue–delay–target trials. This percentage was not significantly dif-
ferent from the 46% of the 78 cells with spatial tuning that were
effector-specific (P > 0.05, chi-square test). In addition, the mag-
nitude of effector-specific activity was similar for spatially tuned
and untuned cells (Fig. 4b).
pared to eye trials in spatially tuned cells will be greater in the Interaction between effector and spatial signals
two- compared to the eight-target condition. Instead, the differ- To understand what computations PRR performs, it is necessary
ence between arm and eye trials was 5.7 ± 2.0 spikes/s with two to know the relationship between spatial and effector-specific
possible target locations, and 7.9 ± 3.3 spikes/s with eight possi- signals. We showed that the presence or absence of spatial activ-
ble target locations (Fig. 4a; 26 and 15 cells tested, respectively). ity did not affect average effector-specific intention within PRR
These two values are not significantly different from one anoth- (Fig. 4b). Further, when the amplitude of spatial activity
er (P > 0.05, one-tailed t-test). In contrast, the latency of eye (target–delay–cue task) is plotted against the amplitude of effec-
movements in the two data sets showed a highly significant dif- tor-specific intention activity (cue–delay–target task), there are
ference (169.8 ± 1.2 ms, n = 486 trials, with two possible target cells with only spatial activity, cells with only effector-specific
locations, and 178.6 ± 1.0 ms, n = 806 trials, with eight possible intention activity, and cells with both types of responses
target locations; P < 0.001, Student’s t-test). This indicates that (Fig. 4c). A statistical analysis reveals that the two properties occur
the animals were indeed using the information contained in the
probability structure to optimize their behavior, and yet the effect
of this optimization was not apparent in the intention activity.
This is direct evidence that intention activity does not reflect the
strength of the prediction that a target will land in the receptive
field of the cell.
As a second test of the spatial prediction hypothesis, we asked
whether cells without spatial tuning showed effector-specific delay
Fig. 5. Effector-specific activity in PRR is not driven by proprioceptive

feedback from the arm or eyes. Left column, average horizontal and ver-
tical arm position during the delay periods of cue–delay–target trials.
For both animals, arm position was influenced little by the effector type
that was cued: reach (dark traces) or saccade (light traces). Right col-
umn, similar plots for eye position. The cued effector had a small influ-
ence on eye position, although the difference between saccade and
reach trials never exceeded 1 degree and is therefore unlikely to
account for our results. Trace thickness represents mean ± s.e.m.

articles
Fig. 6. Effector-specific activity in PRR is not driven by color-selective

sensory responses. Receptive-field plots for two typical PRR cells (top).
Filled circles represent the response to a flashed stimulus at each of 72
locations from 3.5 to 40 degrees from the fovea. Filled circle diameter
indicates mean firing rate, and shading (grey to black) indicates the reli-
ability of each mean. Traces (below) show mean ± s.e.m. of the
responses on cue–delay–target trials. When the receptive field excludes
the fovea (left), effector-specific activity begins ∼200 ms after cue onset.
This activity is different for saccade and reach instructions. When the
receptive field includes the fovea (right), there is an increase in activity

beginning ∼100 ms after cue onset. In this case, the initial response is
similar for saccade and reach instructions and probably reflects a simple
sensory response to the foveal cue. After 200–300 ms, however, effec-
tor-specific activity emerges. The similarity of effector-specific
responses in cells with and without foveal receptive fields suggests that
effector specificity does not result from a learned color preference.
independently of one another. In particular, the probability of a

cell having significant spatial tuning was independent of the pres-
ence or absence of effector-specific activity, and, as noted in the A third reason for rejecting a causal relationship between dif-
previous section, the probability of having significant effector- ferential eye movement and differential firing is that the qualita-
specific activity was independent of the presence or absence of tive effects are dissimilar. In both animals, firing rate increased
spatial tuning (P > 0.05; chi-square tests). Thus, spatial infor- sharply on arm trials, but only slightly on eye trials (Fig. 2c). In
mation and effector information are coded by two partially over- M1, eye position shows the reverse pattern: a sudden sharp change
lapping sets of cells in PRR. on eye trials, but only a slight change on arm trials. Eye position in
M2 shows yet another pattern: equal and opposite changes for eye
Alternative explanations: delay period movement and arm trials. In neither animal does the pattern of differential
In addition to implicit spatial information, we considered other eye movement match the pattern of differential activity.
alternative explanations for effector-specific intention activity.
Animals might make subtle movements during the delay period Alternative explanations: learned color preferences
of cue–delay–target trials. These differences might differ sys- It is possible that what we have identified as effector-specific
tematically on reach compared to saccade trials. We therefore intention instead reflects a learned receptive field preference for
investigated whether a proprioceptive response to differential different colored stimuli35. To rule out this possibility, we care-
movement might explain effector-specific delay-period activity. fully mapped sensory receptive field borders in 10 PRR cells
Arm movements greater than 3 degrees, even if made slowly, showing effector-specific intention. Six cells had receptive fields
would be detected on-line automatically and result in the trial being that clearly excluded the fovea and parafovea, yet nonetheless
aborted. Smaller movements, however, could be accepted. To test showed clear effector-specific intention (Fig. 6, left). Effector-
for slight but systematic differences in movement, we pooled behav- specific (and instruction-specific) effects typically emerged
ioral data (absolute change in arm and eye position) across all trials 200–300 ms after the appearance of the cue. The remaining four
(Fig. 5). This analysis provided a very large amount of statistical cells had receptive fields that included the fovea or parafovea
power, as thousands of trials were averaged together in each case. (Fig. 6, right). We would expect that learned color preferences
Average arm position changed little over the course of the would manifest sooner and more strongly when the stimulus
delay period, with the difference in arm position between reach appears within the receptive field, compared to outside the recep-
and saccade trials always less than 0.2 degrees in each animal. For tive field. Yet this was not the case. Cells with central receptive
this change in position to account for the observed changes in fields, unlike those with peripheral fields, typically showed an
firing (5.5 and 2.7 spikes/s in M1 and M2) would require sensi- initial non-specific increase in activity starting ∼100 ms after cue
tivities of 27 and 13 spikes/s per degrees of arm movement, onset, followed by an effector-specific divergence after 200–300
respectively. Such a high sensitivity is extremely unlikely. ms (Fig. 6, right). Thus effector-specific divergences occurred
In contrast to the nearly identical arm movements, there was with similar frequency and at similar times in cells with and with-
a small amount of differential eye movement (0.8 and 0.5 degrees out parafoveal responses. This strongly suggests that the diver-
in M1 and M2, respectively). In M1, the eyes were displaced up gence does not reflect a simple learned color preference.
and to the right 200 ms after cue onset, with more upward move- Across the population of cells, divergences in firing on
ment on saccade trials. In M2, the eyes drifted down and to the cue–delay–target saccade and cue–delay–target reach trials
left on saccade trials, and up and to the right on reach trials. In almost never occurred within 100 ms of cue onset, and had a
both animals, the drift was gradual. median latency of 233 ms. (For comparison, the visual response
Might the difference in PRR delay activity reflect this differ- latencies in PRR, to targets presented in the receptive field, had
ence in eye movement, either as a result of proprioceptive feed- a median latency of 104 ms). The long latencies for the diver-
back or efference copy? We believe this is unlikely for three reasons. gence of activity on reach and saccade trials are consistent with
First, the required eye movement sensitivity (5 spikes/s per degrees the time required for processing symbolic cues, not color-
of eye movement) is large for parietal cortex33,34. Second, it is selective sensory responses36.
unlikely that there was any systematic relationship between pre-
ferred gain-field directions and differences in eye position; for DISCUSSION
example, when the eye position traces were realigned according Our results show that PPC neurons can be activated not only by
to the preferred direction of each cell, all differences in eye position spatial information without a motor plan, but also by the plan to
between reach and saccade trials disappeared (data not shown). use a specific effector without spatial information (Fig. 2). Some

articles
cells carry both spatial information and effector-specific inten- haps this reciprocal connectivity explains in part why action
tion signals, but many cells carry just one signal or the other selection occurs in both dorsal premotor and posterior parietal
(Fig. 4c). The coding of spatial information without a motor cortices. The finding that some PRR cells code effector-
plan is hardly surprising, given previous ideas of PPC as the specific instructions, others code spatial instructions, and still
region in which spatial information is processed15. In addition, others code both instructions, suggests that PRR, like dorsal
PPC is organized into regions with privileged connections to premotor cortex, is involved in specifying how the animal will
particular effectors 37 such as eye movements 12, arm move- respond to a particular target.
ments12,20–26,38,39 and grasping movements of the hand40,41. Even Although PPC seems to be well organized with respect to
so, the current finding that effector-specific intentions are encod- effector system12,13, there is only a coarse organization of spatial
ed without spatial information is unexpected. On average, signals42. Thus, whereas effector-specific signals seem to be well
effector-specific intention signals drove cells about one-third as localized within particular cortical regions such as PRR (Fig. 1b;
much as spatial signals. These findings strongly suggest that PPC Table 1), similar responses to visual targets can be found through-
is involved in specifying how an organism will respond to a par- out much of the intraparietal sulcus (Table 1). This suggests that,
ticular target (motor intention). without an effector-specific plan to respond to a target, the
Because we generally did not map receptive fields with high appearance of a target will result in diffuse, widespread activa-
spatial resolution, we are likely to have underestimated the effects tion of PPC. Such activation in multiple regions could be viewed
of spatial information. However, the few cells that we did map as contingency plans for multiple potential movements43.
in high resolution showed broad tuning (Fig. 6), and therefore
it is unlikely that finer-resolution mapping would greatly increase Localization
our estimate of population-averaged spatial modulation; hence Neurons with arm-specific intention activity lie on both banks
we expect that it would not greatly change our estimates of the of the proximal portion of the IPS. The proximal one-third of
relative magnitudes of intention and spatial signals. the medial bank is area MIP (medial intraparietal)44,45, although
Conceivably, the effector-specific intention signals may be this name is often misapplied to refer to the entire medial bank46.
modulating implicit rather than explicit spatial information, in The proximal one-third of the lateral bank has been named area
the form of an expectation or prediction about where a target cIPS (caudal intraparietal sulcus)47 or, more recently, zone LOP
will appear. Predictive spatial activity has been described in (lateral occipitoparietal)48. However, the boundaries of both areas
LIP6,38,39. However, this activity appeared when there was near are vague 48. It is not known, for example, whether LIP and
certainty regarding where the target would appear, whereas we cIPS/LOP tile the entire proximal IPS or whether other areas are
observed activity in the presence of as many as eight possible tar- also present. Therefore, for the present we prefer the term PRR.
get locations (Fig. 4a). Predictive activity in the presence of eight Our use of the term ‘region’ rather than ‘area’ is intended to
possible target locations is observed in the superior colliculus28–30. highlight that these neurons have been grouped by function, and
However, the magnitude of predictive activity in the colliculus not by connectivity or histology. Their exact relationship to par-
depends on the probability that a target will appear in the recep- ticular cortical areas remains uncertain, although they clearly over-
tive field of the cell being recorded. This was not the case in PRR lap at least parts of both MIP and cIPS/LOP. Based on the clear
(Fig. 4a). Indeed, even cells without spatial tuning showed effec- and consistent differences in connectivity between the two banks
tor-specific activity, inconsistent with the idea that this activity of the proximal IPS20, we do not expect PRR neurons to form a
is associated with any sort of implicit spatial information, includ- single homogenous functional population. For example, it would
ing a spatially tuned predictive signal (Fig. 4b). be worthwhile to investigate tuning for visual disparity, which has
Finally, the timing of the effector-specific response in PRR is been demonstrated on the lateral bank47 but has not been tested
inappropriate for predictive spatial activity. In LIP, FEF and the on the medial bank. In addition, even properties held in common
superior colliculus, activity predictive of a target appearance aris- by neurons on both banks may prove to have different function-
es shortly before the target is expected to appear6,28–31. In our al consequences for the organism. However, neurons on both
experiments, cue–delay–target trials were interleaved with tar- banks share many properties in common. Both MIP and cIPS
get–delay–cue trials. As a result, on half of all trials, the peripheral neurons are visually responsive, and both are hypothesized to be
target was the first instructional stimulus to appear. Therefore involved in reaching or grasping44,47. Reach-selective memory
one would expect a predictive signal to arise shortly before the activity for spatial locations has been demonstrated12, and we now
first instructional stimulus on every trial. This expectation was show arm-specificity without spatial information in both areas.
not fulfilled. There was no increase in activity shortly before the In conclusion, we have recorded from a localized region of PPC
first stimulus in cue–delay–target and target–delay–cue trials that codes not only spatial information about the goal of an
(Fig. 2c). Thus the delay period activity we observed is quite dif- upcoming movement, but also non-spatial information about the
ferent from spatial predictive activity, and is unlikely to reflect a effector to be used to achieve that goal. These cells lie on both banks
modulation of implicit spatial information by effector-specific of the proximal portion of the IPS. The presence of non-spatial,
intentions. Instead, PRR neurons encode effector-specific inten- task-specific information in PPC is at odds with its classical role
tions even without spatial information. as a sensory association cortex, and instead supports the notion
Cells that can encode effector-specific information and tar- that this region is active in sensorimotor transformations.
get information in isolation or in combination have been
reported elsewhere in the brain. In particular, activity in dor- METHODS
sal premotor cortex reflects both the position of a reach target Recording procedures. Animals sat in a custom-designed monkey chair
(spatial information) and the arm to be used for the reach (Crist Instrument, Hagerstown, Maryland) with an open front that
allowed reaches toward a visual stimulus. Stimuli were back-projected
(effector information)18. As in PRR, either spatial information by a CRT projector onto a touch panel 25 cm in front of the animal. All
or effector-specific information can evoke activity without the experiments complied with the relevant laws and institutional guide-
other. Dorsal premotor cortex has extensive and reciprocal con- lines, and were approved by the Washington University Institutional Ani-
nections with many parts of PPC involved in reaching19–21. Per- mal Care and Use Committee.

