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

com

contents

volume 3 no 2 february 2000

http://neurosci.nature.com

Kim and colleagues report a tech-

nical advance that gives sufficient
spatial resolution to image cortical
editorial
column organization by fMRI in
2000 Nature America Inc. http://neurosci.nature.com

cats. Instead of the traditional How experts communicate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

positive BOLD fMRI signal, the
authors used an earlier, negative
phase of the signal to obtain the
map of cortical orientation
columns shown (image filtered
news and views
and rearranged for artistic
purposes). See pages 105 and 164. A taste for umami . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Bernd Lindemann
SEE ARTICLE, PAGE 113

Presynaptic facilitation by hyperpolarization-activated pacemaker channels. . . . 101

Steven A. Siegelbaum
SEE ARTICLE, PAGE 133

Neurodegeneration in the polyglutamine diseases: Act 1, Scene 1 . . . . . . . . . . . 103

Robert Nussbaum and Georg Auburger
SEE ARTICLE, PAGE 157

Non-invasive visualization of cortical columns by fMRI . . . . . . . . . . . . . . . . . . . . 105

Amiram Grinvald, Hamutal Slovin and Ivo Vanzetta
Expertise recruits
SEE ARTICLE, PAGE 164
face-selective brain areas.
Page 191.

brief communications
Rapid, synaptically driven increases in the intrinsic excitability of
cerebellar deep nuclear neurons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
CD Aizenman and DJ Linden

A taste receptor for

monosodium glutamate.
Pages 99 and 113.

Nature Neuroscience (ISSN 1097-6256) is published monthly by Nature America Inc., headquartered at 345 Park Avenue South, New York, NY 10010-1707. Editorial Office: 345 Park
Avenue South, New York, NY 10010. Telephone 212 726 9200, Fax (212) 696 9635. North American Advertising: Nature Neuroscience, 345 Park Avenue South, New York, NY 10010-
1707. Telephone (212) 726-9200. Fax (212) 696-9006. European Advertising: Nature Neuroscience, Porters South, Crinan Street, London N1 9SQ. Telephone (0171) 833 4000. Fax
(0171) 843 4596. New subscriptions, renewals, changes of address, back issues, and all customer service questions in North America should be addressed to Nature Neuro-
science Subscription Department, PO Box 5054, Brentwood, TN 37024-5054. Telephone (800) 524-0384, Direct Dial (615) 377 3322, Fax (615) 377 0525. Outside North America:
Nature Neuroscience, Macmillan Magazines Ltd., Houndsmill, Brunel Road, Basingstoke, RG21 6XS, U.K.. Tel: +44-(0)1256-329242. Fax: +44-(0)1256 812358. Email:
subscriptions@nature.com. Annual subscription rates: U.S./Canada: U.S. $595, Canada add 7% for GST (institutional/corporate), U.S. $195, Canada add 7% for GST (individual
making personal payment BN: 14091 1595 RT); U.K./Europe:435 (institutional/corporate), 185 (individual making personal payment), 99 (student); Rest of world (excluding
Japan): 480 (institutional/corporate), 195 (individual making personal payment), 110 (student); Japan: Contact Japan Publications Trading Co. Ltd., 2-1 Sarugaku-cho 1 chome,
Chiyoda-ku, Tokyo 101, Japan, phone (03) 292-3755. Back issues: U.S./Canada, $45, Canada add 7% for GST; Rest of world: surface U.S. $43, air mail U.S. $45. Reprints: Nature
Neuroscience Reprints Department, 345 Park Avenue South, New York, NY 10010-1707. Subscription information is available at the Nature Neuroscience homepage at http://neu-
rosci.nature.com. POSTMASTER: Send address changes to Nature Neuroscience Subscription Department, P.O. Box 5054, Brentwood, TN 37024-5054. Application to mail periodicals
postage rate is paid at New York, NY. Executive Officers of Nature America Inc: Nicholas Byam Shaw, Chairman of the Board; Jan Velterop, President; Edward Valis, Secretary-Trea-
surer. Printed by Publishers Press, Shepherdsville, KY, USA. Copyright 2000 Nature America Inc.

nature neuroscience volume 3 no 2 february 2000 i

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

contents

articles
A metabotropic glutamate receptor variant functions as a taste receptor . . . . . . 113
N Chaudhari, AM Landin and SD Roper
SEE NEWS AND VIEWS, PAGE 99

Identification and characterization of the high-affinity choline transporter . . . . . 120

T Okuda, T Haga, Y Kanai, H Endou, T Ishihara and I Katsura

Receptors with opposing functions are in postsynaptic microdomains

under one presynaptic terminal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
G Tsen, B Williams, P Allaire, YD Zhou, O Ikonomov, I Kondova and MH Jacob

Enhancement of synaptic transmission by cyclic AMP modulation of

Gene regulation in
spinocerebellar ataxia type 1.
Pages 103 and 157.
2000 Nature America Inc. http://neurosci.nature.com
presynaptic Ih channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
V Beaumont and RS Zucker
SEE NEWS AND VIEWS, PAGE 101

Synaptic activity modulates presynaptic excitability . . . . . . . . . . . . . . . . . . . . . . 142

TA Nick and AB Ribera

A new form of long-term depression in the perirhinal cortex. . . . . . . . . . . . . . . . 150

K Cho, N Kemp, J Noel, JP Aggleton, MW Brown and ZI Bashir

Polyglutamine expansion down-regulates specific neuronal genes before

pathologic changes in SCA1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
X Lin, B Antalffy, D Kang, HT Orr and HY Zoghbi
SEE NEWS AND VIEWS, PAGE 103

High-resolution mapping of iso-orientation columns by fMRI . . . . . . . . . . . . . . . 164

Psychophysics of DS Kim, TQ Duong and SG Kim
Marroquin patterns. SEE NEWS AND VIEWS, PAGE 105
Page 170.
Dynamics of perceptual oscillations in form vision . . . . . . . . . . . . . . . . . . . . . . . . 170
HR Wilson, B Krupa and F Wilkinson

Motion perception during saccadic eye movements . . . . . . . . . . . . . . . . . . . . . . 177

E Castet and GS Masson

Thermosensory activation of insular cortex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184

AD Craig, K Chen, D Bandy and EM Reiman
Inhibition and excitation at
the same axon terminal. Expertise for cars and birds recruits brain areas involved in face recognition. . . . 191
Page 126. I Gauthier, P Skudlarski, JC Gore and AW Anderson

classified advertising
see back pages

nature neuroscience volume 3 no 2 february 2000 ii

 2000 Nature America Inc. http://neurosci.nature.com
How experts communicate
In Darwins time, it was possible to write a book that was both
a primary scientific report and a popular bestseller. Today, how-
ever, that seems like a remote ideal1. Not only is it difficult to
2000 Nature America Inc. http://neurosci.nature.com
communicate scientific ideas to the general public, but scien-
tists seem to have increasing difficulty communicating with
each other. Even within biology, researchers in different areas of
specialization are often unable to understand each others
papers. This is a particular problem for a field such as neuro-
science, whose advances have often depended on the exchange
of ideas across disciplines. Nature Neuroscience seeks to encour-
age clear writing and to make everything we publish accessi-
ble to as many readers as possible. Yet even if this seems
uncontroversial in principle, it is surprisingly difficult to
achieve in practice.
Despite the obvious advantages of communicating clearly, sci-
entists are often resistant to the suggestion that their articles
should be comprehensible to readers outside their own field. For
one thing, there is a tendency to equate plain language with over-
simplification. As science becomes more complex, the argument
goes, an ever-increasing amount of specialist jargon is required
to describe it precisely. Even if this is true, however, technical ter-
minology can be explained, and it need not present an insur-
mountable problem to the scientifically literate reader. A more
important deterrent to clear expressionalthough people are
less willing to admit itis that plain language, no matter how
precise, strikes many scientists as somehow unprofessional. It is
often seen as a badge of academic credibility to express short sim-
ple ideas in long ponderous phrases; why else would anyone
choose to write a sentence such as "To elucidate these issues, we
utilized the caprine model" instead of "We studied these ques-
tions using goats"?
This type of pomposity is easy to avoid once it is recognized,
and fortunately many other common problems in scientific writ-
ing are similarly easy to correct. One of the most obvious is exces-
sive use of abbreviations. People within the field are likely to be
familiar with common abbreviations and process them as if they
were words. However, every unfamiliar abbreviation makes an
additional demand on the readers memory. Individually, such
problems are minor nuisances, but as they accumulate, they can
severely impair understanding. Another common barrier to com-
munication is to describe experimental results in ways that
emphasize the method of analysis rather than the phenomenon
being studied. For example, "ANOVA revealed a significant main
effect of age and a significant interaction effect" is much less infor-
mative than "Protein levels decreased significantly with age, and
this decline was more pronounced in adrenalectomized animals."
Even when making the effort to write for a wide readership,
many authors adopt solutions that are ineffective. For example,
nature neuroscience volume 3 no 2 february 2000
editorial
vapid introductory statements like "Much recent research has
been aimed at understanding synaptic plasticity" are as useless
to nonspecialists as they are to anyone else. Concluding para-
graphs present a similar temptation to vagueness; saying "this
work provides insights" into some problem is less informative
than explaining what those insights were. In fact, there is nothing
mysterious about writing for nonspecialists. The key is to exam-
ine each sentence for hidden assumptions and unfamiliar con-
cepts, and to ensure that they are clearly explained in a way that
minimizes the demands on the readers memory.
The problems posed by a poorly written article are greater
for nonspecialists, but even experts comprehend more easily
if they do not have to waste mental resources on parsing diffi-
cult sentences. Research in linguistics and cognitive psycholo-
gy shows that sentence structure creates expectations about
content and emphasis. Writing that violates these expectations
is difficult to read 2. Clear writing reduces the demands on
working memory by presenting information where readers
expect to find it. Unfortunately, scientific writing often does
the opposite. One common mistake is to separate the sentences
subject from its verb with a long clause that contains impor-
tant information (for example, "An increase in mRNA, which
resulted from transcriptional upregulation by factors binding
to the AP1 site, was observed"). Because the reader is distract-
ed by the need for syntactic closure, material between subject
and verb receives less attention than it should. The opposite
problem occurs when unimportant material is placed in a loca-
tion that readers naturally emphasize. Each sentence contains
at least one stress position near the end, at the point when
readers comprehend how the various parts of the sentence
relate to each other. Indeed, behavioral studies indicate that
readers slow down as they reach the end of a sentence or
clause3. Material at this location is perceived as being impor-
tantwhether the author intended it to be or not. Thus, read-
ers are most comfortable when familiar information at the
beginning of the sentence creates a context for important new
information introduced at the end. These rules do not require
writers to avoid complicated ideas or long sentences, only to
construct them carefully.
Because young scientists learn by imitating their elders, a cul-
ture of bad writing tends to be self-perpetuating. Perhaps the
solution is for graduate programs to place more emphasis on for-
mal instruction in scientific writing, but this will only happen if
scientists appreciate the need for better communication and
understand the steps that can be taken to achieve it.
1 Gould, S. J. Science 286, 899 (1999).
2 Gopen, G. D. & Swan, J. A. Am. Scientist 78, 550558 (1990).
3 Just, M. A. & Carpenter, P. A. Psych. Rev. 87, 329354 (1980).
97
 2000 Nature America Inc. http://neurosci.nature.com

news and views

A taste for umami

Bernd Lindemann

Chaudhari and colleagues identify the taste receptor for L-glutamate, also known as umami,
found in protein-rich foods. The protein they describe is a new G-protein-coupled receptor that
corresponds to a truncated form of the metabotropic glutamate receptor mGluR4.

Molecular biologist have been trying for
many years to clone taste receptors and
identify their ligands. Now at last, they
seem to have succeeded. Chaudhari and
colleagues have identified a receptor for
umami, the taste corresponding to L-glu-
tamate1. The umami receptor is a G-pro-
tein-coupled receptor, which binds
2000 Nature America Inc. http://neurosci.nature.com
ity of taste receptor cells directly, either by
contributing to current flow through ion
channels or by modulating channel con-
ductances4. In contrast, sweet, bitter and
umami are all thought to depend on the
activation of G-protein-coupled receptors
(GPCRs), but until now, none of these
receptors has been definitively identified.
downregulating the second messenger
cAMP. This receptor is expressed on
presynaptic terminals of both gluta-
matergic and GABAergic neurons, where
it mediates glutamate-dependent regu-
lation of neurotransmitter release7. In
addition, mGluR4 is expressed in taste
tissue, making it an obvious candidate

extracellular glutamate and signals for the umami receptor. Howev-

through a G protein to regulate the er, one problem with this idea is
level of intracellular cAMP and the that glutamate activates mGluR4
firing of taste receptor cells. at micromolar concentrations,
Until recently, the textbook wis- far below the taste threshold.
dom was that humans could detect Chaudhari and colleagues
four primary tastes: sweet, bitter, have now resolved this puzzle by
salty and sour. Yet as early as 1908, showing that rat taste tissue also
Kikunae Ikeda at the Tokyo Imperi- expresses an alternative tran-
al University had identified L-gluta- script of mGluR4 (which may
mate as the principal source of a fifth arise by differential transcription
taste. This taste quality, which can- and/or splicing), in which the
not be mimicked by any combina- first 300 amino acids of the
tion of the other four tastes, Ikeda amino (N) terminus are absent.
called umami. Taste researchers have They go on to show that this iso-
long been aware of Ikedas work, but form, when expressed in a cul-
it is only recently that umami has tured cell line, can transduce a
gained widespread recognition in the response to extracellular gluta-
West, perhaps in part because of the mate, over a concentration range
increasing popularity of oriental that is consistent with it being
food 2. Monosodium glutamate the umami taste receptor.
(MSG), of course, is widely used as The loss of the N terminus
a flavoring additive in Asian cuisine, would be predicted to decrease
but the most abundant amino acid, the receptors affinity for gluta-
glutamate, is also an important mate. The sequence of mGluR4 is
Image courtesy of Umami Information Center, Tokyo
nutrient, and it is presumably for this homologous to that of bacterial
reason that some animals have Fig. 1. Umami is an important flavor constituent of many amino-acid binding proteins, for
evolved the ability to taste it. Free protein-rich foods. which crystal structures are avail-
glutamate is found in many protein- able. Based on this homology, the
rich foods, including meat, milk and A report last year described two new N terminus of mGluR4 is predicted to
seafood (Fig. 1); it is particularly concen- GPCRs, TR1 and TR2, that are expressed form a clamshell structure, in which the
trated in aged cheese. Rats and humans can in taste buds. They were suggested to be two halves of the shell are connected by a
both recognize the taste of glutamate, and candidates for sweet and bitter receptors, hinge, forming a cleft in which glutamate
for adult humans the detection threshold but at present there is no functional evi- can bind 810 (Fig. 2). The high-affinity
is about 0.7 mM (ref. 3). dence for this proposal . 5,6 binding site has been mapped by site-
The key to further advance in under- Chaudhari and colleagues1 have pro- directed mutagenesis of the mGluR4
standing taste perception will be the iden- vided such evidence for what seems to sequence, with the hydroxyl groups of Ser
tification of taste receptors. Salty and sour be the umami receptor. The molecule 159 and Thr 182 predicted to form
tastes are produced by small cations such they describe is a new GPCR that corre- hydrogen bonds with glutamate (or its
as Na+, K+ and H+, which affect the activ- sponds to a truncated form of the agonists) and Arg 78 to provide electro-
metabotropic glutamate receptor static attraction. These highly conserved
Bernd Lindemann is in the Department of mGluR4, which they term taste- residues lie within the first half of the
Physiology, Bldg 58, Saar University, D-66421 mGluR4. The mGluR4 receptor was shell structure, and mutating any of them
Homburg, Germany. originally described in the brain, where to alanine causes a major (over 95%)
e-mail: phblin@med-rz.uni-sb.de it responds to extracellular glutamate by reduction in glutamate binding10.

nature neuroscience volume 3 no 2 february 2000 99

 2000 Nature America Inc. http://neurosci.nature.com
news and views
Fig. 2. The mGluR4 protein is a seven-transmembrane itself11. As with glutamate, the
receptor, and the N terminus of the brain isoform (center)
forms a 'clamshell' structure, homologous to bacterial
amino-acid-binding proteins (left), which contains the high-
affinity binding site for glutamate. The taste isoform (right) consistent with the known taste
is truncated at the N terminus.
2000 Nature America Inc. http://neurosci.nature.com
Given the importance of these N-ter-
minal residues, which are absent in the
taste isoform, how can the truncated taste
variant of mGluR4 recognize glutamate
at all? Presumably it must contain an
additional binding site, of lower affinity
than the one at the N terminus. It is
unclear whether a similar site exists in the
brain isoform, but it is interesting to note
that one of the bacterial amino-acid-
binding proteins was proposed to contain
a second binding site on the opposite side
of the cleft from the first site8. It remains
to be determined whether glutamate
binds to the taste mGluR4 isoform at the
corresponding site or at some other posi-
tion, either in the extracellular domains
or in the bundle of seven helices that span
the membrane.
By expressing the two isoforms in a
cultured cell line, Chaudhari and col-
leagues1 show that activation of the taste
isoform (as measured by the reduction in
intracellular cAMP) requires a much
higher concentration of glutamate than
does the brain isoform. This was not due
to any difference in the level of expression
of the two isoforms on the cell surface, so
it seems likely to reflect a reduced affini-
ty for glutamate. This would be consis-
tent with the truncation of the presumed
high-affinity binding site, but it will be
important to confirm the difference in
affinities by direct binding measurements.
Certainly, it would make sense for the
umami receptor to have a lower affinity,
given the high concentrations of gluta-
mate that exist in certain foods. The con-
centrations required to activate the
receptor are also in the same range as the
known behavioral detection threshold
(about 100 micromolar in juvenile and 1
mM in adult rodents).
100
A skeptic might still argue that
taste-mGluR4 is not a taste recep-
tor but is instead involved in
synaptic transmission or some
other function of taste tissue, but
several further lines of evidence
are consistent with a role in taste
transduction. First, the authors
confirm that taste-mGluR4 is acti-
vated by the glutamate agonist L-
AP4, which to rats is
indistinguishable from glutamate
concentrations of L-AP4 required
to produce a cAMP response are
thresholds in rats. Second, the
same group has previously
shown by in situ hybridization
that mGluR4 expression in taste
tissue is concentrated in the taste buds
themselves, and that about 40% of buds are
positive for the label. Although the probe
used in that study did not distinguish
between the two isoforms, which may both
be present in taste tissue, it is significant
that neither isoform was detected outside
the taste buds, either by in situ hybridiza-
tion or by RT-PCR assays11.
Finally, the effect of mGluR4 activation
in cultured cells is to decrease cAMP lev-
els, and glutamate is known to reduce
cAMP in isolated taste buds (X. Zhou & N.
Chaudhari, Chem. Senses 22, 834, 1997).
Although the experiments on taste buds
were performed under somewhat artificial
conditionsthe baseline concentration of
cAMP was raised by treatment with
forskolinthey nevertheless seem to reflect
the normal process of taste transduction.
For instance, the effect of glutamate was
enhanced by inositol monophosphate,
which, like other ribonucleotides that are
found in meat and other umami foods,
enhances the taste of glutamate24.
How does the decrease in cAMP
modulate membrane potential and cause
the taste receptor cell to signal the pres-
ence of ligand? Electrophysiological
recordings of isolated taste cells from the
posterior part of the tongue have shown
two types of responses to L-glutamate12.
Most cells (60%) respond with a sus-
tained hyperpolarization, possibly due to
closure of nonselective cation channels.
A few cells (4%) respond with a transient
depolarization, probably due to the
opening of such channels.
It seems likely that the sustained
hyperpolarizing response is what leads to
taste signaling, because L-AP4, an ago-
nist that also evokes the taste of umami,
also caused the sustained response but
nature neuroscience volume 3 no 2 february 2000
not the transient depolarization. This
suggests a model in which glutamate
triggers a decrease in cAMP, resulting in
the closure of cyclic nucleotide-gated
channels13 and hyperpolarization of taste
receptor cells. This would be analogous
to visual transduction, in which photons
trigger the breakdown of cGMP, result-
ing in closure of cGMP-gated cation
channels and hyperpolarization of pho-
toreceptors. In the case of taste receptors,
it is still not known whether hyperpolar-
ization modulates tonic release of trans-
mitter, as it does in vertebrate
photoreceptors, or whether it has some
other effect such as inhibiting the
response to other tastants.
An additional point of interest is that a
small number of taste cells from the ante-
rior part of the rat tongue have been found
to respond to L-AP4 with a depolarization
rather than a hyperpolarization14. The basis
for this difference is unknown; perhaps
they express channels that are closed rather
than opened by cAMP15. The role of these
cells in taste transduction is also unknown,
but it is possible that they, rather than the
hyperpolarizing cells, may be the real
umami receptors.
Even though it cannot clarify all these
issues, the paper by Chaudhari and col-
leagues is an important advance. The
cloning of a taste receptor with an iden-
tified ligand should add meat to the field
of taste research. Bon apptit!
1. Chaudhari, N., Landin, A. M. & Roper, S. D.
Nat. Neurosci. 3, 113119 (2000).
2. Umami Company Report. Umami
Information Center, 1-15-1 Kyobashi, Chuo-
ku, Tokyo 104, Japan (1985).
3. Yamaguchi, S. Physiol. Behav. 49, 833841
(1991).
4. Lindemann, B. Physiol. Rev. 76, 719766
(1996).
5. Hoon, M. A. et al. Cell 96, 541551 (1999).
6. Lindemann, B. Nat. Med. 5, 381382 (1999).
7. Bradley, S. R. et al. J. Comp. Neurol. 407, 3346
(1999).
8. Sack, J. S., Saper, M. A. & Quiocho, F. A.
J. Mol. Biol. 206, 171191 (1989).
9. OHara, P. J. et al. Neuron 11, 4152 (1993).
10. Hampson, D. R. et al. J. Biol. Chem. 274,
3348833495 (1999).
11. Chaudhari, N. et al. J. Neurosci. 16, 38173826
(1996).
12. Bigiani, A., Delay, R. J., Chaudhari, N.,
Kinnamon, S. C. & Roper, S. D.
J. Neurophysiol. 77, 30483059 (1997).
13. Misaka, T. et al. J. Biol. Chem. 272,
2262322629 (1997).
14. Lin, W. & Kinnamon, S. C. J. Neurophysiol. 82,
20612069 (1999).
15. Kolesnikov, S. S. & Margolskee, R. F. Nature
376, 8588 (1995).
 2000 Nature America Inc. http://neurosci.nature.com
Presynaptic facilitation by
hyperpolarization-activated
pacemaker channels
Steven A. Siegelbaum
Hyperpolarization-activated potassium channels presynaptically
facilitate transmitter release in a calcium-independent fashion.
The past few years have seen rapid
advances in our understanding and appre-
ciation of hyperpolarization-activated non-
selective cation channels, also known as Ih
channels. Although the first well-charac-
2000 Nature America Inc. http://neurosci.nature.com
terized role of these channels was the con-
trol of pacemaking activity in sinoatrial
node cells of the heart1, they are also wide-
ly expressed in peripheral and central neu-
rons 2. Neuronal I h channels control
rhythmic electrical activity in sponta-
neously active cells and regulate membrane
excitability in quiescent cells. On page 133
of this issue, Beaumont and Zucker3 report
a new role for Ih channelsthey contribute
to presynaptic facilitation of transmitter
release. Surprisingly, this facilitation does
not seem to involve an increase in intracel-
lular calcium; instead, the authors raise the
provocative possibility that it may involve a
direct coupling between these channels and
the vesicular release machinery.
Ih channels have several distinctive fea-
tures. Unlike most voltage-gated channels,
they open in response to negative-going
voltage steps to potentials within the range
of the normal resting potential (Fig. 1b).
They conduct both potassium (K+) and
sodium (Na+) ions, with a three-fold greater
permeability to K +, yielding a reversal
potential of 30 to 40 mV under physio-
logical conditions. As a result, the opening
of Ih channels near the resting potential
(~ 60 mV) generates an inward, depolar-
izing current that is largely carried by Na+.
In heart pacemaker cells, these channels
contribute to the slow phase of depolariza-
tion that follows the repolarization phase
of the action potential. As the membrane
potential reaches threshold, voltage-gated
Na+ channels and T-type Ca2+ channels are
activated, generating the rapid rising phase
of the next action potential, during which
the Ih channels shut (Fig. 1c). Another
Steven Siegelbaum is at the Center for Neurobiology,
Department of Pharmacology, Howard Hughes
Medical Institute, Columbia University, 722 W. 168
St., New York, New York 10032, USA.
e-mail: sas8@columbia.edu
nature neuroscience volume 3 no 2 february 2000
news and views
the K + channel signature sequence
(glytyrgly). In addition, the intracellu-
lar C terminus contains a 120-amino-acid
sequence that is homologous to the cAMP-
and cGMP-binding domains of other pro-
teins, and is thus the likely site for cAMP
regulation of channel opening8. All four
mammalian genes are expressed in brain,
with differing expression patterns.
In neurons, Ih channels have diverse
functions. They were initially shown to be
inward rectifiers; they are active near the
resting potential and pass inward current
unusual property of these channels is their more readily than outward current, there-
regulation by cyclic nucleotides1, which by helping to control resting potential and
speed the rate of channel activation by input resistance2. In photoreceptors, Ih
binding to an intracellular site on the chan- channels help to damp the hyperpolarizing
nel (Fig. 1b). In the heart, this speeds the effect of light; they are activated by hyper-
slow phase of spontaneous depolarization polarization, causing the voltage response
(and hence the heart rate) in response to to light to fade during the first 100200 ms,
-adrenergic receptor stimulation. thus producing sensory adaptation. In many
The genes encoding Ih channels were CNS neurons, activation of Ih channels fol-
recently cloned from both mammals46 and lowing inhibitory postsynaptic potentials
sea urchins7 (see ref. 8 for review). These contributes to a rebound afterdepolariza-
genes, called HCN14 in mammals, are tion (ADP), which can trigger an action
members of the voltage-gated K+ channel potential. Ih can also contribute to sponta-
family. The encoded proteins contain six neous activity in neurons, similar to its role
transmembrane segments, including a pos- in the heart. For instance, in thalamocortical
itively charged S4 voltage sensor (Fig. 1a) relay neurons, activation of Ih underlies a
and a pore-forming P region that includes burst firing pattern associated with slow-
a P
b
out -50 mV
s1 s2 s3 s4 s5 s6
in -90 mV
CNBD
Control
+cAMP
N C
d 50 Hz
control
prox
c +cAMP
dist
+40 mV
Control
10 mV
+ZD7288
50 ms
-60 mV
dist
prox
Ih ICa,T
Amy Center
Fig. 1. Ih channels and their role in excitability. (a) Schematic of transmembrane topology of the
HCN hyperpolarization-activated channels. (b) Hyperpolarizing voltage step from 50 mV to 90
mV activates an inward Ih current with slow kinetics. On returning to the holding potential, there is
an inward tail current as the Ih channels deactivate. Activation of Ih under basal conditions (blue)
and after elevation of cAMP (red). (c) Contribution of Ih current to spontaneous action potential
generation under basal conditions (blue) and after elevation of cAMP (red). (d) Role of Ih in den-
dritic integration. EPSPs recorded in cell body of hippocampal CA1 pyramidal neuron in response
to stimulation of distal or proximal Schaffer collateral inputs. Top traces, control. Bottom traces,
after blockade of Ih with ZD7288 (from ref. 9). Ih channels, localized at a high density in the distal
dendrite, normally act to speed the rate of decay of distal EPSPS, perhaps by decreasing the local
membrane time constant. On blockade of Ih channels, there is a preferential slowing of the decay
phase of the distal EPSP, which increases temporal summation.
101
 2000 Nature America Inc. http://neurosci.nature.com

news and views

wave sleep. Regulation of Ih in these cells by
cAMP is important in the sleepwake cycle2.
Ih channels are also expressed in dendrites,
where they influence the cable properties of
the dendrite and shape the time course of
the EPSP as it is propagated to the soma9
(Fig. 1d).
In their new study, Beaumont and
Zucker3 extend the role of Ih in neurons by
showing that these channels can alter neu-
rotransmitter release from presynaptic ter-
minals. Recording from the motor terminals
at crayfish neuromuscular synapses, the
authors demonstrate a pronounced ADP fol-
lowing a hyperpolarizing voltage step, a hall-
mark of Ih activation. To confirm that the
ADP is indeed due to Ih, the authors show
that it is blocked by inorganic and organic
2000 Nature America Inc. http://neurosci.nature.com
ing, given that they have been detected elec-
trophysiologically in the large presynaptic
terminals of the chick giant calyx10 and by
antibody staining on the basket cell termi-
nals surrounding cerebellar Purkinje neuron
soma4 (Fig. 2a). What is new about the pre-
sent study is their proposed function and the
unusual mechanism by which these chan-
nels regulate synaptic transmission.
Previous studies on the crayfish neuro-
muscular junction have shown that the
neurotransmitter serotonin (5-HT), acting
through both cAMP and phosphatidyl
inositol second-messenger pathways, caus-
es a facilitation of glutamate release from
excitatory presynaptic terminals11. Beau-
mont and Zucker now show that the effect
of 5-HT depends, at least in part, on the
(Fig. 2b). However, a previous study from
the Zucker lab using Ca2+ indicator dyes
failed to detect any increases in Ca2+ levels
in response to 5-HT at crayfish presynap-
tic terminals13. Instead, measurements of
synaptic vesicle cycling with the membrane
dye FM1-43 from the same group14 showed
that 5-HT increases the pool of readily
releasable vesicles.
To explain their findings, Beaumont and
Zucker suggest a provocative hypothesis in
which opening of Ih channels enhances the
mobilization of synaptic vesicles to a readi-
ly releasable pool (Fig. 2). The authors sug-
gest that this effect could be due to a direct
interaction of Ih channels with the release
machinery, perhaps mediated by the
cytoskeleton. This idea is not without prece-

blockers of Ih. The presence of Ih channels at direct activation of presynaptic Ih by cAMP.
these terminals is not particularly surpris- Application of 5-HT, or direct elevation of
cAMP by forskolin, depolar-
izes the presynaptic terminal
by 510 mV and enhances the
magnitude of the postsynap-
tic response by potentiating
transmitter release. Although
cAMP often works by activat-
ing protein kinase A (PKA),
its effect at these terminals
could not be attributed to
PKA, suggesting that cAMP
may instead activate some
other target protein by a direct
mechanism. Because Ih is acti-
vated directly by cAMP, it was
a plausible candidate, and the
authors confirmed its role by
showing that pharmacologi-
cal agents that inhibit Ih par-
tially block the ability of 5-HT
or forskolin to depolarize the
presynaptic terminal and
enhance transmitter release.
How might activation of Ih
facilitate release? One possi-
bility is that it could enhance
Ca2+ influx into the presynap-
tic terminal, for example by
causing depolarization and
Bob Crimi
thus opening voltage-gated
2+
Fig. 2. Models for presynaptic facilitation (a, b) Presynaptic Ca channels or by increasing
the rate of presynaptic action
terminal without (a) and with (b) 5-HT, showing different
modes of regulation of transmitter release during cAMP- potential firing. There is a
dependent presynaptic facilitation. Closure of K+ channels in precedent for this in the mol-
Aplysia neurons, via phosphorylation by protein kinase A, lusk Aplysia, where 5-HT (act-
broadens the action potential and increases Ca2+ influx ing through PKA- and
through voltage-gated calcium channels, which enhances PKC-dependent pathways)
release. There is also a more direct effect of protein kinase A enhances transmitter release
phosphorylation on the release machinery. At the crayfish
at a sensorimotor synapse by
neuromuscular junction presynaptic terminals, 5-HT acts in a +
Ca -independent manner to enhance the pool of readily closing presynaptic K chan-
2+
releasable synaptic vesicles. Modulation of Ih contributes to nels, thereby prolonging the
this effect by an unknown mechanism that does not involve presynaptic action potential
enhanced Ca2+ influx. and enhancing Ca2+ influx12
dent, in that voltage-gated calcium channels
interact directly with the presynaptic
SNARE protein syntaxin, regulating both
channel function and vesicle fusion15. Alter-
natively, increased influx of Na+ through
activated Ih channels could influence some
local signaling cascade. Whatever the mech-
anism, it seems to involve the stable activa-
tion of some downstream signal; the authors
show that direct activation of Ih (by a one-
minute hyperpolarization of the presynap-
tic axon) is sufficient to induce an
enhancement of transmitter release that per-
sists for as long as 20 minutes. Although fur-
ther work is needed to pin down this
mechanism, the new findings clearly point
to an unforeseen role for Ih in the control of
neuronal function.
1. DiFrancesco, D. Annu. Rev. Physiol. 55, 455472
(1993).
2. Pape, H. C. Annu. Rev. Physiol. 58, 299327 (1996).
3. Beaumont, V. & Zucker, R. S. Nat. Neurosci. 3,
133141 (2000).
4. Santoro, B., Grant, S. G., Bartsch, D. & Kandel,
E. R. Proc. Natl. Acad. Sci. USA 94, 1481514820
(1997).
5. Santoro, B. et al. Cell 93, 717729 (1998).
6. Ludwig, A., Zong, X., Jeglitsch, M., Hofmann, F.
& Biel, M. Nature 393, 587591 (1998).
7. Gauss, R., Seifert, R. & Kaupp, U. B. Nature 393,
583587 (1998).
8. Santoro, B. & Tibbs, G. R. Ann. NY Acad. Sci.
868, 741764 (1999).
9. Magee, J. C. Nat. Neurosci. 2, 508514 (1999).
10. Fletcher, G. H. & Chiappinelli, V. A. Brain Res.
575, 103112 (1992).
11. Dixon, D. & Atwood, H. L. J. Neurophysiol. 62,
12511259 (1989).
12. Byrne, J. H. & Kandel, E. R. J. Neurosci. 16,
425435 (1996).
13. Delaney, K., Tank, D. W. & Zucker, R. S.
J. Neurosci. 11, 26312643 (1991).
14. Wang, C. & Zucker, R. S. Neuron 21, 155167
(1998).
15. Catterall, W. A. Ann. NY Acad. Sci. 868, 144159
(1999).