articles
Recordings were made from the left hemispheres of two adult rhesus three-dimensional datasets (Siemens multiplanar rapid acquisition gra-
monkeys. Recording chambers were centered at 5 mm posterior and dient echo, TR = 14.3 ms, TE = 7.0 ms, TI = 15 ms, 0.8 mm isotropic
12 mm lateral (Horsley–Clarke coordinates) and placed flush to the skull. voxels), which were averaged online for signal-to-noise enhancement.
While we searched for cells, animals performed non-delayed, center-out To determine the orientation, depth, and scale of our recording grid
combined eye and arm movements to 20 degrees peripheral targets in each relative to the MRI, we placed a custom-designed plastic cylinder (5 cm
of eight directions. Cells that changed firing rate at any point in the search height, 1.5 cm diameter) containing MR contrast agent (0.015 mM
task (for example, at target appearance or movement onset) were tested gadoversedamide) into the animal’s recording chamber. Bars at known
further. To map PRR, we tested 220 cells (40 M1, 180 M2) on a standard locations within the cylinder (2 mm × 2 mm × 8 mm, 4.5 mm spacing)
interleaved memory reach and saccade task12 (Fig. 1a). After 500 ms of displaced the contrast agent (Fig. 3a), allowing us to reconstruct the posi-
central fixation and touch, a red or green peripheral target appeared at one tion and orientation of the recording chamber and grid (Crist Instru-
of eight positions, instructing either a reach or saccade to the target. After ments) through which our electrodes passed. From this, we were able to
a delay (800 ms), the central stimulus disappeared, and the animal made a project our recording sites onto the MR image with an estimated accuracy
reach without moving its eyes, or a saccade without moving its arms, to of 1 mm or better (Fig. 3c).
the remembered location of the peripheral target. We delineated PRR based
on neural activity during the delay period (Fig. 1b; M1, dimensions,
ML × AP × DV, 5 × 6 × 8 mm; M2, 5 × 6 × 4 mm). Acknowledgments
We thank Y.E. Cohen, G.C. DeAngelis and J.H.R. Maunsell for comments on an
Behavioral tasks. Most cells were tested on cue–delay–target and early version of the manuscript, J. Baker, A. Snyder and D.C. Van Essen for the
target–delay–cue tasks (Fig. 1c and d). Each trial began with the mon- anatomical MR and T. Harper for technical assistance. This work was supported
key fixating and reaching for a blue central square. On cue–delay–target by the Klingenstein Fund, the Sloan Foundation, the McDonnell Center for
trials, after a variable delay of 500–800 ms, the color of the fixation point Higher Brain Research and the NIH.
changed to green or red to instruct an arm or eye movement (color
assignment was reversed between the two animals) and remained on for Competing interests statement
the duration of the trial. After a second delay of 600, 900 or 1,200 ms, a
The authors declare that they have no competing financial interests.
blue peripheral target appeared at one of eight equally spaced positions
located 20° from the central reach/fixation stimulus. Animals were free to
RECEIVED 29 JANUARY; ACCEPTED 1 MAY 2002
acquire the peripheral target as soon as it appeared, according to the pre-
vious movement-type instruction. They received a liquid reward after
holding the target for 400 ms, and so were motivated to respond at short 1. Hyvarinen, J. & Poranen, A. Function of the parietal associative area 7 as
latency. Eye-movement latencies for the two monkeys averaged revealed from cellular discharges in alert monkeys. Brain 97, 673–692 (1974).
171 ± 22 ms and 240 ± 74 ms (mean ± s.d.). Reach latencies averaged 2. Mountcastle, V. B., Lynch, J. G., Georgopoulos, A., Sakata, H. & Acuna, C.
292 ± 90 ms and 384 ± 93 ms, respectively. Posterior parietal association cortex of the monkey: command functions for
operations within extrapersonal space. J. Neurophysiol. 38, 871–908 (1974).
Target–delay–cue trials were similar to cue–delay–target trials, except 3. Robinson, D. L., Goldberg, M. E. & Stanton, G .B. Parietal association cortex
that the spatial target was presented first, and the foveal color change in the primate: sensory mechanisms and behavioral modulation.
occurred second. The variable delay period was similar to that of J. Neurophysiol. 41, 910–932 (1978).
cue–delay–target trials, and the animal was free to acquire the target once 4. Bushnell, M. C., Goldberg, M . E. & Robinson, D. L. Behavioral enhancement
of visual responses in monkey cerebral cortex. I. Modulation in posterior
the foveal color change had occurred. Target–delay–cue and parietal cortex related to selective visual attention. J. Neurophysiol. 46,
cue–delay–target trials were interleaved for 90% of recorded cells. 755–772 (1981).
Arm and eye positions were monitored by a 43.2 cm touch panel 5. Kusunoki, M., Gottlieb, J. & Goldberg, M. E. The lateral intraparietal area as a
(Keytec, Richardson, Texas) and a scleral search coil (CNC Engineering, salience map: the representation of abrupt onset, stimulus motion, and task
relevance. Vision Res. 40, 1459–1468 (2000).
Seattle, Washington), respectively. Trials were terminated if animals failed 6. Colby, C. L, Duhamel, J. R. & Goldberg, M. E. Visual, presaccadic, and
to maintain touch and fixation within a square window (side length ± cognitive activation of single neurons in monkey lateral intraparietal area.
6° for touch, ± 3° for fixation) or moved their arm or eyes more than 3° J. Neurophysiol. 76, 2841–2852 (1996).
in any direction during the delay period. Trials were also terminated and 7. Gottlieb, J. P., Kusunoki, M. & Goldberg, M. E. The representation of visual
salience in monkey parietal cortex. Nature 391, 481–484 (1998).
the data not analyzed if the eyes moved before the reward was delivered 8. Mazzoni, P., Bracewell, R. M., Barash, S. & Andersen, R. A. Spatially tuned
on a reach trial, or if the arm moved before the reward was delivered on auditory responses in area LIP of macaques performing delayed memory
a saccade trial. In practice, animals almost never moved the uninstruct- saccades to acoustic targets. J. Neurophysiol. 75, 1233–1241 (1996).
ed body part after receiving the reward, but instead returned the periph- 9. Linden, J. F., Grunewald, A. & Andersen, R. A. Responses to auditory stimuli
erally displaced limb directly back to the central target to start the next in macaque lateral intraparietal area. II. Behavioral modulation.
J. Neurophysiol. 82, 343–58 (1999).
trial. All tasks were performed in the dark. Animals typically performed 10. Platt, M. L. & Glimcher, P. W. Neural correlates of decision variables in
over 80% of trials successfully. parietal cortex. Nature 400, 233–238 (1999).
Each task (cue–delay–target saccade, cue–delay–target reach, 11. Mazzoni, P., Bracewell, M., Barash, S. & Andersen, R. A. Motor intention
target–delay–cue saccade, target–delay–cue reach) was repeated 8– activity in the macaque’s lateral intraparietal area. I. Dissociation of motor
plan from sensory memory. J. Neurophysiol. 76, 1439–1456 (1996).
10 times using a peripheral target in one of two, four or eight interleaved 12. Snyder, L. H., Batista, A. P. & Andersen, R. A. Coding of intention in the
directions (292, 9 and 67 cells, respectively). When testing in only two posterior parietal cortex. Nature 386, 167–170 (1997).
or four directions, we used a pilot run of two target–delay–cue trials in 13. Colby, C. L. Action-oriented spatial reference frames in cortex. Neuron 20,
each of eight directions to establish the best direction. Receptive fields 15–24 (1998).
14. Platt, M. L. & Glimcher, P. W. Responses of intraparietal neurons to saccadic
were plotted with finer resolution in only ten cells (Fig. 5). Unless oth- targets and visual distractors. J. Neurophysiol. 78, 1574–1589 (1997).
erwise mentioned, baseline and delay period firing rates were measured 15. Snyder, L. S., Batista, A. P. & Andersen, R. A. Intention-related activity in the
0–300 ms before and 300–600 ms after cue onset, respectively, and sig- posterior parietal cortex: a review. Vision Res. 40, 1433–1441 (2000).
nificance was tested with a two-tailed t-test and a criterion value of 16. Ungerleider, L. & Mishkin, M. in Analysis of Visual Behavior (eds. Ingle, D. J.
P < 0.05. Traces were smoothed using a 191 point digital low-pass filter Goodale, M. A. & Mansfield, R. J. W.) 549–586 (MIT Press, Cambridge,
Massachusetts, 1982).
with a transition band spanning 20 to 32 Hz. 17. Goodale, M. A. & Milner, A. D. Separate visual pathways for perception and
action. Trends Neurosci. 15, 20–25 (1992).
Anatomical localization. Anatomical MR images of M2 were acquired 18. Hoshi, E. & Tanji, J. Integration of target and body-part information in the
using a custom-designed surface coil (NOVA Medical, Wakefield, Mass- premotor cortex when planning action. Nature 408, 466–470 (2000).
achusetts) in a Siemens Allegra MR system (Siemens Medical Systems, 19. Kurata, K. Corticocortical inputs to the dorsal and ventral aspects of the
premotor cortex of macaque monkeys. Neurosci. Res. 12, 263–280 (1991).
Erlangen, Germany) operating at 3 T. The anesthetized animal lay prone 20. Tanne, J., Boussaoud, D., Boyer-Zeller, N. & Rouiller, E. M. Direct visual
with his neck partially extended and supported by foam. With the sur- pathways for reaching movements in the macaque monkey. Neuroreport 7,
face coil positioned around the animal’s chamber, we acquired eight 267–272 (1995).

articles
21. Johnson, P. B., Ferraina, S., Bianchi, L. & Caminiti, R. Cortical networks for frontal eye fields induced by experience in mature macaques. Nature 381,
visual reaching: physiological and anatomical organization of frontal and 697–699 (1996).
parietal lobe arm regions. Cereb. Cortex 6, 102–119 (1996). 36. Kustov, A. A. & Robinson, D. L. Shared neural control of attentional shifts
22. Ferraina, S. et al. Combination of hand and gaze signals during reaching: and eye movements. Nature 384, 74–77 (1996).
activity in parietal area 7 m of the monkey. J. Neurophysiol. 77, 1034–1038 37. Colby, C. L. & Goldberg, M. E. Space and attention in parietal cortex. Annu.
(1997). Rev. Neurosci. 22, 319–349 (1999).
23. Battaglia-Mayer, A. et al. Early coding of reaching in the parietooccipital 38. Assad, J. A. & Maunsell, J. H. Neuronal correlates of inferred motion in
cortex. J. Neurophysiol. 83, 2374–2391 (2000). primate posterior parietal cortex. Nature 373, 518–521 (1995).
24. Ferraina, S. et al. Early coding of visuomanual coordination during reaching 39. Eskandar, E. N. & Assad, J. A. Dissociation of visual, motor and predictive
in parietal area PEc. J. Neurophysiol. 85, 462–467 (2001). signals in parietal cortex during visual guidance. Nature Neurosci. 2, 88–93
25. Fattori, P., Gamberini, M., Kutz, D. F. & Galletti, C. ‘Arm-reaching’ neurons (1999).
in the parietal area V6A of the macaque monkey. Eur. J. Neurosci. 13, 40. Murata, A., Gallese, V., Kaseda, M. & Sakata, H. Parietal neurons related to
2309–2313 (2001). memory-guided hand manipuation. J. Neurophysiol. 75, 2180–2186 (1996).
26. Kalaska, J. F. & Crammond, D. J. Deciding not to GO: neuronal correlates of 41. Obayashi, S. et al. Functional brain mapping of monkey tool use. Neuroimage
response selection in a GO/NOGO task in primate premotor and parietal 14, 853–861 (2001).
cortex. Cereb. Cortex 5, 410–428 (1995). 42. Ben Hamed, S., Duhamel, J. R., Bremmer, F. & Graf, W. Representation of the
27. Wise, S. P. & Kurata, K. Set-related activity in the premotor cortex of rhesus visual field in the lateral intraparietal area of macaque monkeys: a
monkeys: effect of triggering cues and relatively long delay intervals. quantitative receptive field analysis. Exp. Brain Res. 140, 127–144 (2001).
Somatosens. Mot. Res. 6, 455–476 (1989). 43. Cisek, P. & Kalaska, J. F. Simultaneous encoding of multiple potential reach
28. Basso, M. A. & Wurtz, R. H. Modulation of neuronal activity by target directions in dorsal premotor cortex. J. Neurophysiol. 87, 1149–1154 (2002).
uncertainty. Nature 389, 66–69 (1997). 44. Colby, C. L. & Duhamel, J. R. Heterogeneity of extrastriate visual areas and
29. Basso, M. A. & Wurtz, R. H. Modulation of neuronal activity in superior multiple parietal areas in the macaque monkey. Neuropsychologia 29,
colliculus by changes in target probability. J. Neurosci. 18, 7519–7534 (1998). 517–537 (1991).
30. Dorris, M. C. & Munoz, D. P. Saccadic probability influences motor preparation 45. Colby, C. L., Gattass, R., Olson, C. R. & Gross, C. G. Topographical
signals and time to saccadic initiation. J. Neurosci. 18, 7015–7026 (1998). organization of cortical afferents to extrastriate visual area PO in the
31. Umeno, M. M. & Goldberg, M. E. Spatial processing in the monkey frontal macaque: a dual tracer study. J. Comp. Neurol. 269, 392–413 (1988).
eye field. II. Memory responses. J. Neurophysiol. 86, 2344–2352 (2001). 46. Paxinos, G., Huang, X. & Toga, A. W. The Rhesus Monkey Brain in Stereotaxic
32. Gold, J. I. & Shadlen, M. N. Representation of a perceptual decision in Coordinates (Academic, San Diego, 2000).
developing oculomotor commands. Nature 404, 390–394 (2000). 47. Sakata, H. et al. Neural coding of 3D features of objects for hand action in the
33. Andersen, R. A., Essick, G. K. & Siegel, R. M. Encoding of spatial location by parietal cortex of the monkey. Phil. Trans. R. Soc. Lond. B Biol. Sci. 353,
posterior parietal neurons. Science 230, 456–458 (1985). 1363–1373 (1998).
34. Boussaoud, D. & Bremmer, F. Gaze effects in the cerebral cortex: reference 48. Lewis, J. W. & Van Essen, D. C. Mapping of architectonic subdivisions in the
frames for space coding and action. Exp. Brain Res. 128, 170–180 (1999). macaque monkey, with emphasis on parieto-occipital cortex. J. Comp.
35. Bichot, N. P., Schall, J. D. & Thompson, K. G. Visual feature selectivity in Neurol. 428, 79–111 (2000).

articles
Neural activity in early visual cortex

reflects behavioral experience and
higher-order perceptual saliency
Tai Sing Lee1,2, Cindy F. Yang1, Richard D. Romero1,2 and David Mumford3
1 Center for the Neural Basis of Cognition and 2 Computer Science Department Carnegie Mellon University, 4400 Fifth Avenue,
Pittsburgh, Pennsylvania 15213, USA
3 Division of Applied Mathematics, 182 George Street, Brown University, Providence, Rhode Island 02912, USA
Correspondence should be addressed to T.S.L. (tai@cnbc.cmu.edu)
We report here that shape-from-shading stimuli evoked a long-latency contextual pop-out response in
V1 and V2 neurons of macaque monkeys, particularly after the monkeys had used the stimuli in a
behavioral task. The magnitudes of the pop-out responses were correlated to the monkeys’ behavioral
performance, suggesting that these signals are neural correlates of perceptual pop-out saliency. The
signals changed with the animal’s behavioral adaptation to stimulus contingencies, indicating that per-
ceptual saliency is also a function of experience and behavioral relevance. The evidence that higher-
order stimulus attributes and task experience can influence early visual processing supports the notion
that perceptual computation is an interactive and plastic process involving multiple cortical areas.
The computation of object saliency is important for directing from-above and lighting-from-below scenarios, the pop-out per-
attention and for guiding eye movements during analysis of a cept is immediate. By contrast, the interpretation of lighting direc-
visual scene. This computation is mediated by both bottom-up tion is ambiguous for stimuli being lit from the side, resulting in
autonomous processes1–2 and top-down attentional selection, a pop-out percept that is much less compelling (Fig. 1b).
which is a function of perception, experience and task demands3. We have coded stimuli according to the way they were gener-
As bottom-up and top-down processes are necessarily inter- ated (shading on a 3D Lambertian sphere with lighting from dif-
twined, it is difficult to cleanly separate their contribution to ferent directions): lighting from above (LA), below (LB), left (LL)
neural activity in visual cortex. Earlier single-unit studies in awake and right (LR) (Fig. 1b). Two 2D contrast elements, white above
and anesthetized monkeys have implicated the primary visual (WA) and white below (WB), were used as controls. Hence, there
cortex in mediating the bottom-up pop-out saliency computa- were six stimulus sets used in the experiments. The 3D elements
tion of oriented bars and contrast gratings4–6. Recent studies (LA, LB, LL, LR) were defined by shading, not stereoscopically.
using texture-contrast stimuli have suggested that top-down feed- Four conditions were used in association with each element type:
back may also be involved7–10. In these studies, however, per- singleton, oddball, uniform and hole (Fig. 1b). Human psycho-
ceptual saliency and orientation contrast were correlated, making logical studies11–13 have shown that among these stimuli, percep-
it difficult to distinguish the relative contribution of feed- tual pop-out saliency is strongest for the LA and LB stimuli,
forward or local mechanisms from that of inter-cortical feed- weaker for the LL and LR stimuli and weakest for the WA and WB
back. Here we studied the influence of two top-down factors— stimuli, even though WA and WB have the strongest luminance
higher-order perceptual inference and behavioral experience—in contrast and should presumably elicit the strongest stimulus-
shaping the saliency computation in early visual cortex. driven response in V1. These observations point to the existence
We used a set of stimuli including shape-from-shading images of neural responses that are correlated with perceptual pop-out
that have been used to demonstrate that parallel pop-out can saliency that can be dissociated from the strength of stimulus con-
occur with ‘high-level’ perceptual constructs11–13. When viewing trast. Higher-order perceptual areas probably participate in the
such a stimulus (Fig. 1a), one readily perceives a convex oddball computation of shape from shading, and these stimuli provide a
popping out from a background of concave distractors. Two- way of testing whether or not the computation of perceptual
dimensional contrast patterns, such as the white above (WA) and saliency involves interaction between early visual areas and
white below (WB) stimuli (Fig. 1b), do not pop out as readily. higher-order extrastriate areas.
The degree of perceptual pop-out seems to depend on the three-
dimensional (3D) interpretation of the elements. The 3D shape RESULTS
inference is influenced by a single global interpretation of light- A series of neuronal recording and behavioral testing experiments
ing direction11. If, instead, we interpret the scene (Fig. 1a) as being was conducted on two rhesus monkeys in nine stages. The data
lit from below, the oddball can be seen as concave, with the dis- presented came from the analysis of 410 units in V1 and 138 units
tractors in the surround appearing convex. In both the lighting- in V2. These included both well-isolated single units and multi-

articles
a b ed the strongest response relative to the other

conditions. That is, when stimulus elements were
added to the surround of the cells’ classical recep-
tive fields, the responses were suppressed from
the very beginning, indicative of the immediate
nature of the competitive lateral inhibition from
the surround (Fig. 2a and b). Moreover, the ini-
tial responses (40–100 ms after stimulus onset)
of V1 neurons to stimuli WA and WB were about

1.5–2 times stronger than were their responses to
the shape-from-shading stimuli, underscoring
the cells’ default sensitivity to stimulus contrast.
These observations were true for all the neural
data recorded from V1 across the various stages
for both monkeys.
Fig. 1. Stimulus sets used in the experiment. (a) A typical stimulus display was composed The enhancement of a neuron’s response to
of 10 × 10 stimulus elements. Each element was 1° visual angle in diameter. The diameter the oddball condition relative to the uniform con-
of the classical receptive field (RF) of a typical cell at the eccentricities tested ranged from dition is called the pop-out response. We quanti-
0.4° to 0.8°. Displayed is the LA (lighting-from-above) oddball condition, with the LA odd- fied it using a normalized pop-out modulation
ball placed on top of the cell’s receptive field (open circle). The solid dot indicates the fix- ratio, given by (Ro – Ru)/(Ro + Ru) where Ro and
ation spot. (b) In stages 1, 3 and 5, each stimulus set had four conditions: singleton, Ru are the responses of the neuron in a certain
oddball, uniform and hole. Shown are the iconic diagrams of all the conditions for the LA time window to the oddball and uniform condi-
set and the LB set, as well as the oddball conditions for the other four sets. The center tions, respectively. The modulation ratio report-
element in the iconic diagram covered the receptive field of the neuron in the experiment.
ed was computed in the time window 120–
The surround stimulus elements were placed outside the RF of the neuron. The compar-
ison was between the oddball condition and the uniform condition; the singleton and the
320 ms after stimulus onset (for comparisons of
hole conditions were controls. The singletons measured the neuronal response to direct modulations across different time windows, see
stimulation of the RF alone; the holes measured the response to direct stimulation of the Supplementary Note (Figs. 7–12) online). The
extra-RF surround only. mean pop-out modulation ratios of monkey A’s
V1 neurons in stage 1 showed marginally signifi-
cant (P < 0.05, t-test) pop-out responses to the
LA and LB stimuli, but not to the other stimuli
ple units. Neural recording at all recording stages was done while (Fig. 2c). Neurons in monkey B showed essentially the same pat-
monkeys were engaged in a fixation task (see Methods). tern: a marginally significant (P < 0.05) pop-out response to LB
The stimulus was presented on the computer screen for and not to the other stimuli (Fig. 3a–c). It seemed that when the
350 ms in each trial. During recording, the test stimuli were not monkeys were naive to the behavioral relevance of the stimuli,
relevant to the monkey’s behaviors. The fixation task was cho- their V1 neurons did not exhibit significant sensitivity to the 3D
sen to allow direct comparison between the neural responses at shape attributes in the stimuli.
the different stages, before and after behavioral training. It also
removed the confounding enhancement effect known to be asso- Behavioral relevance
ciated with saccadic eye movement toward the receptive fields of To assess the impact of behavior on perceptual processing, we
the neurons in early visual areas14. trained the monkeys to perform an oddball detection task in
Only V1 cells that (i) showed a marked response to at least stage 2 (see Methods). They learned the task rapidly but were
one of the shape-from-shading stimuli, and (ii) produced a neg- each trained for 15 sessions (1,200 trials per session) to attain
ligible response to all hole stimuli were included in our analysis. proficiency. We separated the stimuli into three pairs in descend-
In the experiment, the probe stimulus was displayed to the ing order of perceptual saliency—LA and LB, LL and LR, WA
receptive field of the cell with identical stimuli or different stim- and WB—and compared the animals’ performance among these
uli in the surround (Fig. 1a). Receptive fields of the cells tested groups in a pair-wise fashion. Consistent with observations in
in this study were located at between 1.5° and 4.0° eccentricity humans11–13, monkey A’s reaction time was significantly short-
in the lower visual field. The size of the V1 receptive fields er for the LA and LB oddballs than it was for the LL and LR odd-
ranged from 0.4° to 0.8° in diameter, and each stimulus element balls (P < 0.007), which in turn was significantly shorter than for
was 1° in diameter. The center-to-center distance between adja- the WA and WB oddballs (P < 0.0003) (Fig. 2g). Monkey A’s per-
cent elements was 1.75°. formance accuracies (percentage correct) for the 3D stimuli (LA
The receptive fields of V2 neurons were generally larger than and LB, LL and LR) were significantly better than they were for
those of V1 neurons, ranging from 0.8° to 2.0° in diameter. About the 2D stimuli (WA and WB) (P < 0.01 for all 3D versus 2D com-
30% of the encountered V1 neurons were discarded because they parisons) (Fig. 2h). The performance of monkey B was similar.
did not respond to any of the shape-from-shading stimuli. On Accuracy was much better for LB than for LL and LR (P < 10–5)
the other hand, all the V2 neurons encountered were included and was somewhat better for LL and LR than for WA and WB (P
in the analysis because they typically responded to these stimuli < 10–5) (Fig. 3h). Reaction time was shorter for LB than for all
(though many of them also responded to the hole stimuli). the other stimuli (P < 10 –5 ) (Fig. 3g). The main difference
between the two monkeys was that whereas monkey A performed
Naive state equally well for both LA and LB, monkey B showed asymmetri-
In the first stage, we recorded from 30 V1 units from monkey A cal performance for the two stimuli (strongly preferring LB at
and 55 V1 units from monkey B. The singleton condition elicit- the expense of LA).

articles
Fig. 2. Monkey A’s V1 neural

responses and behavioral perfor- a b c
mance in stages 1–5. (a, b) Temporal
evolution of the normalized popula-
tion average response of 30 V1 units
to the LA set (a) and to the WA set
(b) in stage 1. Each unit’s response
was smoothed by a running average
within a 15 ms window. Then the
responses were averaged across the

population. A very small difference
(pop-out response) was seen d e f
between the population average
response to the oddball condition
and that to the uniform condition in
the LA set. No pop-out response
was seen in the WA set. Note that in
these plots, as well as in other tem-
poral response plots presented, the
firing rate at a given time was nor-
malized against the maximum instan-
taneous response of the population g h i
to any stimulus of the tested sets at
each stage, which typically was the
response to either the WA or WB
stimuli. The neural response to the
shape-from-shading stimuli can thus
be gauged relative to the neural
response to the WA and WB con-
trast stimuli. (c) Mean pop-out mod-
ulation ratios of 30 units for all six
stimulus sets in stage 1. Pop-out
enhancements were significant for j k l
LA (P = 0.011) and LB (P = 0.007),
but not for the other stimuli. The
error bars represent the standard
errors of the means (s.e.m.).
(d, e) Temporal evolution of the nor-
malized population average response
of 45 V1 units to the LA set (d) and
WA set (e) in stage 3. Significant
pop-out response was seen in LA (as
well as in LB, LL and LR) starting at
100 ms after stimulus onset. No pop-
out response was seen for WA or WB. (f) Scatter plot of modulation ratios of the neuronal population for LA and LB pop-out responses in stage 3. The
population means of the modulation ratios for LA and LB were both significantly positive. (g, h) Behavioral performance of monkey A in detecting the
different oddballs from 15 testing sessions at the end of stage 2. In general, stimuli with higher pop-out modulations were associated with shorter reac-
tion times and higher accuracy (percentage correct). (i) Mean pop-out modulation ratios of 45 units for all six stimulus sets in stage 3. Pop-out enhance-
ments were highly significant for stimuli LA, LB, LL and LR (P = 10–6, 10–6, 0.0045, 10–4) but not for WA and WB. (j, k) Consequence of the LB-biased
training. Reaction time and percentage correct in the behavioral performance of monkey B, measured in five sessions at the beginning of stage 6,
reflected an improvement in LB oddball detection. (l) Mean population modulation ratios of 56 neurons of monkey A at stage 5 subsequent to LB-
biased training. A strong asymmetry in pop-out modulation was seen for LB over the other stimuli, in parallel to the change in behavior.
We recorded from 45 V1 units from monkey A and 47 V1 V1 neurons for these higher-order stimuli. Overall, the monkeys’
units from monkey B in stage 3 after the behavioral training, behavioral performance in stage 2 and their V1 neural pop-out
again using the fixation task paradigm. Relative to stage 1, there responses in stage 3 were stronger for 3D stimuli than they were
was a significant increase in V1 pop-out responses for both mon- for 2D stimuli.
keys to most of the shape-from-shading stimuli. The pop-out
responses were significant in both monkeys (P < 0.005 monkey A, Adaptability to changes
P < 0.01 monkey B) starting at about 100 ms after stimulus onset There were, however, individual differences in the behavioral
(Figs. 2d and 3d). There were no significant pop-out responses performance between the two monkeys, evidenced by a parallel
for WA and WB (Figs. 2e and 3e). The increases in population difference between their neural pop-out modulations. Specifi-
mean modulation were significant for all the shape-from-shading cally, whereas monkey A’s behavioral performances to LA and
stimuli relative to stage 1 (monkey A: P < 0.03 for LA, LL, LR, P LB stimuli were roughly the same, monkey B had a much
= 0.07 for LB; monkey B: P < 0.001 for all) but not for WB and stronger preference for LB over LA—an asymmetry that was
WA (Figs. 2i and 3i). The data indicate that experience with using mirrored in their pop-out responses. Perhaps monkey B had
the stimuli activated, or enhanced, the neural pop-out effect in developed a habit of looking for the LB pop-out target in a field

articles
a b c
e f
d
g h i
j k l
Fig. 3. Monkey B’s V1 neural responses and behavioral performance in stages 1–5. Figure parts are as in Fig. 2 except that the responses to the LB
set and to the WB set are shown here as examples, because the LB oddball evoked the strongest pop-out response in monkey B. Fifty-five V1 units
were recorded in stage 1, 47 in stage 3, and 47 in stage 5. The only stimulus that evoked significant pop-out responses in stage 1 was the LB oddball.
The LB oddball also evoked the strongest significant pop-out response in stage 3 (P < 10–7), followed by LA, LL and LR (P < 0.01, 10–4 and 10–4,
respectively). Pop-out was weak for WA and insignificant for WB. The positive shifts in modulation between stages 1 and 3 were significant for all the
shape-from-shading stimuli (P < 10–2 in all cases).
of LA distractors, resulting in a facilitation of response to the LB change in the pop-out modulation patterns as measured in
stimuli at the expense of the LA stimuli. This suggests that the stage 5: the pop-out response became significantly stronger for
pop-out response patterns, and hence the perceptual saliency of LB oddball over all the other stimuli (P < 0.01).
the stimuli, may be a function of individual cognitive strategy We subjected monkey B to the opposite biased training to off-
or behavioral experience. set the original asymmetry in favor of LB. Thirty training ses-
To test this hypothesis, we carried out a biased training exper- sions with only LA oddball stimuli were carried out, interleaved
iment to modify the monkeys’ behavior by manipulating the fre- with 15 sessions in which oddballs of all stimulus types were pre-
quency of occurrence of the different oddball stimuli in stage 4. In sented with equal frequency. Combined, the presentation fre-
30 training sessions (1,200 trials per session), monkey A prac- quency of LA oddball relative to any other type of oddball was
ticed solely on LB oddball detection. A preference for LB was 12:1. We found that the change in stimulus contingencies did
developed in the monkey’s behavior as measured in five sessions remove and even reverse the asymmetry in monkey B’s behav-
at the beginning of stage 6. This was accompanied by a parallel ioral performance. V1 neural pop-out responses followed suit as

articles
Fig. 4. Stage 7: spatial extent and durability of the effect. a

(a–d) There were four conditions for each of the six stimulus
b c d
sets tested in stage 7. Shown here are the oddball ON, oddball
NEAR, oddball FAR and oddball absent (uniform) conditions of
the LA set. In the actual display, the full screen was covered by
distractors as in Fig. 1a. The solid black dot is the fixation
spot. The open black circle indicates the spatial extent of the
receptive field of the tested neuron. These stimuli are grouped f
in the LA set because the RF is being probed by an LA stimulus
e
element. (e) Population mean modulation ratios of the

responses to the three oddball conditions computed against
the response to the uniform condition for 30 V1 units of mon-
key A. Significant pop-out response was seen only when the
oddball was exactly on the receptive field. The absence of pop-
out response in both the NEAR and FAR conditions suggests
that the effect was localized spatially in V1 at the location of
the oddball. (f) Population mean neural pop-out modulations
were correlated between stages 5 and 7 in monkey A. Each
data point was the pop-out modulation for one of the six stim- g h
ulus sets. The correlation of modulation before and after a
long recess suggests that the stimulus-specific pop-out modu-
lation was stable over time, or encoded in memory.
(g) Population mean modulation ratios of three oddball condi-
tions computed against the uniform condition for 25 V1 units
in monkey B. The oddball ON condition elicited the strongest
pop-out response in the V1 units. Slight and marginally signifi-
cant enhancements were present in both the NEAR and FAR
conditions of some of the shape-from-shading stimuli.
(h) Similar to (f); population neural pop-out modulations were
significantly correlated between stages 5 and 7 in monkey B.
well (Fig. 3j–l). We thus showed that the pop-out modulations keys A and B were not tested behaviorally or neurophysiologi-
in V1 for the different stimuli could be manipulated with changes cally for at least one and two months, respectively. In stage 7, we
in stimulus contingencies. tested the monkeys on the fixation task with the following four
conditions for each of the six stimulus sets: oddball ON, NEAR,
Spatial extent and durability FAR (from the receptive field), or absent (Fig. 4a–d). In the con-
Is this enhancement effect confined to the location of the odd- ditions of this set, an LA stimulus element was always placed on
ball stimulus, or does it extend to nearby stimuli as well? To the receptive field. In the NEAR condition, an LB oddball was
answer this question, we carried out another recording experi- placed 1.7° away from the receptive field stimulus (center-to-cen-
ment in stage 7 after a long recess (stage 6), during which mon- ter). In the FAR condition, an LB oddball was placed 3.6° away.
a b Fig. 5. Correlation between behavioral performance and V1

neural modulation across multiple stages. Behavior perfor-
mance measurements (percentage correct and reaction time) in
stage 2 (red), 6 (green) and 8 (blue) were paired with neural
pop-out modulation data in stage 3 (red), 5 (green) and 7 (blue).
Each pair of stages produced six data points, corresponding to
the six stimulus types. Eighteen points are shown in each graph
relating a behavioral measure with neural pop-out modulation.
Reaction time and percentage correct was regressed on the
pop-out modulation independently. A linear regression line,
with equation and statistical significance, is shown in each plot.
An outlier, which was >2.5 standard deviations away from the
regression line, was discarded in each graph (red dot with blue
c d outline). The outlier could have arisen from interference result-
ing from other top-down influences. R2 indicates the portion
(percentage) of the variance in the specific behavioral measure
that can be explained by the neural pop-out modulation. The
negative correlation between reaction time and neural modula-
tion was highly significant for both monkeys (statistics and fitted
equation are shown on graphs), as was the positive correlation
between accuracy and neural modulation. The correlations
remained significant when the outlier was included and when
the modulations were evaluated in the following time windows:
100–180 ms, 120–250 ms and 120–320 ms (see Supplemen-
tary Note (Figs. 11–12) online for further details).
articles
a b c Fig. 6. V2 neural pop-out response in monkey

A. Data are from 15 V2 units in stage 1 (a, b)
and 22 V2 units in stage 5 (d, e). Temporal
responses of the V2 population showed signifi-
cant pop-out response for shape-from-shading
stimuli starting at 100 ms, even in stage 1.
Response to LB shown as example. No pop-
out response was seen for the WA and WB
stimuli before or after behaviors. Note the
change in the scale of modulation axes relative

d e f to the graphs for V1 modulation. (c) Neural
modulations in stages 1 and 5. Pronounced
increase in enhancement was seen in LB subse-
quent to LB-biased training. (f) V2 modulations
were correlated to V1 modulations, but were
twice as strong.
The oddball targets were positioned at roughly the same eccen- stage and behavioral performance in the stages immediately
tricity away from the fovea as that of the receptive field location. before or after that recording stage. We grouped the data into
The respective pop-out modulation ratios for the two mon- three pairs: stage 2 behavior + stage 3 neural response, stage 6
keys in stage 7 consistently showed that the neural pop-out effects behavior + stage 5 response, and stage 8 behavior + stage 7
were still present after 1–2 months (Fig. 4e and 4g). The stimulus- response. A significantly positive correlation between the V1
specificity of the pop-out modulation was stable over time, sig- neural pop-out modulation signal and performance accuracy
nificantly correlating with that of stage 5 (Fig. 4f and h). More (Fig. 5a and c), as well as a significant negative correlation
importantly, when the oddball appeared near the receptive fields between the pop-out modulation signal and the reaction time
of the neurons, no significant pop-out effect was seen in the neu- (Fig. 5b and d) were found for both monkeys, suggesting that
rons of monkey A across all stimulus sets (Fig. 4e). Neurons in the neural pop-out signal could be considered a neural correlate
monkey B showed a slight enhancement in some oddball NEAR of perceptual saliency.
and FAR conditions (Fig. 4g), but the magnitude of the enhance-
ment was much smaller in these conditions than it was in the Neural activity in V2
oddball ON conditions. These results show that the pop-out sig- We recorded from a total of 138 V2 neurons from the two mon-
nals were stable over time and were spatially localized. Subse- keys combined, to ascertain the role of cortical interaction in
quently in stage 8, we tested the behavioral performance of both mediating these effects. Most V2 neurons were recorded at rough-
monkeys using the oddball detection task and found the behav- ly the same eccentricities as the V1 neurons were. Several major
ior was again correlated with the neural responses in stage 7. differences and similarities in the neural pop-out responses
between the two areas were seen (Figs. 6 and 7). First, the basic
Perceptual saliency patterns of V2 responses were very similar to those of V1 respons-
The accurate performance of the monkeys in detecting shape- es. In both areas, the singleton stimulus elicited the strongest
from-shading oddballs in the various behavioral testing stages response, and the suppressive effect of the surround was imme-
where chance rate was 25% (4 target locations), indicates that diate. However, V2 neurons had larger receptive fields at the same
the monkeys were most likely perceiving the stimuli rather than eccentricity and tended to respond more strongly to the hole
guessing randomly. To confirm that the neural pop-out signal stimuli than did V1 neurons. Second, the V2 neural pop-out
was truly a physiological measure of subjec-
tive perceptual saliency, we performed a a b c
regression analysis between the neural pop-
out modulation ratio from each recording
Fig. 7. V2 neural pop-out response in monkey B.