102 nature neuroscience volume 3 no 2 february 2000

 2000 Nature America Inc. http://neurosci.nature.com
Neurodegeneration in the
polyglutamine diseases:
Act 1, Scene 1
Robert Nussbaum and Georg Auburger
A mouse model of the neurodegenerative disease
spinocerebellar ataxia type 1 reveals that changes in gene
expression begin many weeks before the onset of symptoms.
The best therapy for neurodegenerative
diseases is likely to be prevention, given
the limited regenerative capacity of cen-
2000 Nature America Inc. http://neurosci.nature.com
tral nervous tissue. To design rational
approaches for the prevention of these
diseases, therefore, we need to identify
the initial pathological events and
understand how they ultimately lead to
neuronal cell death. In this issue of
Nature Neuroscience, Lin et al.1 provide
an enticing glimpse of the first cellular
changes in the hereditary neurodegen-
erative disease spinocerebellar ataxia
type 1 (SCA1). They have used a mouse
model to identify genes whose expres-
sion is altered in the earliest stages of the
disease; moreover, their preliminary
results suggest that many of these genes
show similar alterations in human SCA1
patients.
SCA1 is one of eight hereditary neu-
ropathies (the best-known being Hunt-
ingtons disease) that are caused by the
pathological expansion of a trinucleotide
repeat that encodes polyglutamine2. The
unraveling of the mechanism underly-
ing these disorders has been one of the
great success stories of the last decade.
Although each of the diseases involves a
different gene and different clinical and
neuropathological features, the under-
lying mechanism is the same in each
case; a piece of DNA consisting of CAG
repeats becomes amplified, leading to a
protein product that contains a patho-
logically expanded string of glutamine
residues. These disorders are relatively
rare compared to (say) Alzheimers or
Parkinsons disease, but their significance
goes beyond their impact on public
health. Because they arise from well-
The authors are at the National Human
Genome Research Institute, National Institutes
of Health, 49 Convent Drive, Bethesda,
Maryland 20892, USA.
email: rlnuss@nhgri.nih.gov or
auburger@nhgri.nih.gov
nature neuroscience volume 3 no 2 february 2000
defined genetic causes that lead repro-
ducibly to a common endpointdeath
of neurons by apoptosisthe polygluta-
mine disorders provide a powerful
model in which to study the pathogen-
esis of neurodegeneration.
SCA1 is characterized by the onset
(usually in adulthood) of cerebellar and
bulbar dysfunction. This is due to a
severe loss of cerebellar Purkinje cells as
well as atrophy and degeneration in var-
ious other regions of the brain and spinal
cord. The disease has been studied over
a number of years, notably in the labo-
ratories of Huda Zoghbi and Harry Orr,
who in 1993 identified ataxin-1 as the
gene carrying the pathological polyglut-
amine expansion 3. Although the wild-
type ataxin-1 protein has no known
function, the mutant form gives rise to
aggregates that accumulate in the nuclei
of affected neurons. Perhaps surprising-
ly, ataxin-1 expression is not confined to
the cells that die in the disease, but is
ubiquitous in both the nervous system
and other tissues. Thus, the expression
pattern alone cannot explain the charac-
teristic neuronal degeneration observed
in SCA1. Zoghbi, Orr and colleagues
went on to develop transgenic mouse
models for SCA1, using a Purkinje-cell
specific promoter to drive the expression
of various forms of mutant ataxin-1, and
created mice with cerebellar degenera-
tion similar to that found in human
patients. By expressing mutant ataxin-1
that lacked either a nuclear localization
signal or an aggregation domain, the
authors were able to show that nuclear
localization of abnormal ataxin-1, and
not aggregation per se, was sufficient to
cause the phenotype4.
In the new paper 1 by Lin et al., the
same groups continue their analysis of
transgenic mouse models for SCA1 to
identify the earliest changes in gene
expression in the cerebellum. By using a
PCR-based subtractive hybridization
news and views
approach, they were able to identify tran-
scripts that are either upregulated or
downregulated in the mutant mice. The
authors identified seven such genes, and
for each one they determined when in
development expression first begins to
differ between mutant and wild-type
mice. Surprisingly, one transcript was
found to be downregulated on postnatal
day 11, only one day after the Purkinje-
cell specific promoter begins to drive
expression of the ataxin-1 transgene, and
at least 56 weeks before the first mani-
festations of the disease. This transcript
encodes prenylcysteine carboxylmethyl-
transferase (PCCMT), an enzyme
enriched in the endoplasmic reticulum
(ER) of the cerebellum, which partici-
pates in post-translational lipid modifi-
cation of many proteins, including the
G protein RAS. Later on, but still 23
weeks before the onset of clinical signs
or Purkinje cell pathology, transcripts
encoding proteins with effects on cellu-
lar calcium and glutamate metabolism
were also downregulated. These includ-
ed the type I ER inositol triphosphate
receptor IP3R1, inositol polyphosphate
5-phosphatase (INPP5A), an ER calcium
pump (SERCA2), the calcium ion chan-
nel TRP3 and the glutamate transporter
EAAT4, which is located in dendritic
spines of Purkinje cells. Only later, when
pathological and clinical signs first start-
ed to appear, did another gene begin to
be upregulated; that gene was the early-
stage acute phase protein and AD aggre-
gation marker alpha1-antichymotrypsin.
Importantly, these changes (apart from
the upregulation of alpha1-antichy-
motrypsin) were only seen in mice
expressing the expanded ataxin-1 in the
nucleus; no change was seen in control
mice expressing either wild-type ataxin-1
or an expanded ataxin-1 that lacked a
nuclear localization domain and was
confined to the cytoplasm.
What about human SCA1 patients?
Although it is difficult to obtain materi-
al for RNA studies, the authors were able
to confirm by immunohistochemistry
that three of the gene products down-
regulated in mutant mice (PCCMT,
SERCA2 and IP3R1) also showed
decreased expression in the brain of an
early-onset SCA1 patient, relative to age-
matched controls. Alpha1-antichy-
motrypsin, which is upregulated in the
mutant mice, was also upregulated in the
human patient. Consistent with the con-
clusions of Lin et al.1, unpublished work
from one of our laboratories (G.A.) has
indicated that a number of transcripts
103
 2000 Nature America Inc. http://neurosci.nature.com
news and views
ronal death later in
the disease process,
and if so, by what
mechanism? Here we
can only speculate.
PCCMT is involved
in the post-transla-
tional modification
of many proteins,
any one of which
might be involved in
neuronal degenera-
tion. The other
downregulated genes
are involved in
regulating cytoplas-
mic calcium and glu-
tamate neurotrans
2000 Nature America Inc. http://neurosci.nature.com
mission; IP3R1 stim-
ulates intracellular
Ca2+ release, INPP5A
may act to terminate
the biological activity
of IP3, SERCA2 and
TRP3 are both
involved in calcium
flux across mem-
branes, and EAAT4 is
involved in clearing
excess glutamate
from the extracellular
space. One plausible
hypothesis is that
downregulation of
Fig. 1. Pre- and postsynaptic neurons, with intervening synaptic cleft. these genes might
Shown are some of the molecules implicated by transcript profiling in promote excitotoxici-
the earliest stages following expression of pathologically expanded ty, a well-character-
ataxin-1 (courtesy of D. Leja). ized phenomenon in
which neurons die by
apoptosis due to over-
excitation of gluta-
involved in regulating calcium, inositol mate receptors and a consequent increase
and glutamate neurotransmission show in cytoplasmic calcium5. Consistent with
altered expression in the striatum of this idea, many of the neurons affected
presymptomatic and early-stage Hunt- in the different polyglutamine disorders
ingtons disease patients. show strong staining with various calci-
The findings of Lin et al.1 raise two um-regulating proteins and receive glu-
fundamental questions that must be tamatergic synaptic input 6. Moreover,
answered before the pathogenesis of excessive activation of NMDA-type glu-
SCA1 can be understood. First, we want tamate receptors can produce a clinical
to know how the mutant ataxin leads to and neuropathological pattern reminis-
altered gene expression. The possibili- cent of Huntingtons disease in animals7,
ties include a direct effect on the tran- and overstimulation of AMPA-type glu-
scription of specific genes, a more global tamate receptors can trigger the selective
effect on chromatin structure, nuclear degeneration of Purkinje neurons of the
architecture or transcription factor sta- cerebellum. Clearly these issues cannot
bility, or a change in the stability of par- be resolved simply by looking at tran-
ticular RNA trancripts. With the scriptional profiles, but the availability
availability of a mouse model for the of mouse models should allow more
disease, this is likely to become an area direct genetic and physiological tests of
of intensive inquiry. the excitotoxicity hypothesis.
Second, are these early changes in As we noted at the outset of this arti-
gene expression the actual cause of neu- cle, the significance of these findings
104 nature neuroscience volume 3 no 2 february 2000
extends beyond the polyglutamine dis-
orders. Various abnormal cellular
processes have been proposed as the first
steps in the eventual neuronal loss
observed in all of the major neurode-
generative diseases, including both
Alzheimers and Parkinsons disease.
Among the leading candidates are pro-
tein aggregation, damage from oxygen
free radicals (oxidative stress)8 and exci-
totoxicity, any of which, alone or in
combination, could result in the induc-
tion of apoptosis. Which of these are late
phenomena, and which are important at
the earliest stages of disease? In princi-
ple, these questions can be addressed in
carefully staged human brain tissue, but
in practice such material is not readily
available, meaning that mutant mice of
the type used by Lin et al. are likely to be
central in future advances.
Although the implications of this
study must still be considered tentative,
it is striking that so many of the early
changes seen by Lin et al. are consistent
with an effect on excitotoxicity, whereas
none seems to implicate oxidative stress
as an early event in the SCA1 mouse
model. The notion that dysregulation of
glutamate and calcium signaling is a very
early event in the progression of neu-
rodegeneration will, if substantiated by
future work, stimulate the design of
therapeutic approaches that may even-
tually make a difference to the clinical
management of many neurodegenera-
tive conditions. Indeed, a number of
pharmacological compounds that influ-
ence calcium and glutamate homeosta-
sis are already available, and one such
compound, riluzole, has proven to be
the most effective therapy so far in
patients with amyotrophic lateral scle-
rosis, a neurodegenerative disease of the
spinal cord with an entirely different eti-
ology than the polyglutamine disorders9.
1. Lin X., Antalffy, B., Kang, D., Orr, H. T. &
Zoghbi, H. Y. Nat. Neurosci. 3, 157163
(2000).
2. Lunkes, A., Trottier, Y. & Mandel, J. L. Essays
Biochem. 33, 149163 (1999).
3. Orr, H. T. et al. Nat. Genet. 4, 221226
(1993).
4. Klement, I. A. et al. Cell 95, 4153 (1998).
5. Schwarcz, R., Foster, A. C., French, E. D.,
Whetsell, W. O. Jr. & Kohler, C. Life Sci. 35,
1932 (1984).
6. Greenamyre, J. T. Arch. Neurol. 43,
10581063 (1986).
7. diFiglia, M. Trends Neurosci. 13, 286289
(1990).
8. Beal, M. F. Ann. Neurol. 38, 357366 (1995).
9. Doble, A. Pharmacol. Ther. 81, 163221
(1999).
 2000 Nature America Inc. http://neurosci.nature.com
Non-invasive visualization of
cortical columns by fMRI
Amiram Grinvald, Hamutal Slovin and Ivo Vanzetta
Functional magnetic resonance imaging can now resolve
individual cortical columns, which should provide insights
into sensory perception and higher cognitive functions.
Mountcastle1 and Hubel and Wiesel2 orig-
inally showed that many complex neural
networks display meticulous order. Neu-
rons with common functional properties
are often clustered together in columns,
2001000 m wide, which span the depth
2000 Nature America Inc. http://neurosci.nature.com
of the cortex. To discover principles under-
lying implementation of the neural code,
we must understand functional organiza-
tion across the sheet of cortical columns
during sensory processing and cognitive
tasks. Brain organization has been visual-
ized3 with positron-emission tomography
(PET), optical imaging of intrinsic signals
and, most recently, functional magnetic
resonance imaging (fMRI). Optical imag-
ing allows detailed visualization of corti-
cal columns (spatial resolution of 20100
m, verified by extensive single-neuron
and histological mapping), but it is inva-
sive and thus unsuitable for use in
humans. PET and fMRI offer spectacular
three-dimensional localization of activity
in humans, but have lacked the spatial res-
olution to reveal cortical columns.
These techniques depend on the pio-
neering finding of Roy and Sherrington4
that changes in electrical activity are cou-
pled to microcirculation responses, whose
regulation is still being explored. Neural
activity locally increases metabolic
demands for glucose and oxygen, and the
brain microcirculation respondsmuch
less locallyby increasing blood volume
and flow. In PET, this enhanced blood
flow is visualized by injecting radioactive
material into the circulation. Blood oxy-
genation level-dependent5 (BOLD) fMRI
instead uses concentration changes in
paramagnetic deoxygenated hemoglobin
as an intrinsic contrast agent. In optical
imaging, this contrast agent yields high-
resolution maps of cortical columns6,7,
and BOLD imaging was hoped to have
similar potential. Kim and colleagues 8
now report a striking technical advance in
The authors are in the Department of
Neurobiology, The Weizmann Institute of
Science, Rehovot 76100, Israel.
e-mail: amiram.grinvald@weizmann.ac.il
nature neuroscience volume 3 no 2 february 2000
fMRI, achieving sufficient spatial resolu-
tion to directly visualize cortical columns
responsible for shape perception in visual
cortex of anesthetized cats. Instead of the
traditional positive BOLD fMRI signal,
the authors used an earlier, negative phase
of the signal, termed the initial dip
(deoxygenation phase; Fig. 1b in ref. 8),
to obtain single-condition maps of corti-
cal columns. They further showed that the
traditional BOLD signal (positive hyper-
oxygenation phase) gives less precise res-
olution than this early phase.
Because it remains controversial exact-
ly how electrical activity is coupled to such
metabolic responses, skeptics question
whether imaging maps based on micro-
circulation responses indeed represent the
brains electrical activity or just hemody-
namics. Thus methods to distinguish
between brain and vein9 are critical. After
Kim and colleagues presented their
results 8 at the 1999 Society for Neuro-
science meeting, some felt that neurosci-
entists fantasy might soon become reality.
For others, questions remained. Exactly
how is cortical electrical activity related to
various hemodynamic responses? Why
was the initial dip not observed in most
previous studies? What are the implica-
tions for standard low-resolution fMRI at
low field strength? Can the impressive sin-
gle-condition orientation columns
obtained by fMRI in the anesthetized cat
also be visualized in awake animals? Are
these results relevant to primates?
The relationship of electrical activity
to hemodynamic responses has been
addressed by optical imaging 6,7,10,11 .
Because the spatial resolution of the optics
is less than 1 m, this method permits
unambiguous identification of blood-ves-
sel-derived artifacts, distinguishing them
from cortical activation per se. In addition,
photons are cheap, so one can collect
plenty of them; thus the signal-to-noise
ratio is higher than for fMRI. This allows
direct visualization of hemodynamic
changes without elaborate image process-
ing or statistics. Below we address these
five remaining questions by presenting
news and views
our own optical imaging data showing
ocular dominance columns in behaving
monkeys (Figs. 1 and 2) and by referring
to previous work using biophysics and
optical imaging.
We determined the activity-dependent
spatiotemporal characteristics of hemody-
namic events, including oxygen delivery,
blood flow and blood volume, at the onset
of cortical activity triggered by sensory
stimulation. The first event is a prolonged
increase in oxygen consumption, caused
by an increase in oxidative metabolism of
activated neurons, leading to an increase
in the deoxyhemoglobin concentration
starting less than 100 ms after activity
onset10,11. Upon illumination at 605 nm,
this early event causes cortical darkening
(Figs. 1b, middle panel, and 2b) known to
be colocalized with the site of electrical
activity in columns (see Fig. 4 in ref. 7).
The second event originates from a delayed
increase in blood volume, starting about
300500 ms later10, caused by capillary
dilation, which also causes cortical dark-
ening, particularly apparent in larger blood
vessels (Figs. 1c and d, middle panel, and 2e).
Blood volume changes occur almost simul-
taneously in the three vascular compart-
ments and are not well regulated within
individual cortical columns (Fig. 4 in ref. 7).
The third event is an activity-related
increase in blood flow, as blood rushes into
capillaries. Starting about 0.51.5 seconds
after the onset of electrical activity, this
decreases deoxyhemoglobin concentration
and increases oxyhemoglobin concentra-
tion. The delayed deoxyhemoglobin
decrease is much larger than the initial
increase, causing delayed cortical whiten-
ing, particularly apparent in larger blood
vessels (Fig. 1b and c, right). Unlike the
early deoxygenation phase, this delayed
increase is not localized with columnar
electrical activity. Thus, changes in deoxy-
hemoglobin concentration follow a bipha-
sic time course: a deoxyhemoglobin
increase because of increased oxidative
metabolism of active neurons the ini-
tial dip and then a decrease due to large
overcompensation by enhanced blood flow
(Figs. 1b and 2b). Obviously, increased
oxidative metabolism must localize with
electrical activity.
With high-field fMRI, Logothetis and
colleagues12 observed the initial dip in cor-
tical regions without large vessels but not
in large vessels, whereas the late BOLD
component was detectable in both com-
partments (see Fig. 5 in ref. 12). These
findings support the claims of Kim and
colleagues8 that the initial fMRI dip shows
better spatial specificity. Thus, cortical
105
 2000 Nature America Inc. http://neurosci.nature.com

news and views

spatial resolution. The point spread of the

fMRI signal is large relative to cortical
column diameter14. However, differential
imaging can resolve two distinct cortical
activation sites at distances much smaller
than half the width of the signal spread,
if the signal-to-noise ratio is adequate.
Indeed, ocular dominance columns in
humans have already been imaged with
fMRI by this method 15 (see also T.W.
James et al. Soc. Neurosci. Abstr. 25,
212.5,1999; Tanaka et al., Soc. Neurosci.
Abstr. 25, 572.3, 1999). As discussed
above, both optical imaging and BOLD
fMRI signals have multiple components.
The local component probably origi-
nates from oxygen extraction from the
capillaries; thus it is most closely coupled
2000 Nature America Inc. http://neurosci.nature.com

to electrical activity and most stimulus-

specific. This component is particularly
enhanced during the initial dip, before
delayed hemodynamic events begin to
obscure it. The global signal7,10 refers to
the sum of all the components, which
Fig. 1. Comparison between differential and single-condition maps of ocular dominance columns, originate from this deoxygenation and
visualized by optical imaging. A cranial window was implanted over the primary visual cortex of a other sources, such as nonlocalized
macaque. The exposed cortical surface of the behaving monkey was illuminated at 605 nm to blood-volume and blood-flow changes
emphasize small changes in deoxyhemoglobin concentration. Each row represents a time series, (black and white vessels, respectively, in
taken at roughly two-second intervals beginning one second before stimulus onset. (a) Differential Fig. 1ad, late frames). Most of the inte-
images, obtained by subtracting images obtained with left- versus right-eye stimulation. To show grated global signal is associated with
time development of the differential map, all images are clipped at the same range. (Clipping is these last two components, which may
determining the range and mean of the gray scale's contrast.) (b) Corresponding single-condition not be stimulus-specific as they reflect
maps obtained by normalizing the cortical image to illumination intensity without any additional
hemodynamic activation of venules and
processing, when the contralateral eye viewed the stimulus. To emphasize the slow global changes,
a fixed-range clipping with fixed mean values was used. (c) As in (b), but to emphasize the small large draining veins far from the electrical
changes, a fixed range was used without a fixed mean. (d) As for (b), but to maximize the dynamic activation site7,10. For this reason, when
range of each map, autoclipping was used. Note that the signal-to-noise ratio in a differential image two stimuli are orthogonalmeaning
is much higher than that obtained for a single-condition map. Scale bar, 1 mm. that they activate neurons in comple-
mentary cortical patchescomparing
their activation patterns enhances the sig-
electrical activity is most tightly coupled evidence suggest that the amplitude of the nals local component, while eliminating
to hemodynamic events during the initial initial dip depends on the second power many, though not all, global components.
dip. Because the initial dip is localized of the field strength (K. Ugurbil, personal Therefore, when the assumptions dis-
with neuronal activity, whereas the sub- communication). The existence of the ini- cussed below are fully justified, differen-
sequent larger decrease in deoxyhemo- tial dip has been confirmed by several tial imaging may reveal the correct pattern
globin is not, there is a time window groups using 4 T fMRI in humans13. For of electrical activation. Such maps may
during the first three or four seconds in those debating whether to purchase ultra- even yield a better signal-to-noise ratio
which functional maps indeed correspond expensive high-field magnets, it will be than maps derived from the shorter peri-
to electrical activity. In contrast, delayed critical to establish the relative amplitude od and smaller amplitude of the initial
microvascular events such as volume and and signal-to-noise ratio of the initial dip dip. However, differential imaging has two
flow changes, which are not localized with as a function of field strength. major limitations. First, it works perfect-
electrical activity, may yield hemody- Kim and colleagues reported single- ly only if stimuli are orthogonal in the
namic maps that do not correspond to condition maps (Fig. 2ad in ref. 8). To above sense, such as orthogonal oriented
electrical activation (compare Fig. 2g and understand the implications of their stimuli activating orientation columns in
h with Figs. 3a and b and 4c in ref. 8). advance for standard low-field fMRI cats and monkeys. Even for ocular domi-
Why have low-field (0.51.5 T) BOLD research (the third question), we must nance maps, the stimuli are not strictly
fMRI measurements failed to detect the clarify the advantages and limitations of orthogonal because contralateral activa-
initial dip observed by Kim and colleagues single-condition mapping compared tion is stronger than ipsilateral activation,
with high-field (4.7 T and 9.4 T) magnets with differential mapping (for details, so that blood vessel-derived artifacts may
and by optical imaging? One possibility is see ref. 3, pages 918921). The spatial res- be the largest signals in differential maps
that low-field measurements are not suf- olution of a functional map depends on computed from late hemodynamic
ficiently sensitive to deoxyhemoglobin the spread of a particular imaged signal responses (Fig. 2c and compare early vas-
concentration changes in cortical capil- beyond the site of electrical activity, sig- cular artifacts in Fig. 2f with late ones in
laries. Both theoretical and experimental nal-to-noise ratio and the instruments Fig. 2i). More generally, functional map

106 nature neuroscience volume 3 no 2 february 2000

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

news and views

Finally, skeptics tional technical improvements are still

wondered whether sin- required before high-resolution single-
gle-condition orienta- condition mapping becomes a gold stan-
tion columns obtained dard procedure easily achieved by fMRI.
in anesthetized cats 8 Such improvements might include exten-
can be visualized in sive averaging, use of different MRI
awake animals, and sequences, improved ways to reject vas-
whether these results cular artifacts or reliance on MRI signals
are relevant to pri- other than BOLD.
mates. Our new results The advance described by Kim and
from awake monkey colleagues8 represents a major step for-
cortex (Figs. 1 and 2), ward in functional brain imaging. It
which agree with those should lay the foundation for exploration
of Kim and col- of the human brain at the fundamental
leagues8, address both level of its columnar architecture. This
these concerns. Hence approach may revolutionize fMRI studies
it seems reasonable to by allowing them to address not only
hope that human where but also how processing and com-
2000 Nature America Inc. http://neurosci.nature.com

brain imaging with putations are carried out by individual

fMRI will also allow cortical columns. The combination of
Fig. 2. Comparison between single-condition and differential maps high-resolution single- fMRI with techniques that reveal cortical
obtained during the initial dip (deoxygenation phase) and later rise condition mapping. dynamics in the millisecond time domain,
(hyperoxygenation phase). (a) Cortical image taken in green light to Kim and colleagues such as electroencephalography and mag-
emphasize the vascular pattern. Scale bar, 1 mm. (b) Time course of accomplished the dif- netoencephalography, is also warranted.
the amplitude of the global signal obtained from the single-condition ficult feat of single- In conjunction with such methods, fMRI
maps. (c) Time course of vascular artifacts and of the mapping signal
condition mapping by will become a powerful non-invasive tool
amplitude calculated from the differential map. (df) Maps from the
early deoxygenation phase. (d) Single-condition map for the con- integrating BOLD sig- both in biomedicine and in basic research
tralateral eye. (e) As in (d), for the ipsilateral eye. (f) The differential nals mainly during the of neurophysiology underlying higher
map. (gi) As for (df), from the late period. Note that during this small initial dip. As brain functions, thus bridging the gap
late time, the most prominent signals in the rather poor single-condi- with any new between psychology and neurobiology.
tion maps (g, h) are artifacts from blood vessels (diameter 50150 approach, additional
m). Although these brain or vein artifacts are minimized by differ- research will undoubt- 1. Mountcastle, V. B. J. Neurophysiol. 20, 408434
(1957).
ential mapping (f), at least with optical imaging, they are larger than edly strengthen this
the columnar signal. 2. Hubel, D. H. & Wiesel, T. N. J. Physiol. (Lond.)
works foundations. 160, 106154 (1962).
First, the authors did 3. Windhorst, U. & Johansson, H. (eds) Modern
not present unequivo- Techniques in Neuroscience Research (Springer,
amplitudes can be three- to tenfold small- cal single-neuron confirmation, or con- New York, 1999).
er than ocular dominance column maps firmation by two-deoxyglucose 4. Roy, C. & Sherrington, C. J. Physiol (Lond.). 11,
shown here, making vessel-derived arti- autoradiography or optical imaging, 85-108 (1890).
facts far more problematic. Furthermore, which would be even better. However, the 5. Ogawa, S., Lee, T. M., Kay, A. R. & Tank, D. W.
Proc. Natl. Acad. Sci. USA 87, 98689872
orthogonal stimuli may be impossible to complementary nature of the single-con- (1990).
design, particularly for columns repre- dition maps for the two sets of orthogo-
6. Grinvald, A., Lieke, E., Frostig, R. D, Gilbert, C.
senting higher cognitive functions. Sec- nal orientations strongly suggests that D. & Wiesel, T. N. Nature 324, 361364
ond, by far the largest signals are from their maps were indeed orientation maps. (1986).
draining veins, even those smaller than Second, their choice of threshold, a criti- 7. Frostig R. D., Lieke, E., Tso, D. Y. & Grinvald, A.
100 m (cortical vasculature image, cal step in obtaining single-condition Proc. Natl. Acad. Sci. USA 87, 60826086
(1990).
Fig. 2a; vascular artifacts Fig. 2c and gi). maps, was arbitrary. A cocktail blank
Such large signals may spread 37 mm (average response to four stimuli, a rough 8. Kim, D.-S., Duong, T. Q. & Kim, S.-G. Nat.
Neurosci. 3, 164169 (2000).
from the sites of electrical activity (A. approximation for uniform cortical acti-
9. Frahm, J. Merboldt, K. D., Hanike, W.,
Shmuel et al., ISMRM abstract, in press; vation elicited by many stimuli) was used Kleinschmisdt, A. & Boeker, H. NMR Biomed.
D. Shoham and A.G., unpublished instead of a real blank (no stimulus). 7, 4553 (1994).
results). Most important, signals from Third, one would expect a small negative 10. Malonek, D. & Grinvald, A. Science 272,
such draining veins may be stimulus-spe- BOLD signal also for the orthogonal ori- 551554 (1996).
cific and highly reproducible. Therefore, entation, rather than a flat line or a small 11. Vanzetta, I. & Grinvald, A. Science 286,
use of differential imaging, high thresh- positive signal (Fig. 2e and f in ref. 8). 15551558 (1999).
olds, signal averaging and statistics does This last reservation may suggest that cur- 12. Logothetis, N. K., Guggenberger, H., Peled, S.
not guarantee removal of such artifacts. rent signal-to-noise ratio is not good & Pauls, J. Nat. Neurosci. 2, 555562 (1999).
Single-condition mapping, in which the enough yet. Such single-condition map- 13. Hu, X., Le, T. H & Ugurbil, K. Magn. Reson.
Med. 37, 877884 (1997).
response to a single stimulus is imaged ping is difficult even with optical imag-
directly, thus has great advantages over ing. (Cocktail blanks are commonly used 14. Engel, S. A., Glover, G. H. & Wandell, B. A.
Cereb. Cortex 7, 181192 (1997).
differential imaging and should be the in optical imaging studies as well 3 .)
15. Menon, R. S., Ogawa, S., Strupp, J. P. &
ultimate goal of high-resolution func- Therefore, future work should emphasize Ugurbil, K. J. Neurophysiol. 77, 27802787
tional brain imaging. increasing signal-to-noise ratio, and addi- (1997).

nature neuroscience volume 3 no 2 february 2000 107

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

brief communications

Rapid, synaptically driven
increases in the intrinsic
excitability of cerebellar
deep nuclear neurons
Carlos D. Aizenman and David J. Linden
Department of Neuroscience, Johns Hopkins University School of Medicine,
725 N. Wolfe Street, Baltimore, Maryland 21205, USA
Correspondence should be addressed to D.L. (dlinden@jhmi.edu)
The neurons of the cerebellar deep nuclei are implicated in certain
forms of motor learning such as associative eyeblink condition-
ing, partly because increases in their firing rates parallel acquisi-
tion of the conditioned response. Here we demonstrate that these
2000 Nature America Inc. http://neurosci.nature.com
provide a flexible and informationally rich engram for cerebel-
lar motor learning.
Long-term synaptic potentiation and depression (LTP and
LTD) are considered computationally powerful, in part because
they allow for the synapse-specific modification of a large array
of inputs. Less attention has been paid to the idea that information
storage may involve activity-dependent changes in the intrinsic
excitability of neurons. This type of plasticity would affect post-
synaptic signals evoked by many (or possibly all) synapses imping-
ing on a neuron, depending on the identity and location of the
intrinsic conductances altered, thereby creating a more global
change in signal integration. Slow, activity-dependent changes in
the intrinsic excitability of both invertebrate and mammalian neu-
rons have been reported in cell culture systems1,2. Persistent
changes in intrinsic neuronal excitability can also be induced
somewhat more rapidly. In the mollusk Hermissenda, associative
pairing of light and rotation produces a long-term increase in the
intrinsic excitability of type B photoreceptors due to a reduction

neurons can show persistent increases in their intrinsic excitabil- of K+ currents3. A similar effect is observed in CA1 pyramidal
ity following a Ca 2+ load imposed by synaptic activation of neurons of the rabbit hippocampus following trace eyeblink con-
NMDA receptors or direct current injection. This phenomenon, ditioning4. In hippocampal LTP, increases in the probability of a
together with use-dependent alterations in synaptic strength, may cell firing an action potential for a given size EPSP accompanies

a b
Fig. 1. Tetanization of excitatory synapses
produces a sustained increase in intrinsic
excitability of DCN neurons (a) Averaged
time course of the number of spikes evoked
during 200 ms, depolarizing current test
pulses. Control experiments (open circles)
represent the baseline. Following a synaptic-
conditioning tetanus (at arrowheads), num-
ber of spikes evoked by the test stimulus
slowly increases (filled circles). (b, c)
Representative traces evoked by a +0.5 nA
test pulse in tetanized and control experi-
ments at the time points indicated by aster- d
isks. Traces evoked by subthreshold test
pulses (0.1 and +0.1 nA; b) demonstrate
that Rinput remains stable when the spiking
evoked by a larger depolarizing test pulse
increases. Examples are shown from two c
different cells, one that fired in regular spik-
ing mode (top traces) and one that fired in
burst mode (bottom traces). (d) A cumula-
tive probability distribution shows the
resulting increase in the number of spikes
measured at 2025 min for tetanized and e
control groups shown in (a). (e) The num-
ber of spikes evoked by depolarizing test
pulses is plotted as a function of the ampli-
tude of injected current for four separate
experiments. Solid lines represent the func-
tion immediately before and dashed lines 20
min after the synaptic tetanus. Error bars for
(a) and (e) represent s.e. Experiments used
coronal slices of 1315 day-old rat cerebel-
lum as described9. Slices recorded using
microelectrodes (80150 M) containing 3
M potassium acetate were maintained at
33C in an interface chamber and perfused
with ACSF containing 126 mM NaCl, 5 mM
KCl, 3 mM CaCl2, 1 mM MgSO4, 26 mM
NaHCO3, 1.25 mM NaH2PO4, 20 mM
D-glucose and 0.2 mM picrotoxin and bub-
bled with 95% O2/5% CO2.