Data are from 17 V2 units in stage 1 (a, b) and 24
V2 units in stage 3 (d, e). Significant pop-out
response was seen in the population temporal
response for LB stimuli but not for WB stimuli.
(c) Pop-out modulation ratios for the six stimu- e f
lus types in stages 1 and 3. Behavioral training in
d
stage 2 apparently hastened the emergence of
pop-out response and increased the neural mod-
ulation for some stimuli. The mean pop-out
modulation ratios for 30 V2 units from stage 5
(not shown) were 0.14, 0.15, 0.16, 0.13, 0.03 and
0.008 for LA, LB, LL, LR, WA and WB, respec-
tively. (f) V2 modulations were correlated with
V1 modulations, but were stronger.

articles
modulation in stage 1 was signifi- a b c

cantly positive for nearly all the
shape-from-shading stimuli (mon-
key A, P < 0.05 all cases; monkey B,
P < 0.002 for LA, LB, LL, not sig-
nificant for LR) but was absent for
the 2D contrast patterns (WA, WB)
(Figs. 6a–c and 7a–c). This indi-
cates that V2 neurons were sensitive

to the shape-from-shading pop-out
Fig. 8. Stage 9: effect of distraction. (a, b) Monkey A’s V1 population mean modulation ratios of stage 7
before behavioral training. Subse- (30 units) and stage 9 (35 units) were compared to illustrate the effect of distraction in stage 9. Significant
quent to behavioral training, the attenuation of pop-out response was seen in LB (P < 10–6) in particular. The effect of distraction thus was
pop-out modulation in V2 became most evident in undermining the effect of the LB-biased training in stage 4. V2 population mean modula-
stronger (Fig. 6c and 7c). Third, the tion ratios remained significantly positive for the shape-from-shading stimuli. (c) A comparison between
stimulus-specific patterns of pop- monkey B’s V1 modulation ratios of stage 7 (25 units) and stage 9 (40 units) shows that pop-out
out modulation between V1 and V2 responses were eliminated for most stimuli except LB when attention was forcefully drawn away from the
were correlated across the multiple receptive field. Note that these modulation ratios were computed within the 100–250 ms window after
stages (Fig. 6f and 7f). Hence, V2’s stimulus onset. A shorter window was used because the monkey started reacting to the target (which
pop-out responses were also corre- appeared at variable times starting at 100 ms) in the distraction task. The earliest reaction time was
lated with perceptual saliency. V2 ∼280 ms after stimulus onset.
neural pop-out modulation for
shape-from-shading stimuli was
much stronger (1.5–2.0 times) than that of V1. Significantly more tive field location (Fig. 8a and c). Most strongly affected were the
(2–5 times) individual V2 neurons had positive pop-out respons- stimuli for which the monkeys were trained to develop a prefer-
es to the shape-from-shading stimuli than did V1 neurons across ence in stage 4. LB pop-out for monkey B, which was its origi-
the various stages (Supplementary Table 1). Fourth, the initial nal favorite, seemed to have been spared. V2 was less affected, as
response (40–100 ms) of V1 neurons to shape-from-shading the pop-out responses for LA, LB and LR remained relatively
stimuli was considerably weaker than it was to the 2D contrast strong and significant (Fig. 8b). The data indicate that the avail-
patterns, but the initial responses of V2 neurons to the two types ability of attentional resources, or the lack of distraction, is impor-
of stimuli were about equal. This indicates that V2 neurons may tant for the emergence of this higher-order perceptual pop-out
have greater feed-forward sensitivity to 3D shapes, compensat- saliency signal at the level of V1, but not in V2.
ing for the weaker luminance contrast of the shape-from-shading
stimuli. Finally, the latency of the pop-out responses in V2 in Eye-movement artifacts
stages 3 and 5 was estimated to be ~100 ms after stimulus onset, The pop-out effects could potentially have arisen as a result of
similar to that of the V1 pop-out responses (Fig. 2d versus micro-saccades and/or eye tremors occurring during fixation15.
Fig. 6d, and Fig. 3d versus Fig. 7d). This suggests that V1 and V2 When a strong pop-out target appeared, the monkeys might have
might be tightly coupled in this computation, and that the made small eye movements toward the target, moving the stimu-
processes are likely to be interactive and concurrent in nature. lation across the receptive field and thereby enhancing the
response. This effect would also have been proportional to the
Plasticity and attention perceptual saliency of the target. We analyzed all the eye move-
The persistence of the stimulus-specific pop-out modulation pat- ment data and ruled out eye tremors as a significant cause of the
terns over months suggests that the behavioral relevance of the pop-out effect (see Supplementary Note (Figs. 1–6) online).
different stimuli was encoded in memory to influence neural pro-
cessing at V1. To assess the possible role of attention in mediating DISCUSSION
the extrastriate cortical feedback, we conducted a ‘divided atten- Why shape-from-shading stimuli (which presumably require pro-
tion’ experiment in stage 9. The monkeys were trained to first cessing by higher-order perceptual areas) pop out ‘pre-attentively’
fixate on a red dot during stimulus presentation. Another small, has been a long-standing mystery. Here we found that long-
faint red dot (0.1° diameter) would flash briefly (3–4 refresh latency neural signals in V1 and V2 were correlated with behav-
frames, or 54–72 ms) in one of the three designated locations far ioral performance in monkeys. This correlation suggests that the
away from the receptive field at a random interval (between long-latency signal might be a neural correlate of the subjective
100–300 ms after stimulus onset, at 50 ms intervals). These des- perceptual pop-out saliency that results from shading in the visu-
ignated locations were constant across all sessions and were in al stimulus. The findings that V2—but not V1—neurons (i)
the upper visual field (4° visual angle away from the fovea), responded vigorously in their initial phase of responses to the
whereas the receptive fields were in the lower visual field. The shape-from-shading stimuli and (ii) showed a significantly high
monkeys had to pay attention to the three designated locations degree of sensitivity to the shape-from-shading stimuli in the
to perform successfully on the task, which was to make a saccade later part of their responses (even before behavioral training)
to the dot within 300 ms. The task was considered attentionally together suggest that V2 may be the first cortical area that is sen-
demanding, as the monkeys performed at only 50% correct. We sitive to or provides the primitives for encoding 3D surface
recorded the responses of 35 V1 units and 30 V2 units from mon- shape16 (see also F. T. Qiu et al. Soc. Neurosci. Abstr. 26, 593.2,
key A, and those of 40 V1 units from monkey B, to the oddball 2000). V1 neurons, in contrast, only showed a marginal sensi-
and the uniform conditions of the six sets of stimuli. We found tivity to the most salient shape-from-shading pop-out stimuli
that the pop-out responses in V1 became insignificant for some before behavioral training. The monkeys became significantly
stimuli when attention was forcefully drawn away from the recep- more sensitive to these stimuli only after they had used the stim-

articles
uli in their behavior. These findings suggest that the 3D shape What is the role of V1 in this computation? We have proposed
sensitivity in V1 may be mediated by recurrent feedback con- elsewhere that V1 serves as a ‘high-resolution buffer’ for visual
nections from V2 and/or other extrastriate areas. Supplemen- processing9. As only V1 neurons provide an explicit representa-
tary Table 1 shows the numbers of V1 and V2 units recorded in tion for precise encoding of orientation and spatial information,
each stage and the percentage of neurons individually showing higher-order perceptual inference involving fine details, curvi-
statistically significant pop-out responses for each type of stimuli. linear geometry and spatial precision would necessarily engage
The magnitude of pop-out response and the percentage of neu- V1 in their computation. The effects of such higher-order per-
rons showing a significant effect were markedly greater in V2 ceptual computations should therefore be reflected in the later
than in V1, indicating that the correlation between neural activ- part of V1 activity. Several recent physiological studies support
ity and subjective perception increases along the visual hierar- this conjecture26,27. As feedback from the extrastriate cortex tends
chy. This increase is consistent with an earlier finding on the to be diffuse and broad in spatial extent, V1 could play an impor-
neural correlates of perception as revealed by binocular rivalry17. tant role in localizing the pop-out target by actively sharpening
Our data also suggested that perceptual saliency was not sta- the pop-out response spatially using its well known lateral inhi-
tic, but dynamic and malleable, contingent on the animal’s expe- bition mechanism.
rience and on the behavioral relevance of the stimuli. The Thus, the higher-order pop-out saliency effect seen here is
stimulus-specific pattern of the modulation was stable over probably sub-served by the same mechanisms that mediate the
months until it was changed by new experience. Both the persis- bottom-up pop-out effect for oriented bars3 and the orientation
tence and the adaptability of the effect indicate that the behav- contrast effect for sine wave gratings seen in anesthetized mon-
ioral relevance of the stimuli must have been encoded in memory, keys in numerous earlier studies4–6. The orientation contrast
exerting an influence over early visual processing. Given that the effect could be supported purely by intra-cortical lateral inhibi-
observed effect was not restricted to the retinotopic location of tion mechanisms in V128,29. In awake behaving monkeys versus
the target during training, we suspect that the plasticity compo- anesthetized ones, orientation contrast effects have been found
nents of the effect were distributed over multiple memory and to be enhanced in both magnitude and spatial extent, resulting
perceptual areas or in the feed-forward/feedback connections in the so-called figure–ground effect7–10,30,31. Here we suggest
between cortical areas, although changes in the V1 intrinsic cir- that many or all of these phenomena may be interpreted as parts
cuitries were also possible18–20. of the same set of bottom-up and top-down mechanisms for
What is the mechanism underlying these changes? One pos- computing perceptual saliency. Notably, the time frame (100–
sibility is covert attention, which might be attracted by the salient 150 ms) in which the target selection signal emerges in the frontal
pop-out target automatically. The attenuation of pop-out sig- eye field during a visual search task32 is roughly the same as the
nals when attention was diverted away from the receptive field time frame for the emergence of the higher-order pop-out sig-
location implicated the involvement of attention, particularly nals in both V1 and V2. Taken together, the present findings indi-
for those stimuli for which the animals had developed a prefer- cate that the representation of perceptual saliency of objects in a
ence during biased training. Because of the late onset of the visual scene is distributed across multiple cortical areas and that
saliency effect, this attention was likely triggered by the input its computation is interactive in nature, involving the concerted
stimulus. Stimuli with stronger perceptual saliency (as deter- action of many areas in the brain9,33–37.
mined by higher-order brain areas) would attract more atten-
tion. Previous studies have manipulated top-down spatial METHODS
attention and top-down feature attention21–23, but our study Recording technique. Recordings were made transdurally with epoxy-
shows a potential interaction between top-down perceptual coated tungsten electrodes through a surgically implanted well overlying
inference and attentional allocation processes and the parallel the operculum of area V1 of the awake behaving monkey9. A protocol cov-
ering these studies was approved by the Institutional Animal Care and Use
computations in the early visual areas. We propose that the input
Committee of Carnegie Mellon University, in accordance with Public Health
stimulus generated an initial representation in V1, which then Service guidelines for the care and use of laboratory animals. The neurons
initiated a cascade of perceptual computations across multiple were isolated on the basis of spike heights using a window discriminator.
extrastriate visual areas for target selection and for deduction of The cells’ classical receptive fields (RFs) were mapped by a small oriented
shape from shading and figure–ground and target selection. This bar. Based on the depth of penetration, most V1 cells studied were esti-
higher-order perceptual and attentional processing interacts with mated to be cells in layers 2 and 3 of V1. V2 cells were drawn from arbi-
the early visual processing to determine the perceptual saliency trary layers of V2. Eye position was measured using the scleral search coil
of the stimuli, and thereby modifies the representations across technique and sampled at 200 Hz during the experimental sessions.
the whole visual hierarchy. The observed phenomena therefore
Fixation task. Throughout recording stages 1, 3, 5 and 7, a fixation task
reflect changes in both covert spatial attention and object atten- was performed by the monkey. In each trial, while the stimuli were pre-
tion in response to the input stimuli. sented on the screen for 350 ms each, the monkey was required to fixate
The idea that attention is involved in pop-out computation on a red dot, maintaining gaze within a fixation window ranging from
seems to be at odds with the conventional notion that pop-out 0.5° to 0.65° of visual angle in diameter. When the presentation was com-
is necessarily a pre-attentive process. This conventional idea, how- plete, the fixation dot disappeared, a second red dot appeared at a dif-
ever, has been challenged by recent psychological studies24–25 ferent location, and the monkeys were required to make a saccade to it
showing that attention may be critical for the covert detection, in order to receive a juice reward. The probe stimulus (the center stimu-
and even the overt perception, of pre-attentive stimulus features. lus in each of the iconic diagrams in Fig. 1b) was placed on the receptive
field of the cell. The position of the second dot target was not correlat-
Further, the interactions between bottom-up and top-down
ed with the stimulus, and hence the test stimulus was irrelevant to the
processes have been shown to be modifiable by perceptual train- monkeys’ behaviors in the recording sessions. Twenty-four conditions
ing25. Our findings are consistent with these psychological obser- were tested in each session: six stimulus sets, each with four conditions.
vations, suggesting that there is a tight coupling between the Each condition was repeated 12–15 times for each cell. The presentation
parallel pop-out computation and the top-down perceptual and of the conditions was randomly interleaved. The distraction task in stage
attentional processes. 9, as described in the text, was a variant of the fixation task.

articles
Oddball detection task. During behavioral training and testing in stages 5. Levitt, J. B. & Lund, J. S. Contrast dependence of contextual effects in primate
2, 6 and 8, the monkeys performed an oddball detection task. In each trial, visual cortex. Nature 387, 73–76 (1997).
the oddball target was randomly drawn from the six basic stimulus types 6. Nothdurft, H. C., Gallant, J. L. & Van Essen, D. C. Response modulation by
texture surround in primate area V1: correlates of ‘pop-out’ under anesthesia.
and placed at one of four random locations, distributed over the four Vis. Neurosci. 16, 15–34 (1999).
quadrants of the visual field and at 4° eccentricity away from the fovea. 7. Lamme, V. A. F. The neurophysiology of figure-ground segregation in
The oddball was embedded in a field of distractors, each of which was the primary visual cortex. J. Neurosci. 10, 649–669 (1995).
reflected image of the oddball. The monkeys had to make a saccade to the 8. Zipser, K., Lamme, V. A. F. & Schiller, P. H. Contextual modulation in the
oddball location to complete the trial correctly. The chance rate was there- primary visual cortex. J. Neurosci. 16, 7376–7389 (1996).
9. Lee, T. S., Mumford, D., Romero, R. & Lamme, V. A. F. The role of the
fore 25% correct. No reward was given for incorrect trials. The order of primary visual cortex in higher level vision. Vis. Res. 38, 2429–2454
presentation of the stimuli was randomly interleaved in each block. At (1998).

least five sessions were carried out in each testing stage to assess behav- 10. Super, H., Spekreijse, H. & Lamme, V. A. F. Two distinct modes of sensory
ior. Fifty trials were tested for each condition per session. During the biased processing observed in monkey primary visual cortex (V1). Nat. Neurosci. 4,
304–310 (2001).
training in stage 4, the same protocol was used except that the frequency 11. Ramachandran, V. S. Perception of shape from shading. Nature 331, 163–166
of oddball occurrence was a function of the stimulus types. (1988).
12. Sun, J. & Perona, P. Early computation of shape and reflectance in the visual
Data analysis. The t-test, ANOVA for two groups, was used throughout system. Nature 379, 165–168 (1996).
to test for significant differences between means, either at the individual 13. Braun, J. Shape from shading is independent of visual attention and may be a
‘texton’. Spat. Vis. 7, 311–322 (1993).
neuron level or at the population level. For an individual neuron, we 14. Wurtz, R. Response of striate cortex neurons to stimuli during rapid eye
compared response distributions to the oddball condition with those to movements in the monkey. J. Neurophysiol. 32, 975–986 (1969).
the uniform conditions. For neuronal populations, we used the t-test to 15. Leopold, D. A. & Logothetis, N. K. Microsaccades differentially modulate
evaluate the significance of (i) the mean pop-out modulation in each neural activity in the striate and extrastriate visual cortex. Exp. Brain Res. 123,
341–345 (1998).
stage and (ii) the change in distribution between stages. In population 16. Bakin, J. S., Nakayama, K. & Gilbert, C. Visual responses in monkey areas V1
analysis, the modulation ratio of each cell was treated as a data point, and V2 to three-dimensional surface configurations. J. Neurosci. 20,
and the significance of the population distribution of modulation ratios 8188–8198 (2000).
was evaluated against either a zero-mean distribution for positive pop 17. Logothetis, N. K. Object vision and visual awareness. Curr. Opin. Neurobiol.
out, or against another distribution for shift in the distribution. P > 0.05 8, 536–44 (1998).
18. Crist, R. E., Li, W. & Gilbert, C. D. Learning to see: experience and attention
indicated that the difference between the two means was not significant. in primary visual cortex. Nat. Neurosci. 4, 519–525 (2001).
The onset time of the pop-out response was estimated by first computing 19. Karni, A. & Sagi, D. The time course of learning a visual skill. Nature 365,
a temporal profile of the pop-out modulation ratio of the neuronal pop- 250–252 (1993).
ulation evaluated within a 15 ms running window for each stimulus set, 20. Ahissar, M. & Hochstein, S. Task difficulty and the specificity of perceptual
learning. Nature 387, 401–406 (1997).
and then determining the time at which the population modulation ratio 21. Motter, B. Focal attention produces spatially selective processing in visual
became significantly and consistently positive (when P dropped below cortical areas V1, V2 and V4 in the presence of competing stimuli.
0.05 in a population t-test). J. Neurophysiol. 70, 909–919 (1993).
22. Ito, M. & Gilbert, C. D. Attention modulates contextual influences in the
primary visual cortex of alert monkeys. Neuron 22, 593–604 (1999).
Note: Supplementary information is available on the Nature Neuroscience website. 23. Luck, S. J., Chelazzi, L., Hillyard, S. A. & Desimone, R. Neural mechanisms of
spatial selective attention in areas V1, V2 and V4 of macaque visual cortex.
J. Neurophysiol. 77, 22–42 (1997).
24. Joseph, J. S., Chun, M. M. & Nakayama, K. Attentional requirements in a
Acknowledgments ‘preattentive’ feature search task. Nature 387, 805–807 (1997).
We thank C. Olson, P. Schiller, J. McClelland, C. Colby, R. Kass, A. Tolias, 25. Ahissar, M., Laiwand, R. & Hochstein, S. Attentional demands following
perceptual skill training. Psychol. Sci. 12, 56–62 (2001).
T. Moore and G. Deco for advice and discussion, and M. Nguyen, S. Yu, 26. Roelfsema, P. R., Lamme, V. A. & Spekreijse, H. Object-based attention in the
X. G. Yan, E. Cassidente, H. J. Qi, K. Medler, J. Rollenhagen and K. Nakamura primary visual cortex of the macaque monkey. Nature 395, 376–381 (1998).
for assistance. T.S.L. was supported by National Science Foundation (NSF) 27. Lee, T. S. & Nguyen, M. Dynamics of subjective contour formation in the
early visual cortex. Proc. Natl. Acad. Sci. USA 98, 1907–1911 (2001).
CAREER 9984706, McDonnell-Pew Foundation, NSF LIS 9720350, National 28. Stemmler, M., Usher, M. & Niebur, E. Lateral interactions in the primary
Institutes of Health (NIH) NEI vision research core grant EY08098 and NIH visual cortex: a model bridging physiology and psychophysics. Science 269,
1877–1880 (1995).
2P41RR06009-11 for biomedical supercomputing. R.D.R. was supported by a
29. Li, Z. Visual segmentation by contextual influences via intra-cortical
NSF graduate fellowship. D.M. was supported by NSF DMS-0074276 and interactions in the primary visual cortex. Network 10, 187–212 (1999).
Burrough’s Wellcomb Foundation grant 2302. 30. Hupe, J. M. et al. Feedback connections act on the early part of the responses
in monkey visual cortex. J. Neurophysiol. 85, 134–145 (2001).
31. Rossi, A. F., Desimone, R. & Ungerleider, L. G. Contextual modulation in
Competing interests statement primary visual cortex of macaques. J. Neurosci. 21, 1698–1709 (2001).
The authors declare that they have no competing financial interests. 32. Bichot, N. P. & Schall, J. D. Effects of similarity and history on neural
mechanisms of visual selection. Nat. Neurosci. 2, 549–554 (1999).
33. Deco, G. & Lee, T. S. A unified model of spatial and object attention based on
RECEIVED 20 MARCH; ACCEPTED 26 APRIL 2002 inter-cortical biased competition. In Neurocomputing (Elsevier, Amsterdam,
in press).
34. Grossberg, S. Competitive learning: from interactive activation to adaptive
1. Julesz, B. Experiments in the visual perception of texture. Sci. Am. 232, 34–43 resonance. Cogn. Sci. 11, 23–63 (1987).
(1975). 35. McClelland, J. L. & Rumelhart, D. E. An interactive activation model of
2. Treisman, A. & Gelade, G. A feature-integration theory of attention. Cognit. context effects in letter perception (Part I). Psychol. Rev. 88, 375–407
Psychol. 12, 97–136 (1980). (1981).
3. Desimone, R. & Duncan, J. Neural mechanisms of selective visual attention. 36. Pollen, D. A. On the neural correlates of visual perception. Cereb. Cortex 9,
Annu. Rev. Neurosci. 18, 193–222 (1995). 4–19 (1999).
4. Knierim, J. J. & Van Essen, D. C. Neuronal responses to static texture patterns 37. Mumford, D. On the computational architecture of the neocortex II. Biol.
in area V1 of the alert macaque monkey. J. Neurophysiol. 67, 961–980 (1992). Cybern. 66, 241–251 (1992).

articles
Motion illusions as optimal percepts

Yair Weiss1, Eero P. Simoncelli2 and Edward H. Adelson3
1 School of Computer Science and Engineering, Hebrew University of Jerusalem, Givat Ram Campus, Jerusalem 91904, Israel
2 Howard Hughes Medical Institute, Center for Neural Science and Courant Institute of Mathematical Sciences, New York University,
4 Washington Place, New York, New York 10003, USA
3 Brain and Cognitive Sciences Department, Massachusetts Institute of Technology, 77 Massachusetts Ave, Cambridge, Massachusetts 02139, USA
Correspondence should be addressed to Y.W. (yweiss@cs.huji.ac.il)
The pattern of local image velocities on the retina encodes important environmental information.
Although humans are generally able to extract this information, they can easily be deceived into see-
ing incorrect velocities. We show that these ‘illusions’ arise naturally in a system that attempts to
estimate local image velocity. We formulated a model of visual motion perception using standard
estimation theory, under the assumptions that (i) there is noise in the initial measurements and (ii)
slower motions are more likely to occur than faster ones. We found that specific instantiation of such
a velocity estimator can account for a wide variety of psychophysical phenomena.
The human ability to analyze visual motion in general scenes far information of both gratings. Graphically, this corresponds to
exceeds the capabilities of the most sophisticated computer vision the point in velocity space that lies at the intersection of both
algorithms. Yet psychophysical experiments show that humans also constraint lines (Fig. 1b, circle). The VA solution is the average
make some puzzling mistakes, misjudging speed or direction of of the two normal velocities. Graphically, this corresponds to the
very simple stimuli. In this paper, we propose that such mistakes of point in velocity space that lies halfway between the two normal
human motion perception represent the best solution of a ratio- velocities (Fig. 1b, square). An FT solution corresponds to the
nal system designed to operate in the presence of uncertainty. velocity of some feature of the plaid intensity pattern (for exam-
In both biological and artificial vision systems, motion analy- ple, the locations of maximum luminance at the grating inter-
sis begins with local measurements such as the output of direc- sections) 15,16 . For plaids, the FT and IOC solutions both
tion-selective cells in primary visual cortex1, or of spatial and correspond to the veridical (true) pattern motion.
temporal derivative operators in artificial systems2,3. These are Which of the three rules best describes human perception? The
then integrated to generate larger, more global motion descrip- answer is not clear: depending on the stimulus, the perceived pat-
tions. The integration process is essential because the initial local tern motion can be nearly veridical (consistent with IOC or FT) or
motion measurements are ambiguous. For example, in the vicin- closer to the VA solution. The relevant stimulus features include
ity of a contour, only the motion component perpendicular to relative grating orientation and speed17–19, contrast20, presenta-
the contour can be determined (a phenomenon referred to as the tion time17 and retinal location17.
‘aperture problem’)2,4–7. Such an integration stage seems to be Similar effects have been reported with stimuli that appear
consistent with much of the psychophysical8–11 and physiologi- quite different from plaids16,21. For a moving rhombus (Fig. 2),
cal8,12–14 data. as for a plaid pattern, the motion of each opposing pair of sides is
Despite the vast amount of psychophysical data published consistent with a constraint line in the space of velocities. As
over the past two decades, the nature of the integration scheme shown in the velocity space diagrams (Fig. 2c and f), IOC or FT
underlying human motion perception remains unclear. This is predicts horizontal motion, whereas VA predicts diagonal motion.
true even for the simple and widely studied ‘plaid’ stimulus, in Perceptually, however, the rhombus appears to move horizon-
which two superimposed oriented gratings translate (move with- tally at high contrast and diagonally at low contrast. To further
out changing shape, size or orientation) in the image plane complicate the situation, the percept depends on the shape. If
(Fig. 1a). Due to the aperture problem, each grating’s motion is the rhombus is fattened (Fig. 2d), it appears to move horizon-
consistent with an infinite number of possible translational veloc- tally at both contrasts. To view these moving stimuli, see
ities lying on a constraint line in the space of all velocities http://www.cs.huji.ac.il/~yweiss/Rhombus.
(Fig. 1b). When viewing a single drifting grating in isolation, One might reason that the visual system uses VA for a thin,
subjects typically perceive it as translating in a direction normal low-contrast rhombus, and IOC/FT for a thin, high-contrast
to its contours (Fig. 1b). When two gratings are presented simul- rhombus and for a fat rhombus. Although a model based on this
taneously, subjects often perceive them as a coherent pattern ad hoc combination of rules certainly fits the data, it is clearly
translating with a single motion5,7. not a parsimonious explanation. Furthermore, each of the ide-
How is this coherent pattern motion estimated? Most expla- alized rules is limited to stimuli containing straight structures
nations are based on one of three rules7: intersection of con- at only two orientations, and does not offer a method for com-
straints (IOC), vector average (VA), or feature tracking (FT). The puting the normal velocities of those structures. One would pre-
IOC solution is the unique translation vector consistent with the fer a single, coherent model that could predict the perceived

articles
Fig. 1. Intersection of constraints. (a) Drifting gratings superimposed in

a bV the image plane produce a translating ‘plaid’ pattern. (b) Dotted lines
y
indicate constraint lines; arrows indicate perceived direction of grating
viewed in isolation. The IOC solution (circle) is the unique velocity con-
IOC sistent with the constraint lines of both gratings. The VA solution
(square) is the average of the two normal velocities. There is experi-
Vx mental evidence for both types of combination rule.
VA
to a ‘fuzzy’ constraint line— velocities on the constraint line have

the highest likelihood, and the likelihood decreases with distance
from the line. The ‘fuzziness’ of the constraint line is governed
velocity of any arbitrary spatiotemporal stimulus that appears by σ, the standard deviation of the assumed noise. At corners,
to be translating. We have developed such a model based on a where local motion measurements are less ambiguous, the like-
simple formulation of the problem of velocity estimation and lihood no longer has the elongated shape of a constraint line but
on a few reasonable assumptions. becomes tightly clustered around the veridical velocity.
In Helmholtz’s view, our percepts are our best guess as to what This model of additive Gaussian noise also resulted in a
is in the world, given both sensory data and prior experience22. dependence of the likelihood on contrast. For a fixed value of σ,
To make this definition more quantitative, one must specify (i) the likelihoods were broader at low contrast (Fig. 3, bottom).
what is ‘best’ about a best guess, and (ii) the way in which prior This makes intuitive sense: at low contrast there is less informa-
experience should influence that guess. In the engineering liter- tion about the exact speed of the stimulus, and therefore more
ature, the theory of estimation formalizes these concepts. The local uncertainty, so the likelihood is more spread out. In the
simplest and most widely known estimation framework is based extreme case of zero contrast, the uncertainty is infinite.
on Bayes’ rule (see ref. 23 for examples of Bayesian models in The second assumption underlying our ideal observer model
perception and refs. 24 and 25 for Bayesian motion models). Fol- is that velocities tend to be slow. Suggestions that human observers
lowing an approach described in previous work26–29, we devel- prefer the ‘shortest path’, or slowest motion consistent with the
oped an optimal Bayesian estimator (known as an ‘ideal observer’ visual input, date back to the beginning of the 20th century (see
in the psychophysics literature) for two-dimensional velocity. ref. 30 and references therein). In particular, Wallach suggested
Here, as in most studies of the aperture problem, we considered that humans prefer to see the normal velocity for a single line seg-
only cases in which humans see a single global translational ment because that is the slowest velocity consistent with the image
motion (no deformation, rotation, occlusion boundaries, trans- data5. Likewise in apparent motion displays, humans tend to
parency, or the like). Elsewhere, we have developed extensions of choose the shortest path or slowest motion that would explain
this model that can handle more complicated scenes29. the incoming information.
Our model begins with the standard principle of intensity We formalized this preference for slow speeds using a prior
conservation: it assumes that any changes in image intensity over probability distribution on the two-dimensional space of veloc-
time are due entirely to translational motion of the intensity pat- ities that is Gaussian and centered on the origin. According to
tern. We then made two basic assumptions: (i) local image mea- this ‘prior’, in the absence of any image data, the most probable
surements are noisy and (ii) image velocities tend to be slow. We velocity is zero (no motion), and slower velocities are generally
formulated these assumptions using probability distributions more likely to occur than fast ones. As with the noise model, we
(see below), and used Bayes’ rule to derive the ideal observer (for have no direct evidence (either from first principles or from
further mathematical details, see Methods). empirical measurements) that this assumption is correct. We will
We instantiated the first assumption using a noise model com- show, however, that it is sufficient to account qualitatively for
monly used in engineering because of the tractability of the solu- much of the perceptual data.
tion: measurements are contaminated with additive, independent, Under the Bayesian framework, the percept of the ideal
Gaussian noise with a known standard deviation (σ). Although observer is based on the posterior probability (the probabili-
this simple noise model is unlikely to be correct in detail, we show ty of a velocity given the image measurements), which is com-
that it is sufficient to account for much of the data. This
noise model provides a functional form for the local like-
lihood: a distribution over the space of velocities that is
based on measurements made in a local image patch. We a b c Vy
depicted this likelihood as a gray-level image (Fig. 3) in
which intensity corresponds to probability. For patches IOC
containing a single edge, the likelihood function is similar Vx
VA
Fig. 2. Insufficiency of either VA, IOC or FT rules as an expla-
nation for human perception of a horizontally moving rhombus.
(a) A ‘narrow’ rhombus at high contrast appears to move hori- d e f Vy
zontally (consistent with IOC/FT). (b) A narrow rhombus at
low contrast appears to move diagonally (consistent with VA). IOC
(c) Velocity space constraints for a narrow rhombus. (d,e) A Vx
‘fat’ rhombus at low or high contrast appears to move horizon-
tally (consistent with IOC/FT). (f) Velocity space constraints for VA
a fat rhombus.
articles
Likelihood Likelihood Likelihood

Stimulus at location a at location b at location c
a
b
Vx Vx Vx
c
Vy Vy Vy
a
b
Vx Vx Vx
c
Vy Vy Vy
Fig. 3. Likelihood functions for three local patches of a horizontally translating diamond stimulus, computed using equation (4). Intensity corresponds
to probability. Top, high-contrast sequence. Bottom, low-contrast sequence, with the same parameter σ. At edges, the local likelihood is a ‘fuzzy’ con-
straint line; at corners, the local likelihood peaks around the veridical velocity. The sharpness of the likelihood decreases with decreasing contrast.
puted from the likelihood and prior using Bayes’ rule (see of a computer mouse. The predictions of equation (1) provide
Methods). We formulated the posterior distribution by mul- an excellent fit to the human experimental data (Fig. 4d). In
tiplying the prior and the likelihoods at all image locations. addition, the qualitative predictions remained unchanged while
This is correct under the assumptions that the noise in the mea- the free parameter was varied over two orders of magnitude
surements is statistically independent, and that the likelihoods (Fig. 4d). In fact, no setting of the free parameter could make
being multiplied correspond to image locations that are mov- the perception of narrow rhombuses more veridical than that
ing at the same velocity. of fat ones. Similarly, there is no setting that would make the
One can calculate the velocity estimate (v∗) of the ideal observ- perception of low-contrast rhombuses more veridical than that
er as the mean or maximum of the posterior distribution. Our of high-contrast rhombuses.
posterior distribution is Gaussian, and the mean (which is also
the most likely) velocity was computed analytically using the fol- RESULTS
lowing matrix equation: We compared the predictions of the ideal observer (the solu-
tion of equation (1)) to previously published psychophysical
data17–20,31,32. The free parameter was adjusted manually for
 