nature neuroscience volume 3 no 2 february 2000 109

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

brief communications

a tion storage for associative eye-blink condi-

tioning 8. Microelectrode recordings were
made from neurons in the DCN of juvenile
rat cerebellar slices as described9,10. Picrotox-
in (200 M) was added to block GABA A
receptors. To record changes in intrinsic
excitability, the cell was tonically hyperpolar-
ized to suppress spontaneous spiking, and a
b short depolarizing test pulse (250 ms, 0.10.5
nA), which evoked a small number of action
potentials (typically 14), was then used as a
baseline measure. Repeated application of this
test pulse alone did not alter the number of
evoked spikes, as stable responses could be
recorded for more than > 25 min (Fig. 1a and
c). After a baseline period, an excitatory
synaptic tetanus was applied to the DCN neu-
c ron via a stimulating electrode placed in the
2000 Nature America Inc. http://neurosci.nature.com

adjacent white matter. The tetanus (10 high-

frequency bursts applied at 4 Hz; each burst
comprising 10 pulses delivered at 100 Hz) was
repeated 5 times. In 8 of 9 cells, the number
of spikes evoked by the test pulse increased in
response to this synaptic-conditioning tetanus,
stabilizing in 1520 minutes (Fig. 1a and b).
The number of spikes evoked by the test pulse
d was significantly increased by the tetanus
(n = 9; before, 4 1; increase, 8 2; p < 0.01,
Students t-test; Fig. 1d).When Rinput was mea-
sured using a hyperpolarizing current pulse,
no significant change was observed 20 min
after the tetanus (114 10% of baseline,
n = 4). Furthermore, synaptic tetanization did
not produce a significant change in the ampli-
tude of the first spike evoked by the depolar-
izing test pulse (98 2% of baseline, n = 9).
Experiments were excluded if the amplitude
of the first evoked spike decreased by >10%
Fig. 2. Sustained increases in intrinsic excitability of DCN cells require Ca2+ influx. (a) Averaged at any point in the recording, if the cell exhib-
time course and sample traces from experiments in which 50 M D-AP5 was applied before ited spontaneous spiking between test stim-
synaptic tetanization. (b) Sustained increases in spike firing could also be induced by applying an uli or if the recording could not be
intracellular conditioning tetanus consisting of depolarizing current pulses (indicated by arrow-
maintained for 20 min following the condi-
head; see text). (c) Bath-applied 200 M Cd2+ blocked increases in spiking induced by the same
intracellular conditioning stimulation given in (b). (d) Changes in the late/peak EPSP amplitude
tioning stimulus. Inputoutput curves were
ratio, measured 1520 minutes after synaptic tetanization under different experimental condi- then plotted relating injected depolarizing
tions. There was no significant change in the amplitude of the EPSP peak (left). Arrows indicate current to the number of evoked spikes
EPSP waveforms sampled 20 minutes post-tetanus. immediately before or 20 minutes after the
synaptic conditioning tetanus (Fig. 1e).
Increased excitability was clearly associated
with reduced spike threshold in three of four
increases in synaptic strength. This phenomenon has been called cells. Additionally, in 2/4 cells, maximum spiking rate evoked by
EPSP-to-spike (ES) potentiation. However, as ES potentiation the largest test pulses increased. These results indicate that exci-
is generally occluded by GABAA receptor antagonists, it is pro- tatory synaptic tetanization produced a change in intrinsic post-
posed not to reflect changes in intrinsic postsynaptic conduc- synaptic excitability that did not merely reflect alteration in the
tances, but rather to represent a decrease in the ratio of microelectrode or the general health of the impaled cell.
feedforward inhibition to excitation yielding a net increase in We wished to determine the initial signals required for this
excitability5. However, based on indirect measures using field change in intrinsic excitability. The synaptic conditioning tetanus
potentials6 or intracellular recording with direct current injection evoked spiking (66 12 spikes per burst, n = 9, range, 17130
(but without blocking GABA receptors, allowing confounding spikes per burst), but the number of spikes evoked during the
changes in tonic inhibitory drive)7, others claim that ES poten- synaptic conditioning tetanus was uncorrelated with resulting
tiation reflects an increase in intrinsic excitability. changes in excitability (increase in number of spikes evoked by
The neurons of the deep cerebellar nuclei (DCN) integrate a test pulse; r = 0.34, p > 0.10). To further characterize the
excitatory and inhibitory inputs representing several streams of requirements for this effect, synaptic conditioning tetani were
sensory/motor information, and are a proposed site of informa- applied during bath application of 50 M D-AP5, an NMDA

110 nature neuroscience volume 3 no 2 february 2000

 2000 Nature America Inc. http://neurosci.nature.com
receptor antagonist. The number of spikes induced during the
synaptic conditioning tetanus in D-AP5 was adjusted so that it
was not significantly different from that without drug (p = 0.498,
Kolmogorov-Smirnov). With D-AP5, the conditioning tetanus
failed to produce an increase in the number of spikes evoked by
the test pulse (0.4 1 spikes, p = 0.83, Students t-test, n = 5;
Fig. 2a), indicating that Ca2+ influx through NMDARs is required.
However, this requirement for NMDAR activation does not pre-
clude contribution of other receptors such as mGluRs or AChRs
to this effect.
Depolarizing pulses that evoke substantial spiking in DCN
neurons cause large intracellular Ca2+ transients in both the soma
and dendritic arbor9,12, sufficient to evoke LTP of IPSPs9. We then
tested whether depolarizing current pulses alone could affect the
excitability of DCN neurons without synaptic stimulation. An
intracellular conditioning stimulus (pulse amplitude, +0.5 nA;
duration, 100 ms; applied at 4 Hz and delivered 5 times through
the recording electrode) caused a marked, gradual increase in the
2000 Nature America Inc. http://neurosci.nature.com
number of spikes evoked by the test pulse (increase of 6 2
spikes; p = 0.016, Students t-test, n = 7, Fig. 2b). When influx of
Ca2+ was prevented by 200 M Cd2+, spiking did not significant-
ly increase (1 0.3; p = 0.097, Students t-test; n = 4, Fig. 2c).
Alterations in dendritic or somatic voltage-gated channels
could alter the kinetics of EPSPs. To test this, we delivered a
synaptic conditioning tetanus after collecting a 10-min baseline
of EPSPs. The synaptic tetanus caused an inconsistent change in
the peak EPSP amplitude ranging from LTP to LTD with many
cases of no change at all. To analyze EPSP kinetics, amplitude
measurements were taken at both the peak and a time at which
the averaged baseline EPSP waveform had decayed to 50% of its
peak amplitude. These measurements were then normalized to
the peak amplitude of each EPSP and analyzed as the ratio of
late/peak EPSP amplitude. Synaptic conditioning significantly
increased this ratio (121 8% of baseline, p = 0.026 Students t-
test, n = 13) even when there was no change in the peak ampli-
tude, indicating that conditioning caused the EPSP to decay more
slowly (Fig. 2d). As in experiments in Fig. 1b and c, Rinput was
unaltered by the synaptic tetanus (106 3% of baseline, n = 13,
Fig. 2d). EPSP broadening could be partially suppressed by
50 M D-AP5 (109 2, p = 0.02 Students t-test, n = 4). As a test
of whether this broadening could result from increases in the
proportion of slower NMDA to faster AMPA currents underly-
ing the EPSP, the experiment was repeated with an AMPA recep-
tor antagonist, NBQX (10 M). A substantial EPSP could be
evoked in the presence of NBQX even when the cell was held at
negative membrane potentials (below 80 mV), probably because
of Mg2+-insensitive NR2D receptor subunits found in the DCN11.
This NMDAR-mediated EPSP also showed a broadening of the
EPSP waveform (116 7%, p = 0.03 Students t-test, n = 12),
suggesting that the observed change in EPSP decay kinetics
involved either a change in intrinsic postsynaptic conductances,
as observed in hippocampal CA1 pyramidal neurons13, or, alter-
natively, a selective change in the NMDAR kinetics.
We noted a high degree of heterogeneity in the firing modes of
the cells sampled for these experiments, which ranged in a con-
tinuum from regular spiking (Fig. 1b, top traces) to burst firing
(Fig. 1b, bottom traces). This heterogeneity may reflect different
modulatory states of the DCN neurons10. We tested whether the
propensity of the cell to burst correlated with the amount of
change in intrinsic excitability resulting from synaptic or depo-
larization conditioning by relating the averaged first inter-spike
interval during a test pulse within the baseline period (cells with
a greater propensity to burst have a shorter first interspike inter-
nature neuroscience volume 3 no 2 february 2000
brief communications
val) with the resulting increase in the number of spikes. We saw
no correlation between these two measures (r = 0.1, p > 0.1),
indicating that the cells firing mode did not correlate with its
ability to undergo long-term changes in intrinsic excitability.
Moreover, we did not observe a strong tendency for cells to switch
firing mode following the conditioning.
Our main finding is that either synaptically driven activation
of NMDA receptors or Ca2+ influx driven by direct depolarization
can trigger a rapid and persistent increase in the intrinsic excitabil-
ity of DCN neurons. This increase took the form of a reduction
in spike threshold and/or an increase in the maximum firing rate.
At present, we have no direct evidence to implicate any particular
conductance in this effect. However, the excitability of these cells is
strongly modulated by a hyperpolarization-activated cation current
(Ih), a low threshold voltage-sensitive Ca2+ current (ICa(T))and an
apamin-sensitive Ca2+-gated K+ current10. As the increase in intrin-
sic excitability seemed to require a Ca2+ transient, it is possible that
the relevant ion channels were altered by a Ca2+-dependent sig-
naling cascade such as PKC or Ca2+-sensitive adenylyl cyclase/PKA.
Activation of these signaling systems produces transient effects on
intrinsic excitability in hippocampal neurons, at least in part,
through actions on Ca2+-gated K+ channels14,15.
How might alterations of intrinsic excitability of DCN neu-
rons contribute to information storage in the cerebellum?
Although not directly tested, the Ca2+ transient produced by a
burst-and-pause stimulus delivered to the GABAergic Purkinje
cellDCN synapses9 probably would also be a sufficient signal to
induce this change. This could provide a framework for het-
erosynaptic interaction between excitatory and inhibitory inputs
to the DCN. In addition, a striking observation regarding asso-
ciative eyeblink conditioning is that extracellular records of spike
firing in the DCN show patterns of increased activity that cor-
relate with the development of the conditioned response8. The
most common explanation for this observation is that these
increases reflect decreased inhibitory drive from Purkinje cells
and/or increased excitatory drive from mossy fibers. Our results
suggest that increases in intrinsic excitability could potentially
complement input-specific alterations in synaptic strength to
give rise to a flexible and informationally rich engram.
ACKNOWLEDGEMENTS
We thank C. Hansel, A. Kirkwood, H.-K. Lee and S. Morris for suggestions. This
work was supported by MH01590 and the Develbiss Fund (D.J.L.) and a
fellowship from the Howard Hughes Medical Institute (C.D.A.).
RECEIVED 25 OCTOBER; ACCEPTED 16 DECEMBER 1999
1. Turrigiano, G., Abbot, L. F. & Marder, E. Science 264, 974977 (1994).
2. Desai, N. S., Rutherford, L. C. & Turrigiano, G. Nat. Neurosci. 2, 515520
(1999).
3. Alkon, D. L. J. Exp. Biol. 112, 95112 (1984).
4. de Jonge, M. C., Black, J., Deyo, R. A. & Disterhoft, J. F. Exp. Brain Res. 80,
456462 (1990).
5. Abraham, W. C., Gustafsson, B. & Wigstrm, H. J. Physiol. (Lond.) 394,
367380 (1987).
6. Jester, J. M, Campbell, L. W. & Sejnowski, T. J. J. Physiol. (Lond.) 484,
689705 (1995).
7. Pugliese, A. M., Ballerini, L., Passani, M. B. & Corradetti, R. Neuroscience 62,
10211032 (1994).
8. Raymond J. L., Lisberger S. G. & Mauk M. D. Science 272,11261131 (1996).
9. Aizenman, C. D., Manis, P. B. & Linden, D. J. Neuron 21, 827835 (1998).
10. Aizenman, C. D. & Linden, D. J. J. Neurophysiol. 82, 16971709 (1999).
11. Cull-Candy, S. G. et al. Neuropharmacology 37, 13691380 (1998).
12. Muri, R. & Knpfel, T. J. Neurophysiol. 71, 420428 (1994).
13. Hess, G. F. & Gustafsson, B. Neuroscience 37, 6169 (1990).
14. Baraban, J. M., Snyder, S. H. & Alger, B. E. Proc. Natl. Acad. Sci. USA 82,
25382542 (1985).
15. Madison, D. V. & Nicoll, R. A. J. Physiol. (Lond.) 372, 221244 (1986).
111
 2000 Nature America Inc. http://neurosci.nature.com

articles

A metabotropic glutamate
receptor variant functions as a
taste receptor
Nirupa Chaudhari1, Ana Marie Landin and Stephen D. Roper1

Dept. of Physiology and Biophysics and 1Program in Neuroscience, University of Miami School of Medicine, P.O. Box 016430 (R430), Miami, Florida 33101, USA
Correspondence should be addressed to N.C. (nchaudhari@miami.edu)

Sensory transduction for many taste stimuli such as sugars, some bitter compounds and amino acids
is thought to be mediated via G protein-coupled receptors (GPCRs), although no such receptors that
2000 Nature America Inc. http://neurosci.nature.com

respond to taste stimuli are yet identified. Monosodium L-glutamate (L-MSG), a natural component
of many foods, is an important gustatory stimulus believed to signal dietary protein. We describe a
GPCR cloned from rat taste buds and functionally expressed in CHO cells. The receptor couples neg-
atively to a cAMP cascade and shows an unusual concentrationresponse relationship. The similarity
of its properties to MSG taste suggests that this receptor is a taste receptor for glutamate.

Chemoreceptor cells in taste buds monitor the chemical envi-
ronment in the oral cavity and generate signals that lead to taste
perceptions. Taste transduction for simple salts involves altered
permeation of the receptor cell membrane by ions such as Na+, K+
or H+ (ref. 1). The resulting receptor currents in taste bud cells
stimulate neurotransmitter release to excite sensory afferents,
ultimately leading to perceptions such as salty or sour. Taste
transduction for larger organic molecules such as sugars, amino
acids or a heterogeneous collection of compounds that elicit per-
ception of bitterness, is thought to include binding at specific
receptors on the taste cell plasma membrane1,2.
Some of the proteins that orchestrate this plethora of sensory
transduction mechanisms have been identified using molecular
biological methods. Taste receptor cells express G proteins, includ-
ing -gustducin3, -transducin4, a number of additional G sub-
units 5,6, several G subunits and a taste-specific G 7.
Phosphodiesterases4 and a cyclic nucleotide-gated channel8 cloned
from mammalian taste buds could potentially participate in sen-
sory transduction pathways. Epithelial sodium channels demon-
strated in taste buds presumably underlie salty and sour
transduction911. Detectable receptor activity for bitter stimuli
is found in membrane preparations from taste tissue12. Although
a number of candidate taste-GPCRs have been proposed1315, their
functional significance in taste transduction has not been estab-
lished2. This report describes the cloning and functional charac-
terization of a taste receptor and its natural taste stimulus.
Sweet, sour, salty, bitter and (arguably) umami constitute basic
taste qualities. Umami denotes the taste of the glutamate moiety
in monosodium L-glutamate (L-MSG), a compound that occurs
naturally in protein-rich and other foods. Taste transduction for
glutamate is hypothesized to entail stimulation of neurotransmit-
ter-like ionotropic and metabotropic glutamate receptors1618. A
number of ionotropic glutamate receptors are expressed in lingual
tissue, although none seems preferentially localized to taste buds17.
Metabotropic glutamate receptors (mGluR18) constitute a fam-
ily of GPCRs that are found in many neuronal cells19. In taste recep-
tor cells, molecular, physiological and behavioral evidence
implicates a metabotropic receptor similar or identical to mGluR4
in taste transduction for L-glutamate20. Such evidence includes the
findings that mGluR4 is expressed in rat taste buds17,21 and that
an mGluR4-selective ligand, L-AP4, mimics the taste of L-MSG in
conditioned taste aversion in rats17 and in human psychophysical
measurements22. Further, both L-MSG and L-AP4 interact syner-
gistically with nucleotide monophosphates to elicit umami taste
responses23,24. Additionally, stimulating taste buds with glutamate
decreases cellular cAMP (X. Zhou & N. Chaudhari, Chem. Senses
22, 834, 1997) and alters membrane conductances25, a signaling
cascade also triggered by mGluR4. Collectively, these findings are
consistent with the transduction of L -glutamate taste by an
mGluR4-like receptor. Nevertheless, several lines of evidence indi-
cate apparent discrepancies between umami taste and the proper-
ties of mGluR4. The concentrations of glutamate needed to elicit
taste and to activate the neurotransmitter receptor mGluR4 differ
markedly. The detection threshold for L-MSG in recordings from
sensory afferents is 0.10.3 mM in juvenile and 13 mM in adult
rodents26,27, whereas mGluR4 requires glutamate in the micro-
molar range. Further, the ability of glutamate agonists to stimu-
late mGluR4 does not correlate fully with their umami taste28.
Additionally, umami taste does not seem to be blocked by a known
antagonist of mGluR423. These observations suggest that the recep-
tor(s) transducing umami taste may differ significantly from
mGluR4, particularly in the glutamate-binding domain.
The glutamate-binding domain of the mGluR is contained
within the large extracellular N terminus. Although detailed
structural information is lacking, a model of the N terminus of
mGluRs is based on the structure of a bacterial periplasmic
leucine-isoleucine-valine binding protein (LIVBP)29. Experi-
mental verification of this model includes mutation of contacting
amino acids29, expression of truncated extracellular domains that
retain binding characteristics30 and chimeric receptors with dis-
tinct agonist sensitivities31. Thus, the extracellular N terminus
might be a plausible site for differences between neurotransmit-
ter and taste receptors for glutamate. We found an unusual vari-
ant of mGluR4, taste-mGluR4, expressed in lingual epithelium.

nature neuroscience volume 3 no 2 february 2000 113

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

articles

Fig. 1. The 5 end of mGluR4

cDNA from taste papillae a
contains novel sequence
derived from an intron.
(a) The 5 end of mGluR4
cDNA from taste papillae
(lower lines) is aligned with
the corresponding region of
mGluR4 cDNA from brain
(upper lines). Amino acid
(bold) and translated
nucleotides (regular) for each
sequence are numbered. A
stop codon (), in frame with
the long open reading frame, b
is found within the novel
region of the cDNA from
taste tissue. Sequence identity
between the two cDNAs
begins abruptly at the codon
2000 Nature America Inc. http://neurosci.nature.com

for R291. This position also

corresponds to the location of
intron a (). (b) Schematic
showing the full-length cDNA
for mGluR4 from brain,
including 5 and 3 untrans-
lated regions (line), translated
region (shaded bar) and
regions encoding seven puta-
tive transmembrane segments
(gray stripes). The 800-bp seg-
ment (double line) expressed c d
in rat taste buds17 as well as
the 3 and 5 RACE products
(**) we characterized are
shown above the brain-
derived mGluR4 cDNA. The
corresponding genomic orga-
nization shows two interven-
ing sequences. The last 60 bp
at the 3 end of intron a (in
black) are identical to the first
60 bp at the 5 end of the
taste-derived cDNA; the taste-derived cDNA includes part of intron a followed by exon N1. (c) mGluR4 mRNA from brain includes several exons
upstream of N0, whereas mGluR4 mRNA derived from taste tissue begins with an extended exon N1 spliced to exon N2. In the taste-derived mRNA, the
presence of the upstream stop codon prevents translation of the first 128 bp (white); translation presumably initiates at the next methionine codon (M309
in brain-derived cDNA). (d) Predicted transmembrane topology of brain- and taste-mGluR4 showing the truncated extracellular N terminal domain fol-
lowed by seven putative transmembrane helices and cytoplasmic C terminus. The LIVBP-like glutamate-binding domain29 and its truncated version are
shown as heavy lines.

The corresponding protein is predicted to lack approximately
half the N terminus, including a large portion of the LIVBP-like
putative glutamate-binding domain. Taste-mGluR4 cDNA
expressed in CHO cells conferred sensitivity to L-glutamate at
concentrations ~100-fold higher than needed for brain-mGluR4,
and its expression level was negatively coupled to cAMP con-
centration. These findings correspond well with the concentra-
tions of glutamate needed to elicit umami taste and resolve the
discrepancy between neurotransmitter receptors for glutamate
and taste receptors for glutamate. Some of these results were pre-
sented in abstracts (A. Fedorov & Chaudhari, N., Chem. Senses
23, 593, 1998; A. M. Landin et al., Chem. Senses 24, 586, 1999).
RESULTS
In-situ hybridization shows that mGluR4 is expressed in rat taste
buds and cannot be detected in surrounding non-sensory epithe-
lium17,21. However, it is unclear whether mGluR4 in taste buds is
identical to that in the brain. Earlier RT-PCR and in-situ hybridiza-
tion analyses focused on an ~800-bp core region conserved among
mGluR1-8 (Double line in Fig. 1b). The N terminus of mGluRs
comprises a large extracellular glutamate-binding domain29. The
cytoplasmic C terminus of mGluRs participates in interactions
with G proteins. Because N and C termini of mGluRs determine
essential functional characteristics, we analyzed the corresponding
sequences for mGluR4 expressed in taste tissue.
Full-length cDNA for mGluR4 from taste tissue
In the brain, mGluR4 mRNA is found in two forms, mGluR4a and
mGluR4b, which differ at the 3 end32. The longer mRNA includes an
exon containing an in-frame stop codon, and thus generates a short-
er protein product, mGluR4a. To analyze the C terminus of mGluR4
in taste cells, we used poly(A)RNA from taste (circumvallate and

114 nature neuroscience volume 3 no 2 february 2000

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

articles

foliate) papillae and gene-specific a b

primers located in the previously
characterized 800-bp core region.
We performed 3 RACE (rapid
amplification of cDNA ends)
reactions and cloned the longest
specifically amplified bands.
DNA-sequence analysis of repre-
sentative clones yielded 100%
identity with mGluR4a cDNA
from the brain33. We also carried
out RT-PCR with a primer pair
straddling the alternatively spliced
exon. The principal amplification
product was confirmed by DNA
sequence analysis to be mGluRa.
A faint band corresponding in
size to mGluR4b was detected Fig. 2. The truncated mRNA in taste papillae is a mature mRNA. (a) RT-PCR with a reverse primer (R) in exon
2000 Nature America Inc. http://neurosci.nature.com

only in occasional PCRs (not N2 and forward primers B (within exon N0) or T1, T2, T3 (within intron a) using poly(A)RNA from taste tissue
shown). Thus, we conclude that or from brain. With taste RNA, PCR product is visible in lanes amplified with T1-R and B-R primer pairs but not
the C-terminal sequence in taste with the T2-R or T3-R primer pairs, demonstrating that only a portion of intron a sequence is present in the
tissue was predominantly of the RNA. With brain RNA used as template, PCR products are clearly seen with all three primer pairs, T1-R, T2-R
mGluR4a type. and T3-R, implying that sequences from intron a are present in unspliced precursor for the known mGluR4
Gene-specific reverse primers mRNA. A 1-kb ladder (BRL) is in the far left lane. Schematics on the right show RNA templates postulated to
yield the PCR products obtained (taste-mGluR4, precursor of brain-mGluR4 and mature brain-mGluR4,
in the 800 bp core region were
respectively). (b) RNase protection assay demonstrates that taste tissue contains significant amounts of the
also used in 5 RACE reactions. truncated mGluR4 RNA. [32P]-labeled probe was hybridized with poly(A)RNA from liver, cerebellum (0.05 g,
Previously cloned cDNAs for rat cer) and taste papillae (23 g) from adult (ad) or juvenile (juv) rats. Lengths are indicated in nucleotides; RNA
mGluR4a include a 5 untrans- markers are on the left, and RNase-protected fragments are on the right. A schematic of the probe used (right)
lated region (5 UTR) of either shows exons (gray) and the segment of intron a found in taste-derived mGluR4 cDNA (black).
69 bp 33 or 854 bp 29. Thus, we
estimated that our 5 RACE
products should range between
1450 and 2235 bp. Using brain poly(A)RNA to validate the (precursor) RNA rather than a mature mRNA. Because introns
method, amplification products extending to 2000 bp were are spliced out intact, a precursor RNA should include the com-
obtained as expected. In contrast, the 5 RACE product obtained in plete sequence of intron a. This prediction was tested by RT-PCR
parallel from taste tissue terminated abruptly at approximately 600 (Fig. 2a). We used three forward primers (T1, T2 and T3) along
bp. Similarly truncated products from taste-derived mRNA result- intron a. The reverse primer (R) was selected from a separate
ed from 5 RACE reactions with at least four different gene-spe- downstream exon to preclude amplification from genomic DNA.
cific reverse primers and two sources of reverse transcriptase. From brain poly(A)RNA, approximately equal amounts of prod-
uct were detected for all three reactions using forward primers
A distinct mGluR4 cDNA found in taste tissue within intron a (Fig. 2a). This result implies that intron a may be
Sequence analysis of the cloned RACE product indicated that the a late-spliced intron and that brain poly(A)RNA contains pre-
5 terminal 4060 bp of the taste-derived cDNA clones were not cursor RNAs that include the complete intron. In contrast, taste
similar to any region of brain mGluR4 cDNA (Fig. 1a). A stop poly(A)RNA showed amplification product only from the far-
codon was found near the 5 end, in-frame with the long open thest-downstream intronic primer, T1, and not from upstream
reading frame, implying that the 5 end is untranslated sequence. primers T2 or T3. Thus, poly(A)RNA in taste tissue included only
We found two potential start codons in frame, 102 bp and 189 bp a short segment from the 3 end of intron a, suggesting that the
downstream of the stop codon. Thus, the sequence of mGluR4 truncated mGluR4 cDNAs obtained in 5 RACE were probably
cDNA from taste tissue is unique for the first ~60 bp and then is derived not from an unspliced precursor RNA, but from a mature
identical for at least 2220 bp to mGluR4a cDNA from brain. mRNA. A forward primer (B) located in the next exon upstream
The point of divergence between unique and identical (N0), amplified from the previously known mGluR4 mature
sequences for taste- and brain-derived cDNAs, located at amino mRNA, served as a control. PCR product from mGluR4 mRNA
acid R291 of mGluR4a, resembles a splice acceptor site. To test lacking this intron was detected in poly(A)RNA from both brain
this, we analyzed a genomic fragment amplified from this region and taste papillae. Precursor RNAs are typically found in tissue
and determined that an intron (intron a, Fig. 1b) interrupts the at considerably lower concentration than their respective mature
codon for R291. The 3 end of intron a is identical to the ~60-bp mRNAs. In taste tissue, the low concentration of mature full-
unique sequence at the 5 end of the taste-derived mGluR4 length mGluR4 mRNA17 precludes detecting its precursor RNA
cDNAs (Fig. 1a and b). One additional intron was also identified (as RT-PCR products with T3 and T2 primers).
further downstream. RNA secondary structure can cause premature termination
of reverse transcripts and yield truncated products in 5 RACE.
Taste-tissue mGluR4 is a mature mRNA This did not seem to be the case for the truncated taste-mGluR4
The presence of intron sequence in the taste-derived mGluR4 cDNA because brain poly(A)RNA did yield long 5 RACE prod-
cDNA raised the possibility that it was amplified from a nuclear ucts, and because precursor RNAs in brain samples were readily

nature neuroscience volume 3 no 2 february 2000 115

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

articles

band, 254263 nucleotides in

a b length, as might be expected for an
RNA that included the 178
nucleotides of exon (N 1 plus N 2)
sequence and extended 7685
nucleotides forward into intron a.
This protected band was consistent
with the mGluR4 cDNA cloned in
the 5 RACE experiments. The band
was not detected in cerebellar RNA
lanes. Thus, we designate this form
of mGluR4 RNA, which includes a
part of intron a, as taste-mGluR4.
We note that the ~178-bp band
Fig. 3. Taste-mGluR4 is activated by glutamate at much higher concentrations than brain-mGluR4. derived from brain-mGluR4 is
(a) Immunoblot of CHO cells stably transfected with brain-mGluR4 (B) or taste-mGluR4 (T) shows a promi- sharp, as expected, whereas the
nent immunoreactive band of predicted molecular weight (~102 kDa and 68 kDa, respectively) when probed broad taste-mGluR4-derived band
with an antibody against the shared C terminus of both forms. CHO cells transfected with non-recombinant indicates more heterogeneity in
2000 Nature America Inc. http://neurosci.nature.com

vector (con) do not show either band. Extracts of cerebellum (cer) and brain-mGluR4 expressing CHO cells length. The 5 ends of mRNAs fre-
also show a higher band, presumably a dimerized receptor at 208 kDa. (b) CHO cells expressing brain- quently show such heterogeneity,
mGluR4 () and taste-mGluR4 () both respond to glutamate by decreasing cAMP levels. Cells were stimu-
spanning 315 nucleotides.
lated with L-glutamate in the presence of forskolin and IBMX (see Methods). Six independent experiments
were carried out, each including triplicate wells of cells for each concentration. Three of the experiments
We observed the broad band
were conducted on lines of clonal transformants and another three on non-clonal lines of stably expressing corresponding to taste-mGluR4
CHO cells. The values represent mean s.e.; the curves are a sigmoidal fit to the data. Control cells, trans- mRNA in three separate protection
fected with the non-recombinant vector did not significantly alter cAMP levels when stimulated with L-gluta- experiments using different batch-
mate (1 mM, 127 16.0%; 10 mM, 132 19.0%; n = 9). The absolute values of cAMP per well in es of poly(A)RNA from taste tissue
forskolin-stimulated cells in the absence of glutamate were: 6.07 0.60 pmole, 6.88 0.78 pmole and of juvenile rats. Densitometric
4.33 0.42 pmole for cells transfected with brain-mGluR4, taste-mGluR4 and vector, respectively. analysis from three experiments
indicated that taste-mGluR4
mRNA is present at 70120% of
the concentration of brain-
reverse transcribed and amplified through intron a (Fig. 2a). mGluR4 mRNA in taste tissue from juvenile rats. Interesting-
Thus, based on the 5 RACE and RT-PCR analyses above, we ten- ly, the taste-mGluR4 band was found at lower concentration
tatively concluded that taste tissue may contain two forms of when poly(A)RNA from adult rather than juvenile rats was
mGluR4 mRNAone similar to the known mGluR4a33 and used for protection (Fig. 2b).
another with a truncated 5 end.
Because RT-PCR can detect RNAs that are present in minor Taste-mGluR4 is a functional mRNA
(nonphysiological) quantities in cells, we tested whether the trun- To determine whether taste-mGluR4 is a functional mRNA, we
cated mGluR4 RNA found in taste tissue was present at significant generated full-length clones for both forms of mGluR4 in
concentration using RNase protection (Fig. 2b), an independent pcDNA3.1 vector, transfected them into CHO cells and selected
method not based on amplification. The probe for this assay con- stable transfectants. When probed with an antibody against the C
sisted of the last ~400 nucleotides of intron a, followed by 178 terminus of mGluR4a, immunoblots (Fig. 3a) of cerebellar extracts
nucleotides in two consecutive exons, N1 and (part of) N2. No contained a strong band of the expected molecular weight, 102
bands were generated with RNA from liver, a control tissue that kDa, as previously reported34. A presumed dimer at 210 kDa was
does not express mGluR4, demonstrating the specificity and also detected. Clones of cells stably transfected with brain-mGluR4
RNase sensitivity of the probe. The known full-length form of showed prominent bands of the same size as in cerebellum (Fig. 3a).
mGluR4 mRNA (henceforth designated brain-mGluR4) should In contrast, CHO cells transfected with the taste-mGluR4 construct
protect a band of 178 nucleotides, as it includes no sequence from consistently showed a prominent band of 68 kDa, corresponding
intron a. Indeed, consistent with an identity as brain-mGluR4, to the size predicted from the cDNA sequence. Neither the 102-
a protected band of ~180 nucleotides was detected in kDa nor 68-kDa bands was present in parallel lanes containing
poly(A)RNA from cerebellum and, at a lower concentration, from lysates of CHO cells transfected with non-recombinant pcDNA
taste papillae. In addition, we detected two bands corresponding vector. Thus, the truncated taste-mGluR4 mRNA characterized
to precursor nuclear RNA when cerebellar poly(A)RNA was used from taste tissue was indeed a functional mRNA that was translat-
in RNase protection. Because exons N1 and N2 are not contiguous ed into immunologically recognizable protein.
in the genome, precursor RNA protected fragments ~460
nucleotides long (400 nucleotides of intron a plus exon N1) and Taste-mGluR4 is negatively coupled to cAMP
~120 nucleotides (fragment of exon N2). The absence of these In transfected CHO cells, activation of group III mGluRs leads to a
precursor bands in taste samples indicated that genomic DNA suppression of forskolin-stimulated cAMP synthesis35. CHO cells
(which would protect the same size bands as precursor RNA) was stably expressing brain-mGluR4 responded predictably to L-gluta-
not a significant contaminant in the taste samples. mate (Fig. 3b). The EC50 for this response was 2 M glutamate, con-
In addition to bands derived from the known mGluR4 sistent with earlier reports on mGluR4a (5 M)35. Cells expressing
mRNA and its precursor RNA, hybridization with poly(A)RNA taste-mGluR4 displayed no response to L-glutamate at concentra-
from taste papillae yielded an additional band. This was a broad tions of 30 M or below. The EC50 for L-glutamate for taste-mGluR4

116 nature neuroscience volume 3 no 2 february 2000

 2000 Nature America Inc. http://neurosci.nature.com
Fig. 4. Taste-mGluR4 is activated by L-AP4 at higher concentrations
than is brain-mGluR4. CHO cells expressing brain-mGluR4 (solid bars)
and taste-mGluR4 (shaded bars) were stimulated with the indicated
concentrations of L-AP4. Six independent experiments were carried out
2000 Nature America Inc. http://neurosci.nature.com
(three each on clonal and non-clonal lines of stably transfected cells).
For each concentration, triplicate wells of cells were used in each exper-
iment. The values represent mean s.e.
was calculated to be 280 M glutamate. We considered the possi-
bility that a low cell-surface density of taste-mGluR4 might explain
its low efficacy (high EC50). Thus, we examined two separate lines
of CHO cells expressing taste-mGluR4. Expression levels of taste-
mGluR4 (as determined by immunoblot) were 100-fold and 2-fold
lower, respectively, than in parallel lines expressing brain-mGluR4.
Nevertheless, the EC50 values for the two lines expressing taste-
mGluR4 were very similar, 300 and 250 M, respectively. The con-
sistently high EC50 for taste-mGluR4 suggests that low receptor
density does not explain its low efficacy. Instead, the data suggest
that taste-mGluR4 is approximately two orders of magnitude less
sensitive to L-glutamate than is brain-mGluR4. Direct measurements
of affinity will be necessary to confirm this interpretation.
Because the concentration of L-glutamate required for taste-
mGluR4 was high, we considered that osmotic or other nonspe-
cific effects might influence the cAMP response. Thus, we
stimulated mGluR4-expressing cells with D-glutamate, an isomer
that does not elicit umami taste36. Relative to controls, cells stim-
ulated with 1 mM D-glutamate yielded cAMP levels of 126 17%
(brain-mGluR4) or 99 3% (taste-mGluR4). Control CHO cells
transfected with non-recombinant vector did not alter cAMP con-
centrations upon treatment with either L- or D-glutamate. Thus,
receptor-independent mechanisms do not seem to decrease cAMP.
Taste-mGluR4 responds to L-AP4
In rat and human behavioral studies, L-AP4 mimics the taste of
L-MSG17,22. Thus, we predicted that the taste receptor for L-MSG
should also be stimulated by L-AP4. We directly tested whether
taste-mGluR4 meets the criterion of a taste receptor by measur-
ing its response to L-AP4 in a concentration range effective for
taste. CHO cells expressing brain-mGluR4 responded to L-AP4
by suppressing forskolin-stimulated cAMP production. The
response seemed to saturate at all concentrations of L-AP4 above
10 M (Fig. 4), as expected from the published EC50 of 0.51.0
M35. In cells expressing taste-mGluR4, 10 M L-AP4 was inef-
fective, whereas 100 M and 1 mM L-AP4 gave progressively larg-
er responses. For rats in a behavioral assay, 100 M L-AP4 is near
the detection threshold17. Thus, taste-mGluR4 responds to L-AP4
over a concentration range similar to that observed for L-AP4
nature neuroscience volume 3 no 2 february 2000
articles
taste. The cAMP levels in control cells transfected with non-
recombinant vector did not change upon treatment with L-AP4.
DISCUSSION
MGluR4 is a metabotropic glutamate receptor originally char-
acterized from the brain. In situ hybridization analyses have
shown that this receptor is also expressed in taste buds17,21. The
present report demonstrates that mGluR4 in taste tissue is
expressed as a structurally and functionally distinct form that we
have termed taste-mGluR4. Taste-mGluR4 is a truncated ver-
sion of the previously characterized brain receptor, and lacks
50% of the receptors extracellular N terminus. This truncation
is particularly significant, because the N terminus of metabotrop-
ic glutamate receptors is believed to contain the glutamate-bind-
ing domain29, and changes in this region are likely to influence
the affinity of the receptor for ligands. Indeed, we report here
that the truncated taste-mGluR4 is much less sensitive to L-glu-
tamate and L-AP4 than the full-length brain form, implying a
reduced affinity for these agonists. Importantly, the reduced sen-
sitivity of taste-mGluR4 corresponds well with the concentra-
tions of L-glutamate and L-AP4 needed to elicit a response in
gustatory receptor cells in situ. The data are fully consistent with
the interpretation that the novel taste-mGluR4 is a taste recep-
tor for umami (the taste quality elicited by L-MSG).
The molecular identification and characterization of taste
receptors has lagged behind research on other sensory receptors,
notably, receptors for vision and olfaction. GPCRs are present
in taste tissue13,14; receptors with sequences related to those of
the mGluRs15. Although mRNA and/or protein for such candi-
date taste receptors has been demonstrated in lingual tissue, the
lack of functional expression has hampered ligand identification
and validation of their physiological significance. One of the chal-
lenges for studying the function of taste receptors is the high con-
centrations of stimuli needed. In the case of sugars, salts, and
glutamate, the detection thresholds of gustatory sensory cells in
nerve recordings or behavioral tests are in the range of a few hun-
dred micromolar and higher. The low sensitivity of taste recep-
tors, which may result from low affinity for ligands, has
complicated binding assays and functional tests. By utilizing a
high concentration of glutamate and a selective glutamate recep-
tor ligand (L-AP4), we have been able to characterize the func-
tion of taste-mGluR4 in transfected cells.
The taste-mGluR4 cDNA in this report was cloned from pos-
terior (circumvallate and foliate) taste papillae of juvenile rats.
The threshold concentration for activating taste-mGluR4 (30 M)
matches well with the threshold (100 M) reported for glutamate
taste responses in the glossopharyngeal nerve of juvenile mice26.
Interestingly, taste nerve thresholds in adult mice and rats, at 210
mM, are significantly higher26,27. In our studies, we found that the
mRNA for taste-mGluR4 is expressed at lower concentration in
adult than in juvenile rats17 (Fig. 2b), which may explain the
decreased sensitivity to MSG taste in adult rodents.
L-AP4 is a highly effective ligand at brain-mGluR4 (ref. 35).
L-AP4 mimics the taste of glutamate in rats17 and is an umami
stimulus in humans22. Here we show that L-AP4 also stimulates
taste-mGluR4, at concentrations effective as taste stimuli. MAP4,
an antagonist of brain-mGluR4, fails to block taste nerve
responses to glutamate and L-AP4 in chorda tympani nerve
recordings23. Tests of MAP4 on cloned taste-mGluR4 may help
to resolve this seeming paradox. Given the drastically altered
N terminus of taste-mGluR4, it is impossible to predict its
response to the various glutamate analogs known to activate or
antagonize brain-mGluR4.
117
 2000 Nature America Inc. http://neurosci.nature.com
articles
A distinctive feature of umami is the potentiation of glutamate
responses by the nucleotide monophosphates of inosine and
guanosine (IMP and GMP). This synergy is well documented in
human psychophysical studies37, animal behavioral experiments24,
gustatory nerve recordings27 and patch-clamp studies25. Synergis-
tic interactions are also found between L-AP4 and nucleotides23,24,
further underscoring the importance of taste-mGluR4 to umami
transduction. The site of interaction between glutamate and
nucleotides remains to be defined, but might involve synaptic con-
vergence of separate gustatory sensory cells onto common affer-
ent fibers, or converging signaling pathways from separate receptors
within the same glutamate-sensing taste bud cell. It is also possible
that both glutamate and nucleotides interact with a common recep-
tor molecule. For instance, ligand-binding studies on bovine taste
membranes suggested an allosteric model for nucleotide effects on
glutamate taste38. The cloned taste-mGluR4 will allow direct tests
of such models of receptor-taste stimulus interactions.
Our RT-PCR and RNase protection analyses indicate that both
2000 Nature America Inc. http://neurosci.nature.com
taste-mGluR4 and brain-mGluR4 are expressed in taste tissue (Fig.
2a and b). Glutamate is implicated as a neurotransmitter for taste
bud afferent synapses39, and brain-mGluR4 could serve as an autore-
ceptor at such synapses. We suggest that taste-mGluR4 is likely to
function as an umami receptor, whereas brain-mGluR4 may serve
as a neurotransmitter receptor at synapses in or near taste papillae.
Glutamate receptors other than mGluR4 may be expressed in
gustatory sensory cells. For example, patch-clamp recordings, Co2+
uptake, and Ca2+ imaging on rat and mouse taste buds suggest that
glutamate activates both ionotropic and metabotropic glutamate
receptors18,25,39,40. However, all these experiments exposed both
apical and basolateral membranes of taste receptor cells to gluta-
mate. Given that bona-fide taste stimulation reaches only the api-
cal membrane of taste receptor cells, it is critical to note that
behavioral studies and nerve recordings (which stimulate only the
apical membrane) indicate a minimal role for ionotropic-like glu-
tamate receptors in taste transduction for L-MSG17,2224. Thus, baso-
lateral synapses on gustatory receptor cells may include glutamate
receptors other than taste-mGluR4 (reviewed in ref. 20).
Transcripts of several mGluRs, including mGluR1 and mGluR5,
are alternatively spliced41,42. For mGluR4, alternative splicing gives
rise to receptors with either long or short C termini32. In spite of
considerable structural diversity, most alternative-splicing variants of
mGluRs show only subtle changes in their functional properties. For
instance, mGluR1c and mGluR1a elicit distinct temporal patterns
of Ca2+ release42. The present case of taste-mGluR4 demonstrates
substantially different function of two receptors derived from the
same gene. We do not presently know the molecular mechanism by
which the truncated mRNA is produced in taste cells. It is possible
that an alternative splice acceptor site is located within intron a and
that our 5 RACE reactions did not progress into additional exons
upstream of the ~60 bp in intron a. Nevertheless, the presence of an
in-frame stop codon within these 60 nucleotides implies that any
additional exons would constitute 5 UTR of the mRNA and would
not affect the sequence of the translated protein. An alternative pos-
sibility is that the origin for transcription of taste-mGluR4 mRNA is
located within intron a. Multiple promoters yielding mRNAs with
distinct 5 exons occur in the mGluR5 gene43.
METHODS
Tissues and RNA. All tissues were from Harlan Sprague-Dawley rats.
Circumvallate and foliate taste papillae were dissected from tongue, and
rapidly frozen on dry ice. Poly(A)mRNA was extracted from tissues by
direct binding to oligo dT-cellulose (FastTrack II kit, Invitrogen, Carlsbad,
California). Unless stated otherwise, taste tissue samples were from juve-
nile (pre-weaning 1620 day-old) rats.
118
5 and 3-RACE and RT-PCR. Initial 5 RACE (rapid amplification of
cDNA ends) reactions were carried out using superscript reverse tran-
scriptase followed by terminal deoxynucleotidyl transferase (both from
Gibco-BRL, Gaithersburg, Maryland)44. Subsequently, the Marathon
RACE system (Clontech Laboratories, Palo Alto, California) was
employed for generating and cloning RACE products. Poly(A)RNA,
extracted from vallate and foliate papilla, was used to synthesize double
strand cDNA, which was then ligated to an adapter-primer. Nested gene-
specific primers were designed in the core region of rat mGluR4 cDNA
sequence within the 800-bp region known to be expressed in taste buds17.
RACE reactions in both directions were carried out using Klentaq DNA
polymerase (Clontech) or Elongase (BRL) to ensure amplification of long
products. Annealing steps were at the highest temperature permitted by
respective primers. Amplification proceeded for 2530 cycles to mini-
mize nonspecific products. Amplification products were electrophoresed
and blot hybridized to identify mGluR4-related bands, which were then
cloned into pGEM-T vector (Promega) and sequenced on an ABI
Sequencer Model 373A.
The following primers based on mGluR4a cDNA33 (with identifying
amino acid positions) were used in reverse transcriptase-polymerase
chain reaction (RT-PCR):
B: 5 (D266) CGACAAGATCATCAAACTGCCTAC 3
R: 5 (F455) GAAGTTGACGTTCCTGATGTACT 3
To isolate a genomic fragment that contained intron a, genomic DNA
was amplified with primers located in cDNA sequence: forward primer B
(above) and a reverse primer at F307N301. The following primers were
based on intron sequence derived from genomic clones, and were used to
map whether the entire intron a was represented in precursor nuclear
RNAs (Fig. 2a):
T1: 5 (48 bp upstream of R292) CAGCTGGGTAGCCTTACATGTCT 3
T2: 5 (400 bp upstream of R292) TCTGGAGTAGGATCAGGTGGATG 3
T3: 5 (2000 bp upstream of R292) AAAGGCTGCTATCTCGTGGACT 3;.
RNase protection assay. The template for mGluR4 probe was constructed
by ligating together a genomic fragment and a cDNA fragment at a shared
AflIII site located 32 bp upstream of the junction between intron a and exon
N1 (Figs. 1b and 2b). The genomic fragment contained the downstream
400 bp of intron a (extended for 368 bp upstream of the AflIII site). The
210 bp cDNA segment of the chimeric probe began at the AflIII site in intron
a and included two sequential exons, N1 and part of N2. The resulting con-
struct in pGEM-T vector was used to transcribe an antisense probe (Fig.
2b) with at least 238 nt complementary to taste-mGluR4 mRNA, as pre-
dicted from the sequence of the longest 5 RACE clone (see Results). [32P]-
labeled antisense RNA was transcribed using T7 RNA polymerase.
Hybridization and RNase digestion were performed using the High-Speed
Hybridization RPA kit from Ambion (Austin, Texas). Protected fragments
were analyzed on denaturing 6% polyacrylamide-urea gels.
Transfection. Full-length cDNAs for brain-mGluR4 and taste-mGluR4
(see Results) were reconstructed from cerebellar or taste poly(A)RNA
respectively, by RT-PCR using Elongase (BRL) for high-fidelity amplifi-
cation. The brain-mGluR4 insert included 7 nucleotides of 5 untrans-
lated region; the taste-mGluR4 insert included 100 nucleotides of
presumed 5 UTR upstream of the first in-frame start codon. Both cDNA
inserts terminated at their common stop codon (following I912), yield-
ing the mGluR4a version of the C terminus33. The inserts were cloned
into the EcoRI site of pcDNA3.1 vector (Invitrogen).
CHO cells were transfected with brain-mGluR4 and taste-mGluR4 con-
structs and with non-recombinant pcDNA3.1 vector, all in parallel, using
a cationic lipid, DMRIE-C (Gibco-BRL). Cells with stably integrated plas-
mid were selected and maintained in 300 g per ml G418 starting 24 h
after transfection. The medium consisted of Dulbeccos Modified Eagle
Medium (D-MEM) supplemented with 10% fetal bovine serum, 2 mM
glutamine, 1.7 mM proline and 100 units per ml penicillin-streptomycin35.
Clones were lifted within two weeks and screened for expression by west-
ern blot analysis. A single clone for each form was propagated for func-
tional assays. Stably transfected but non-clonal lines of cells from
independent transfections with both forms of mGluR4 and non-recom-
binant vector were also maintained and used in functional assays.
nature neuroscience volume 3 no 2 february 2000
 2000 Nature America Inc. http://neurosci.nature.com
Immunoblot analysis. Anti-peptide anti-serum directed against the C-
terminal 18-amino-acid sequence of mGluR4a was raised in rabbits and
antibody was purified by affinity chromatography (Zymed Laboratories,
San Francisco, California). This epitope is shared between both brain-
mGluR4 and taste-mGluR4 described here. Lysates of transfected CHO
cells were electrophoresed and tested for expression of brain- and taste-
mGluR4 by immunoblot analysis45 on PVDF membrane. Detection was
with alkaline phosphatase-conjugated secondary antibody and CSPD
chemiluminescent substrate (both from Tropix, Bedford, Massachusetts).
Bands on autoradiographs were densitometrically quantified as needed.
Functional assay. CHO cells, stably transfected with brain-mGluR4, taste-
mGluR4 or non-recombinant pcDNA3.1, were maintained as sub-con-
fluent cultures and refed every two days to minimize chronic stimulation
of expressed receptors by glutamate released from dead cells. Cells were
plated 20 h before assay in a 96-well microtiter plate at 2 104 cells per
200 l well. Fresh medium was replaced for one hour immediately before
assaying receptor function as described35. Briefly, cells were incubated in
Dulbeccos phosphate buffered saline containing 1 mM IBMX for 20 min
followed by stimulation for 10 min in 10 M forskolin and 1 mM IBMX,
2000 Nature America Inc. http://neurosci.nature.com
with or without agonists. Stimulation buffer was rapidly removed, and
cells were lysed in 200 l 0.25% dodecyltrimethyl-ammonium bromide
in 50 mM acetate buffer (Amersham, Piscataway, New Jersey). Released
cAMP in 20% of each lysate was assayed directly using an Amersham EIA-
based kit and plotted as mean s.e. of 36 experiments, each performed
with triplicate wells of cells.
ACKNOWLEDGEMENTS
This work was supported by grants from NIH/NIDCD (DC 03013) and from Cultor
Food Science, Inc. We are also grateful for support from the Umami Manufacturers
Association of Japan during the early stages of this study. We acknowledge technical
assistance from Cynthia Lamp and Helena de Carvalho.
RECEIVED 20 SEPTEMBER; ACCEPTED 14 DECEMBER 1999
1. Lindemann, B. Taste reception. Physiol. Rev. 76, 719766 (1996).
2. Lindemann, B. Receptor seeks ligand: on the way to cloning the molecular
receptors for sweet and bitter taste. Nat. Med. 5, 381382 (1999).
3. McLaughlin, S. K., McKinnon, P. J. & Margolskee, R. F. Gustducin is a taste-
cell-specific G protein closely related to the transducins. Nature 357, 563569
(1992).
4. McLaughlin, S. K., McKinnon, P. J., Spickofsky, N., Danho, W. & Margolskee,
R. F. Molecular cloning of G proteins and phosphodiesterases from rat taste
cells. Physiol. Behav. 56, 11571164 (1994).
5. Kusakabe, Y., Abe, K., Tanemura, K., Emori, Y. & Arai, S. GUST27 and closely
related G-protein-coupled receptors are localized in taste buds together with
Gi-protein -subunit. Chem. Senses 21, 335340 (1996).
6. Kusakabe, Y. et al. Identification of two alpha-subunit species of GTP-binding
proteins, Ga15 and Gaq, expressed in rat taste buds. Biochim. Biophys. Acta
1403, 265272 (1998).
7. Huang, L. et al. G13 colocalizes with gustducin in taste receptor cells and
mediates IP3 responses to bitter denatonium. Nat. Neurosci. 2, 10551062 (1999).
8. Misaka, T. et al. Taste buds have a cyclic nucleotide-activated channel,
CNGgust. J. Biol. Chem. 272, 2262322629 (1997).
9. Kretz, O., Barbry, P., Bock, R. & Lindemann, B. Differential expression of
RNA and protein of the three pore-forming subunits of the amiloride-
sensitive epithelial sodium channel in taste buds of the rat. J. Histochem.
Cytochem. 47, 5164 (1999).
10. Lin, W., Finger, T. E., Rossier, B. C. & Kinnamon, S. C. Epithelial Na+ channel
subunits in rat taste cells: localization and regulation by aldosterone. J. Comp.
Neurol. 405, 406420 (1999).
11. Ogawa, S. et al. Receptor that leaves a sour taste in the mouth. Nature 395,
555556 (1998).
12. Ming, D., Ruiz-Avila, L. & Margolskee, R. F. Characterization and
solubilization of bitter-responsive receptors that couple to gustducin. Proc.
Natl. Acad. Sci. USA 95, 89338938 (1998).
13. Matsuoka, I., Mori, T., Aoki, J., Sato, T. & Kurihara, K. Identification of novel
members of G-protein coupled receptor superfamily expressed in bovine
taste tissue. Biochem. Biophys. Res. Commun. 194, 504511 (1993).
14. Abe, K., Kusakabe, Y., Tanemura, K., Emori, Y. & Arai, S. Primary structure
and cell-type specific expression of a gustatory G protein-coupled receptor
related to olfactory receptors. J. Biol. Chem. 268, 1203312039 (1993).
15. Hoon, M. A. et al. Putative mammalian taste receptors: a class of taste-specific
GPCRs with distinct topographic selectivity. Cell 96, 541551 (1999).
nature neuroscience volume 3 no 2 february 2000
articles
16. Faurion, A. Are umami taste receptor sites structurally related to glutamate
CNS receptor sites? Physiol. Behav. 49, 905912 (1991).
17. Chaudhari, N. et al. The taste of monosodium glutamate: Membrane
receptors in taste buds. J. Neurosci. 16, 38173826 (1996).
18. Hayashi, Y., Zviman, M. M., Brand, J. G., Teeter, J. H. & Restrepo, D.
Measurement of membrane potential and [Ca2+ ]i in cell ensembles:
Application to the study of glutamate taste in mice. Biophys. J. 71, 10571070
(1996).
19. Conn, P. J. & Pin, J. P. Pharmacology and functions of metabotropic
glutamate receptors. Annu. Rev. Pharmacol. Toxicol. 37, 205237 (1997).
20. Chaudhari, N. & Roper, S. D. Molecular and physiological evidence for
glutamate (umami) taste transduction via a G protein-coupled receptor. Ann.
NY Acad. Sci. 855, 398406 (1998).
21. Yang, H., Wanner, I. B., Roper, S. D. & Chaudhari, N. An optimized method
for in situ hybridization with signal amplification that allows the detection of
rare mRNAs. J. Histochem. Cytochem. 47 431446 (1999).
22. Kurihara, K. & Kashiwayanagi, M. Introductory remarks on umami taste.
Ann. NY Acad. Sci. 855, 393397 (1998).
23. Sako, N. & Yamamoto, T. Analyses of taste nerve responses with special
reference to possible receptor mechanisms of umami taste in the rat.
Neurosci. Lett. 261, 109112 (1999).
24. Delay, E. R. et al. Taste preference synergism between glutamate receptor
ligands and IMP in rats. Chem. Senses (in press).
25. Lin, W. & Kinnamon, S.C. Physiological evidence for ionotropic and
metabotropic glutamate receptors in rat taste cells. J. Neurophysiol. 82,
20612069 (1999).
26. Ninomiya, Y., Tanikukai, T., Yoshida, S., Funakoshi, M. & Tanimukai, T.
Gustatory neural responses in preweanling mice. Physiol. Behav. 49, 913918
(1991).
27. Yamamoto, T. et al. Electrophysiological and behavioural studies on the taste
of umami substances in the rat. Physiol. Behav. 49, 919925 (1991).
28. Monastyrskaia, K. et al. Effect of the umami peptides on the ligand binding
and function of rat mGlu4a receptor might implicate this receptor in the
monosodium glutamate taste transduction. Br. J. Pharmacol. 128, 10271034
(1999).
29. OHara, P. J. et al. The ligand-binding domain in metabotropic glutamate
receptors is related to bacterial periplasmic binding proteins. Neuron 11,
4152 (1993).
30. Han, G. & Hampson, D. R. Ligand binding to the amino-terminal domain of
the mGluR4 subtype of metabotropic glutamate receptor. J. Biol. Chem. 274,
1000810013 (1999).
31. Takahashi, K., Tsuchida, K., Tanabe, Y., Masu, M. & Nakanishi, S. Role of the
large extracellular domain of metabotropic glutamate receptors in agonist
selectivity determination. J. Biol. Chem. 268, 1934119345 (1993).
32. Thomsen, C. et al. Cloning and characterization of a metabotropic glutamate
receptor, mGluR4b. Neuropharmacology 36, 2130 (1997).
33. Tanabe, Y., Masu, M., Ishii, T., Shigemoto, R. & Nakanishi, S. A family of
metabotropic glutamate receptors. Neuron 8, 169179 (1992).
34. Bradley, S. R., Levey, A. I., Hersch, S. M. & Conn, P. J. Immunocytochemical
localization of group III metabotropic glutamate receptors in the
hippocampus with subtype-specific antibodies. J. Neurosci. 16, 20442056
(1996).
35. Tanabe, Y. et al. Signal transduction, pharmacological properties and
expression patterns of two rat metabotropic glutamate receptors, mGluR3
and mGluR4. J. Neurosci. 13, 13721378 (1993).
36. Hettinger, T. P., Frank, M. E. & Myers, W. E. Are the tastes of polycose and
monosodium glutamate unique? Chem. Senses 21, 341347 (1996).
37. Rifkin, B. & Bartoshuk, L.M. Taste synergism between monosodium
glutamate and disodium 5-guanylate. Physiol. Behav. 24, 11691172 (1980).
38. Torii, K. & Cagan, R. H. Biochemical studies of taste sensation. IX.
Enhancement of L-[3H]glutamate binding to bovine taste papillae by
5-ribonucleotides. Biochim. Biophys. Acta 627, 313323 (1980).
39. Caicedo, A., Kim, K. & Roper, S. Glutamate-induced cobalt uptake reveals
non-NMDA receptors in rat taste cells. J. Comp. Neurol. (in press).
40. Bigiani, A., Delay, R. J., Chaudhari, N., Kinnamon, S. C. & Roper, S. D.
Responses to glutamate in rat taste cells. J. Neurophysiol. 77, 30483059
(1997).
41. Pin, J. P. & Duvoisin, R. The metabotropic glutamate receptors: structure and
functions. Neuropharmacology 34, 126 (1995).
42. Pin, J.-P., Waeber, C., Prezeau, L., Bockaert, J. & Heinemann, S. F. Alternative
splicing generates metabotropic glutamate receptors inducing different
patterns of calcium release in Xenopus oocytes. Proc. Natl. Acad. Sci. USA 89,
1033110335 (1992).
43. Yamaguchi, S. & Nakanishi, S. Regional expression and regulation of
alternative forms of mRNAs derived from two distinct transcription
initiation sites of the rat mGluR5 gene. J. Neurochem. 71, 6068 (1998).
44. Frohman, M. A., Dush, M. K. & Martin, G. R. Rapid production of full-
length cDNAs from rare transcripts: amplification using a single gene-
specific oligonucleotide primer. Proc. Natl. Acad. Sci. USA 85, 89989002
(1988).
45. Towbin, H., Staehelin, T. & Bordon, J. Electrophoretic transfer of proteins
from polyacrylamide gels to nitrocellulose sheets: Procedure and some
applications. Proc. Natl. Acad. Sci. USA 76, 43504354 (1979).
119
 2000 Nature America Inc. http://neurosci.nature.com

articles

Identification and characterization

of the high-affinity choline
transporter
Takashi Okuda1, Tatsuya Haga1, Yoshikatsu Kanai2,3, Hitoshi Endou2, Takeshi Ishihara4 and
Isao Katsura4

1 Department of Neurochemistry, Faculty of Medicine, University of Tokyo and CREST of Japan Science and Technology Corporation,
Bunkyo-ku, Tokyo 113-0033, Japan
2 Department of Pharmacology and Toxicology, Kyorin University School of Medicine, Mitaka, Tokyo 181-8611, Japan
3 PRESTO of Japan Science and Technology Corporation, Mitaka, Tokyo 181-8611, Japan
2000 Nature America Inc. http://neurosci.nature.com

4 Structural Biology Center, National Institute of Genetics and Department of Genetics, School of Life Science, Graduate University for Advanced Studies,
Mishima 411-8540, Japan
Correspondence should be addressed to T.O. (okuda@m.u-tokyo.ac.jp)

In cholinergic neurons, high-affinity choline uptake in presynaptic terminals is the rate-limiting step
in acetylcholine synthesis. Using information provided by the Caenorhabditis elegans Genome
Project, we cloned a cDNA encoding the high-affinity choline transporter from C. elegans (cho-1). We
subsequently used this clone to isolate the corresponding cDNA from rat (CHT1). CHT1 is not
homologous to neurotransmitter transporters, but is homologous to members of the Na+-dependent
glucose transporter family. Expression of CHT1 mRNA is restricted to cholinergic neurons. The
characteristics of CHT1-mediated choline uptake essentially match those of high-affinity choline
uptake in rat brain synaptosomes.

Cholinergic neurons are vital for cognitive functions of the
brain1,2 and are known to be vulnerable in Alzheimers disease3.
Because cholinergic neurons lack the capacity to synthesize
choline de novo, their function depends upon choline uptake4.
Brain synaptosome studies demonstrate two carrier-medi-
ated transport systems for choline uptake59. At high concen-
trations, choline is transported primarily by a low-affinity and
Na+-independent system that is inhibited by hemicholinium-3
(HC3)10 with a high Ki of approximately 50 M. This system
is thought to be ubiquitously present in cells and to be required
for phosphatidylcholine synthesis. At low concentrations,
choline is transported by a high-affinity, Na+-dependent sys-
tem that is inhibited by HC3 with a low Ki of 10100 nM. The
high-affinity system is supposed to be present specifically in
cholinergic neurons, because a substantial proportion of
choline is converted to acetylcholine only when taken up
through the high-affinity system5,7,8.
The proposal that the high-affinity choline transport system
is unique to cholinergic neurons is supported by the selective loss
of the high-affinity choline uptake following depletion of cholin-
ergic terminals in a variety of denervation studies8,9. Choline
uptake is generally believed to be the rate-limiting step in acetyl-
choline synthesis49. In addition, the high-affinity choline uptake
is regulated by neuronal activity11,12, suggesting that neuronal
activity also regulates acetylcholine synthesis. In Alzheimers dis-
ease, cholinergic neurons selectively degenerate. Consistent with
the above hypothesis, the high-affinity choline transporter is
reduced in Alzheimers disease13. On the other hand, some reports
suggest upregulation of the high-affinity choline transporter or
[3H]HC3-binding sites in Alzheimers disease14.
Although complementary DNAs for transporters for major
neurotransmitters such as GABA, noradrenaline, dopamine,
serotonin, glycine and glutamate have been isolated 15, the
cDNA encoding the high-affinity choline transporter remains
to be isolated despite its physiological importance. Moreover,
the cDNAs encoding cholinergic neuronal markers, choline
acetyltransferase16 and the vesicular acetylcholine transporter17
have been isolated.
Here we report the functional identification and charac-
terization of a cDNA encoding the high-affinity choline trans-
porter (cho-1) in the nematode Caenorhabditis elegans. We also
describe the isolation and characterization of the rat homolog
of cho-1, designated CHT1. Identification of this high-affinity
choline-transporter molecule may further understanding of
cholinergic neurons and suggest new therapeutic strategies to
counter cholinergic-selective neurodegenerative disorders such
as Alzheimers disease3,14,18.
RESULTS
To isolate the cDNA encoding the high-affinity choline trans-
porter, we examined cDNAs predicted by sequences from the C.
elegans Genome Project19 to be members of the Na+-dependent
transporter family. Xenopus laevis oocytes were injected with
cRNA prepared from each candidate full-length clone and exam-
ined for induction of high-affinity choline uptake. We used inhi-
bition by 1 M hemicholinium-3 (HC3) as the criteria for
high-affinity choline uptake, because this uptake in mammalian
brain synaptosomes is completely inhibited by 1 M HC3
(K i = 10100 nM) 7,8,10. By contrast, the low-affinity choline
uptake, which is ubiquitously distributed, is inhibited only at