2
σ –1
Σ Ix2 + —2 Σ Ix Iy each experiment but held constant for all conditions within
 ΣI I 
σp

each experiment. Different observers probably make different
 ΣI I
x t
v =–*
 ‘assumptions’ regarding noise, and indeed, substantial individ-
σ2
 y t ual differences for these illusions have been reported17. As with
Σ Ix Iy Σ Iy2 + —2
σp
the rhombus example, the value of the free parameter did not
(1) change the qualitative predictions of the model for any of the
stimuli discussed here.
where Ix, Iy, It refer to the spatial (two dimensions) and tem-
poral derivatives of the image sequence. The sums were taken Influence of contrast on perceived grating speed
over all locations that translate together (here, we assumed this The perceived speed of a single grating depends on con-
included the entire image). This equation allowed us to predict trast31,33–35, with lower-contrast patterns consistently appear-
the ideal observer’s velocity estimate for any image sequence. The ing slower than higher-contrast patterns34. This may underlie
solution of equation(1) has only one free parameter: the ratio of the tendency of automobile drivers to speed up in the fog36. In
σ to σp. Changing both of these while holding the ratio constant a psychophysical experiment quantifying this effect31, subjects
changes the width, but not the peak, of the posterior. were asked to compare the apparent speed of two gratings of
We calculated the posterior for the moving rhombus stimuli different contrast (Fig. 5a). The low-contrast grating was con-
(Fig. 4a–c), holding the free parameter (σ/σp) constant. Con- sistently perceived to be moving slower. This illusion depend-
sistent with human data, the ideal observer predicts horizontal ed primarily on the ratio of contrasts of the two gratings: the
motion for a narrow, high-contrast rhombus, diagonal motion perceived speed was an approximately linear function of the
for a narrow, low-contrast rhombus and nearly horizontal contrast ratio, and was approximately independent of absolute
motion for a fat, low-contrast rhombus. For a more quantitative contrast. The ideal observer shows a qualitatively similar con-
comparison of the ideal observer and human perception, we trast dependence. At low contrasts, the likelihood is broader
showed three subjects a continuum of low-contrast rhombuses and the prior has a stronger influence on the estimate. Con-
that varied between the extremes of ‘thin’ and ‘fat’, and asked sistent with human perception, the ideal observer also esti-
them to report the perceived direction by positioning the cursor mates the low-contrast grating as moving slower (Fig. 5a).

articles
Image Image Fig. 4. Predictions of ideal observer for rhombus stimuli.

a b (a–c) Construction of the posterior distribution for the
rhombus stimuli. For clarity, likelihood functions for only
two locations are shown; the estimator used in our study
incorporated likelihoods from all locations. (d) Circles
show perceived direction for a single human subject as
rhombus angle was shifted gradually from thin to fat
rhombuses (all three subjects showed a similar effect, and
Vy Vy Vy Vy Vy Vy all gave informed consent to participate in the study).
Each subject was given 100 presentations. Solid line

Vx Vx Vx Vx Vx Vx shows the predictions of the Bayesian estimator com-
puted using equation (1), where the free parameter was
Prior Likelihood 1 Likelihood 2 Prior Likelihood 1 Likelihood 2 varied manually to fit the data. Dotted lines indicate the
predictions when the free parameter was decreased by a
X X factor of 10 (top dotted line) or increased by a factor of
10 (bottom line).
Vy Vy
x
Vx x Vx
higher uncertainty and hence the low-contrast grat-

Posterior Posterior
ing has less influence on the estimate.
c Image Contrast influence on perceived line direction

Subjects tend to misperceive the direction of a mov-
ing line at low contrasts, even when its endpoints are
visible32. We replotted data from an experiment in
which subjects reported the perceived direction of a
d 0 ‘matrix’ of lines (Fig. 5c). The matrix was con-
structed by replicating a single line at multiple loca-
Vy V y V y
tions in the visual field. The line was oriented such
Direction (degrees)
10
that its normal velocity was downward even when
V x V x V x
20
the line was moving upward. At low contrasts, sub-
jects performed far below chance, indicating that
Prior Likelihood 1 Likelihood 2
30 they perceived upward motion while the line actu-
ally moved downward. The authors proposed two
X 40
separate mechanisms to explain this finding, one
50
dealing with terminator (line endpoint) motion and
V y 0 10 20 30
Rhombus angle (degrees)
40
other with line motion. The terminator mechanism
xV
x
was assumed to be active primarily at high contrasts
and the line strategy primarily at low contrasts.
Posterior We found that at low contrast, the ideal observer
also misperceived the direction of motion because the
likelihoods are broader and the estimator prefers the
The simple ideal observer presented here does not predict normal velocity (which is slower than the true velocity). To obtain
the quasilinear shape of the perceived relative speeds, nor does a percentage of correct responses for the ideal observer, we assumed
it predict the lack of dependence on total contrast (it makes that v* was corrupted by decision noise, and we calculated the prob-
slightly different predictions for maximum contrasts of 40% ability that the corrupted v* was in the upward direction. The deci-
and 70%, Fig. 5a). We also constructed a slightly more elabo- sion noise was Gaussian in velocity space. The standard deviation of
rate model that can account for these effects in a more quanti- the decision noise determines the sharpness of the psychometric
tative manner (see Discussion). function and was adjusted manually. The predicted percentage cor-
rect for the ideal observer was in accordance with human perception
Influence of contrast on perceived plaid direction (Fig. 5c, solid line).
The perceived direction of a plaid depends on the relative con-
trast of the two constituent gratings20. We replotted data from Type I versus type II plaids: perceived direction
an experiment in which subjects reported the perceived direc- In the plaid literature, a distinction is often made between two
tion of motion of symmetric plaids while the contrast ratio of types of configuration: for a ‘type I’ plaid, the direction of the
the two components was varied (Fig. 5b). Perceived direction veridical velocity lies between that of the two normal velocities;
was always biased toward the normal direction of the higher- for a ‘type II’ plaid, the veridical direction lies outside the two
contrast grating. The magnitude of the bias changed as a func- normals17. In the latter case, the vector average is quite different
tion of the total contrast of the plaid (the sum of the contrasts from the veridical velocity.
of the two gratings). Increasing the contrast of both gratings At low contrast, the perceived direction for type II plaids is
(while the ratio of contrasts is held fixed) resulted in a smaller strongly biased in the direction of the vector average, and the
bias. The ideal observer shows a similar effect (E. P. Simoncelli & perceived direction of type I plaids is largely veridical. We replot-
D. J. Heeger, Invest. Opthal. Vis. Sci. Suppl. Abstr. 33, 954, 1992), ted data from a single subject who reported the perceived direc-
which again follows from the fact that at low contrast, there is tion of a plaid under five different conditions17 (Fig. 5d, circles).

articles
a 1.1 b 25 c 100
Feature motion
1 Max contrast 70% 20

5% 80
Percentage correct
Bias (degrees)
Relative speed
0.9 15
60
0.8 10 40%
40
0.7
5
20
0.6 Max contrast 40%
0
Normal motion
0
0.5
2 1. 5 1 0. 5 0 0.5 5
0 1 2 3 4 0 20 40 60 80
Log contrast ratio Log2 contrast ratio Contrast
d 80 e 300
f
VA VA
Judged plaid direction (degrees)

60 100
Percentage in VA direction
Direction (degrees)
40 290
80
20
280 60
IOC
0
20 270 40
IOC
40 20
260
60 IOC
0
80 250
1 2 3 4 5 0 10 20 30 40 50 0.4 0.5 0.6 0.7 0.8
Condition Plaid component separation (degrees) Ratio of component speeds
Fig. 5. Comparison of ideal observer (solid lines) to a variety of published psychophysical data (circles). (a) Contrast influence on perceived grating
speed. Circles indicate the perceived speed of the lower-contrast grating relative to the higher-contrast grating, as a function of the contrast ratio. Solid
lines show the predictions of the ideal observer for two different maximal contrasts (data from ref. 31). (b) Relative contrast influence on perceived plaid
direction (data from ref. 20). (c) Contrast influence on perceived line direction (data from ref. 32). (d) Perceived direction of type I (conditions 1, 3, 5)
versus type II (conditions 2, 4) plaids (data from ref. 17). Dotted line shows the IOC prediction. (e) Influence of relative orientation on perception of type
II plaid motion (data from ref. 18). (f) Influence of relative speed on perception of type II plaids (data from ref. 19).
In all five conditions, the angular separation between the two of plaids as a function of this angle, while pattern velocity was
gratings was 22.3°. In some conditions the two normal velocities held constant18 (Fig. 5e). The perceived direction is not con-
were on different sides of the veridical motion (type I), whereas sistent with a pure VA mechanism or a pure IOC mechanism.
in others they were on the same side of the veridical motion (type Instead, it shows a gradual shift from the VA to the IOC solu-
II). Subjects saw type I plaids moving in the IOC direction and tion as the angle between the components increases. The solid
type II plaids moving in approximately the VA direction (∼55° line shows the prediction of the ideal observer (the direction
away from veridical direction). The authors of the original study of v* in equation (1)). This situation is similar to the ‘narrow’
explained their findings using a contrast-dependent combina- versus ‘fat’ rhombuses (Fig. 4). When two likelihoods whose
tion of first-order and second-order motion analyzers37. constraint lines are nearly identical are multiplied, their prod-
The ideal observer also predicted different directions of uct will be broad and hence have less of an influence on the
motion for the two types of plaids at low contrast (Fig. 5d, solid posterior. By contrast, when two likelihoods have widely dif-
line). The ‘misperception’ of type II plaids is similar to the per- fering constraint lines, their product will be narrow and hence
ception of the narrow rhombus: the VA velocity is much slower have greater influence on the posterior.
than the IOC solution and hence it is favored at low contrasts.
In the ideal observer, this bias toward the VA solution weakens Influence of relative speed on type II plaids
with increasing contrast, as the likelihoods become narrower. The perceived direction of a plaid also depends on the relative
It has also been reported that the VA bias is more pronounced speeds of the components. We plotted data from a single sub-
with shorter presentation durations17. We based our ideal observ- ject19 who viewed a plaid with IOC and VA directions on oppo-
er on instantaneous measurements, so it is not affected by dis- site sides of upward, and reported whether the motion appeared
play duration. The formulation can easily be extended so that the to be more leftward or rightward (Fig. 5f). When the speeds of
ideal observer integrates information over time. This way, the two components were similar, the subject answered right-
increased duration acts in a similar fashion to increased contrast: wards (consistent with the VA solution), but when the speeds
the longer the duration, the narrower the likelihood. Such an were dissimilar, the subject answered leftwards (consistent with
extended formulation predicts that the VA bias would decrease the IOC solution). We found that the ideal observer described
with increased duration. A similar effect of duration has been by equation (1) shows a similar shift from leftward to rightward
reported elsewhere 32, which would also be predicted by this velocities. We again calculated a ‘percentage correct’ value for the
extension of our model. ideal observer by assuming decision noise (Fig. 5f, solid line).
Influence of relative orientation on type II plaids DISCUSSION

The perceived direction of a type II plaid depends strongly on Research on visual motion analysis has yielded a tremendous
the angle between the components. We replotted data from an amount of experimental data. When viewed in the context of
experiment in which subjects reported the perceived direction existing rules such as IOC and VA, these data seem contradic-

articles
tory, requiring an arbitrary combination scheme that applies cepts reflect the shape of the full posterior distribution.
the right rule in the right conditions. Such an approach can We have focused on an ideal observer for estimating a single
successfully fit the data, but is typically lacking in predictive two-dimensional translation. This model cannot estimate more
power: with a complicated enough combination scheme one complicated motions such as rotations and expansions, nor can
can model any experiment. More importantly, because these it handle scenes containing multiple motions. Elsewhere, we
rules are not formulated directly on image measurements, it is describe an extended ideal observer for more general scenes with
not clear how one should generalize them for application to multiple motions29. We show that an ideal observer that assumes
arbitrary spatiotemporal stimuli. that velocity fields are ‘slow and smooth’42 can explain an even
Here we have taken an alternative approach. We derived an wider range of motion phenomena. In particular, the bias toward
optimal estimator for local image velocity using the standard slower motions can sometimes account for one of the most crit-
assumption of intensity constancy and two additional assump- ical issues in motion perception: the question of whether to com-
tions: measurement noise and an a priori preference for slower bine measurements into a single coherent motion or assume that
velocities. We found, consistent with results in humans, that the there are actually multiple motions (H. Farid & E. P. Simoncel-
motion estimates of this model include apparent biases and illu- li, Invest. Opthal. Vis. Sci. Suppl. Abstr. 35, 1271, 1994).
sions. Moreover, the predicted non-veridical percept is quite sim- Although the details of our model should certainly be refined
ilar to that exhibited by humans under the same circumstances. and extended to handle more complicated phenomena, we believe
Although the model does not account for all of the existing data the underlying principle will continue to hold: that many motion
quantitatively, it correctly predicted a wide range of effects. ‘illusions’ are not the result of sloppy computation by various com-
Our model does not provide a good quantitative fit to the ponents in the visual system, but rather a result of a coherent com-
data of Fig. 5a (see Results), which suggest a quasilinear depen- putational strategy that is optimal under reasonable assumptions.
dence of perceived grating speed on contrast, and minimal
dependence on total contrast. Our model has been extended by METHODS
including a nonlinear ‘gain control’ function to map stimulus Most models of early motion extraction rely on an assumption of ‘inten-
contrast into perceived contrast (F. Hurlimann, D. Kiper & M. sity conservation’. Under this assumption, the points in the world, as
measured in the image, move but do not change their intensity over time.
Carandini, Invest. Opthal. Vis. Sci. Suppl. Abstr. 40, 794, 2000). Mathematically, this is expressed as:
For each subject in that study, the authors measured a gain con-
trol function from contrast-discrimination experiments. They I(x,y,t) = I(x + vx∆t, y + vy∆t, t + ∆t) (2)
then used the perceived contrast rather than stimulus contrast
as input to our model, and found that when these realistic rep- where vx and vy are the components of the vector, v, describing the image
resentations of contrast were used, the quantitative predictions velocity. If we assume that the observed image is noisy, then intensity is
not conserved exactly. Thus, equation (2) becomes
of the Bayesian model were in general agreement with the data.
We also found, using a numerical search procedure, that a I(x,y,t) = I(x + vx∆t, y + vy∆t, t + ∆t) + η (3)
monotonic nonlinear gain control function enabled our model
to better fit the results reported here and in ref. 31 (see Supple- where η is a random variable representing noise.
mentary Results online). We used equation (3) to derive the likelihood at location i,
One result33 that is not predicted by our model is the find- P(I(xi,yi,t)|vi). This required additional assumptions. We assumed the
noise, η, is Gaussian with standard deviation σ. We further assumed
ing that low-contrast gratings actually appear to move faster
that the velocity is constant in a small window around xi,yi and that the
than high-contrast gratings for temporal frequencies above intensity surface I(x,y,t) is sufficiently smooth that it can be approxi-
8 Hz. However, the same author later was unable to reproduce mated by a linear function for small temporal durations. We thus
this result using a forced-choice task31, and concluded that the replaced I(x + vx∆t, y + vy∆t, t + ∆t) with its first-order Taylor series
original finding was probably “an artifact of the experimental expansion, which gives:
method with subjects making ‘speed’ matches based on some
other criterion”. P(I(xi,yi,t)|vi) ∝
Our Bayesian estimator is meant as a perceptual model, and 
exp  –
∫
1 wi(x,y) (Ix(x,y,t)vx + Iy(x,y,t)vy + It(x,y,t))2 dx dy
does not specify a particular implementation. Nevertheless, the — 2 
2σ x,y
solution can be instantiated using so-called motion energy mech- (4)
anisms28,38, and detailed models of the physiology of the motion
where {Ix,Iy,It} denote the spatial and temporal derivatives of the intensity
pathway24,25,28,39–41 suggest that a population of MT cells may
function I, and wi(x,y) is a window centered on (xi,yi). The likelihoods
be forming a representation of the local likelihood of velocity. In shown in Fig. 3 and Fig. 4 are computed from equation (4) with w(x,y)
addition, we believe it should be possible to refine and justify the a small Gaussian window.
assumptions we have made. In particular, the prior distribution Finally, we assumed a prior favoring slow speeds:
on velocity could be estimated empirically from the statistics of
motion in the world. In a physiological implementation, the noise P(v) ∝ exp(–||v||2/2σp2). (5)
model should be replaced by one that more accurately reflects
the uncertainties of neural responses. The posterior probability of a velocity was computed by combining the
Our model also suggests some future experiments. First, if the likelihood and prior using Bayes’ rule. Because we assumed that the noise
single free parameter is observer dependent (but otherwise con- is independent over spatial location, the total likelihood function is just
stant), the magnitude of different illusions for the same subject a product of likelihoods:
should be correlated. For example, observers who greatly under-
estimate the speed of low-contrast gratings should also show a P(v|I) ∝ P(v) Π P(I(x i,yi ,t) |v), (6)
larger bias towards VA in type II plaids. Second, in all of our sim- i
ulations we used only the maximum (or mean) of the posterior where the product is taken over all locations i that are moving with a com-
distribution. It would be interesting to test whether human per- mon velocity (vi = v). Substituting equations (4) and (5) into equaion (6),

articles
13. Movshon, J. A. & Newsome, W. T. Visual response properties of striate