120 nature neuroscience volume 3 no 2 february 2000

 2000 Nature America Inc. http://neurosci.nature.com
a b c C. elegans cho-1. (a) [3H]choline uptake in Xenopus
Choline uptake (cpm)
Water C48D1.3 Water C48D1.3
d neurons. However, some DA and DB neurons do
2000 Nature America Inc. http://neurosci.nature.com
higher concentrations of HC3 (Ki of approximately 50 M). One
cDNA clone corresponding to the predicted gene C48D1.3
induced a small but significant choline uptake that was inhibited
by 1 M HC3 (Fig. 1a). This HC3-sensitive uptake was Na +
dependent, and the apparent Ki for HC3 was estimated to be 50
nM (Fig. 1b and c). Hence we designated this clone cho-1 (high-
affinity choline transporter-1).
A comparison of the cDNA and the genomic sequence indi-
cates that the cho-1 gene is composed of nine exons (data not
shown). The cho-1 cDNA sequence predicts a protein of 576
amino acids (Fig. 2a). A search of available databases with the
amino-acid sequence revealed weak but significant homology
with members of the Na+-dependent glucose transporter fami-
ly20 (Fig. 2b). Analysis of hydrophobicity and comparison with
other transporters suggests a topological configuration contain-
ing twelve transmembrane regions (Fig. 2c).
To identify the cells that express cho-1, we introduced the
green fluorescent protein (GFP) gene linked to the 5.1-kb
upstream region of cho-1 gene into worms. Expression of GFP
was detected in the nervous system, including the nerve ring and
cholinergic motor neurons of the ventral nerve cord, supporting
the idea that cho-1 encodes the high-affinity choline transporter
in cholinergic neurons (Fig. 1d).
We next focused on the vertebrate homolog of cho-1. A data-
base search using the predicted amino-acid sequence of cho-1 iden-
tified an entry (GenBank accession number AQ316435) in the
human genomic survey sequence (GSS). Based on the sequence
homology between cho-1 and this human genomic clone, we
amplified a cDNA fragment from the rat spinal cord cDNA using
degenerate PCR primers. This fragment was used to screen a rat
spinal cDNA library and to isolate positive clones. The sequence
of the longest open reading frame predicted a protein of 580 amino
nature neuroscience volume 3 no 2 february 2000
articles
Fig. 1. Identification, expression and localization of
oocytes injected with C48D1.3 cRNA or water.
Black and gray bars indicate choline uptake values in
the absence and presence of 1 M hemicholinium-3
(HC3), respectively. Each bar represents the mean
s.e. (n = 68 oocytes). (b) Effects of Na+ on choline
uptake. Black bars indicate choline uptake measured
in the standard medium ([Na+] = 100 mM); gray bars
indicate uptake in the absence of Na+. (Na+ was
replaced by Li+.) (c) Inhibition of choline uptake by
HC3. (d) Distribution of neurons that express cho-
1::gfp in the nervous system in C. elegans. An image
of a L1 larvae of N2 carrying an extrachromosomal
array of cho-1::gfp reporter constructs. The arrow-
log[HC3] (M) head indicates the nerve ring. In the ventral nerve
cord, GFP is expressed only in the cholinergic motor
not express GFP, probably because of the loss of the
extrachromosomal array. Scale bar, 50 m.
acids with 51% identity and 70% similarity to
cho-1 (Fig. 2a). The rat cDNA clone was desig-
nated CHT1. The amino-acid sequence of
CHT1 has significant homology with members
of the family of Na+-dependent glucose trans-
porters20 (2025%; Fig. 2b), but no significant
homology with those of the choline transporter
of yeast21, the mammalian creatine transporter
originally identified as a high-affinity choline
transporter22 or any known neurotransmitter
transporters15. The topological model predicted for CHT1 is essen-
tially the same as that of C. elegans CHO-1 (Fig. 2c).
Expression of CHT1 mRNA was examined by Northern blots
and insitu hybridization. Northern blots of various rat tissues
for CHT1 mRNA showed expression of 5-kb transcripts
(Fig. 3a). High mRNA expression was observed in basal fore-
brain, brainstem and spinal cord, and low expression was seen
in striatum. Each of these regions contains cholinergic neurons.
In contrast, no transcripts were detected in other areas of brain or
the non-neuronal tissues examined. Consistent with the North-
ern blots, insitu hybridization demonstrated expression of CHT1
mRNA in major cholinergic cell groups examined, including the
striatum, ventral spinal cord and scattered cell groups in the basal
forebrain (Fig. 3b and c). This distribution was virtually identi-
cal to those for choline acetyltransferase23 and vesicular acetyl-
choline transporter24 mRNA, indicating that CHT1 mRNA was
expressed exclusively in cholinergic neurons.
Choline uptake in Xenopus oocytes injected with CHT1 cRNA
was two- to fourfold greater than the background uptake in con-
trol oocytes injected with water (Fig. 4a). Choline uptake medi-
ated by CHT1 saturated with increasing concentrations of choline
with a Km of 2.2 0.2 M (n = 3; Fig. 4b). The apparent Km of
choline for endogenous uptake in control oocytes was greater
than 10 M (data not shown). Choline uptake mediated by
CHT1 was completely inhibited by less than 0.1 M HC3 with
a Ki of 25 nM (Fig. 4c). By contrast, choline uptake was only
slightly inhibited by 10 M HC3 in control oocytes. Examina-
tion of the ionic dependency of CHT1-mediated choline uptake
revealed that the uptake is not only Na+ dependent, but also Cl
dependent (Fig. 4d). These results show that CHT1-mediated
choline uptake had the expected characteristics of the high-affin-
ity choline uptake in mammalian synaptosomes studies58 (high
121
 2000 Nature America Inc. http://neurosci.nature.com

articles

a b

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

Fig. 2. Deduced amino-acid sequence and topological model for the high-affinity choline transporter. (a) Alignment of the predicted amino-
acid sequence of rat CHT1 and C. elegans CHO-1. Numbers in the right column correspond to amino-acid residues. Identical residues are in
bold. The putative transmembrane domains IXII are underlined. (b) Phylogenetic tree of Na+-dependent glucose transporter family. The tree
was constructed by the neighbor-joining method37 using the CLUSTALW program of National Institute of Genetics (Mishima, Japan). The per-
cent amino-acid identity between rat CHT1 and related proteins is shown to the right. Abbreviations: rbSNST1, rabbit sodium/nucleoside
cotransporter 1 (SwissProt accession number P26430); rSGLT1, rat sodium/glucose cotransporter 1 (P53790); hSMIT1, human sodium/myo-
inositol cotransporter 1 (P53794); putP, E. coli sodium/proline symporter (P07117); panF, E. coli sodium/pantothenate symporter (P16256);
rNIS, rat Na/I symporter (PIR accession number S68513); rSMVT, rat Na+-dependent vitamin transporter (PRF accession number 2413365A).
(c) Predicted topology of rat CHT1 in the membrane. Circles represent individual amino acids. Filled circles indicate residues identical (black)
or highly conserved (gray) with C. elegans CHO-1, and open circles indicate divergent residues. Branched lines indicate potential N-linked gly-
cosylation sites. Circled P indicates consensus site for protein kinase C phosphorylation.

affinity for choline, high sensitivity to HC3 and dependence on
Na+ and Cl ions).
We also examined [3H]HC3 binding activity of membranes
prepared from COS7 cells, which were transfected with CHT1
cDNA or vector only (control). Na +-dependent binding of
[3H]HC3 was detected in membranes from CHT1-transfected
cells but not in membranes from control cells. (Fig. 5a). The equi-
librium dissociation constant (Kd) was estimated to be 1.6 0.2
nM (n = 3; Fig. 5b), which is similar to the values reported for
brain synaptosomes10,25,26. Specific binding of [3H]HC3 was dis-
placed by tenfold lower concentrations of choline than of acetyl-
choline (Fig. 5c). These results indicate that CHT1 is not only a
high-affinity choline transporter, but also a HC3 binding site.
DISCUSSION
In this study, we identified and characterized cDNAs encoding
high-affinity choline transporters in C. elegans and rat, and we
confirmed that the CHT1 has the same characteristics as the
transporter found in cholinergic nerve endings. We isolated the
cDNA clone for the high-affinity choline transporter by using
information provided by the C. elegans Genome Project19 and
systematically expressing putative transporters one by one in
Xenopus oocytes. We then extended this use of C. elegans genom-
ic sequences to isolate a mammalian homolog. A similar strategy
may be applicable to expression cloning of unidentified cDNAs of
biological importance.
Our results may explain why the high-affinity choline trans-
porter was not previously cloned. First, CHT1 has no significant
homology with members of neurotransmitter transporter fami-
ly, precluding the use of homology-based cloning to obtain CHT1
cDNA. CHT1 is not expected to belong to the Na+-dependent glu-
cose transporter family because the Na+-dependent glucose trans-
porter depends on Na+ but not Cl ions20, whereas the high-affinity
choline uptake in synaptosomes as well as other neurotransmitter
transporters depends on both Na+ and Cl ions27. Second, the
transport rate by CHT1 is not very high compared with the rates

122 nature neuroscience volume 3 no 2 february 2000

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

articles

a b

Fig. 3. Expression studies of rat CHT1. (a) Northern analysis of CHT1

mRNA transcripts in rat tissues. The size of standards (0.249.5 kb) are c
shown on the left. (b, c) Insitu hybridization of CHT1 transcripts in rat
brain and spinal cord. Bright-field micrographs of cryosections
hybridized to digoxigenin-labeled antisense cRNA probe. CHT1 mRNA
2000 Nature America Inc. http://neurosci.nature.com

transcripts were detected in the vertical and horizontal limbs of the

diagonal band (VDB, HDB), medial septal nucleus (MS), caudate and
putamen (CPu) and olfactory tubercle (Tu; b). In the spinal cord, expres-
sion was detected in the ventral horn (VH; c). Adjacent sections
hybridized with a probe for the vesicular acetylcholine transporter
showed essentially the same pattern (data not shown). Sense probe
yielded no signal. Scale bars, 1 mm.

of the high-capacity, low-affinity transporters present in Xenopus
oocytes or cultured cells, possibly explaining previous failures to
detect the transporter by expression cloning.
Choline acetyltransferase (ChAT) and vesicular acetylcholine
transporter (VAChT) are two proteins identified as markers for
cholinergic neurons. Their genes are present at the same locus,
with the VAChT gene positioned within the first intron of the
ChAT gene, and their expression is regulated by the same pro-
moter28,29. We identified CHT1 as a third marker protein for
cholinergic neurons. The human CHT1 gene, however, occurred
at a locus different from the one for ChAT and VAChT (data not
shown). It will be interesting to determine the similarities of the
CHT1 promoter region and CHT1 mRNA distribution to those
for ChAT and VAChT.
Earlier studies using mammalian synaptosomes show that
choline taken up through the high-affinity uptake system is effi-
ciently converted to acetylcholine, although choline taken up
through the low-affinity system is not significantly converted to

a b c
CHT1-mediated choline uptake

(fmol/oocyte/30 min)
(pmol/oocyte/30 min)
(pmol/oocyte/30 min)
[3H]Choline uptake

Choline uptake

Water CHT1 [Choline] (M)

log[HC3] (M)

Fig. 4. Functional expression of rat CHT1 in Xenopus oocytes. d

(a) [3H]choline uptake into Xenopus oocytes injected with CHT1 cRNA or
water. Black and gray bars indicate choline uptake measured in standard
(pmol/oocyte/30 min)

medium containing 100 mM NaCl or LiCl, respectively. Each bar represents the
Choline uptake

mean s.e. (n = 68 oocytes). (b) Effect of choline concentrations on CHT1-

mediated choline uptake. The uptake in water-injected oocytes was subtracted
from that in cRNA-injected oocytes, yielding CHT1-mediated choline uptake.
The uptake was fitted to the Michaelis-Menten curve. (c) Inhibition of choline
uptake by hemicholinium-3 (HC3). (d) Na+ and Cl dependence of choline
uptake. Gray and black bars indicate choline uptake into water- or cRNA-
injected oocytes, respectively. NaCl (100 mM) in the standard uptake solution
was replaced by equimolar amounts of the indicated salts. NaCl LiCl KCl RbCl NaBr NaF Nal

nature neuroscience volume 3 no 2 february 2000 123

 2000 Nature America Inc. http://neurosci.nature.com
articles
a b
Specific Z[3H]HC3 binding
(fmol/mg protein)
(fmol/mg protein)
[3H]HC3 binding
pcDNA pcDNACHT1 Free [3H]HC3 (nM)
Fig. 5. Functional expression of rat CHT1 in COS7 cells. (a) Binding of [3H]HC3 to membranes prepared from COS7 cells transfected with
CHT1 cDNA or vector (pcDNA3.1) only. (b) Saturation analysis of specific [3H]HC3 binding. (c) Displacement of specific [3H]HC3 binding by
HC3, choline (Cho) and acetylcholine (ACh). Acetylcholine was tested in the presence of 1 M physostigmine.
2000 Nature America Inc. http://neurosci.nature.com
acetylcholine5,7,8. These findings suggest a close relationship
between high-affinity choline uptake and acetylcholine synthe-
sis30. The cloning of CHT1 provides a means to directly examine
if and how CHT1 and ChAT interact.
Cholinergic neurons are crucial in learning and memory1,2,
and cholinergic deficits are correlated with the severity of
dementia3. High-affinity choline uptake is the rate-limiting
step in acetylcholine synthesis, and its activity is regulated by
neuronal activity11,12 and other stimuli3133. Furthermore, high-
affinity choline uptake or HC3 binding are upregulated in
Alzheimers disease14. The cloning of CHT1 may be important
in discovering the molecular mechanisms underlying these reg-
ulations and in developing new therapeutic strategies for treat-
ing Alzheimers disease.
METHODS
Cloning of transporter cDNAs. Candidate cDNAs for the choline trans-
porter were isolated by reverse transcription-PCR from mixed-stage poly(A)+
RNA, using MarathonTM cDNA Amplification Kit (Clontech, Palo Alto, Cal-
ifornia), following the manufacturers protocols. PCR oligonucleotide for-
ward primers based on DNA sequences obtained from the C. elegans Genome
Project19 were designed to construct putative translation-initiation sites of
the predicted genes. These amplified products were cloned into the Nco I
(filled-in) and Not I sites of modified pSPUTK (Strategene, La Jolla, Cali-
fornia), and the sequences of the inserted DNA were determined. Sequence
data reported in this paper will appear in the DDBJ/EMBL/GenBank
nucleotide sequence databases with the accession numbers AB030946,
AB030947. The rat CHT1 cDNA was isolated from a rat spinal cord cDNA
library using the GeneTrapper cDNA Positive Selection System (Gibco BRL,
Life Technologies, Rockville, Maryland) and a primer based on the sequence
of the cDNA fragment, following the manufacturers protocols. After char-
acterizing the resulting cDNA clones, we subcloned a positive clone into
pSPUTK and pcDNA3.1+ (Invitrogen, Carlsbad, California).
Expression in Xenopus oocytes. SP6 or T7 RNA polymerase in the
presence of cap analog was used to prepare cRNA in vitro. Stage VVI
Xenopus laevis oocytes were injected with 2030 ng capped cRNA. The
uptake was assayed essentially as described34. Choline uptake (23 d
after injection) was measured by incubating 68 oocytes for 3045
min with [3H]choline chloride in 750 l standard medium (100 mM
NaCl, 2 mM KCl, 1 mM MgCl2, 1 mM CaCl2, 10 mM HEPES, 5 mM
Tris, pH 7.4). Oocytes were solubilized with 10% SDS, and the [3H]
content was measured by liquid scintillation counting.
GFP expression construct. The cho-1::gfp transcriptional fusion was con-
structed by PCR essentially as described35. The green fluorescent protein
gene preceded by a nuclear localization signal (NLS) sequence was insert-
ed in-frame, 3 amino acids downstream of the cho-1 translation start
124
c
Specific [3H]HC3 binding
(%)
log[Competitor] (M)
site. NLS and gfp sequences were amplified from pPD104.53 (provided by
A. Fire). To prepare the 5.1-kb region upstream from the cho-1 transla-
tion initiation site, we used a PCR primer that was designed to include the
in-frame third amino-acid codons of cho-1. The rol-6 (su1006) marker
and the constructed DNA were co-injected into the gonads of animals
as described36.
Northern analysis. Poly(A)+RNA (6 g) prepared from different rat tissues
was resolved by formaldehyde-agarose gel electrophoresis, transferred
onto nylon membranes and hybridized to a random-primed [32P]-labeled
cDNA fragment of CHT1 for 16 h at 42C in 50% formamide, 5 SSPE,
5 Denhardts solution, 0.5% SDS and 100 g per ml salmon sperm DNA.
The filters were washed to a final stringency of 0.1 SSPE, 0.1% SDS at
65C and exposed to film with an enhancing screen for 7 days.
Insitu hybridization. Digoxigenin-labeled antisense transcripts were
synthesized in vitro. Transcripts were alkali-hydrolyzed to an average
length of 200400 base pairs and hybridized in situ on cryostat sections
(1020 m) of fresh-frozen tissues. For hybridization, sections were incu-
bated at 45C for 20 h with the labeled cRNA probes (1 g per ml) dis-
solved in 1 Denhardts solution containing 50 mM Tris-HCl (pH 8.0),
2.5 mM EDTA, 0.3 M NaCl, 50% formamide, 10% dextran sulphate and
1 mg per ml E. coli tRNA. Sections were then washed twice in 2
SSC/50% formamide and once in 1 SSC/50% formamide at 45C. The
hybridized probes were visualized using anti-digoxigenin Fab fragments
(Boehringer Mannheim, Mannheim, Germany) and NBT/BCIP sub-
strate. Sections were developed in substrate solution for 2448 h.
Binding assays. [3H]hemicholinium-3 (HC3; 128 Ci per mmol) was
purchased from NEN Life Science Products (Boston, Massachusetts).
COS7 cells were transiently transfected with pcDNA3.1-CHT1 or
pcDNA3.1 using TransFast Reagent (Promega, Madison, Wisconsin)
and following the manufacturers protocols. To prepare membranes,
cells were homogenized in 0.32 M sucrose and centrifuged for 1 h at
200,000 g, and the pellet was resuspended. Binding assays were done
essentially as described 26. Specific binding was calculated by sub-
tracting non-specific binding (determined in the presence of 10 M
HC3) from total binding. Binding-saturation data were analyzed by
nonlinear transformation of the calculated specific [3H]HC3 bind-
ing to generate a Kd value.
ACKNOWLEDGEMENTS
We thank Y. Iino for the C. elegans N2 strain, Y. Kobayashi for help with basal
forebrain preparations, A. Fire for pPD104.53, S. Yamashita for yeast choline-
transporter cDNA, T. Suzuki and Y. Kirino for Torpedo electric lobe and its cDNA
library, Y. Koyama for laterodorsal tegmental nucleus, K. Kameyama for
suggestions and D. Saffen for English corrections. This work was supported by
grants from Japan Science and Technology Corporation (CREST) and the Ministry
of Japan Society for the Promotion of Science (Research for Future Program).
nature neuroscience volume 3 no 2 february 2000
 2000 Nature America Inc. http://neurosci.nature.com
RECEIVED 1 NOVEMBER; ACCEPTED 23 NOVEMBER 1999
1. Winkler, J. et al. Essential role of neocortical acetylcholine in spatial memory.
Nature 375, 484487 (1995).
2. Dutar, P., Bassant, M. H., Senut, M. C. & Lamour, Y. The septohippocampal
pathway: structure and function of a central cholinergic system. Physiol. Rev.
75, 393427 (1995).
3. Bierer, L. M. et al. Neurochemical correlates of dementia severity in
Alzheimers disease: relative importance of the cholinergic deficits. J.
Neurochem. 64, 749760 (1995).
4. Tucek, S. Regulation of acetylcholine synthesis in the brain. J. Neurochem. 44,
1124 (1985).
5. Haga, T. Synthesis and release of [14C]acetylcholine in synaptosomes. J.
Neurochem. 18, 781798 (1971).
6. Yamamura, H. I. & Snyder, S. H. Choline: high-affinity uptake by rat brain
synaptosomes. Science 178, 626628 (1972).
7. Haga, T. & Noda, H. Choline uptake systems of rat brain synaptosomes.
Biochim. Biophys. Acta 291, 564575 (1973).
8. Kuhar, M. J. & Murrin, L. C. Sodium-dependent, high-affinity choline
uptake. J. Neurochem. 30, 1521 (1978).
9. Kuhar, M. J., Sethy, V. H., Roth, R. H. & Aghajanian, G. K. Choline: selective
accumulation by central cholinergic neurons. J. Neurochem. 20, 581593
(1973).
2000 Nature America Inc. http://neurosci.nature.com
10. Happe, H. K. & Murrin, L. C. High-affinity choline transport sites: use of
[3H]hemicholinium-3 as a quantitative marker. J. Neurochem. 60, 11911201
(1993).
11. Simon, J. R. & Kuhar, M. J. Impulse-flow regulation of high affinity choline
uptake in brain cholinergic nerve terminals. Nature 255, 162163 (1975).
12. Murrin, L. C. & Kuhar, M. J. Activation of high-affinity choline uptake in
vitro by depolarizing agents. Mol. Pharmacol. 12, 10821090 (1976).
13. Pascual, J. et al. High-affinity choline uptake carrier in Alzheimers disease:
implications for the cholinergic hypothesis of dementia. Brain Res. 552,
170174 (1991).
14. Bissette, G., Seidler, F. J., Nemeroff, C. B. & Slotkin, T. A. High affinity choline
transporter status in Alzheimers disease tissue from rapid autopsy. Ann. NY
Acad. Sci. 777, 197204 (1996).
15. Nelson, N. The family of Na+/Cl neurotransmitter transporters. J.
Neurochem. 71, 17851803 (1998).
16. Berrard, S. et al. cDNA cloning and complete sequence of porcine choline
acetyltransferase: in vitro translation of the corresponding RNA yields an
active protein. Proc. Natl. Acad. Sci. USA 84, 92809284 (1987).
17. Alfonso, A. et al. The Caenorhabditis elegans unc-17 gene: a putative vesicular
acetylcholine transporter. Science 261, 617619 (1993).
18. Wurtman, R. J. Choline metabolism as a basis for the selective vulnerability of
cholinergic neurons. Trends Neurosci. 15, 117122 (1992).
19. The C. elegans Sequencing Consortium. Genome sequence of the nematode
C. elegans: a platform for investigating biology. Science 282, 20122018
(1998).
20. Hediger, M. A. & Rhoads, D. B. Molecular physiology of sodium-glucose
cotransporters. Physiol. Rev. 74, 9931026 (1994).
nature neuroscience volume 3 no 2 february 2000
articles
21. Nikawa, J., Hosaka, K., Tsukagoshi, Y & Yamashita, S. Primary structure of
the yeast choline transport gene and regulation of its expression. J. Biol.
Chem. 265, 1599616003 (1990).
22. Schloss, P., Mayser, W. & Betz, H. The putative rat choline transporter
CHOT1 transports creatine and is highly expressed in neural and muscle-
rich tissues. Biochem. Biophys. Res. Commun. 198, 637645 (1994).
23. Oh, J. D. et al. Cholinergic neurons in the rat central nervous system
demonstrated by in situ hybridization of choline acetyltransferase mRNA.
Neuroscience 47, 807822 (1992).
24. Roghani, A. et al. Molecular cloning of a putative vesicular transporter for
acetylcholine. Proc. Natl. Acad. Sci. USA 91, 1062010624 (1994).
25. Vickroy, T. W., Roeske, W. R. & Yamamura, H. I. Sodium-dependent high-
affinity binding of [3H]hemicholinium-3 in the rat brain: a potentially
selective marker for presynaptic cholinergic sites. Life Sci. 35, 23352343
(1984).
26. Sandberg, K. & Coyle, J. T. Characterization of [3H]hemicholinium-3
binding associated with neuronal choline uptake sites in rat brain
membranes. Brain Res. 348, 321330 (1985).
27. Kuhar, M. J. & Zarbin, M. A. Synaptosomal transport: a chloride dependence
for choline, GABA, glycine and several other compounds. J. Neurochem. 31,
251256 (1978).
28. Erickson, J. D. et al. Functional identification of a vesicular acetylcholine
transporter and its expression from a cholinergic gene locus. J. Biol. Chem.
269, 2192921932 (1994).
29. Bejanin, S., Cervini, R., Mallet, J. & Berrard, S. A unique gene organization
for two cholinergic markers, choline acetyltransferase and a putative
vesicular transporter of acetylcholine. J. Biol. Chem. 269, 2194421947
(1994).
30. Barker, L. A. & Mittag, T. W. Comparative studies of substrates and
inhibitors of choline transport and choline acetyltransferase. J. Pharmacol.
Exp. Ther. 192, 8694 (1975).
31. Vogelsberg, V., Neff, N. H. & Hadjiconstantinou, M. Cyclic AMP-mediated
enhancement of high-affinity choline transport and acetylcholine synthesis
in brain. J. Neurochem. 68, 10621070 (1997).
32. Beeri, R. et al. Enhanced hemicholinium binding and attenuated dendrite
branching in cognitively impaired acetylcholinesterase-transgenic mice. J.
Neurochem. 69, 24412451 (1997).
33. Kar, S. et al. Amyloid -peptide inhibits high-affinity choline uptake and
acetylcholine release in rat hippocampal slices. J. Neurochem. 70, 21792187
(1998).
34. Kanai, Y. & Hediger, M. A. Primary structure and functional
characterization of a high-affinity glutamate transporter. Nature 360,
467471 (1992).
35. Cassata, G. et al. Rapid expression screening of Caenorhabditis elegans
homeobox open reading frames using a two-step polymerase chain reaction
promoter-gfp reporter construction technique. Gene 212, 127135 (1998).
36. Mello, C. C., Kramer, J. M., Stinchcomb, D. & Ambros, V. Efficient gene
transfer in C. elegans extrachromosomal maintenance and integration of
transforming sequences. EMBO J. 10, 39593970 (1991).
37. Saitou, N. & Nei, M. The neighbor-joining method: a new method for
reconstructing phylogenetic trees. Mol. Biol. Evol. 4, 406425 (1987).
125
 2000 Nature America Inc. http://neurosci.nature.com

articles

Receptors with opposing functions

are in postsynaptic microdomains
under one presynaptic terminal
Guoshan Tsen1, Brian Williams1, Pauline Allaire2, Yu-Dong Zhou1, Ognian Ikonomov3,
Ivanela Kondova4 and Michele H. Jacob1

1 Department of Neuroscience, Tufts University School of Medicine, 136 Harrison Ave., Boston, Massachusetts 02111, USA
2 Blackstone-Millville Regional High School, 175 Lincoln St., Blackstone, Massachusetts 01504, USA
3 Dept. of Psychiatry, Wayne State University Med. Ctr., 540 E. Canfield St., Detroit, Michigan 48201, USA
4 Department of Infectious Disease, Tufts University Veterinary School, 200 Westboro Rd., Grafton, Massachusetts 01536, USA
2000 Nature America Inc. http://neurosci.nature.com

The first two authors contributed equally to this work.

Correspondence should be addressed to M.H.J. (mjacob@opal.tufts.edu)

Fast excitatory synaptic transmission through vertebrate autonomic ganglia is mediated by postsynaptic
nicotinic acetylcholine receptors (nAChRs). We demonstrate a unique postsynaptic receptor microhetero-
geneity on chick parasympathetic ciliary ganglion neuronsunder one presynaptic terminal, nAChRs and
glycine receptors formed separate but proximal clusters. Terminals were loaded with [3H]glycine via the
glycine transporter-1 (GlyT-1), which localized to the cholinergic presynaptic terminal membrane;
depolarization evoked [3H]glycine release that was calcium independent and blocked by the GlyT-1
inhibitor sarcosine. Ganglionic synaptic transmission mediated by nAChRs was attenuated by glycine.
Coexistence of separate clusters of receptors with opposing functions under one terminal contradicts
Dales principle and provides a new mechanism for modulating synaptic activity in vivo.

Nicotinic acetylcholine receptors mediate fast excitatory synap- RESULTS

tic transmission through all ganglia in the vertebrate periph-
eral nervous system (PNS). It is thought that this synapse
functions as a simple relay, and that modulation occurs cen-
trally through the integration of excitatory and inhibitory
inputs to the preganglionic neurons. However, functional
inhibitory glycine receptors (GlyRs) were found on embry-
onic chick parasympathetic ciliary ganglion (CG) neurons1,
raising questions as to their spatial distribution and physio-
logical significance. These issues are particularly intriguing,
as each ciliary neuron is innervated by only one presynaptic
terminal, a large calyx-type ending2,3. Moreover, the terminal
is cholinergic and is derived from the sole source of presy-
naptic input to the CG, the accessory oculomotor nucleus
(AON)28. Multiple synaptic contacts are formed in the region
of apposition between the calyx ending and the postsynaptic
neuron. According to the current interpretation of Dales prin-
ciple, all of the active zones of the presynaptic terminals
derived from one axon release the same combination of neu-
rotransmitters, and all of the underlying specialized postsy-
naptic membrane regions on one neuron are predicted to
contain the same receptor types9,10. We show here that this is
not the case. We demonstrate a unique postsynaptic receptor
microheterogeneity: under a single presynaptic terminal, exci-
tatory nAChR clusters and separate but proximal inhibitory
GlyR clusters are localized in discrete membrane
microdomains. These findings are surprising, considering that
these receptors respond to different fast-acting transmitters
and have opposing actions.
On chick ciliary neurons, nAChRs are localized predominantly
in the postsynaptic membrane (Fig. 1)1113. We showed this at
the ultrastructural level with immunocytochemistry using mon-
oclonal antibody (mAb) 35 to nAChRs, visualized with horse-
radish peroxidase (HRP). In this study, nAChR refers strictly to
the 3-containing nAChRs, which are concentrated in the post-
synaptic membrane and are specifically recognized by mAb 35,
and does not include the 7-containing nAChRs, which are
perisynaptic, excluded from the postsynaptic membrane and not
recognized by mAb 35 (refs. 13-16). We defined synapses by char-
acteristic morphological featuresa thickened and parallel
arrangement of the pre-and postsynaptic membranes, an accu-
mulation of synaptic vesicles adjacent to the presynaptic density,
an enhanced postsynaptic density and a widened synaptic cleft.
Concentrations of nAChRs are found in patches of postsynaptic
membrane demarcated by the postsynaptic density along their
cytoplasmic face and lying opposite the presynaptic density of
the active zone. However, labeling for nAChRs was not found in
all of the specialized postsynaptic membrane microregions under
one calyx ending (Fig. 1a). This selective staining pattern was
unlikely to result from limited penetration of labeling reagents, as
unlabeled postsynaptic membrane microregions were close to
labeled ones.
GlyRs were concentrated in similar specialized postsynaptic
membrane microdomains on late-stage embryonic ciliary neu-
rons as shown by immunolabeling with either mAb 4a or mAb
2b and ultrastructural analysis (Fig. 1b). On each neuron exam-
ined, some, but not all, membrane microregions demarcated by

126 nature neuroscience volume 3 no 2 february 2000

 2000 Nature America Inc. http://neurosci.nature.com
Fig. 1. Ultrastructural localization of nAChRs, GlyRs
and gephyrin at discrete postsynaptic membrane a
microdomains under the calyx-type presynaptic ter-
minal on E1517 chick ciliary neurons in vivo. The
three components were present at only a subset of
the postsynaptic membrane microdomains demar-
cated by the postsynaptic density along their cyto-
plasmic faces and lying opposite the synaptic vesicle
accumulations at the active zone (labeled synapse, b c
black arrowhead; unlabeled synapse, white arrow-
head). (a) nAChRs detected by immunolabeling with
mAb 35. (b) GlyRs labeled with mAb 2b. (c) gephyrin
labeled with mAb 7a (visualized with HRP). Scale bar,
1 m. (d) Schematic representation of the relative
distribution of nAChRs, GlyRs and gephyrin at spe-
cialized postsynaptic membrane microdomains under d
the calyx terminal (nt; synaptic vesicles, sv; short den-
drites, d; based on data in Figs. 1 and 2).
2000 Nature America Inc. http://neurosci.nature.com
the postsynaptic density were labeled for GlyRs.
The presence of both GlyR and nAChR accu-
mulations in postsynaptic membrane
microdomains under the cholinergic calyx ter-
minal was unexpected.
As gephyrin is a peripheral membrane pro-
tein known to codistribute with GlyRs at synapses and is required
for GlyR clustering on rodent CNS neurons1820, we used it as a
positive control to examine the relative distribution of GlyRs and
nAChRs on avian ciliary neurons. We first determined that
gephyrin was expressed in avian CG neurons and that the avian
and rodent sequences were highly conserved, suggesting that
mAbs to rodent gephyrin could be used for localization studies
in chick. Specifically, we isolated the full-length gephyrin coding
sequence from a chick brain cDNA library (GenBank accession
number AF174130). The avian and rodent sequences showed
87% identity at the nucleotide level and 98% identity at the
amino-acid level. We then examined gephyrin mRNA expression
in the chick CG by using RT-PCR amplification of total gan-
glionic RNA. Because five alternatively spliced variants of
gephyrin are expressed in the rodent in a tissue-specific pattern21,
we used seven different pairs of gephyrin-specific primers to flank
all of the known alternatively spliced cassettes identified in rodent
cDNA (see Methods). We exclusively detected the gephyrin
mRNA isoform containing the C2 cassette in the chick CG, as
determined by size and sequence analysis of the PCR amplifica-
tion products (n = 14 separate experiments; data not shown).
This is also the dominant isoform expressed in the rodent CNS21.
Immunocytochemical labeling of avian ciliary neurons with
anti-gephyrin mAb 5a or mAb 7a20,22 and ultrastructural analysis
demonstrated gephyrin distribution similar to that of the recep-
tors, with labeling present at only a subset of the specialized post-
synaptic membrane microdomains under the calyx (Fig. 1c). A
minor difference was that gephyrin staining was concentrated at
the postsynaptic density rather than at the postsynaptic mem-
brane. To determine the synaptic distribution of nAChRs, GlyRs
and gephyrin relative to one another, we used immunofluores-
cent double labeling with laser-scanning confocal microscopy. In
both late-stage embryonic and adult CGs, GlyR (red) and
gephyrin (green) clusters were colocalized on the neuron surface,
as indicated by the predominance of yellow fluorescent patches
(Fig. 2a). In contrast, nAChR (red) and gephyrin (green) clus-
ters did not overlap, as indicated by distinct patches of red or
green fluorescence (Fig. 2b). Separate nAChR and GlyR/gephyrin
nature neuroscience volume 3 no 2 february 2000
articles
clusters were present in approximately equal proportions on all
CG neurons at stages after pre-and postganglionic synapse for-
mation, ranging from late-stage embryonic ages to reproduc-
tively mature adults. Functional GlyRs were present on CG
neurons at earlier embryonic ages as well, but their expression
was developmentally delayed relative to functional nAChRs1. As
independent confirmation of the GlyR and gephyrin colocaliza-
tion results, we showed by the yeast two-hybrid genetic assay that
chick gephyrin specifically interacts with the chick GlyR -sub-
unit, but not with nAChR subunits (Fig. 2c). We tested for
gephyrin interactions with the major cytoplasmic loop of the
receptor subunits, as the gephyrin binding motif in the rodent
GlyR is a linear 18-amino-acid (aa) sequence in the -subunit
long cytoplasmic loop23,24. We established that this 18-aa sequence
was completely conserved in the chick GlyR -subunit (GenBank
accession number AF181717) as determined by RT-PCR ampli-
fication of CG total RNA using specific primers corresponding
to the mammalian GlyR -subunit sequence. The gephyrin bind-
ing motif was absent from neuronal nAChR subunits. Only yeast
transformants co-expressing the chick GlyR -subunit loop and
gephyrin showed transcriptional activation of the two reporter
genes, HIS3 and LacZ (Fig. 2c). Altogether, the colocalization
studies revealed a synaptic microheterogeneity consisting of sep-
arate but proximal clusters of nAChRs and of GlyRs in discrete
postsynaptic membrane microdomains under one presynaptic
terminal (Fig. 1d). The protein-interaction data suggest that dif-
ferent molecules organize the nAChR and GlyR clusters in the
distinct postsynaptic microdomains.
To determine whether this synaptic heterogeneity was a
unique feature of a calyx-type synapse, we examined the choroid
neuron subpopulation in the chick CG, as they are innervated
by multiple typical bouton-type endings. Each choroid neuron
receives inputs from two to three preganglionic neurons in the
AON3. Separate clusters of nAChRs and of GlyRs (and the GlyR-
associated protein gephyrin) were detected in distinct postsy-
naptic membrane regions under separate bouton terminals
(Fig. 2d). Although there may be separate glycinergic and cholin-
ergic inputs to the choroid neurons, this seems unlikely, as all
127
 2000 Nature America Inc. http://neurosci.nature.com

articles

presynaptic terminals to the CG are derived from the AON and
are cholinergic28. Thus, as for the calyceal synapse, heteroge-
neous microdomains occur under axonal branches terminating in
multiple bouton-type endings.
We examined the functional effects of GlyR activation in
CG neurons. Using acutely isolated, intact E1517 ganglia, we
stimulated the preganglionic nerve and extracellularly record-
ed postsynaptic responses with or without exogenous glycine
(0.51 mM, Fig. 3). Glycine reduced initial slope of the nAChR-
mediated excitatory postsynaptic potential (EPSP; Fig. 3b and c)
by 30% and peak-to-peak amplitude of the postganglionic
compound action potential (a.p.) by 65% (data not shown),
but did not change the preganglionic a.p. volley as indicated
by the electrical coupling potential (Fig. 3b). This inhibition
was completely reversed following washout of glycine. Strych-
nine (110 M), a GlyR antagonist, was not used to test for
GlyR involvement because it also blocks nAChRs on CG neu-
rons (data not shown)25. Compared with glycine, hexametho-
2000 Nature America Inc. http://neurosci.nature.com
To study the functional significance of inhibitory GlyR accu-
mulations at these largely cholinergic synapses, we investigated
potential sources of endogenous glycine in the CG and conditions
that lead to its release. Glycine is present in blood at a high con-
centration (0.5 mM)29. Glycine transporters (GlyTs) provide an
avenue for glycine uptake, but operate in reverse to release glycine
under certain physiological conditions (such as strong depolar-
ization and high internal sodium concentration)30. We established
that the two GlyTs, GlyT-1 and GlyT-2, were present in the chick
CG at both late embryonic and adult stages based on immuno-
labeling and uptake studies. The two transporters had different
distributions as determined by anti-GlyT-1 and anti-GlyT-2 spe-
cific antisera and immunofluorescence microscopy (Fig. 4a and b).
GlyT-1 (green) was localized predominantly in patches on the
presynaptic terminal surface membrane and colocalized with
synaptic-vesicle immunolabeling (red) in the late-stage embry-
onic and adult ganglia (Fig. 4a). GlyT-1 labeling was also present
on neuronal cell bodies, but not on Schwann cells. In contrast,

nium (0.5 mM), anAChR antagonist, decreased initial slope of
the EPSP by 70% (Fig. 3a and c) and the compound a.p. ampli-
tude by 80%, consistent with earlier reports of partial block by
hexamethonium but complete inhibition of nAChR-mediated
excitatory responses in the CG by other nicotinic cholinergic
antagonists26. Tetrodotoxin (1 M) abolished all postsynaptic
responses (Fig. 3a and c). Our results were similar to those of
previous electrophysiological studies of CG neurons showing
that activation of the chloride-selective GlyR channel drives
the cell toward the chloride equilibrium potential (near the
resting potential and away from a.p. threshold) and that acti-
vation of the GABAA receptor, another chloride-selective chan-
nel, completely blocks synaptic transmission through the CG at
the embryonic age tested here1,27,28. In sum, GlyR activation
inhibits excitatory synaptic activity in the late-stage embryon-
ic chick CG.
GlyT-2 labeling (red) was detected only on Schwann cells (Fig. 4b).
It should be noted that the localization of GlyT-1 on presynaptic
nerve terminals and GlyT-2 on Schwann cells in this avian periph-
eral ganglion is reversed from their typical distribution in the
mammalian CNS. However, reported exceptions include the reti-
na, where GlyT-1 is expressed on nerve terminals but not on glia,
and the cerebellum, where GlyT-2 is on both glia and terminal
boutons31. In addition, we determined that, in the chick CG, GlyT-
1 and GlyT-2 facilitated uptake of exogenous glycine. Acutely iso-
lated, intact E1517 ganglia were treated with collagenase to
increase reagent access, incubated with [3H]glycine (1 M), rinsed
and processed either for scintillation counting or light-micro-
scopic autoradiographic analysis. In addition, the intact ganglia
were incubated with FM1-43 to label synaptic vesicles and there-
by identify presynaptic terminals. Adjacent sections were processed
for fluorescence or autoradiography. We observed dense clusters

a b c d

Yeast two-hybrid assay

Gephyrin interaction with

Fig. 2. Segregation of nAChR clusters from clusters of GlyRs and their

directly associated protein gephyrin in postsynaptic membrane
microdomains, including those under separate cholinergic bouton-type
terminals on CG choroid neurons. (a) Confocal micrograph of a single
E21 CG neuron showing colocalization of GlyR (red) and gephyrin (green) clusters on the neuron surface (predominance of yellow fluorescent
patches) by immunofluorescent double labeling. In addition, there were a few gephyrin clusters (green patches) lacking in GlyR staining. Either these
sites may represent a distinct subset of gephyrin clusters or the plane of section in these regions may favor the intracellular postsynaptic density
(gephyrin localization) and show less of the postsynaptic surface membrane (GlyR localization). Only 2 optical sections (each 0.5 m) were summed
in this image; including additional optical sections resulted in complete overlap, yielding only yellow patches. (b) In contrast, nAChR (red) and gephyrin
(green) clusters did not overlap on the E21 CG neuron surface (distinct red or green patches). Scale bar, 10 m. (c) Yeast two-hybrid genetic assay
showing chick gephyrin specifically interacts with chick GlyR -subunit long cytoplasmic loop, but not with nAChR subunit loops. For the -gal assay,
+++, blue color within 1 h; , not blue after overnight. For the HIS3 assay, +, overnight growth; no growth on histidine-deficient media. (d) Electron
micrograph showing nAChRs at the specialized postsynaptic membrane microdomains under a subset of bouton-type endings (nt) on a choroid neu-
ron in the E1517 CG in vivo (black arrowhead, labeled synapse; white arrowhead, unlabeled synapse). Scale bar, 1.2 m.

128 nature neuroscience volume 3 no 2 february 2000

 2000 Nature America Inc. http://neurosci.nature.com
a b
0.2 mV
3 ms
Fig. 3. Inhibitory GlyR response in CG neurons. (a) Schematic showing
the position of the recording electrode used to measure postsynaptic
responses extracellularly following direct electrical stimulation of the
preganglionic nerve in an acutely isolated, intact E1517 CG. Top trace
2000 Nature America Inc. http://neurosci.nature.com
(control) shows a typical postsynaptic response composed of the cou-
pling potential (); due to direct electrical connection between pre- and
postsynaptic cells, followed with a slight delay by the chemically medi-
ated field EPSP ()4,5. Bath-applied hexamethonium (0.5 mM) decreased
the initial slope of the field EPSP, with no detectable alteration in the
coupling potential. Tetrodotoxin (1 M) abolished all postsynaptic
responses. (b) Representative trace showing that exogenous glycine (0.5
mM) reduced the initial slope of the nAChR-mediated EPSP (), with no
change in the electrical coupling potential (). (c) Histogram of EPSP
initial slope values expressed as the ratio of test over control (mean
s.e.; number of postsynaptic responses measured is shown in parenthe-
ses above the bars; **p < 0.005, ***p < 0.001, Students t-test.
of silver grains representing [3H]glycine uptake over presynaptic
terminals and Schwann cells, with fewer grains over the neuronal
cell bodies (Fig. 4c). Pre- and co-incubation of CGs with sarco-
sine, a selective GlyT-1 inhibitor 32,33, dramatically reduced
[3H]glycine uptake in a concentration-dependent manner; at 0.5,
1 and 5 mM, sarcosine decreased [3H]glycine levels per CG by
50%, 60% and 70%, respectively. Moreover, in the presence of 5
mM sarcosine, fewer silver grains were observed over presynap-
tic terminals and neuronal somata, whereas label intensity over
Schwann cells was not detectably altered (Fig. 4d). In compari-
son, the addition of 1,000-fold excess cold glycine reduced
[3H]glycine levels per CG by 80% and drastically lowered the
number of silver grains over terminals, neuronal somata and
Schwann cells (Fig. 4e). Depolarization of these radiolabeled gan-
glia with 30 mM KCl induced the release of [3H]glycine (Fig. 4f).
The release was Ca2+ independent and blocked by the GlyT-1
antagonist sarcosine or an excess of cold glycine (Fig. 4g), consis-
tent with a non-vesicular, transport-mediated mechanism. Alto-
gether, these results suggest that GlyT-1 on the presynaptic
terminals probably mediates depolarization-induced glycinergic
synaptic transmission in the chick CG.
DISCUSSION
We demonstrated separate clusters of excitatory nAChRs and
inhibitory GlyRs in distinct postsynaptic membrane
microdomains under one presynaptic terminal in vivo. As these
receptors respond to different fast-acting transmitters and gen-
erally have opposing functions, the presence of both GlyRs and
nAChRs under one presynaptic ending is surprising. Glycine
nature neuroscience volume 3 no 2 february 2000
articles
0.1 mV
1 ms
c
attenuated synaptic transmission mediated by nAChRs in the
CG. We propose that two distinct release mechanisms, vesicular
acetylcholine release and non-vesicular, transport-mediated
glycine efflux, both occur at the calyceal synapse on ciliary neu-
rons in vivo and jointly regulate ganglionic transmission.
Our data suggest that GlyT-1 on the cholinergic presynaptic ter-
minal mediates depolarization-induced glycinergic synaptic trans-
mission in the CG. Amino-acid transporters on presynaptic
terminals release high levels of transmitter in response to depolar-
ization and can function as a relatively fast release mechanism34. The
depolarization protocol required to activate GlyT-1-mediated efflux
in the CG in vivo is not yet defined. It should be noted that nAChRs
composed of 7 subunits are present on the presynaptic calyx end-
ing and may contribute to depolarization of the terminal35. CG GlyR-
rich synaptic microregions do not lack presynaptic vesicles, as
previously reported for synapses with strictly transporter-mediated
release. These microregions lack classical structural features of
inhibitory synapses such as pleomorphic or flattened vesicles; instead,
they have small, clear, round presynaptic vesicles, resembling the
neighboring cholinergic synaptic microregions30,36. Strikingly simi-
lar to the co-existence of vesicular acetylcholine release and trans-
port-mediated glycine efflux suggested by our observations at
synapses in the CG is the demonstration that retinal starburst
amacrine cells corelease ACh and GABA via vesicular and carrier-
mediated mechanisms, respectively37. The two release mechanisms
differ in their extracellular Ca2+ requirements and in the ranges of
membrane voltages over which they are active30,34,37, suggesting that
dynamic shifts in the release of transmitters with opposing func-
tions from single terminals in the CG are possible.
129
 2000 Nature America Inc. http://neurosci.nature.com

articles

Fig. 4. Glycine transporter localization

and glycine uptake and release in the CG. a b c d
(a) Immunofluorescence micrograph of a
single double-labeled E16 CG neuron
showing GlyT-1 (green) predominantly
on presynaptic terminals, where it par-
tially overlapped with synaptic vesicles
(SV2 antigen, red; overlap, yellow). GlyT-1
was on the neuronal soma, but not on
Schwann cells. (b) In contrast, GlyT-2 e
(red) was localized only on Schwann cells
and did not overlap with synaptic-vesicle
staining (green). Scale bar, 10 m.
(ce) Autoradiograms showing exoge-
f g
nous [3H]glycine uptake sites in the
acutely isolated E1517 CG. (c) Dense
clusters of silver grains were found over

Release per ganglion

Release per ganglion

(ratio test: control)

(ratio test: control)

presynaptic terminals and Schwann cells,

with fewer grains over neuronal soma.
(d) Pre- and co-incubation with sarco-
2000 Nature America Inc. http://neurosci.nature.com

sine, a selective GlyT-1 inhibitor, sharply

reduced the number of silver grains over
terminals and neuronal somata, whereas
the grain density over Schwann cells was
unaltered. (e) Addition of 1,000-fold
excess cold glycine drastically lowered
silver grain levels over all CG areas.
Sections were Nissl stained to mark neu-
rons and Schwann cells. Scale bar, 30 m.
(f) Depolarization (using 30 mM KCl)
induced [3H]glycine release from a pre-
loaded ganglion. (g) Amount of release
was unaltered by adding Cd2+ (0.1 mM) or reducing Ca2+ levels (from 5.4 mM to 0.2 mM), whereas sarcosine (5 mM) and excess cold glycine (1 mM)
decreased high-K+-induced [3H]glycine release. Data are expressed as the mean s.e. of the ratio of the proportion of total ganglionic [3H]glycine
released into the medium in test over control conditions. For each condition, n = 6. *p < 0.02, Students two-tailed t-test.

The receptor arrangement reported here represents an unex-
pected degree of complexity and heterogeneity of functionally
specialized subdomains in the neuronal postsynaptic membrane.
Postsynaptic receptor microheterogeneity may have relevance to
the mammalian CNS as well, as suggested by the demonstration
that two fast-acting excitatory and inhibitory neurotransmitters
are coreleased by a single presynaptic CNS neuron, possibly at
the same synapse, in vitro38. The presence under one terminal of
two distinct receptor types responding to different fast trans-
mitters with opposing functions may represent an important
mechanism for modulating activity at a monosynaptic connec-
tion in vivo.
METHODS
Electron microscopy. The localization of nAChRs and GlyRs in postsy-
naptic membrane microdomains on the CG neuron surface was estab-
lished at the ultrastructural level. Briefly, CGs were dissected from White
Leghorn chick embryos in ovo (Spafas) at E1517, incubated in 0.1 M
primary mAb followed by 0.1 M biotinylated secondary antibody and
strepavidinbiotinHRP complex (Vector Labs, Burlingame, California),
fixed, reacted for peroxidase activity, and processed for electron
microscopy1113. The nAChRs were detected by using rat mAb 35, and
GlyRs were detected using mouse mAb 4a or 2b (provided by Heinrich
Betz, Max-Planck Institute for Brain Research, Frankfurt, Germany)17.
As a control, CGs were incubated in the absence of primary mAb. Thin
sections were stained with 5% aqueous uranyl acetate and examined with
a Philips CM10 electron microscope (FEI Company, Hillsboro, Oregon).
Anti-nAChR mAb 35 from a hybridoma isolated by Jon Lindstrom and
colleagues (University of Pennsylvania, Philadelphia, Pennsylvania) was
purified as described39.
The ultrastructural distribution of gephyrin in the chick CG was estab-
lished using the anti-rat gephyrin mAbs 5a and 7a (gift from Heinrich
Betz, Max-Planck Institute)20,22. To gain intracellular access for gephyrin
immunolabeling, freshly dissected E1517 CGs were lightly fixed, deter-
gent-permeabilized with saponin12, incubated with mAb 5a or 7a and
processed as described above.
Laser-scanning confocal microscopy. Double-labeling immunofluores-
cence and laser-scanning confocal microscopy (Leica TCS) were done as
described40,41. Both late-stage embryonic (E15-E21) and adult CGs were
analyzed. (Freshly decapitated heads of reproductively mature adult
White Leghorn chickens were obtained from a local fresh poultry sup-
plier.) The nAChRs were detected by mAb35 and Cy-3-conjugated anti-
rat IgG, gephyrin by mAb 7a and FITC-conjugated anti-mouse IgG, and
GlyRs by mAb 2b and Cy3-conjugated anti-mouse IgG (primary mAbs
used at 1:1001:200 dilution; secondary IgGs used at 1:1000; Vector Labs;
Jackson ImmunoResearch Labs, West Grove, Pennsylvania). The anti-
rat and anti-mouse IgGs were affinity purified to eliminate cross-reac-
tivity with mouse and rat primary mAbs, respectively. Double-labeling
with the GlyR and gephyrin mouse mAbs was carried out using a previ-
ously described protocol for multiple staining with antibodies raised in
the same species42. As controls to test for specific binding, we omitted
the first or second primary mAb and reversed the antibody sequence in
separate tests.
Light microscopy. To establish the cellular localization of GlyTs in
E15E17 and adult CGs, frozen ganglionic sections (68 m thickness)12
were labeled with goat anti-GlyT-1 or guinea pig anti-GlyT-2 specific
antisera (1:200 and 1:500 dilution, respectively; Chemicon Internation-
al, Temecula, California) and FITC-conjugated anti-goat IgG (1:500; Vec-
tor Labs) and Cy3-conjugated anti-guinea pig IgG (1:1000; Jackson
ImmunoResearch Labs). To determine the distribution of GlyTs relative

130 nature neuroscience volume 3 no 2 february 2000

 2000 Nature America Inc. http://neurosci.nature.com
to synapses, synaptic vesicles were stained using anti-SV2 mAb (1:200
dilution; gift from Kathy Buckley, Harvard Medical School)43 and FITC-
or TRITC-conjugated anti-mouse IgG (1:1000 dilution; Vector Labs).
Isolation and sequencing of chick gephyrin cDNA. The full-length chick
gephyrin clone was isolated by screening a White Leghorn E13 chick
brain gt10 cDNA library (2 106 pfu) with a [32P]dCTP-labeled rat
gephyrin cDNA probe (908 bp) using standard procedures44 (chick brain
cDNA library was provided by Barbara Ranscht, The Burnham Institute,
La Jolla, California and the rat gephyrin clone by Heinrich Betz, Max-
Planck Institute. We isolated a single positive plaque with an insert size of
3.1 kb. The identity of the insert as gephyrin was established by Southern
blot hybridization to rat gephyrin cDNA as probe and by sequence analy-
sis (Sequenase Version 2.0 DNA sequencing kit, United States Biochem-
ical Corporation, Cleveland, Ohio) after subcloning the full-length chick
cDNA and smaller restriction enyzme digestion fragments into Blue-
script II SK vector (Stratagene, La Jolla, California).
RT-PCR analysis of gephyrin alternative-splice variants. Total RNA was
extracted from CGs (at selected ages ranging from E9 to E15) using Tri
2000 Nature America Inc. http://neurosci.nature.com
Reagent (Molecular Research Center, Cincinnati, Ohio) according to the
manufacturers instructions. Using an amount of total RNA equivalent
to that present in a single CG, RT-PCR was done as described45. Seven
different pairs of gephyrin specific primers were used to characterize the
alternatively spliced variants expressed in the CG: to flank the C1 and
C2 cassettes, we used sense (nt 315334) and antisense (nt 552571)21
primers; to flank C3, sense (nt 615634 or nt 780797) and antisense
(nt 10111028) primers; to flank C4, sense (nt 10111028) and antisense
(nt 12041223 or nt 12151232) primers; to flank C3 and C4, sense
(nt 615634 or 780797) and antisense (nt 12041223) primers. Ampli-
fication products were separated by 2% agarose-gel electrophoresis and
stained with ethidium bromide. Identities of the products were confirmed
by Southern blot hybridization with a chick gephyrin cDNA probe and
sequence analysis.
Yeast two-hybrid assay. Yeast two-hybrid genetic assay was done as
described46,47. Bait constructs were expressed as recombinant peptide
fragments fused to the DNA-binding protein lexA, using pBTM116 vec-
tor (gift from Peter Pryciak, University of Massachusetts Medical Cen-
ter). Target constructs were expressed as a fusion peptide with the Gal4
activation domain protein, using the pAD-GAL4 vector (Stratagene). To
test for interactions, chick full-length gephyrin and the major cytoplasmic
loop of either chick GlyR or nAChR 3, 5, 4, 2 or 7 were co-
expressed in the yeast L40 reporter strain as bait and target constructs,
respectively, and in the reverse orientations in separate experiments. As
a negative control, single transformants were tested for transcriptional
activation of reporter genes, HIS3 and LacZ.
Electrophysiology. Extracellular recordings were obtained from the intact
CG preparation using described techniques with minor modifications27.
Briefly, the E1517 CG was dissected with the pre-and postganglionic
nerve stumps attached, mounted on a perfusion recording chamber and
continuously perfused at about 2 ml per min with oxygenated recording
buffer. A bipolar tungsten stimulation electrode was used to stimulate
the preganglionic nerve stump with current pulses (1020 V, 150 s) at
a frequency of 0.1 Hz. A 35 M glass microelectrode filled with nor-
mal Krebs solution was used to extracellularly record the field EPSP and
compound a.p. triggered by synaptic transmission. The effect of drugs
on the initial slope of the field EPSP and peak-to-peak amplitude of the
compound a.p. were measured. Drugs were added directly into the per-
fusate. Each trace was averaged 510 times. Glycine, strychnine, hexam-
ethonium and tetrodotoxin were obtained from Sigma.
Rough calculations predict that, in the presence of 1 mM exogenous
glycine, CG neuron [Na+]i may increase at most by 1.5 mM due to co-
transport via GlyT-1. This change in [Na+]i is not expected to affect the
driving force for the nAChR current.
Glycine uptake and autoradiographic analysis. For [3H]glycine uptake
studies, E1517 CGs were dissected into oxygenated recording buffer,
incubated with collagenase type A (1 mg per ml, Boehringer
nature neuroscience volume 3 no 2 february 2000
articles
Mannheim Biochemicals, Indianapolis, Indiana) in divalent-free
buffer 28 at 37C for 10 min, rinsed and incubated with 1 10 6 M
[3H]glycine (41.1 Ci per mmol; New England Nuclear, Boston, Mass-
achusetts) in oxygenated recording buffer at 37C for 1 h followed by
rinsing in buffer four times. To determine the total number of counts
per CG, single ganglia were dried on Whatman filter paper (2.4 cm)
and counted using a scintillation counter (LS 5000TD, Beckman Coul-
ter, Fullerton, California). To localize [3H]glycine uptake sites, other
radiolabeled CGs were fixed in 3% glutaraldehyde in PBS (pH 7.4) at
RT for 1 h or at 4C overnight, and processed for embedding in either
paraffin or Embed 812 (Electron Microscopy Sciences, Fort Washing-
ton, Pennsylvania) and for light microscopic autoradiographic analy-
sis 48,49. To label synaptic vesicles in presynaptic terminals, freshly
dissected CGs were incubated in oxygenated recording buffer with
90 mM KCl and 10 m FM1-43 (Molecular Probes, Eugene, Oregon)
at 37C for 1 min 50 , rinsed, then incubated with [ 3 H]glycine and
processed as described above except that alternate sections of the paraf-
fin-embedded ganglia were processed for fluorescence analysis with a
Zeiss Axioskop light microscope.
Glycine release assays. For release assays, E1517 CGs were labeled with
[3H]glycine as detailed above and incubated singly at 37C for 15 min in
a 24-well plate in 200 l of recording buffer alone (as a negative control),
30 mM KCl in recording buffer (to depolarize) or high-K+ buffer con-
taining either Cd2+ (0.1 mM), low Ca2+ levels (0.2 mM in contrast to
5.4 mM), sarcosine (5 mM) or excess cold glycine (1 mM). CGs were
pre-incubated in the test condition without high K+ for 10 min. The
buffer and the CG were counted separately in a scintillation counter (LS
5000TD, Beckman).
ACKNOWLEDGEMENTS
We acknowledge Heinrich Betz for providing glycine receptor and gephyrin
antibodies and clones, Yimen Ge (Massachusetts General Hospital, Harvard
Medical School) for assistance with the laser-scanning confocal microscope and
Kathleen Dunlap, Daniel Jay and Tim Turner for advice and comments on the
manuscript. This work was supported by NIH grant 21725 to M.H.J.
RECEIVED 9 SEPTEMBER; ACCEPTED 9 DECEMBER 1999
1. Zhang, Z. W. & Berg, D. K. Patch-clamp analysis of glycine-induced
currents in chick ciliary ganglion neurons. J. Physiol. (Lond.) 487, 395405
(1995).
2. Landmesser, L. & Pilar, G. The onset and development of transmission in the
chick ciliary ganglion. J. Physiol. (Lond.) 222, 691713 (1972).
3. Landmesser, L. & Pilar, G. Synaptic transmission and cell death during
normal ganglionic development. J. Physiol. (Lond.) 241, 737749 (1974).
4. Martin, A. R. & Pilar, G. Dual mode of synaptic transmission in the avian
ciliary ganglion. J. Physiol. (Lond.) 168, 443463 (1963).
5. Martin, A. R. & Pilar, G. Transmission through the ciliary ganglion of the
chick. J. Physiol. (Lond.) 168, 464475 (1963).
6. Reiner, A. et al. Neurotransmitter organization of the nucleus of Edinger-
Westphal and its projection to the avian ciliary ganglion. Vis. Neurosci. 6,
451472 (1991).
7. Engisch, K. L. & Fischbach, G. D. The development of ACh- and GABA-
activated currents in embryonic chick ciliary ganglion neurons in the absence
of innervation in vivo. J. Neurosci. 12, 11151125 (1992).
8. Arenella, L. S., Oliva, J. M. & Jacob, M. H. Reduced levels of acetylcholine
receptor expression in chick ciliary ganglion neurons developing in the
absence of innervation. J. Neurosci. 13, 45254537 (1993).
9. Dale, H. Pharmacology and nerve-endings. Proc. R. Soc. Med. (Lond.) 28,
319332 (1935).
10. Nicoll, R. A. & Malenka, R. C. A tale of two transmitters. Science 281, 360361
(1998).
11. Jacob, M. H., Lindstrom, J. M. & Berg, D. K. Surface and intracellular
distribution of a putative neuronal nicotinic acetylcholine receptor. J. Cell
Biol. 103, 205214 (1986).
12. Jacob, M. H. Acetylcholine receptor expression in developing chick ciliary
ganglion neurons. J. Neurosci. 11, 17011712 (1991).
13. Williams, B. M. et al. The long internal loop of the alpha 3 subunit targets
nAChRs to subdomains within individual synapses on neurons in vivo. Nat.
Neurosci. 1, 557562 (1998).
14. Jacob, M. H., & Berg, D. K. The ultrastructural localization of -
bungarotoxin binding sites in relation to synapses on chick ciliary ganglion
neurons. J. Neurosci. 3, 260271 (1983).
131
 2000 Nature America Inc. http://neurosci.nature.com
articles
15. Jacob, M. H., Berg, D. K. & Lindstrom, J. M. Shared antigenic determinant
between the Electrophorus acetylcholine receptor and a synaptic component
on chick ciliary ganglion neurons. Proc. Natl. Acad. Sci. USA 81, 32233227
(1984).
16. Vernallis, A. B., Conroy, W. G. & Berg, D. K. Neurons assemble acetylcholine
receptors with as many as three kinds of subunits while maintaining subunit
segregation among receptor sites. Neuron 10, 451464 (1993).
17. Schroder, S., Hoch, W., Becker, C. M., Grenningloh, G. & Betz, H. Mapping of
antigenic epitopes on the alpha 1 subunit of the inhibitory glycine receptor.
Biochemistry 30, 4247 (1991).
18. Kirsch, J., Wolters, I., Triller, A. & Betz, H. Gephyrin antisense
oligonucleotides prevent glycine receptor clustering in spinal neurons.
Nature 366, 745748 (1993).
19. Colin, I., Rostaing, P., Augustin, A. & Triller, A. Localization of components
of glycinergic synapses during rat spinal cord development. J. Comp. Neurol.
398, 359372 (1998).
20. Feng, G. et al. Dual requirement for gephyrin in glycine receptor clustering
and molybdoenzyme activity. Science 282, 13211324 (1998).
21. Prior, P. et al. Primary structure and alternative splice variants of gephyrin, a
putative glycine receptor-tubulin linker protein. Neuron 8, 11611170
(1992).
22. Pfeiffer, F., Simler, R., Grenningloh, G. & Betz, H. Monoclonal antibodies and
peptide mapping reveal structural similarities between the subunits of the
glycine receptor of rat spinal cord. Proc. Natl. Acad. Sci. USA 81, 72247227
2000 Nature America Inc. http://neurosci.nature.com
(1984).
23. Meyer, G., Kirsch, J., Betz, H. & Langosch, D. Identification of a gephyrin
binding motif on the glycine receptor beta subunit. Neuron 15, 563572
(1995).
24. Kins, S., Kuhse, J., Laube, B., Betz, H. & Kirsch, J. Incorporation of a
gephyrin-binding motif targets NMDA receptors to gephyrin-rich domains
in HEK 293 cells. Eur. J. Neurosci. 11, 740744 (1999).
25. Zhang, Z. W., Vijayaraghavan, S. & Berg, D. K. Neuronal acetylcholine
receptors that bind alpha-bungarotoxin with high affinity function as ligand-
gated ion channels. Neuron 12, 167177 (1994).
26. Marwitt, R., Pilar, G. & Weakly, J. N. Characterization of two ganglion cell
populations in avian ciliary ganglia. Brain Res. 25, 317334 (1971).
27. McEachern, A. E., Margiotta, J. F. & Berg, D. K. Gamma-Aminobutyric acid
receptors on chick ciliary ganglion neurons in vivo and in cell culture.
J. Neurosci. 5, 26902695 (1985).
28. Sorimachi, M., Rhee, J. S., Shimura, M. & Akaike, N. Mechanisms of GABA-
and glycine-induced increases of cytosolic Ca2+ concentrations in chick
embryo ciliary ganglion cells. J. Neurochem. 69, 797805 (1997).
29. McGale, E. H., Pye, I. F., Stonier, C., Hutchinson, E. C. & Aber, G. M. Studies
of the inter-relationship between cerebrospinal fluid and plasma amino acid
concentrations in normal individuals. J. Neurochem. 29, 291297 (1977).
30. Attwell, D., Barbour, B. & Szatkowski, M. Nonvesicular release of
neurotransmitter. Neuron 11, 401407 (1993).
31. Zafra, A. F. et al. Glycine transporters are differentially expressed among CNS
cells. J. Neurosci. 15, 39523969 (1995).
32. Liu, W., Leibach, F. H. & Ganapathy, V. Characterization of the glycine
transport system GLYT 1 in human placental choriocarcinoma cells (JAR).
Biochim. Biophys. Acta 1194, 176184 (1994).
132
33. Pow, D. V. Transport is the primary determinant of glycine content in retinal
neurons. J. Neurochem. 70, 26282636 (1998).
34. Schwartz, E. A. Depolarization without calcium can release gamma-
aminobutyric acid from a retinal neuron. Science 238, 350355 (1987).
35. Coggan, J. S., Paysan, J., Conroy, W. G. & Berg, D. K. Direct recording of
nicotinic responses in presynaptic nerve terminals. J. Neurosci. 17, 57985806
(1997).
36. Dumoulin, A. et al. Presence of the vesicular inhibitory amino acid
transporter in GABAergic and glycinergic synaptic terminal boutons. J. Cell
Sci. 112, 811823 (1999).
37. OMalley, D. M., Sandell, J. H. & Masland, R. H. Co-release of acetylcholine
and GABA by the starburst amacrine cells. J. Neurosci. 12, 13941408 (1992).
38. Jo, Y. H. & Schlichter, R. Synaptic corelease of ATP and GABA in cultured
spinal neurons. Nat. Neurosci. 2, 241245 (1999).
39. Smith, M. A., Stollberg, J., Lindstrom, J. M. & Berg, D. K. Characterization of
a component in chick ciliary ganglia that cross-reacts with monoclonal
antibodies to muscle and electric organ acetylcholine receptor. J. Neurosci. 5,
27262731 (1985).
40. Horch, H. L. & Sargent, P. B. Perisynaptic surface distribution of multiple
classes of nicotinic acetylcholine receptors on neurons in the chicken ciliary
ganglion. J. Neurosci. 15, 77787795 (1995).
41. Horch, H. L. & Sargent, P. B. Effects of denervation on acetylcholine receptor
clusters on frog cardiac ganglion neurons as revealed by quantitative laser
scanning confocal microscopy. J. Neurosci. 16, 17201729 (1996).
42. Wang, B. L. & Larsson, L. I. Simultaneous demonstration of multiple
antigens by indirect immunofluorescence or immunogold staining. Novel
light and electron microscopical double and triple staining method
employing primary antibodies from the same species. Histochemistry 83,
4756 (1985).
43. Buckley, K. & Kelly, R. B. Identification of a transmembrane glycoprotein
specific for secretory vesicles of neural and endocrine cells. J. Cell Biol. 100,
12841294 (1985).
44. Sambrook, J., Fritsch, E. F. & Maniatis, T. Molecular Cloning: a Laboratory
Manual (Cold Spring Harbor Press, New York, 1989).
45. Ikonomov, O. C., Kulesa, M. C., Shisheva, A. C. & Jacob, M. H. Innervation
and target tissue interactions induce Rab-GDP dissociation inhibitor (GDI)
expression during peripheral synapse formation in developing chick ciliary
ganglion neurons in situ. J. Neurosci. 18, 63316339 (1998).
46. Bartel, P., Chien, C. T., Sternglanz, R. & Fields, S. Elimination of false
positives that arise in using the two-hybrid system. Biotechniques 14, 920924
(1993).
47. Hollenberg, S. M., Sternglanz, R., Cheng, P. F. & Weintraub, H. Identification
of a new family of tissue-specific basic helix-loop-helix proteins with a two-
hybrid system. Mol. Cell Biol. 15, 38133822 (1995).
48. Jacob, M. H. & Berg, D. K. The distribution of acetylcholine receptors in
chick ciliary ganglion neurons following disruption of ganglionic
connections. J. Neurosci. 8, 38383849 (1988).
49. Boyd, R. T., Jacob, M. H., McEachern, A. E., Caron, S. & Berg, D. K. Nicotinic
acetylcholine receptor mRNA in dorsal root ganglion neurons. J. Neurobiol.
22, 114 (1991).
50. Ryan, T. A. et al. The kinetics of synaptic vesicle recycling measured at single
presynaptic boutons. Neuron 11, 713724 (1993).
nature neuroscience volume 3 no 2 february 2000
 2000 Nature America Inc. http://neurosci.nature.com

articles

Enhancement of synaptic
transmission by cyclic AMP
modulation of presynaptic Ih channels
Vahri Beaumont and Robert S. Zucker

Division of Neurobiology, Department of Molecular and Cell Biology, University of California, Berkeley, California 94720, USA
Correspondence should be addressed to V.B. (vahri@socrates.berkeley.edu)

Presynaptic activation of adenylyl cyclase and subsequent generation of cAMP represent an

important mechanism in the modulation of synaptic transmission. In many cases, short- to medium-
2000 Nature America Inc. http://neurosci.nature.com

term modulation of synaptic strength by cAMP is due to activation of protein kinase A and
subsequent covalent modification of presynaptic ion channels or synaptic proteins. Here we show
that presynaptic cAMP generation via serotonin receptor activation directly modulated
hyperpolarization-activated cation channels (Ih channels) in axons. This modulation of Ih produced an
increase in synaptic strength that could not be explained solely by depolarization of the presynaptic
membrane. These studies identify a mechanism by which cAMP and Ih regulate synaptic plasticity.