cortical neurons projecting to area MT in macaque monkeys. Vis. Neurosci.
P(v|I) ∝ exp  –||v||2/2σ p2 – ∫ x,y wi(x,y) (Ix(x,y)vx + Iy(x,y)vy + It)2 dx dy  .
1
— 2 Σ 16, 7733–7741 (1996).
2σ

i 14. Okamoto, H. et al. MT neurons in the macaque exhibited two types of
bimodal direction tuning as predicted by a model for visual motion
detection. Vision Res. 39, 3465–3479 (1999).
Here we assumed the entire image moves according to a single trans- 15. Ferrera, V. & Wilson, H. Perceived direction of moving two-dimensional
lational velocity, and so summed over all spatial positions. In this case, patterns. Vision Res. 30, 273–287 (1990).
∑i wi(x,y) is a constant, so the posterior probability is given by: 16. Mingolla, E., Todd, J. & Norman, J. The perception of globally coherent
motion. Vision Res. 32, 1015–1031 (1992).
17. Yo, C. & Wilson, H. Perceived direction of moving two-dimensional patterns
P(v|I) ∝ exp  –||v|| /2σ p – ∫ x,y (I(x,y) vx + Iy(x,y)vy + It) dx dy depends on duration, contrast, and eccentricity. Vision Res. 32, 135–147 (1992).
2 2 1 2
—
2σ 2
 18. Burke, D. & Wenderoth, P. The effect of interactions between one-
dimensional component gratings on two dimensional motion perception.
Vision Res. 33, 343–350 (1993).
To find the most probable velocity, we replaced the integral with a dis- 19. Bowns, L. Evidence for a feature tracking explanation of why type II plaids
move in the vector sum directions at short durations. Vision Res. 36,
crete sum, took the logarithm of the posterior, differentiated it with 3685–3694 (1996).
respect to v and set the derivative equal to zero. The logarithm of the 20. Stone, L., Watson, A. & Mulligan, J. Effect of contrast on the perceived
posterior is quadratic in v so that the solution can be written in closed direction of a moving plaid. Vision Res. 30, 1049–1067 (1990).
form using standard linear algebra. The result is given in equation (1). 21. Rubin, N. & Hochstein, S. Isolating the effect of one-dimensional motion
signals on the perceived direction of moving two-dimensional objects. Vision
Res. 33, 1385–1396 (1993).
Note: Supplementary information is available on the Nature Neuroscience website. 22. Helmholtz, H. Treatise on Physiological Optics (Thoemmes, Bristol, UK, 2000;
original publication 1866).
23. Knill, D. & Richards, W. Perception as Bayesian Inference (Cambridge Univ.
Acknowledgments Press, Cambridge, 1996).
Y.W. and E.H.A. were supported by US National Eye Institute R01 EY11005 to 24. Ascher, D. & Grzywacz, N. A Bayesian model for the measurement of visual
velocity. Vision Res. 40, 3427–3434 (2000).
E.H.A. E.P.S. was supported by the Howard Hughes Medical Institute and the 25. Koechlin, E., Anton, J. L. & Burnod, Y. Bayesian inference in populations of
Sloan-Swartz Center for Theoretical Visual Neuroscience at New York cortical neurons: a model of motion integration and segmentation in area
University. We thank J. McDermott, M. Banks, M. Landy, W. Geisler and the MT. Biol. Cybern. 80, 25–44 (1999).
26. Simoncelli, E., Adelson, E. & Heeger, D. in Proc. IEEE Conf. Comput.
anonymous referees for comments on previous versions of this manuscript. Vision Pattern Recog. 310–315 (IEEE, Washington DC, 1991).
27. Heeger, D. J. & Simoncelli, E. P. in Spatial Vision in Humans and Robots Ch.
Competing interests statement 19 (eds. Harris, L. & Jenkin, M.) 367–392 (Cambridge Univ. Press, 1994).
28. Simoncelli, E. P. Distributed Representation and Analysis of Visual Motion.
The authors declare that they have no competing financial interests. Thesis, Massachusetts Institute of Technology (1993).
29. Weiss, Y. Bayesian Motion Estimation and Segmentation. Thesis,
RECEIVED 19 FEBRUARY; ACCEPTED 15 APRIL 2002 Massachusetts Institute of Technology (1998).
30. Ullman, S. The Interpretation of Visual Motion (MIT Press, Cambridge,
Massachusetts, 1979).
1. Nakayama, K. Biological image motion processing: a review. Vision Res. 25, 31. Stone, L. & Thompson, P. Human speed perception is contrast dependent.
625–660 (1985). Vision Res. 32, 1535–1549 (1990).
2. Horn, B. K. P. & Schunck, B. G. Determining optical flow. Artif. Intell. 32. Lorenceau, J., Shiffrar, M., Wells, N. & Castet, E. Different motion sensitive
units are involved in recovering the direction of moving lines. Vision Res. 33,
17(1–3), 185–203 (1981).
1207–1217 (1992).
3. Lucas, B. D. & Kanade, T. An iterative image registration technique with an
33. Thompson, P. Perceived rate of movement depends on contrast. Vision Res.
application to stereo vision. in Proceedings of the 7th International Joint 22, 377–380 (1982).
Conference on Artificial Intelligence 674–679 (Morgan-Kaufmann, San 34. Thompson, P., Stone, L. & Swash, S. Speed estimates from grating patches are
Fransisco, 1981). not contrast normalized. Vision Res. 36, 667–674 (1996).
4. Wuerger, S., Shapley, R. & Rubin, N. On the visually perceived direction of 35. Blakemore, M. & Snowden, R. The effect of contrast upon perceived speed: a
motion by Hans Wallach: 60 years later. Perception 25, 1317–1367 (1996). general phenomenon? Perception 28, 33–48 (1999).
5. Wallach, H. Ueber visuell whargenommene bewegungrichtung. Psychol. 36. Snowden, R. N., Stimpson, N. & Ruddle, S. Speed perception fogs up as
Forsch. 20, 325–380 (1935). visibility drops. Nature 392, 450 (1998).
6. Marr, D. & Ullman, S. Directional selectivity and its use in early visual 37. Wilson, H., Ferrera, V. & Yo, C. A psychophysically motivated model for two-
processing. Proc. R. Soc. Lond. B Biol. Sci. 211, 151–180 (1981). dimensional motion perception. Vis. Neurosci. 9, 79–97 (1992).
7. Adelson, E. & Movshon, J. Phenomenal coherence of moving visual patterns. 38. Weiss, Y. & Fleet, D. in Probabilistic Models of the Brain Ch. 4 (eds. Rao, R.,
Nature 300, 523–525 (1982). Olshausen, B. & Lewicki, M.) 77–96 (MIT Press, Cambridge, Massachusetts,
8. Movshon, A., Adelson, E., Gizzi, M. & Newsome, W. The analysis of moving 2002).
visual patterns. Exp. Brain Res. 11, 117–152 (1986). 39. Nowlan, S. J. & Sejnowski, T. J. A selection model for motion processing in
9. Welch, L. The perception of moving plaids reveals two processing stages. area MT of primates. J. Neurosci. 15, 1195–1214 (1995).
Nature 337, 734–736 (1989). 40. Simoncelli, E. & Heeger, D. A model of neuronal responses in visual area MT.
10. Morgan, M. Spatial filtering precedes motion detection. Nature 355, 344–346 Vision Res. 38, 743–761 (1998).
(1992). 41. Pouget, A., Dayan, P. & Zemel, R. Information processing with population
11. Schrater, P., Knill, D. & Simoncelli, E. Mechanisms of visual motion codes. Nat. Rev. Neurosci. 1, 125–32 (2000).
detection. Nat. Neurosci. 3, 64–68 (2000). 42. Grzywacz, N. & Yuille, A. Theories for the visual perception of local velocity
12. Rodman, H. & Albright, T. Single-unit analysis of pattern motion selective and coherent motion. in Computational Models of Visual Processing (eds.
properties in the middle temporal visual area MT. Exp. Brain Res. 75, 53–64 Landy, J. & Movshon, J.) 231–252 (MIT Press, Cambridge, Massachusetts,
(1989). 1991).

articles
Stable perception of visually

ambiguous patterns
David A. Leopold, Melanie Wilke, Alexander Maier and Nikos K. Logothetis
Max Planck Institut für biologische Kybernetik, Spemannstraβe 38, 72076 Tübingen, Germany
Correspondence should be addressed to D.A.L. (david.leopold@tuebingen.mpg.de)
During the viewing of certain patterns, widely known as ambiguous or puzzle figures, perception
lapses into a sequence of spontaneous alternations, switching every few seconds between two or
more visual interpretations of the stimulus. Although their nature and origin remain topics of
debate, these stochastic switches are generally thought to be the automatic and inevitable
consequence of viewing a pattern without a unique solution. We report here that in humans such
perceptual alternations can be slowed, and even brought to a standstill, if the visual stimulus is peri-
odically removed from view. We also show, with a visual illusion, that this stabilizing effect hinges on
perceptual disappearance rather than on actual removal of the stimulus. These findings indicate that
uninterrupted subjective perception of an ambiguous pattern is required for the initiation of the
brain-state changes underlying multistable vision.
Visual perception involves coordination between sensory sampling appearance of the pattern, not on intermittent disappearance of
of the world and active interpretation of the sensory data. Human the sensory representation itself.
perception of objects and scenes is normally stable and robust, but
it falters when one is presented with patterns that are inherently RESULTS
ambiguous or contradictory. Under such conditions, vision lapses The initial aim of this study was to examine the influence of
into a chain of continually alternating percepts, whereby a viable recent visual history on the perceptual organization of an ambigu-
visual interpretation dominates for a few seconds and is then ous pattern. To address this, we repeatedly presented a rotating
replaced by a rival interpretation. This multistable vision, or ‘mul- sphere (RS) stimulus (Fig. 1a), whose three-dimensional structure
tistability’, is thought to result from destabilization of fundamen- is ambiguous17, in a train of 3-s presentations separated by 5-s
tal visual mechanisms, and has offered valuable insights into how periods during which the screen was blank. Whenever the stim-
sensory patterns are actively organized and interpreted in the ulus was present, subjects reported whether it appeared to be
brain1,2. Despite a great deal of recent research and interest in mul- rotating upward or downward around the horizontal axis. For
tistable perception, however, its neurophysiological underpinnings intermittent viewing, perception tended to become ‘stuck’ in a
remain poorly understood. Physiological studies have suggested particular configuration, often for several minutes at a time
that disambiguation of ambiguous patterns draws on activity with- (Fig. 2). As compared with the continuous viewing condition,
in the visual cortex3–10, but how this activity ultimately contributes phases of perceptual dominance were markedly lengthened, and
to perceptual solution is not yet known. Even less clear is the nature some subjects saw no reversals in rotation for the duration of the
of the perceptual alternation process itself. Traditional views hold 10-min session. Stabilization for the RS was present in 22 of 23
that it is an automatic consequence of incompatible, antagonistic subjects tested in different experiments throughout the study,
stimulus representations in the sensory visual cortex11,12. Recent with no consistent difference between horizontal and vertical
evidence challenges this notion, suggesting instead that perceptu- rotation. This effect was not the result of a permanent perceptu-
al alternations are initiated outside the primarily sensory areas13,14 al bias or reset mechanism following each new stimulus presen-
(for a review, see ref. 15). tation, as there was often sequential stabilization of both possible
In the present study, we report that the spontaneous changes percepts (subjects BL, AM, MW, HH & AH, Fig. 2). In general,
of multistable perception can be greatly slowed, and even brought subjects with shorter mean dominance phases during continu-
to a standstill, when the inducing patterns are viewed intermit- ous presentation exhibited less stabilization during the intermit-
tently rather than continuously. Specifically, when we introduced tent condition. As each presentation time was fixed at 3 s for all
blank periods of several seconds into an extended period of con- subjects, we attribute this trend to an increased probability for
tinuous viewing, we consistently slowed the rate of alternation ‘fast-switchers’ to experience perceptual reversal during a single
by two orders of magnitude. Individual periods of perceptual stimulus presentation.
dominance often exceeded 10 minutes. The stabilization effect Stabilization increased when the stimulus was absent for longer
was present for a variety of bistable patterns, and could not be periods of time (Fig. 3a), indicating that a perceptual configura-
attributed to a permanent perceptual bias. In addition, we used a tion, once established, could survive even relatively long epochs
visual illusion called motion-induced blindness16 to show that in which the stimulus was gone. There was no apparent decline
the stabilization effect depended on intermittent subjective dis- in the survival of a percept for blank times as long as 40 s

articles
Fig. 1. Stimuli used in the study. (a) Rotating sphere (RS), which turned
a b about either the vertical or horizontal axis; (b) quartet dots (QD);
(c) Necker cube (NC); (d) binocular rivalry (BR); and (e) rotating star
(RSt) amid randomly moving dots used to provoke the motion-induced
blindness illusion. (f) Stimuli were presented either continuously or
intermittently, with variable on-times and blank-screen durations.
c d mittent/blinks presentation, whereas a mix of higher rates was

seen during continuous presentation.
To examine whether this stabilization effect holds for multi-
stable stimuli other than the RS, we repeated the experiment with
three other common, but inherently different, bistable patterns
(Fig. 1b–d): the Necker cube18 (NC, ambiguous static depth),
quartet dots19 (QD, ambiguous motion correspondence) and
binocular rivalry20 (BR, introcular conflict). For each stimulus,
we first adjusted the time the stimulus was present (on-time) to
e f be markedly shorter than the mean dominance phase during con-
tinuous viewing. Subjects again viewed each pattern either con-
tinuously or intermittently, reporting which of the two perceptual
configurations they saw whenever the stimulus was present.
Results for these three additional stimuli were similar to those
obtained with the RS (Fig. 4).
Finally, we asked whether physical removal of the stimulus
was necessary to achieve stabilization, or whether its mere sub-
jective disappearance would give similar results. To test this, we
(Fig. 3b). Notably, when alternations did occur, the probability exploited the recently described visual illusion of motion-induced
of a configuration persisting until the subsequent presentation blindness (MIB)16 to provoke frequent illusory fading of an
was highly dependent on how long it was seen before the stimulus ambiguous rotating star (RSt) pattern (Fig. 1e). The RSt was sim-
was removed. Only when a perceptual configuration was seen ilar in many respects to the RS: during continuous viewing, it
consistently (without reversals) for 2 s or longer before stimulus produced perceptual reversals in the direction of rotation. Instead
disappearance was there a high probability that it would persist of comparing continuous with intermittent presentation as
to the next stimulus presentation (Fig. 3c). before, we overlaid the RSt with a field of randomly moving black
Given the reliable persistence of a perceptual configuration dots, which caused the star to periodically fade from perception
across blank periods, could the increased stabilization with longer (several times per minute; Fig. 5). As with the RS, subjects were
blank times be accounted for simply by the decreased overall time required to track the rotational direction of the moving star
of exposure to the stimulus? This cannot be the case, because the whenever it was visible. These epochs of illusory fading had the
actual reduction in total alternations far exceeded that expected same stabilizing effect on the perceived direction of rotation as
from stimulus presentation time alone (Fig. 3d). This also was did the physical blanking described above for the RS. This result
true for the blink condition, where subjects mimicked the dis- indicates that the subjective disappearance of the pattern, rather
appearance and reappearance of the stimulus by periodically than the actual sensory representation, ultimately underlies this
shutting and opening their eyes when they heard appropriately stabilization phenomenon.
timed tones. For blinking, even higher stabilization
was observed. Finally, alternation rates were mea-
sured for a 1-minute sliding window shifted by 1-s
increments, pooled for all subjects and compiled
into a histogram for each condition (with blanks,
Fig. 3e; with blinks, Fig. 3f). Note that for both con-
ditions, very low reversal rates dominated the inter-
Fig. 2. Effects of intermittent presentation on percep-

tion of a RS rotating around the horizontal axis (11 sub-
jects). Alternation in the perceived motion of the front
surface is shown here as fluctuation between the upper
(gray) and lower (black) levels, corresponding to upward
and downward rotation, respectively. Each row com-
pares the condition when the stimulus was continuously
present (left) to that when it was only intermittently pre-
sent (right). Intermittent viewing consisted of 3-s on-
periods interleaved with 5-s blank periods, illustrated at
the bottom. For each subject, ordered from top to bot-
tom according to their inherent switch rate during con-
tinuous viewing, the total number of perceptual reversals
during each stimulus condition is shown at the right.

articles
Fig. 3. Factors contributing to stabilization of the RS.

(a) Mean phase duration (± s.e.m.) for intermittent pre-
a b c
sentation as a function of blank duration (six subjects, on-
time 5 s). (b) Probability that a given perceptual state
survived variable-length blanking period (same data as
in a). Arrows indicate the blank durations for the experi-
ment shown in Fig. 2. (c) Probability that a spontaneous
perceptual reversal persists to the next stimulus presen-
tation, as a function of post-reversal interval (time
between the perceptual change and stimulus blanking).