Serotonin is an important neuromodulator of synaptic trans-
mission in many phyla17. In crustaceans, serotonin acts as a neu-
rohormone; on binding to a Gs-coupled receptor8, it strongly
enhances neuromuscular transmission by increasing the num-
ber of quanta released per action potential911. Studies on both
the inhibitor (GABAergic) and excitor (glutamatergic) nerve
innervating the muscle have implicated two distinct processes:
serotonin increases the absolute number of vesicles available for
transmitter release12, and serotonin increases the kinetics of
release from the inhibitor nerve13.
Serotonin alters neither resting presynaptic [Ca2+]i nor cal-
cium influx during an action potential in crayfish nerve termi-
nals 13,14 . Its actions seem to be mediated by at least two
second-messenger systems, one involving phospholipase C
(PLC) and another involving adenylyl cyclase8,15. Serotonin
enhancement of transmission is blocked by presynaptic injec-
tion of a selective PLC inhibitor8. The downstream messenger
and its target that mediate this enhancement are uncertain, but
this pathway may involve cAMP. Compounds increasing
[cAMP]i can mimic and potentiate serotonin action15,16, and
serotonin increases cAMP levels when added to lobster neuro-
muscular preparations, although a pre- or postsynaptic locus
for cAMP production was undetermined17. In addition, presy-
naptic injection of an adenylyl cyclase inhibitor reduces sero-
tonin enhancement15.
Here we show that cAMP generation is an important com-
ponent of serotonin enhancement of synaptic transmission, and
furthermore, that the target for cAMP is presynaptically locat-
ed Ih channels.
RESULTS
Serotonin-induced enhancement involves cAMP
We recorded excitatory junction potentials (EJPs) from crayfish
muscle cells at the neuromuscular junction while stimulating
innervating glutamatergic axons. Superfusion of serotonin (300
nM, 25 min) resulted in a maximal increase of 310% in EJP
amplitude (Fig. 1a) with an EC50 of 60 19 nM and a Hill coef-
ficient of 1.14 0.22 (n = 4; Fig. 1d). Simultaneous intracellular
recording of axon membrane potential revealed an associated
depolarization of the resting potential by 10 mV (Fig. 1a and
c) and a slight reduction in action potential amplitude and the
area beneath its voltage trace, consistent with previous reports8,15.
Forskolin, an activator of adenylyl cyclase, partially mimicked
the effects of serotonin (Fig. 1b and c), maximally increasing EJP
amplitude by 120% (EC50, 19 10 M; Hill slope, 1.09 0.26;
n = 3), and depolarizing the axon membrane by 7 mV. Ampli-
tude of the action potential and area beneath its trace were also
slightly reduced. The membrane-permeable cAMP analog 8-Br-
cAMP (300 M) mimicked the effect of forskolin, enhancing EJP
amplitude by 80 12% (n = 6).
We next sought to determine the role of endogenous cAMP
generation in serotonin-induced synaptic enhancement. The phos-
phodiesterase inhibitor IBMX prevents breakdown of cAMP, so
we would expect it to potentiate serotonin enhancement if cAMP
were generated after serotonin receptor activation. Application of
a maximal concentration of serotonin (300 nM) increased EJP
amplitude by 304 46%. After washout (2 h) to allow EJP ampli-
tude to return to pre-drug levels, subsequent incubation in a low
concentration of IBMX (1 M) increased EJP amplitude by
30 19%. When this EJP amplitude was taken as the new base-
line, application of serotonin in the continued presence of IBMX
potentiated EJP enhancement (457 122% increase in EJP ampli-
tude, n = 5, p < 0.1, Fig. 1e). A second application of serotonin in
the absence of IBMX produced no increase in EJP amplitude over
that seen during the first application (p > 0.1, n = 4).
As IBMX is also a nonselective adenosine-receptor antagonist,
we investigated whether the observed potentiation might be due
to inhibition of adenosine receptors rather than inhibition of
phosphodiesterase. At a concentration 100-fold higher than that
used for IBMX, application of the nonselective adenosine recep-
tor antagonist, 8-(p-sulfophenyl)-theophylline (8-PST, 100 M;
n = 5), similarly increased basal EJP amplitude by 43 16%

nature neuroscience volume 3 no 2 february 2000 133

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

articles

Fig. 1. Serotonin and forskolin a Control Serotonin (300 nM)

c
enhance synaptic transmission.

Axon depolarization
(a, b) Intracellular recording of
axonal action potentials (APs,
Pre) and muscle excitatory

serotonin

8-BrcAMP
(mV)

forskolin
junction potentials (EJPs; Post)
Post
before (control) and after
application of serotonin (300
nM, 25 min; a) and forskolin Pre
(30 M, 25 min; b). Each trace
is the average of all EJPs/APs
recorded for 1 min at 2-Hz
Control Forskolin (30 M)
b

Percent increase
stimulation. Scale bars, 25 ms

8-BrcAMP
forskolin
EJP amplitude
and 1 mV (Post) or 30 mV

serotonin
(Pre). (c) Bar charts summariz-
ing the axonal depolarization
(top) and enhancement of EJP Post
amplitude (bottom) induced by
serotonin (300 nM), forskolin Pre
(30 M) or 8-Br cAMP (300
2000 Nature America Inc. http://neurosci.nature.com

M); n = 412. (d) Cumulative

concentrationresponse curves
for serotonin () and forskolin
(). The lower maximal d e
increase in EJP amplitude with
IBMX (1 M)
forskolin than with serotonin
Fold increase in EJP amplitude

demonstrates that adenylyl serotonin (300 nM)

% increase control EJP

cyclase activation cannot

entirely account for serotonin-
induced enhancement. Each
amplitude

point represents the

mean s.e. of four experi-
ments. (e) Time course of
serotonin (300 nM)-induced
enhancement of EJP amplitude
(). After EJPs returned to
control amplitudes following
washout of serotonin, each
log10 [drug] (log M) Time (min)
preparation was incubated for
25 min with a low concentra-
tion of the phosphodiesterase inhibitor IBMX (1 M). IBMX alone resulted in a small enhancement of EJP amplitude (normalized to control in graph). A sec-
ond serotonin application was potentiated in the presence of IBMX (n = 5, ). This effect might be explained by increased cAMP generation over control
levels in response to serotonin application.

(IBMX response, 30 19%, n = 5). However, in contrast to IBMX,
8-PST did not potentiate the serotonin (300 nM) response; the
increase in EJP amplitude induced by serotonin in the presence
of 8-PST was not different from that elicited by serotonin in the
absence of 8-PST (3 11% difference; p > 0.05, n = 5). Therefore,
IBMX potentiation of the serotonin-induced enhancement seems
to result from phosphodiesterase inhibition, although the effect
of IBMX alone on EJP amplitude may be due to a nonspecific
action, or even to inhibition of adenosine receptors.
Serotonin and forskolin act presynaptically
The locus of action of both serotonin and forskolin (and hence
cAMP generation) was examined using analysis of miniature EJPs
(mEJPs). In Normal Van Harrevalds solution containing TTX
(1 M), the mean frequency of mEJPs recorded over a 30-minute
period was 0.31 0.1 Hz with a mean amplitude of 254 36 V
(n = 12). After incubation for 20 minutes in either serotonin
(1 M) or forskolin (30 M), the same muscle fiber was record-
ed for 30 minutes in the presence of the drug. In 5 of 6 paired
experiments, serotonin induced a significant increase in frequency
(p < 0.05, Kolmogorov-Smirnov test) but not amplitude of mEJPs
(Fig. 2a and b), with a mean increase in frequency of 64 11%
(Fig. 2b; p < 0.05, Students t-test). Forskolin significantly
increased frequency in 3 of 6 experiments (Kolmogorov-Smirnov
test, p < 0.05) with a mean increase in frequency of 42 33% and
no increase in amplitude of mEJPs (p > 0.1, Students t-test,
Fig. 2b). Consistent with serotonins enhancement of the size of
the vesicle pool12, this finding supports the assumption that the
locus of action of both serotonin and forskolin is presynaptic.
PKA is not involved
The downstream effects of cAMP at the crayfish neuromuscu-
lar junction have previously been attributed to activation of
PKA 15 . We tested the effects of two membrane-permeable
inhibitors of PKA on the enhancement induced by serotonin
and 8-Br-cAMP. A submaximal concentration of serotonin (100
nM, 25 min) elicited an increase in EJP amplitude of 144 24%.
In the presence of 30 M H-7, which should fully inhibit PKA,
PKC and PKG (Ki = 36 M18,19), a second application of sero-
tonin resulted in an amplitude increase no different from the
control response (148 33%, n = 4, p > 0.05). H-7 was simi-
larly ineffective against the increase induced by 8-Br-cAMP (8-
Br-cAMP alone, 74 23%; 8-Br-cAMP plus H-7, 95 38%;
n = 3; p > 0.05). However, H-7 (30 M) prevents an unrelated

134 nature neuroscience volume 3 no 2 february 2000

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

articles

presynaptic phorbol ester-induced potentiation of transmis-
sion at the crayfish neuromuscular junction (V.B. and R.S.Z.,
unpublished observation), demonstrating that this compound
permeates the axon sufficiently to inhibit PKC.
In addition to H-7, we tested the highly specific, cell-perme-
able PKA inhibitor Rp-8-Br-cAMPS. Incubation with Rp-8-Br-
cAMPS (300 M) alone caused a 40 13% increase in EJP
amplitude, suggesting that this cAMP analog may mimic cAMP
itself. When serotonin was also added, EJP amplitude increased by
292 68% of pre-drug amplitude, not significantly different from
the paired response with serotonin alone (324 125%, n = 3;
Fig. 3a). Similar effects of Rp-8-Br-cAMPS were found against
8-Br-cAMP-induced increases in EJP amplitude (Fig. 3b). Because
Rp-8-Br-cAMPS is a cAMP analog, and it increased EJP ampli-
tude significantly, these results suggest that the target of cAMP
may be a cyclic nucleotide-binding effector other than PKA.
Presynaptic Ih channels are modulated by cAMP
2000 Nature America Inc. http://neurosci.nature.com
shifts of the activation curve to more depolarized potentials2025
would explain depolarization on addition of cAMP analogs8,15.
The presence of Ih channels in the crayfish excitor axon was
investigated by intracellular injection of hyperpolarizing current
pulses (500 ms, 5 to 50 nA) into the axon. Upon hyperpolar-
ization of the membrane beyond resting potential, Ih should
become activated, typically producing a depolarization sag back
toward the resting potential, whereas termination of the pulse
results in a transient depolarizing overshoot of the original resting
potential (after-depolarization potential; ADP)25,26. ADPs were
produced by injecting current pulses of different amplitudes
(Fig. 4). With large current injections, the ADP was of sufficient
amplitude to initiate firing of action potentials, as demonstrated
for Ih channels in many neurons25. When the Ih blocker Cs+ (1
mM) was applied to the preparation, resting membrane poten-
tial hyperpolarized by approximately 4 mV, and ADP amplitude
was significantly reduced (Fig. 4a). Injection of between 5 and
50 nA of current elicited ADPs (Fig. 4b; n = 6) that were blocked

Depolarization of the axon by forskolin, 8-Br-cAMP and serotonin
(Fig. 1ac) suggests that presynaptic Ih channels may be targets for
cAMP. Indeed, modulation of Ih channels by the direct binding of
cyclic nucleotides (including Rp-8-Br-cAMPS20) and subsequent
a
Control
by Cs+ (n = 3) or the specific Ih blocker ZD7288 (ref. 27; n = 3).
These data not only demonstrate the presence of Ih channels in
excitor axons, but also suggest that these channels contribute to
the axons resting potential. However, it should be noted that,

Serotonin (1 M)

b
Fig. 2. Serotonin and forskolin act
at a presynaptic locus to enhance
synaptic transmission. (a) 20-s
sample traces of miniature EJPs,
Cumulative percent

measured in NVH solution con-

taining TTX (1 M) before (con-
trol) and after a 20-min incubation
in serotonin (1 M). Scale bars,
0.5s, 0.3 mV. (b) Left, cumulative
probability plot for inter-event Control 100 V
interval after analysis of mEJPs
Serotonin (1 M)
recorded for 30 min in NVH solu-
20 ms
tion alone (____) or in the presence
of serotonin (----) indicates signifi-
cantly higher event frequency in Inter-event interval (s)
serotonin (0.57 Hz versus 0.33 Hz,
control). Right, superimposed
average mEJPs with and without
serotonin show similar amplitudes.
(c) Percent change in both mean c Percent change in mean frequency Percent change in mean amplitude
frequency (left) and mean ampli-
tude (right) of mEJPs after applica-
tion of either forskolin (open bars)
or serotonin (filled bars) Each bar
represents the mean s.e. of six Forskolin
experiments. Both forskolin and
serotonin application increased Serotonin
mEJP frequency without signifi-
cantly altering amplitude.

nature neuroscience volume 3 no 2 february 2000 135

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

articles

Fig. 3. Rp-8Br-cAMPS elevates

EJP amplitude and does not sig- a b
nificantly block serotonin Rp-8-Br-cAMPS Rp-8-Br-cAMPS

Percent increase EJP amplitude

Percent increase EJP amplitude

enhancement of transmission. serotonin 8-Br-cAMPS
(a) Time course of the increase
of EJP amplitude in response to
serotonin (100 nM; ). After
serotonin was washed out and
EJP amplitude reached a steady
baseline, preparations were incu-
bated with the membrane-per-
meable specific PKA inhibitor
Rp-8Br-cAMPS (300 M).
Surprisingly, this incubation
enhanced EJP amplitude (,*) by
approximately 50% over control. Time (min) Time (min)
Serotonin-induced enhancement
was affected little by the PKA inhibitor (). (b) Time course of the increase of EJP amplitude by the membrane-permeable cAMP analog 8-Br-cAMP
(, 300 M). After washout of 8-Br-cAMP, EJP amplitude reached a steady baseline, and preparations were incubated with Rp-8Br-cAMPS (300 M).
Again, this incubation increased EJP amplitude 48 4% over control (,*) Addition of 8-Br-cAMP together with Rp-8Br-cAMPS resulted in a slightly
2000 Nature America Inc. http://neurosci.nature.com

smaller increase in EJP amplitude (78 7%; ) than 8-Br-cAMP applied alone (120 30%, ).

whereas ADPs elicited by current injection up to and including
30 nA could be blocked effectively, ADPs elicited by larger current
injections were not completely blocked by either Cs+ or ZD7288
(Fig. 4a and b), suggesting that ADPs evoked by injection of more
than 30 nA may involve conductances other than Ih channels.
To determine whether Ih channels were modulated by cAMP,
we attempted to block the serotonin- and forskolin-induced depo-
larization of the axon (Fig. 1a and b) using Cs+ and ZD7288.
Application of serotonin (300 nM) resulted in a maximum axon
depolarization of 9.5 0.75 mV from a mean resting membrane
potential of 71 1 mV (n = 12; Fig. 5a) within 16 minutes of
serotonin application. Addition of Cs+ (3 M3 mM) reversed
this depolarization in a concentration-dependent manner (IC50,
260 20 M; Hill coefficient, 1.26 0.05; n = 5, Fig. 5b). A slight
broadening of the action potential and an increase in area beneath
the voltage trace was observed in Cs+, consistent with additional
block of a potassium conductance. However, addition of the potas-
sium-channel blocker Ba2+ (1 mM) to the nerve reproduced the
effects on the action potential seen with Cs+, but did not affect
serotonin-induced depolarization (Fig. 5a). ZD7288
(100 nM100 M; 25 min) potently reduced serotonin-induced
depolarization of the axon, with an IC50 similar to that for its
action on I h channels 27,28 (IC 50, 6 2 M; Hill coefficient,
1.19 0.01; n = 4, Fig. 5b).
Serotonin seemed to alter Ih activity through cAMP genera-
tion, as application of forskolin (30 M) also produced axon
depolarization of 6.3 0.6 mV from a mean resting potential of
71 1.4 mV (n = 8, Fig. 5c and d). The significantly smaller size
of the depolarization with forskolin than with serotonin
(10 mV) probably resulted from the barely submaximal con-
centration of forskolin used (Fig. 1d). Incubation with Cs +
(1 mM, n = 4) or ZD7288 (30 M, n = 4) abolished the depolar-
ization induced by a second application of forskolin (Fig. 5c and
d). In summary, we demonstrated that cAMP generated in the
axon acts on Ih, resulting in depolarization with a time course
paralleling that of synaptic enhancement.

a b
20 nA 40 nA
Iinj
Peak amplitude ADP (mV)

RMPaxon

Hyperpolarizing pulse (nA)

Fig. 4. Ih channels are involved in maintenance of resting membrane potential, and Ih activation results in anomolous membrane rectification in exci-
tatory axons. (a) Injection of axons with hyperpolarizing current (500 ms) resulted in after-depolarizing potentials (ADPs; traces in black). Resting
membrane potential of the axon (dashed line) was 66 mV. Membrane responses during the pulse were truncated. With large hyperpolarizing pulses
(typically 40 nA), the ADP was sufficient to reach threshold for action potential generation (APs truncated in trace). Scale bar, 2 s, 10 mV. In the
same axon, resting membrane potential became relatively hyperpolarized in the Ih channel blocker Cs+ (1 mM; compare gray traces with dashed line),
and ADP amplitudes in response to hyperpolarizing pulses were reduced (gray traces). (b) Plots of peak ADP amplitude versus size of injected hyper-
polarizing current in controls (, n = 6), with Cs2+ (, 1 mM, n = 3) or with ZD7288 (, 30 M, n = 3).

136 nature neuroscience volume 3 no 2 february 2000

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

articles

Modulation of Ih by cAMP results in synaptic enhancement
To investigate the involvement of Ih modulation in synaptic
enhancement, we studied the increase of EJP amplitude by sero-
tonin, forskolin and 8-Br-cAMP in the absence and presence of
ZD7288 (30 M) or Cs+ (1 mM), with Ba2+ (1 mM) as a control.
In the absence of any other drug, serotonin (100 nM) caused
a 2.9 0.3-fold increase in EJP amplitude (control response;
n = 13). Two hours later, recordings from the same muscle cells
incubated in Cs+ (1 mM) 10 minutes before and during a second
serotonin application showed a 43 7% reduction compared with
control responses to serotonin (1.8 0.2-fold increase in EJP ampli-
tude; p < 0.05, n = 6), whereas application of 1 mM Ba2+ before
and during a third serotonin application insignificantly increased
EJP amplitude 17 13% over controls, consistent with the slight
widening of the action potential we observed (n = 6; Fig. 6a and
b). A 25-minute application of ZD7288 before and during sero-
tonin application was as effective as Cs+ in reducing synaptic
enhancement, decreasing the response to serotonin by 42 13%
2000 Nature America Inc. http://neurosci.nature.com
blocked in the presence of ZD7288 (30 M), which decreased
synaptic enhancement by 67 6% (Fig. 6c) and 70 18%
(Fig. 6d) relative to the respective control responses. Enhance-
ment of EJP amplitude by 8-Br-cAMP was blocked by Cs+ equal-
ly well (82 2% reduction of the control response, n = 3).
However, Cs+ was less effective in blocking forskolin enhance-
ment, reducing the response to only 31 15% of the control
response (n = 3). Although the reason for this is not obvious,
Cs+ blockade of a potassium conductance in these cells may have
induced a depolarization that counteracted the effect of block-
ing Ih to a greater extent than observed in other experiments.
It is conceivable that the reduction of the serotonin and
forskolin response by ZD7288 and Cs+ ions may have been a
result only of the membrane hyperpolarization produced by
these agents, rather than by a specific block of an Ih conduc-
tance. We felt this was unlikely, as hyperpolarizing the axon
by 5 mV by manipulating extracellular potassium has no
effect on either serotonin-induced axonal depolarization or

from the control (n = 7, p < 0.05; Fig. 6a). Neither Cs+ nor ZD7288
applied on its own reduced basal EJP amplitude (n = 4 each).
Both forskolins (30 M) and 8-Br-cAMPs (300 M)
enhancement of EJP amplitude (1.4 0.2-fold increase, n = 9
and 0.8 0.1-fold increase, n = 6, respectively) were markedly
serotonin-induced synaptic enhancement14,29. Nevertheless,
we eliminated hyperpolarization of the axon in the presence
of ZD7288 (30 M) by slightly elevating extracellular potas-
sium concentration to 6.757.0 mM (Fig. 7a). Even with the
offset of hyperpolarization afforded by increased [K + ] e ,

a c Cs+ (1 mM)
Cs+ Ba2+ forskolin forskolin
serotonin serotonin serotonin
RMP axon(mV)

RMP axon(mV)

Time (min) Time (min)

d ZD7288 (30 M)
b
forskolin forskolin
Depolarization of axon

RMP axon(mV)
(mV)

log10 [Ih inhibitor] (log M) Time (min)

Fig. 5. Ih modulation by cAMP generation results in axon depolarization. (a) Application of serotonin (300 nM) to axons resulted in an approximately
10-mV membrane depolarization (n = 5, ). After washout of serotonin, membrane depolarization slowly reversed to pre-serotonin levels.
Application of Cs+ (1 mM) to the axon resulted in a rapid membrane hyperpolarization of approximately 5 mV. Subsequent application of serotonin
in the presence of Cs+ was ineffective in eliciting a membrane depolarization. After washout of Cs+, incubation with Ba2+ (1 mM) affected neither
resting membrane potential nor subsequent depolarization in response to serotonin (n = 4; typical s.e., 1.5 mV). (b) Cumulative concentrationinhi-
bition curves for Cs+ (; n = 3 separate experiments) and the irreversible Ih blocker ZD7288 (; n = 4) against axon depolarization induced by sero-
tonin (300 nM). Each point represents the mean s.e. (c, d) Forskolin (30 M) mimics serotonin at the axon, suggesting that cAMP generation is
responsible for increased Ih activation and resulting depolarization. The effects of forskolin can be blocked by either Cs2+ (c; 1 mM) or ZD7288
(d; 30 M).

nature neuroscience volume 3 no 2 february 2000 137

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

articles

forskolin-induced synaptic enhancement after ZD7288 treat-
ment was reduced to the same extent as without elevating
[K+]e (68 23% reduction, n = 3, Fig. 7b).
In summary, pooled results from the above experiments show
that Ih activation by cAMP was responsible for 62 7% (n = 18)
of the cAMP-induced synaptic enhancement, and activation of Ih
via cAMP generation accounted for 43 6% (n = 13) of the sero-
tonin-induced enhancement of synaptic strength.
Axonal hyperpolarization enhances synaptic transmission
Our results thus far implicated cAMP modulation of Ih channels
as a vital component of the increase in synaptic strength observed
on addition of either serotonin or forskolin. We extended this
observation to see whether voltage-gated activation of Ih channels
per se was sufficient to increase synaptic transmission at the cray-
fish neuromuscular junction. First, the fluorescent dye Fura-2 (17
mM in 200 mM KCl) was iontophoresed (10 nA, 10 min) into
the axon via a microelectrode. Diffusion of the dye along the
2000 Nature America Inc. http://neurosci.nature.com
several hundred microns of boutons innervating a muscle fiber,
which was also impaled with a microelectrode. We then simulta-
neously recorded both evoked action potentials (2 Hz) in the exci-
tor nerve and subsequent EJPs in the muscle (Fig. 8a). After 10
minutes, stimulation was stopped and the axon hyperpolarized
by injection of 30 nA for 1 minute via the intracellular electrode,
a protocol that elicits a large ADP, but is not sufficient to induce
spontaneous firing of action potentials (Fig. 4). The increased
time of current injection used in this experiment (1 min versus
0.5 s in Fig. 4) did not significantly affect the amplitude of the
ADP (13 6 mV after 1-min and 16 4 mV after 0.5-s current
injection) or its sensitivity to ZD7288 (79 21% block and 87
18% block, respectively; n = 3), suggesting that this prolonged
current protocol elicited an ADP indicative mainly of Ih channel
activation. During axon hyperpolarization, we often noted an
increase in the number of spontaneous EJPs (Fig. 8a) which was
not observed previously using methods that polarize large axon-
al branches30. If no increase in spontaneous events was detected

length of the axon allowed visualization of small tertiary branch- during hyperpolarization, it was assumed that the hyperpolariz-
es and boutons innervating individual muscle fibers. Using a new ing electrode was too far from boutons to elicit a response in the
intracellular electrode (3 M KCl), we reimpaled the axon within recorded muscle fiber, and the experiment was discounted (4 of

a Percent of control 5HT increase

b
1 mM Cs+/Ba2+
Percent increase EJP amplitude

serotonin

control 5HT enhancement

1 mM Ba2+

1 mM Cs + Fig. 6. Inhibition of Ih channels

resulted in a reduction of
ZD7288 (30 M) synaptic enhancement by sero-
c Time (min)
tonin, forskolin and 8-Br-cAMP.
(a) Effects of Ba2+ (1 mM) and
forskolin (30 M)
the Ih blockers Cs+ (1 mM) or
Percent increase EJP amplitude

Percent of control forskolin increase

ZD7288 (30 M) on EJP ampli-
tude enhancement by sero-
tonin (100 nM; upper),
forskolin (30 M; middle, gray
forskolin, control bars) or 8-Br cAMP (300 M;
lower) To allow visual compari-
son of the percent inhibition of
1 mM Ba2+
synaptic enhancement by either
Cs+ or ZD7288, enhancements
1 mM Cs + with serotonin, forskolin or 8-
Br-cAMP alone (control
enhancement) were normal-
ZD7288 d Time (min) ized to 100%. (bd) Data from
which bar charts were gener-
Percent of control 8-Br-cAMP 8-Br-cAMP (300 M) ated. (b) Time course of the
Percent increase EJP amplitude

increase
percent increase in EJP ampli-
8-Br-cAMP, control tude with serotonin alone (100
nM; ) or after concurrent
incubation with Cs+ (1 mM; )
1 mM Ba2+
or Ba2+ (1 mM, _____).
(c, d) Percent increase in EJP
Cs + amplitude during application of
forskolin (c; ) or 8-Br-cAMP
(d; ) alone or after a 30-min
ZD7288 pre-incubation with ZD7288
(30 M, ). Each point repre-
sents the mean s.e. of 47 dif-
Time (min) ferent experiments.

138 nature neuroscience volume 3 no 2 february 2000

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

articles

a b sistent with the notion that EJP

enhancement is a result of hyperpo-

Percent increase in EJP amplitude

forskolin larization per se. Indeed, if hyperpo-
KCI 5.4 mM 6.5 mM 7 mM 5.4 mM
larization independent of I h
ZD7288 activation mediated the enhance-
RMPaxon (mV)

ment of EJP amplitude, one would

expect a greater increase after
ZD7288 application, where blockade
of Ih conductance would increase the
hyperpolarization induced by cur-
rent injection. We therefore conclude
that a relatively brief Ih channel acti-
vation, independent of cAMP mod-
Time (min) Time (min) ulation, is also capable of eliciting a
Fig. 7. Reversal of the ZD7288-mediated axonal hyperpolarization does not affect ZD7288 block of more prolonged synaptic enhance-
forskolin-induced synaptic enhancement. (a) The axon was impaled, and ZD7288 (30 M) was applied ment.
after determining the postsynaptic control response to forskolin (b, ). Resulting presynaptic membrane
hyperpolarization was then offset by stepwise increases in extracellular potassium to return the axon to DISCUSSION
2000 Nature America Inc. http://neurosci.nature.com

its original resting potential. The recovery of hyperpolarization when the [K+]e was returned to normal
Neuronal Ih channels subserve fun-
(5.4 mM) demonstrates the irreversibility of Ih block by ZD7288. (b) After determining the postsynaptic
control response to forskolin (), ZD7288 was applied and the resulting hyperpolarization of the axon
damental physiological functions
membrane potential offset back to the original resting potential (see a) before recording the response to such as contribution to membrane
forskolin from a different muscle fiber in the same preparation (, n = 3). ZD7288 still reduced forskolin- potential3134, integration of synaptic
induced synaptic enhancement under these conditions. input to neurons 35 and the rhyth-
micity of various brain regions3639.
Binding of cAMP to Ih results in a
concentration-dependent shift of the
12). After axon hyperpolarization, action potentials and subse- activation curve to more depolarized voltages through a phos-
quent EJPs were again evoked at two Hz. EJP amplitude signifi- phorylation-independent allosteric interaction with the chan-
cantly increased following hyperpolarization, increasing an average nel40,41. Thus receptors both positively and negatively coupled to
of 54 11% by 1020 minutes after current injection (Fig. 8a and adenylyl cyclase can shift Ih activation by up to 20 mV25, and
b; p < 0.05, n = 8). -adrenoceptors23, serotonin receptors23,42,43, A1 adenosine recep-
In separate experiments, ZD7288 (30 M) was applied for tors44 and -opioid receptors21 regulate Ih in this way. The chan-
30 minutes before and during the 30 nA
hyperpolarization of the axon (Fig. 8b). a Before hyperpolarization Before hyperpolarization 10 min after hyperpolarization
Again, individual experiments were aban-
doned if spontaneous release was not seen During hyperpolarization
during the hyperpolarizing pulse, although Post Post
this was rare (1 of 7) in the presence of
ZD7288. The increase in spontaneous Pre Pre
release observed during hyperpolarization 1 mV
with ZD7288 probably indicated that hyper- 0.5 s
polarizing current pulses applied during Ih
block resulted in greater changes in poten-
tial compared with those in control trials. b
EJP amplitude sometimes increased slight- Hyperpolarization
(30 nA, 1 min)
ly (average 11 8%) and transiently
Percent increase EJP

between 1 and 5 minutes after the termina-

amplitude

tion of the pulse, correlating positively with

the level of spontaneous activity during the
hyperpolarization, although this was not
quantified further. However, ZD7288 com-
pletely prevented the enhancement of EJP
amplitude between 10 and 20 minutes after
current injection (2 5% reduction from
basal amplitude, n = 6, significantly different Time (min)
from the control trials; p < 0.001, Students
Fig. 8. Brief activation of Ih channels by hyperpolarization is sufficient to induce long-lasting
unpaired t-test).
enhancement of EJP amplitude. (a) Intracellular recording of axonal action potentials (APs; pre)
It should be noted that, although this and muscle excitatory junction potentials (EJPs; post) before (control) and 10 minutes after
experiment is unavoidably compromised by hyperpolarizing current injection (1 min, 30 nA) into the axon to activate I channels. Each trace
h
the probable difference in axon membrane is the average of all EJPs/APs recorded for 1 min at 2-Hz stimulation. Scale bars, 25 ms and 1 mV
potential change elicited by the current pulse (post) or 25 mV (pre). Spontaneous EJPs increase upon hyperpolarization (middle panel).
with or without Ih channel block, the pre- (b) Time course of EJP amplitude before, during and after hyperpolarization (30 nA, 1 min) to
vention of any increase by ZD7288 is incon- activate presynaptic Ih channels with ZD7288 (; 30 M; n = 6) or without ZD7288 (; n = 8).