The number of instances contributing to each bin is
shown above the corresponding bars (same data as in a).
(d) Reversal rate for intermittent condition, normalized
to the reversal rate under continuous viewing, shown for d e f
periodic removal of the stimulus (, five subjects, on-
time 2 s) and timed eye closure (, six subjects, eyes
open 2 s). The dotted line represents the rate decrease
expected if the diminished overall presentation time
were the only factor. (e) Distribution of reversal rates for
continuous versus intermittent (on-time 2 s, blank 5 s)
viewing of RS (five subjects), calculated using a sliding
window (see Results). Probability density plotted for
continuous (light gray) and intermittent (dark gray) con-
ditions. (f) Same as (e), but with eye blinking instead of
stimulus blanking (six subjects, pooled).
DISCUSSION reversal is governed by an autonomous oscillator that operates

The inevitability of perceptual alternations during ambiguous independently of the visual stimulus14. If that were the case, a
vision has often been taken as evidence for a stimulus-induced simple stimulus manipulation probably would not bring per-
but autonomously maintained reaction of the sensory process- ceptual alternation to a standstill. Second, our results indicate
ing apparatus to multistable patterns, prompting models whose that continuous, prolonged viewing of an ambiguous stimulus
basis was limited to the visual system11,12,21. More recently, the is a requirement for the initiation of the fluctuations underlying
switching mechanism has been postulated to lie outside of exclu- multistable perception. Moreover, the MIB illusion experiment
sively visual areas. Several lines of evidence have suggested that showed that the continuous presence of the sensory stimulus does
frontoparietal areas associated with selective visual attention are not automatically lead to perceptual instability as predicted by
centrally involved in the initiation of perceptual alternations13,15. models of alteration based on reciprocal sensory connectivity
Other evidence has suggested that alternation in hemispheric and adaption. Instead, the periodic subjective fading of the
dominance, controlled by a free-running oscillator circuit in the ambiguous stimulus had approximately the same stabilizing effect
brainstem, might underlie visual bistability14,22. as its physical removal. Perceptual switching thus appears to be
Although the present results cannot pinpoint specific struc- governed by an active mechanism that continuously monitors
tures responsible for perceptual reversal, they do offer new the conscious perceptual representation of the stimulus itself.
insights into the process itself. First, the stabilization effect Perceptual changes may involve the reorganization of neural net-
described here is evidence against the notion that perceptual works related to sensory processing, but they do not seem to be
triggered by competition in such networks, nor do
a they reflect the workings of a separate, endogenous
b bistable circuit whose impact is merely probed by
an ambiguous pattern. The fact that the greatest sta-
bilization was obtained when subjects made the
stimulus disappear by purposefully closing their
eyes adds further support to the notion that active,
endogenous mechanisms ultimately govern the
occurrence of a perceptual reversal.
Finally, ambiguous perception has often been
described as ‘memoryless’ because of the lack of
c d
Fig. 4. Stabilization of different bistable patterns.
(a–c) Effects of continuous (left) versus intermittent
(right) presentation on the perception of NC, QD and
BR stimuli (two subjects shown for each stimulus).
(d) Total number of reversals for continuous versus
intermittent viewing (means ± s.e.m. shown). Blank dura-
tions were 5 s in all cases, and on-times were RS, 3 s
(eight subjects); NC, 1 s (seven subjects); QD, 5 s (eight
subjects); BR, 1.2 s (eight subjects).
articles
Fig. 5. Effect of subjective disappearance on perceptual

a b reversal of the RSt. (a) Comparison of continuous visi-
bility (left) versus intermittent visibility (right).
Intermittent visibility was achieved by motion-induced
blindness (see Methods). In both cases, the physical
stimulus was continuously present. Data is shown for
two subjects (CW and VM). The pattern of subjective
appearance and disappearance is shown at the bottom
for each subject, with gray periods representing times
when the stimulus was visible. (b) Mean reversal rate

(± s.e.m.) with and without epochs of perceptual disap-
pearance (five subjects).
statistical correlation between successive reversal intervals23,24. posed of blue lines (0.12° in thickness) and covered an area of 4.2° ×
Nonetheless, we found that once the ambiguous stimulus was 4.2°. The QD consisted of solid blue circles (0.36° in diameter) appear-
ing in pairs on opposite corners of an imaginary square (2.7° × 2.7°)
removed, recent perceptual history was the dominant factor in
centered on the middle of the screen. At each moment, only two circles
determining how that stimulus was interpreted on subsequent were displayed. The stimulus alternated between the two possible con-
viewings. Thus, with each presentation, perceptual organiza- figurations every 311 ms, generating the perception of apparent motion
tion seems to be guided by some sort of implicit perceptual between either vertically or horizontally corresponding points. The
memory that does not decline over several seconds. This ‘mem- BR stimulus consisted of two dichoptically presented Gabor patches
ory’, which may be important during normal vision, was abol- (radius, 2.25°; spatial frequency, 2.7 cycles/°), with the left eye view-
ished by the continuous presentation of an ambiguous ing a 45° leftward-tilted greenish patch and the right eye viewing a 45°
stimulus. Whereas a number of previous studies have demon- rightward-tilted pinkish patch. The RSt was similar to the RS in that it
strated that external stimuli can act to deterministically bias was ambiguous in its three-dimensional structure from motion. It was
1.15° in extent, composed of dots (0.044° in diameter) that were red
the perception of an ambiguous pattern25–28, the stabilization
(CIE x = 0.547, y = 0.319, 7.61 cd/m2) and rotated once every 1.3 s. In
observed here depended entirely on the persistence of an inter- the MIB condition16, the RSt was presented amid a large field of small
nally generated perceptual state. Further experiments may black dots moving in random directions (dot diameter, 0.072; density,
reveal whether the same frontoparietal cortical areas that have 9.9 dots/°; speed, 5.4°/s; lifetime, 330 ± 30 ms). While the RSt was pre-
been implicated in perceptual alternation during bistable pat- sented monocularly to the left eye, the dots were shown binocularly
tern viewing 13,29 also contribute to perceptual stabilization in correspondence. Pilot experiments showed that binocular presen-
during intermittent viewing. tation of the dots nearly doubled the disappearance time of the star,
an effect that may be facilitated, in part, by binocular rivalry. The
METHODS eccentricity of presentation ranged from 0.0–1.6° and was determined
Subjects. Thirty-nine subjects (19 female, 20 male) between the ages of initially for each subject based on the position providing the highest
15 and 36 years (median 25) participated in the study. The experi- frequency of stimulus disappearance.
ments were done in accordance with guidelines of the local authori-
ties (Regierungspraesidium), and all subjects gave informed written Experimental task. Participants rested their chins on a padded bar and
consent. Each subject had normal or corrected-to-normal vision, and were instructed to inspect the stimulus without special regard to fixa-
most had prior experience as a psychophysical subject. Apart from two tion, except for the MIB experiment, where they were instructed to
authors (subjects MW and AM), each subject was completely naïve to maintain fixation on a small cross. For each bistable stimulus, buttons
the hypotheses and goals of the experiment, and was paid for partici- were assigned beforehand to each of the two potential percepts. Sub-
pation. An interview after each session revealed that most subjects jects were required to press the button corresponding to the perceived
(>80%) were unaware that the perceptual changes they experienced configuration of each pattern when it appeared, and to release the but-
were entirely subjective. ton when it disappeared. In the case of BR, where percepts could be
mixed, they were instructed to respond according to the dominant
Visual stimuli. Stimuli were generated on a computer (Intergraph Zx10 pattern, even if its dominance was incomplete. All relevant events,
PC, Huntsville, Alabama; Intense3D Graphics, Sunnyvale, California) including stimulus presentations and subject responses, were record-
and presented in color on two 21-inch monitors, presented separately to ed on a computer for analysis.
each eye by a mirror stereoscope. The spatial resolution of each moni-
tor was 1,280 × 800 pixels, with an eye–screen distance of 123 cm and a Acknowledgments
refresh rate of 90 Hz. Unless otherwise mentioned, all stimuli were drawn The authors would like to thank M. Sereno for suggestions and help with the
on the center of a gray screen (5.50 cd/m2) with no fixation spot. Four structure from motion stimuli, A. Gail for discussion regarding the binocular
white, radially protruding bars (0.14° × 3.6°), starting 2.8° from the cen- rivalry experiment and J. Werner for technical assistance. This work was
ter of the screen and extending outward, were used to ensure proper supported by the Max Planck Society.
binocular vergence for all conditions.
The RS, QD and NC stimuli were blue (CIE x = 0.250, y = 0.208, Competing interests statement
7.30 cd/m2) and were presented monocularly to the left eye while the The authors declare that they have no competing financial interests.
right eye viewed a blank screen. Subjects verified that under this con-
dition the stimuli did not subjectively fade (indicating that there was no RECEIVED 7 FEBRUARY; ACCEPTED 9 APRIL 2002
binocular rivalry). The RS stimulus consisted of an orthographic pro-
jection of 450 blue dots (0.044° in diameter) uniformly covering a vir-
tual sphere with a diameter of 1.72° in diameter. The sphere rotated 1. Rock, I. Perception (Scientific American Library, New York, 1984).
rigidly with a period of 4.0 s, giving the appearance of three-dimen- 2. Gregory, R. Eye and Brain: The Psychology of Seeing 5th edn. (Princeton Univ.
Press, Princeton, NJ, 1997).
sional structure. Because no size or perspective cues differentiated the 3. Logothetis, N. K. & Schall, J. D. Neuronal correlates of subjective visual
front from rear surfaces, the direction of rotation was ambiguous. The perception. Science 245, 761–763 (1989).
NC, containing ambiguous information about static depth, was com- 4. Leopold, D. A. & Logothetis, N. K. Activity changes in early visual cortex

articles
reflect monkeys’ percepts during binocular rivalry. Nature 379, 549–553 17. Wallach, H. & O’Connell, D. N. The kinetic depth effect. J. Exp. Psychol. 45,
(1996). 205–217 (1953).
5. Sheinberg, D. L. & Logothetis, N. K. The role of temporal cortical areas in 18. Necker, L. A. Observations on some remarkable optical phenomena seen in
perceptual organization. Proc. Natl. Acad. Sci. USA 94, 3408–3413 Switzerland; and on an optical phenomenon which occurs on viewing a
(1997). figure of a crystal or geometrical solid. Lond. Edinburgh Phil. Magazine J. Sci.
6. Bradley, D. C., Chang, G. C. & Andersen, R. A. Encoding of three- 1, 329–337 (1832).
dimensional structure-from-motion by primate area MT neurons. Nature 19. von Schiller, P. Stroboskopische Alternativversuche. Psychol. Forsch. 17,
392, 714–717 (1998). 179–214 (1933).
7. Dodd, J. V., Krug, K., Cumming, B. G. & Parker, A. J. Perceptually bistable 20. Dutour, E. F. Discussion d’une question d’optique [Discussion on a question
three-dimensional figures evoke high choice probabilities in cortical area MT. of optics]. l’Academie des Sciences. Memoires de Mathematique et de physique
J. Neurosci. 21, 4809–4821 (2001). presentes par Divers Savants 3, 514–530 (1760).
8. Polonsky, A., Blake, R., Braun, J. & Heeger, D. J. Neuronal activity in human 21. Lehky, S. R. An astable multivibrator model of binocular rivalry. Perception
primary visual cortex correlates with perception during binocular rivalry. 17, 215–228 (1988).
Nat. Neurosci. 3, 1153–1159 (2000). 22. Miller, S. M. et al. Interhemispheric switching mediates perceptual rivalry.
9. Tononi, G., Srinivasan, R., Russell, D. P. & Edelman, G. M. Investigating Curr. Biol. 10, 383–392 (2000).
neural correlates of conscious perception by frequency-tagged 23. Borsellino, A., De Marco, A., Allazetta, A., Rinesi, S. & Bartolini, B. Reversal
neuromagnetic responses. Proc. Natl. Acad. Sci. USA 95, 3198–3203 time distribution in the perception of visually ambiguous stimuli. Kybernetik
(1998). 10, 139–144 (1972).
10. Brown, R. J. & Norcia, A. M. A method for investigating binocular rivalry in 24. Fox, R. & Herrmann, J. Stochastic properties of binocular rivalry
real-time with the steady-state VEP. Vision Res. 37, 2401–2408 (1997). alternations. Percept. Psychophys. 2, 432–436 (1967).
11. Attneave, F. Multistability in perception. Sci. Am. 225, 63–71 (1971). 25. Carlson, V. R. Satiation in a reversible perspective figure. J. Exp. Psychol. 45,
12. Mueller, T. J. A physiological model of binocular rivalry. Vis. Neurosci. 4, 442–448 (1953).
63–73 (1990). 26. Wolfe, J. M. Reversing ocular dominance and suppression in a single flash.
13. Lumer, E. D., Friston, K. J. & Rees, G. Neural correlates of perceptual rivalry Vision Res. 24, 471–478 (1984).
in the human brain. Science 280, 1930–1934 (1998). 27. Nawrot, M. & Blake, R. Neural integration of information specifying
14. Pettigrew, J. D. Searching for the switch: neural bases for perceptual rivalry structure from stereopsis and motion. Science 244, 716–718 (1989).
alternations. Brain Mind 2, 85–118 (2001). 28. Blake, R., Westendorf, D. & Fox, R. Temporal perturbations of binocular
15. Leopold, D. A. & Logothetis, N. K. Multistable phenomena: changing views rivalry. Percept. Psychophys. 48, 593–602 (1990).
in perception. Trends Cogn. Sci. 3, 254–264 (1999). 29. Lumer, E. D. & Rees, G. Covariation of activity in visual and prefrontal cortex
16. Bonneh, Y. S., Cooperman, A. & Sagi, D. Motion-induced blindness in associated with subjective visual perception. Proc. Natl. Acad. Sci. USA 96,
normal observers. Nature 411, 798–801 (2001). 1669–1673 (1999).

Nature Neuroscience June 2002

Uploaded by

Copyright:

Available Formats

You might also like

Nature Neuroscience June 2002

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Nature Neuroscience June 2002

Uploaded by

Copyright:

Available Formats

contents

volume 5 no 6 june 2002

Wnt signals lead cells down the caudal path . . . . . . . . . . . . . . . . . . . . . . . . . . . . 510

nature neuroscience • volume 5 no 6 • june 2002 i

Progressive induction of caudal neural character by

Ion channel properties underlying axonal action potential initiation

Ethanol elicits and potentiates nociceptor responses via

Decreasing hypothalamic insulin receptors causes hyperphagia

Stable perception of visually ambiguous patterns . . . . . . . . . . . . . . . . . . . . . . . . 605

classified advertising . . . . . . . . . . . . . . . . . . . . . . . . . . . . . see back pages

nature neuroscience • volume 5 no 6 • june 2002 ii

Of apes and men

nature neuroscience • volume 5 no 6 • june 2002 501

Quick-change artist: from excitatory

nature neuroscience • volume 5 no 6 • june 2002 503

504 nature neuroscience • volume 5 no 6 • june 2002

principal neurons, the investigators 3

tions, concentrations and repetition rates

in the glomerular population1. A second many different imaging methods, these

nature neuroscience • volume 5 no 6 • june 2002 505

salient spatial cues and our responses to

506 nature neuroscience • volume 5 no 6 • june 2002

neurons responded more strongly

ment in different behavioral con-

nature neuroscience • volume 5 no 6 • june 2002 507

observers’. As one might guess intuitive-

Bayes each possible true state of the environ-

508 nature neuroscience • volume 5 no 6 • june 2002

Fig. 1. Bayesian ideal observers for tasks involving the perception

given the stimulus. Finally, the posterior probability distribution is

nature neuroscience • volume 5 no 6 • june 2002 509

Wnt signals lead cells down the caudal path

510 nature neuroscience • volume 5 no 6 • june 2002

um-level sensing mechanism in the brain8.

Fig. 1. Nax is a sodium concentration–sensitive sodium channel.

observed in wild-type and Nax-null cells. *P < 0.005 by one-tailed Mann-

Whitney tests; n = 10.

nature neuroscience • volume 5 no 6 • june 2002 511

a b 40 a Control 170 mM Anti-Nax

Fig. 2. Nax cDNA transfection conferred [Na ]o sensitivity on Nax- +

512 nature neuroscience • volume 5 no 6 • june 2002

Acute versus chronic NMDARs are crucial in the developmental acquisition of

Fig. 1. Effects of acute and chronic blockade of NMDA receptors on

genetic9,10 experiments. Together, these findings indicate that

514 nature neuroscience • volume 5 no 6 • june 2002

Concurrent measurement of the concentrations in blood did not

nature neuroscience • volume 5 no 6 • june 2002 515

even in more interior brain regions, such as the amygdala10.

516 nature neuroscience • volume 5 no 6 • june 2002

Thalamcortical optimization of tactile

Miguel A. L. Nicolelis1,2,3,4 and Erika E. Fanselow1

nature neuroscience • volume 5 no 6 • june 2002

Fig. 1. Schematic diagram of the main rat

A shift from bursting mode to tonic

for 50–100 ms so that I T will be suffi-

518 nature neuroscience • volume 5 no 6 • june 2002

Fig. 3. Neural activity in VPM thalamus

(bold dotted lines, stimulus presentation;

standing or sitting still but twitch