nature neuroscience volume 3 no 2 february 2000 139

 2000 Nature America Inc. http://neurosci.nature.com
articles
nels we describe were similar to other reported Ih channels based
on their activation by hyperpolarization, the contribution made
to axonal resting potential, their increased activation by cAMP
and their selective pharmacological block by both Cs + and
ZD7288. The Ih current is a mixed cation current (Na+/K+)25, and
under physiological conditions, current is carried predominant-
ly by the movement of sodium ions. In the crayfish axon, this
also seems to be the case, as serotonin-induced depolarization of
membrane potential was prevented when external sodium was
replaced with choline15.
The mechanism by which serotonin-induced activation of
Ih channels in the axon results in an increase in the total num-
ber of vesicles available for release12 remains elusive. It is clear
from previous studies that the amplitude of depolarization per
se elicited by serotonin or forskolin is insufficient to account
for the enhancement of neurotransmission, as depolarizing the
axon by the same amount using either high [K+]14,29 or current
injection through intracellular recording electrodes45 could not
2000 Nature America Inc. http://neurosci.nature.com
mimic the increase in synaptic strength. In addition, the obser-
vation that EJP amplitude remains elevated for up to 20 min-
utes after a relatively brief (1 minute) axonal hyperpolarization
would suggest that Ih activation is required only for initiating
a downstream process resulting in synaptic enhancement.
Synaptic enhancement after axon hyperpolarization is observed
not only in crayfish30 but also in vertebrate junctions46. Before
knowledge of Ih channels, either a prolongation of the action
potential30 or a hyperpolarization-dependent mobilization of
vesicles was assumed responsible30,47. As we did not stimulate
the nerve and record EJPs during hyperpolarization, and action
potentials were unaltered after hyperpolarization, the former
hypothesis can be rejected. However, active Ih-dependent mobi-
lization of vesicles from non-releasable to releasable pools rep-
resents an attractive possibility, and the role of microtubule and
actin transport pathways in this cAMP-induced enhancement
are being investigated.
Ih clearly can be included with other presynaptic voltage- and
ligand-gated ion channels modulated by neurotransmitters to
inhibit or facilitate transmission. The many reports of Ih chan-
nels in vertebrate and invertebrate axons42,4749 suggest physio-
logical importance of cAMP-induced Ih modulation in regulating
synaptic strength in a variety of organisms.
METHODS
Crayfish (Procambarus clarkii, 23 inches) were obtained from Niles Bio-
logical (Sacramento, California). Preparation of the innervated dactyl
opener muscle of the first walking leg is described14. Autotomized legs
were continuously superfused by a gravity-fed perfusion system with
Normal Van Harrevalds solution, containing 195 mM NaCl, 13.5 mM
CaCl2, 5.4 mM KCl, 2.6 mM MgCl2 and 10 mM Na-HEPES at pH 7.4
and 1417C. Serotonin, H-7, 8-(phenylsulfonyl)-theophylline and 8-
Br-cAMP were obtained from Sigma; Fura-2 from Molecular Probes
(Eugene, Oregon); and ZD7288 from Tocris Cookson (Ballwin, Mis-
souri). All other drugs were from Calbiochem (La Jolla, California). All
paired control experiments contained final concentrations of DMSO
equal to those used for drug delivery.
Electrophysiology. Sharp electrodes were used to impale both proximal
muscle fibers (electrode resistance, 1225 M) and/or either primary or
secondary branches of the excitor nerve axon (beveled electrode resis-
tance, 2545 M). Excitor nerves were stimulated (2 Hz) during record-
ing using a suction electrode containing the excitor axon freed from the
meropodite segment of the leg. Signals were amplified (Neuroprobe 1600
Amplifier, A-M systems Carlsborg, Washington), filtered at 2 kHz, and
digitized at 5 kHz and the average of all EJPs elicited each minute was
saved to computer using pClamp7 software (Axon Instruments Foster
140
City, California). EJP amplitudes were measured off-line (Clampfit 6.05,
Axon Instruments). For miniature EJP recording, beveled sharp electrodes
(resistance 510 M) were used to continuously acquire recordings sub-
sequently filtered at 1 kHz, digitized at 2.5 kHz and analyzed off-line using
Minianalysis Program version 4.0.1 (Synaptosoft, Leonia, New Jersey).
Data presentation and statistical analysis. As control EJP amplitudes were
extremely variable from fiber to fiber, results were expressed as percent
change from control EJP amplitude, taken as the average EJP amplitude
over ten minutes of continuous recording in the absence of drug. Data
are plotted as mean s.e. percent change from this control level. When
effects of different drugs were tested within a single muscle fiber, results
were analyzed by Students paired t-test; the Kolmogorov-Smirnov test
was used to compare differences in cumulative probability for analysis of
mEJPs. Significance was assumed if p < 0.05, unless otherwise stated.
ACKNOWLEDGEMENTS
We thank Russell English for technical assistance. This work was supported by
NSF Grant IBN-9722826.
RECEIVED 24 AUGUST; ACCEPTED 30 NOVEMBER 1999
1. Sugita, S., Goldsmith, J. R., Baxter, D. A. & Byrne, J. H. Involvement of pro-
tein kinase C in serotonin-induced spike broadening and synaptic facilitation
in sensorimotor connections of Aplysia. J. Neurophysiol. 68, 643651 (1992).
2. Hori, Y., Endo, K. & Takahashi, T. Long-lasting synaptic facilitation induced
by serotonin in superficial dorsal horn neurones of the rat spinal cord.
J. Physiol. (Lond.) 492, 867876 (1996).
3. Sun, Z. Y. & Schacher, S. J. Binding of serotonin to receptors at multiple sites
is required for structural plasticity accompanying long-term facilitation of
Aplysia sensorimotor synapses. J. Neurosci. 18, 39914000 (1998).
4. Schuman, E. M. & Clark, G. A. Synaptic facilitation at connections of
Hermissenda type B photoreceptors. J. Neurosci. 14, 16131622 (1994).
5. Stubli, U. & Otaky, N. Serotonin controls the magnitude of LTP induced by
theta bursts via an action on NMDA-receptor-mediated responses. Brain Res.
643, 1016 (1994).
6. Mintz, I. & Korn, H. Serotonergic facilitation of quantal release at central
inhibitory synapses. J. Neurosci. 11, 33593370 (1991).
7. Mitoma, H., Kobayashi, T., Song, S. Y. & Konishi, S. Enhancement by serotonin
of GABA-mediated inhibitory synaptic currents in rat cerebellar Purkinje cells.
Neurosci. Lett. 173, 127130 (1994).
8. Dixon, D. & Atwood, H. L. Phosphatidylinositol systems role in serotonin-
induced facilitation at the crayfish neuromuscular junction. J. Neurophysiol. 62,
239246 (1989).
9. Dudel, J. Facilitatory effects of 5-hydroxy-tryptamine on the crayfish neuromuscular
junction. Naunyn Schmiedebergs Arch. Pharmacol. 249, 515528 (1965).
10. Fischer, L. & Florey, E. Modulation of synaptic transmission and excitation-
contraction coupling in the opener muscle of the crayfish, Astacus leptodactylus,
by 5-hydroxytryptamine and octopamine. J. Exp. Biol. 102, 187198 (1983).
11. Glusman, S. & Kravitz, E. A. The action of serotonin on excitatory nerve
stimulation in lobster nerve-muscle preparations. J. Physiol. (Lond.) 325,
223241 (1982).
12. Wang, C. & Zucker, R. S. Regulation of synaptic vesicle recycling by calcium and
serotonin. Neuron 21, 155167 (1998).
13. Vyshedskiy, A., Delaney, K. & Lin, J.-W. Neuromodulators enhance transmitter
release by two separate mechanisms at the inhibitor of crayfish opener muscle.
J. Neurosci. 18, 51605169 (1998).
14. Delaney, K., Tank, D. W. & Zucker R. S. Presynaptic calcium and serotonin-
mediated enhancement of transmitter release at crayfish neuromuscular
junction. J. Neurosci. 11, 26312643 (1991).
15. Dixon, D. & Atwood, H. L. Conjoint action of phosphatidylinositol and
adenylate cyclase systems in serotonin-induced facilitation at the crayfish
neuromuscular junction. J. Neurophysiol. 62, 12511259 (1989).
16. Enyeart, J. Cyclic AMP, 5-HT, and the modulation of transmitter release at the
crayfish neuromuscular junction. J. Neurobiol. 12, 505513 (1981).
17. Goy, M. F. & Kravitz, E. A. Cyclic AMP only partially mediates the actions of
serotonin at lobster neuromuscular junctions. J. Neurosci. 9, 369379 (1989).
18. Boulis, N. M. & Davis M. Blockade of the spinal excitatory effect of cAMP on the
startle reflex by intrathecal administration of the isoquinoline sulfonamide H8:
comparison to the protein kinase C inhibitor H7. Brain Res. 525, 198204 (1990).
19. Quick, J., Ware, J. A. & Driedger P. E. The structure and biological activities of
the widely used protein kinase inhibitor, H7, differ depending on the
commercial source. Biochem. Biophys. Res. Commun. 187, 657663 (1992).
20. Ingram, S. L. & Williams, J. T. Modulation of the hyperpolarization-activated
current (Ih) by cyclic nucleotides in guinea pig primary afferent neurones.
J. Physiol. (Lond.) 492, 97106 (1996).
nature neuroscience volume 3 no 2 february 2000
 2000 Nature America Inc. http://neurosci.nature.com
21. Ingram, S. L & Williams, J. T. Opioid inhibition of Ih via adenylyl cyclase. Neuron
13, 179186 (1994).
22. Tokimasa, T. & Akasu, T. Cyclic AMP regulates an inward rectifying sodium-
potassium current in dissociated bull-frog sympathetic neurones. J. Physiol.
(Lond.) 420, 409429 (1990).
23. Pape, H.-C. & McCormick, D. A. Noradrenaline and serotonin selectively
modulate thalamic burst firing by enhancing a hyperpolarization-activated
cation current. Nature 340, 715718 (1989).
24. Di Francesco, D. & Mangoni, M. Modulation of single hyperpolarization-
activated channels (if) by cAMP in the rabbit sinoatrial node. J. Physiol. (Lond.)
474, 473482 (1994).
25. Pape, H.-C. Queer current and pacemaker: the hyperpolarization-activated
cation current in neurons. Annu. Rev. Physiol. 58, 299327 (1996).
26. Lthi, A. & McCormick, D. A. Periodicity of thalamic synchronized oscillations:
the role of calcium mediated upregulation of Ih. Neuron 20, 553563 (1998).
27. BoSmith, R. E., Briggs, I. & Sturgess, N. C. Inhibitory actions of ZENECA
ZD7288 on whole-cell hyperpolarization activated inward current (If) in guinea-
pig dissociated sinoatrial node cells. Brit. J. Pharmacol. 110, 343349 (1993).
28. Harris, N. C. & Constanti A. Mechanism of block by ZD7288 of the
hyperpolarization-activated inward rectifying current in guinea pig substantia
nigra neurons in vitro. J. Neurophysiol. 74, 23662378 (1995).
29. Dixon, D. & Atwood, H. L. Crayfish motor nerve terminals response to serotonin
examined by intracellular microelectrode. J. Neurobiol. 16, 409424 (1985).
30. Dudel, J. The effect of polarizing current on action potential and transmitter
2000 Nature America Inc. http://neurosci.nature.com
release in crayfish motor nerve terminals. Pflgers Arch. 324, 227248 (1971).
31. McCormick, D. A. & Pape, H.-C. Properties of a hyperpolarization-activated
cation current and its role in rhythmic oscillation in thalamic relay neurones.
J. Physiol. (Lond.) 431, 319342 (1990).
32. Maccaferri, G., Mangoni, M., Lazzari, A. & DiFrancesco, D. Properties of the
hyperpolarization-activated cation current in rat hippocampal CA1 pyramidal
cells. J. Neurophysiol. 69, 21292136 (1993).
33. Pape, H.-C. Specific bradycardic agents block the hyperpolarization-activated
cation current in central neurons. Neuroscience 59, 363373 (1994).
34. Travagli, R. A. & Gillis, R. A. Hyperpolarization-activated currents, Ih and IKIR,
in rat dorsal motor nucleus of the vagus neurons in vitro. J. Neurophysiol. 71,
13081317 (1994).
35. Magee, J. C. Dendritic Ih normalizes temporal summation in hippocampal
CA1 neurons. Nat. Neurosci. 2, 508514 (1999).
nature neuroscience volume 3 no 2 february 2000
articles
36. Bal, T. & McCormick, D. A. What stops synchronized thalamocortical
oscillations? Neuron 17, 297308 (1996).
37. Leresche, N., Lightowler, S., Soltesz, I., Jassik-Gerschenfeld, D. & Crunelli, V.
Low-frequency oscillatory activities intrinsic to rat and cat thalamocortical
cells. J. Physiol. (Lond.) 441, 155174 (1991).
38. Dekin, M. S. Inward rectification and its effects on the repetitive firing
properties of bulbospinal neurons located in the ventral part of the nucleus
solitarius. J. Neurophysiol. 70, 590601 (1993).
39. Golowasch, J. & Marder, E. Ionic currents of the lateral pyloric neuron of the
somatogastric ganglion of the crab. J. Neurophysiol. 67, 318331 (1992).
40. DiFrancesco, D. Dual allosteric modulation of pacemaker (f) channels by
cAMP and voltage in rabbit SA node. J. Physiol. (Lond.) 515, 367376
(1999).
41. Lthi, A. & McCormick D. Modulation of a pacemaker current through
Ca2+-induced stimulation of cAMP production. Nat. Neurosci. 2, 634641
(1999).
42. Bobker, D. H. & Williams, J. T. Serotonin augments the cationic current Ih in
central neurons. Neuron 2, 15351540 (1989).
43. Kiehn, O. & Harris-Warrick, R. M. 5-HT modulation of hyperpolarization-
activated inward current and calcium-dependent outward current in a
crustacean motor neuron. J. Neurophysiol. 68, 496508 (1992).
44. Pape, H.-C. Adenosine promotes burst activity in guinea-pig
geniculocortical neurones through two different ionic mechanisms.
J. Physiol. (Lond.) 447, 729753 (1992).
45. Wojtowicz, J. M. & Atwood, H. L. Presynaptic membrane potential and
transmitter release at the crayfish neuromuscular junction. J. Neurophysiol.
52, 99113 (1984).
46. Hubbard, J. I. & Willis, W. D. Hyperpolarization of mammalian motor nerve
terminals. J. Physiol. (Lond.) 163, 115137 (1962).
47. Birch, B. D., Kocsis, J. D., DiGregorio, F., Bhisitkul, R. B. & Waxman, S. G. A
voltage and time-dependent rectification in rat dorsal spinal root axons.
J. Neurophysiol. 66, 719728 (1991).
48. Eng, D. L., Gordon, T. R., Kocsis, J. D. & Waxman, S. G. Current-clamp
analysis of a time-dependent rectification in rat optic nerve. J. Physiol.
(Lond.) 421, 185202 (1990).
49. Grafe, P., Quasthoff, S., Grosskreutz, J. & Alzheimer, C. Function of the
hyperpolarization-activated inward rectification in nonmyelinated
peripheral rat and human axons. J. Neurophysiol. 77, 421426 (1997).
141
 2000 Nature America Inc. http://neurosci.nature.com

articles

Synaptic activity modulates

presynaptic excitability
Teresa A. Nick1,2 and Angeles B. Ribera1

1 Department of Physiology and Biophysics, The University of Colorado Health Sciences Center, Denver, Colorado 80262, USA
2 Present address: Division of Biology, 216-76, California Institute of Technology, Pasadena, California 91125, USA
Correspondence should be addressed to T.A.N. (teresa@etho.caltech.edu)

Synaptic activity modulates synaptic efficacy and is important in learning and development. Here we
show that development of excitability in presynaptic motor neurons required synaptic activation of
postsynaptic muscle cells. Synaptic blockade broadened action potentials and decreased repetitive
2000 Nature America Inc. http://neurosci.nature.com

firing of presynaptic neurons. Consistent with these findings, synaptic blockade also decreased
potassium-current density in the presynaptic cell. Application of neurotrophin-3, but not related
neurotrophins, prevented these changes. Recordings from patches of somatic membrane indicated
that modifications of presynaptic potassium and sodium currents occurred in a remote, nonsynaptic
compartment. Thus, activity-dependent postsynaptic signals modulated presynaptic excitability,
potentially regulating transmission at all synapses of the presynaptic cell.

Electrical activity regulates cell death, neurite outgrowth, neu-
ronal differentiation and synaptic competition in the developing
nervous system14. Although developmental changes in postsy-
naptic activity during synapse formation have been examined5,
regulation of excitability in presynaptic cells during development
and other forms of plasticity is not well understood. Moreover,
it is completely unknown whether activation of postsynaptic cells
regulates presynaptic excitability. This lack of understanding is
particularly striking in light of the huge body of data on synapse-
specific plasticity610, which cannot explain the spread of synap-
tic modulation that occurs in many neural systems1114.
Studies suggest that the intrinsic excitability of a single neuron
may be regulated according to its history of firing action poten-
tials1517. Real and model neurons can respond actively to exper-
imentally induced alterations in input frequency by changing
expression of voltage-dependent ion channels. The mechanism
underlying a neurons ability to calculate how these properties
should be altered and what aspect of its activity is important in
this calculation are not known. Brain-derived neurotrophic fac-
tor (BDNF) is proposed to be involved18, although the source of
this factor (pre-, post-, auto- or non-synaptic) and the mecha-
nism underlying its regulation of excitability are unclear. Others
propose that the neuron averages over time a sequence of guess-
es regarding how it should respond to given inputs16. Alterna-
tively, a neuron may respond to signals relayed from the
postsynaptic cell that indicate successful synaptic transmission.
Here we report that one aspect of activity important in regulating
neuronal excitability is the extent to which such activity induces
synaptic transmission and subsequent retrograde signaling from
activated targets.
Although synaptic transmission critically depends on the shape
and frequency of presynaptic action potentials, nothing is known
concerning the role of synaptic activity in their regulation. Here we
present evidence that presynaptic action potentials can be mod-
ulated by synaptic activation of efferent targets. Control of net-
work properties through feedback regulation of presynaptic
excitability by postsynaptic cells would allow targets to modulate
the responsiveness of their presynaptic neurons to afferent inputs.
Such regulation could have profound implications for the function
of neural networks in development and learning. Aspects of this
work have previously appeared in abstract form (T.A.N. and
A.B.R., Soc. Neurosci. Abstr. 25, 406.18, 1999).
RESULTS
We used neuromuscular junctions (NMJs) in Xenopus neu-
ron/myocyte co-cultures19 to investigate the role of the postsy-
naptic cell in the differentiation of presynaptic excitability. These
cultures are essentially homogeneous during the initial differen-
tiation of excitability, when duration of action potentials dra-
matically decreases20,21. However, these cultures contain a variety
of neuronal subtypes22. We hypothesized that, as neurons mature
in these co-cultures, they become heterogeneous with regard to
electrical excitability. In addition, we proposed that synaptic acti-
vation induces myocytes to release a retrograde signal that alters
motor neuron excitability. Myocytes were the ideal postsynaptic
cell for this investigation because they are morphologically dis-
tinguishable from other cells and contain a variety of factors that
affect motor neurons2327. To test our hypotheses, we first com-
pared the excitability of neurons that formed NMJs with
excitability of those without NMJs. We then examined neuron
excitability following blockade of synaptic activity with the nico-
tinic acetylcholine receptor blocker -bungarotoxin (-BgTx)
during NMJ formation and differentiation.
Neuronal properties correlate with synaptic contact
We found that neurons that did not contact muscle (solitary)
and those that contacted muscle but did not form an NMJ (no
NMJ) were significantly different from cells that formed func-
tional NMJs (Table 1; Figs. 1 and 2). Compared with neurons
lacking NMJs (solitary, no NMJ analyzed separately), cells with
NMJs had higher membrane capacitance (p < 0.0001), hyper-
polarized resting potential (p < 0.0002), shorter falling phases

142 nature neuroscience volume 3 no 2 february 2000

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

articles

Fig. 1. Neurons with NMJs had

a b shorter-duration action potentials
than other neurons. Traces show
sample whole-cell current-clamp

falling duration (ms)

recordings of motor neurons with

Action potential
functional NMJs (a) or myocyte
contact but no NMJ (b). Action
potentials had significantly shorter
falling durations in neurons with
NMJs compared with cells lacking
80 mV NMJs. *Significantly different from
control NMJ.

Control Control
Control NMJ Control no NMJ NMJ (29) no NMJ (28)

of the action potential (p < 0.04; Fig. 1), shorter refractory peri- NMJs fundamentally differ from cells that do not contact mus-
ods (compared only with no NMJ; p < 0.005; Fig. 2) and cle and from those that contact muscle but do not form func-
increased potassium-current (IK) density (stepped from 80 to 0 tional synapses. We next investigated whether these differences
2000 Nature America Inc. http://neurosci.nature.com

mV; p < 0.04). These findings indicate that neurons that form were due to intrinsic, cell-autonomous mechanisms of motor

Table 1. Properties of neurons vary with synaptic contact and function.

Control, with NMJ Control, no NMJ Control, solitary -BgTx, NMJ NT-3 + -BgTx, NMJ
Soma diameter (m) 23.0 0.3 23.2 0.3 23.2 0.3 22.5 0.3 23.9 0.4
n = 97 n = 85 n = 85 n = 136 n = 54

Membrane capacitance (pF) 64 4 30 1* 25 1* 50 3* 47 5*

n = 126 n = 136 n = 29 n = 175 n = 50

Input resistance (M) 723 64 836 132 689 137 681 61 646 75
n = 28 n = 13 n=6 n = 36 n = 15

Resting potential (mV) 51 2 41 2* 31 2* 43 1* 45 2*

n = 96 n = 54 n = 12 n = 111 n = 36

Mean rheobase (nA)

at 40 mV 0.65 0.03 0.60 0.04 0.54 0.05 0.80 0.05* 0.68 0.04
n = 28 n = 27 n=9 n = 45 n = 13
at 80 mV 1.64 0.05 1.52 0.04 1.55 0.07 1.57 0.03 1.48 0.05*
n = 30 n = 28 n = 10 n = 54 n = 20

Action potential falling duration

at 40 mV (ms) 1.2 0.1 1.7 0.2* 1.7 0.2* 1.7 0.1* 1.2 0.1
n = 28 n = 27 n=9 n = 45 n = 13
at 80 mV (ms) 1.2 0.1 1.8 0.2* 1.8 0.3* 2.1 0.2* 1.1 0.1
n = 29 n = 28 n = 10 n = 54 n = 20

Action potential amplitude

at 40 mV (mV) 58 2 58 3 49 4* 53 2* 53 2
n = 28 n = 27 n=9 n = 45 n = 13
at 80 mV (mV) 102 1 103 2 99 2 96 1* 96 2*
n = 29 n = 28 n = 10 n = 54 n = 20

Refractory period (ms) 13.2 1.0 20.4 2.7* 17.0 2.1 17.7 1.5* 12.2 1.3
n = 27 n = 15 n=6 n = 32 n = 10

Ik density at 0 mV (step 89 13 50 9* 51 9* 59 7* 109 15

from 80 mV; A/F) n = 17 n = 17 n = 11 n = 30 n = 17
Abbreviations: NMJ, neuromuscular junction; -BgTx, -bungarotoxin; NT3, neurotrophin-3. *Significantly different from control NMJ.

nature neuroscience volume 3 no 2 february 2000 143

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

articles

Fig. 2. Neurons with NMJs showed enhanced ability to fire repetitively.

(a) Refractory period was measured with a twin-pulse protocol. Three a
superimposed episodes from a sample cell are shown for each condi-
tion. Dotted lines indicate the end of the second current pulse for each
episode. (b) Refractory period (minimum time between two pulses that
each evoke an action potential) was shorter in neurons with NMJs com-
pared with those lacking NMJs. *Significantly different from control NMJ.

40 mV

neurons or whether signals from the postsynaptic muscle cell

induce a specific set of changes in innervating neurons.

Synaptic blockade alters motor neuron excitability Control NMJ Control no NMJ
Chronic blockade of the NMJ with rhodamine-conjugated -BgTx
changed passive (Table 1) and active (Table 1; Figs. 37) electro-
physiological properties of motor neurons. Compared with
untreated controls, motor neuron resting potential was signifi- b
cantly depolarized by chronic -BgTx treatment
2000 Nature America Inc. http://neurosci.nature.com

Refractory period (ms)

(p < 0.0001). In addition, synaptic blockade with -BgTx increased
the duration (p < 0.006; Fig. 3) and decreased the amplitude (p <
0.003) of the action potential. Similar changes in resting potential
and action potentials are observed during NMJ formation and dif-
ferentiation in vivo28,29. Thus, synaptic blockade seems to prevent
normal electrophysiological development of motor neurons.
Capacitance of motor neurons was also decreased by -BgTx
treatment (p < 0.0001; Table 1). Decreased capacitance seemed
to reflect changes in the process and/or synaptic terminal, as soma
diameter was unaffected. NMJ blockade with curare modestly
reduces the number of synapses formed at the earliest stages
examined30. Because we examined the initial stages of NMJ for-
mation, our results parallel these previous findings.
NT-3 prevents effects of synaptic blockade
If synaptic blockade prevents release of a retrograde signal and sub-
sequent differentiation of presynaptic excitability, then co-application
of the putative signal to -BgTx-treated junctions should overcome
the effects of synaptic inactivity. Because this neurotrophin alters
presynaptic secretion23,25,26,31 and is upregulated by myocyte depo-
larization31, NT-3 was the first candidate retrograde signal investi-
gated. Co-application of NT-3 blocked effects of -BgTx on resting
potential (versus -BgTx alone; p < 0.002) and action potential
falling duration (versus -BgTx alone; p < 0.008; Table 1; Fig. 3) but
not capacitance or action potential amplitude. Thus NT-3, a poten-
tial retrograde signal released from myocytes, alters a specific sub-
set of presynaptic electrophysiological properties.
Control Control
NMJ (27) no NMJ (15)
The increase in action potential falling duration resulting from -
BgTx-induced synaptic blockade (Fig. 3) revealed a regulatory mech-
anism that may explain previous reports of synaptic depression upon
caged-calcium release in the myocyte14,32. During normal NMJ devel-
opment without -BgTx or with the putative retrograde signal NT-
3 in addition to -BgTx, the spike substantially narrows, decreasing
the amount of transmitter released per action potential. Upregulation
of calcium in the myocyte increases NT-3 transcription31 and, pre-
sumably, release26,31. Thus, caged-calcium release in the myocyte
may induce release of NT-3, which would tend to narrow the presy-
naptic action potential and seem to produce synaptic depression.
Ability of motor neurons to fire repetitively increases during
development33,34. Spike narrowing may allow neurons to recover
from action potentials faster and thus enhance repetitive firing.
Consistent with this hypothesis, we found that synaptic blockade

a b
falling duration (ms)
Action potential

80 mV

Control -BgTx NT3 +

Control -BgTx NT3 + -BgTx
(29) (54) -BgTx (20)

Fig. 3. Chronic synaptic blockade with -BgTx induced broadening of action potentials that was prevented with neurotrophin-3. (a) Whole-cell cur-
rent-clamp recordings of motor neurons with functional NMJs following chronic -BgTx revealed dramatic broadening of action potentials that was
prevented with simultaneous application of NT-3. (b) Action potential broadening induced by -BgTx was due to an increase in the falling duration.
NT-3 prevented the -BgTx-induced increase in falling duration (indistinguishable from controls). *Significantly different from controls.

144 nature neuroscience volume 3 no 2 february 2000

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

articles

Fig. 4. Synaptic blockade reduced the ability to fire

repetitively, whereas co-application of NT-3 a 10 mV
1.5 nA
increased repetitive-firing capacity to control levels.
(a) Refractory period was measured with a twin- 5 ms
pulse protocol. Three episodes from a sample cell
are superimposed for each condition. The dotted
lines indicate the end of the second current pulse for
each episode. Note that action potentials overshoot
the stimulus artifact in control and NT-3 + -BgTx
conditions. (b) Refractory period was increased by 40 mV
-BgTx but unchanged by co-applied -BgTx and
NT-3. *Significantly different from controls.
Control -BgTx NT3 + -BgTx

with -BgTx, which produces spike broaden-

ing, does indeed decrease capacity for repetitive
b

Refractory period (ms)

firing as measured by an increase in refractory
period (Fig. 4, p < 0.02). NT-3, which prevents
2000 Nature America Inc. http://neurosci.nature.com

-BgTx-induced spike broadening, also pre-

vents the increase in refractory period (versus
-BgTx; p < 0.05). These changes in repetitive
firing globally regulate neuronal output. Thus, a
change initiated by postsynaptic feedback at one
synapse may modulate the efficacy of all synaps- Control -BgTx NT3 +
es of a given presynaptic neuron. (27) (32) -BgTx (10)

NT-3 effects on IK are specific

Changes in action potential duration suggest changes in IK, which To further test the hypothesis that IK was modulated by neu-
is the ionic current primarily responsible for repolarization during an rotrophins, we exposed cultures to K252a (0.2 M), which, at
action potential. We found that chronic synaptic blockade with - this concentration, is a relatively specific inhibitor for the neu-
BgTx decreased IK density in motor neurons (measured from 20 rotrophin-receptor protein-tyrosine kinases (Trks)25,35. In the
to 0 mV; p < 0.04; Fig. 5). Moreover, NT-3 prevented this decrease in absence of synaptic blockade, K252a decreased IK compared with
current density (versus -BgTx, 20 to +10 mV; p < 0.007). untreated controls (p < 0.01; Fig. 5c). This further indicates the
role of neurotrophins in the modulation of
a I K. To investigate potential non-specific
neurotrophic effects of NT-3, we attempted
to prevent -BgTx effects on IK with two
2 nA related neurotrophins, nerve growth factor
(NGF) and brain-derived neurotrophic fac-
20 ms tor (BDNF). Neither of these factors was
effective in preventing -BgTx-induced
changes in IK (NGF versus control at 0 mV,
p < 0.04; BDNF, p < 0.05; Fig. 5c). Thus,
NT-3 was the only neurotrophin tested that
was capable of reversing the effects of
synaptic blockade on IK.
Control -BgTx NT3 + -BgTx

b c
Current density (A/F)

Fig. 5. Decreases in delayed-rectifier IK density

induced by synaptic blockade were prevented by
Current density
at 0 mV (A/F)

co-application of NT-3, but not other neu-

rotrophins. (a) Sample whole-cell patch-clamp
NT3 + -BgTx (17) recordings of IK in motor neurons with NMJs
Control (17) revealed a substantial decrease in IK induced by
-BgTx (30) chronic -BgTx that was prevented by NT-3.
(b) Current-densityvoltage plots show that at
several voltages -BgTx induced a current
decrease that was prevented by NT-3.
NT3 + -BgTx (17)

K252a (10)
NGF + -BgTx (5)
BDNF + -BgTx (7)
Control (17)

-BgTx (30)

(c) Although prevented by NT-3, decreases in

current density induced by -BgTx were unaf-
fected by BDNF or NGF. K252a alone decreased
current densities to levels comparable to those
Voltage (mV)
of motor neurons treated with -BgTx alone.
*Significantly different from controls.

nature neuroscience volume 3 no 2 february 2000 145

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

articles

Fig. 6. -BgTx and NT-3 shifted the voltage dependence of activation of

IK. (a) The relative conductancevoltage curve for motor neuron IK a
shifted to more depolarized values with -BgTx-induced synaptic block-

Relative tail current

ade. This shift was prevented by NT-3. (b) Boltzmann fits of GV plots
revealed a significant change in V1/2 with -BgTx that was prevented by
co-application of NT-3.

The observed changes in IK density could be due to changes

in the number of channels expressed, the properties of channels or
both. If the number of channels in the plasma membrane changed,
the maximum conductance (and, thus, maximum current den-
sity) of the neuron should also change. We found that the number
of channels, as measured by maximum whole-cell tail-current
density, was not significantly different across groups (control, 56 Voltage (mV)
6.2 A/F, n = 16; -BgTx, 44 3.9 A/F, n = 27; NT-3 + -BgTx,
53 6.4 A/F, n = 12). We next examined changes in channel prop-
erties through comparison of normalized conductancevoltage
2000 Nature America Inc. http://neurosci.nature.com

(GV) relationships for control, -BgTX and NT-3 + -BgTx b

groups. Channel properties changed in response to -BgTx, as
indicated by a shift in the GV curve compared with controls
(measured from 20 to +10 mV; p < 0.02; Fig. 6a). Moreover, NT-
3 prevented this shift (versus -BgTx, 10 to +10 mV; p < 0.05). V1/2(mV)
To determine whether this shift was due to a change in the voltage
that gave a half-maximal conductance (V1/2) and/or a change in
the slope (k) of the GV curve, we fit data from each neuron with
the Boltzmann equation. Slope values were not significantly dif-
ferent (control, 15 1.2 mV; -BgTx, 13 0.6 mV; NT-3 + -
BgTx, 14 1.2 mV). In contrast, we found that V1/2 was shifted
to a more depolarized potential by -BgTx (p < 0.0005; Fig. 6b)
and that this shift was prevented by co-application of NT-3 (ver-
sus -BgTx; p < 0.03). The finding that -BgTx and NT-3 both
affected the same IK parameters further indicates that they act to whole-cell experiments, -BgTx shifted the GV curve in
upon the same pathway. macropatches to more depolarized potentials (measured from 0
to +30 mV; p < 0.03; Fig. 7b). In these experiments, we also
Synaptic blockade alters channels in a cell-wide manner noted an overall depolarizing shift in both groups compared to
To determine whether the effects of synaptic blockade extend whole-cell data, most likely due to a difference in C-type inac-
beyond the synapse, we examined IK properties in outside-out tivation between excised membrane patches and whole-cell
macropatches pulled from somata. We found that, comparable patches 36,37. The finding that macropatches pulled from the
soma show an effect initiated at the synapse suggests
a that this particular postsynaptic effect on presynap-
tic differentiation is cell-wide.
We next asked whether synaptic blockade might
globally affect other ionic currents. One candidate was
sodium current (I Na), as the observed decrease in
action potential amplitude with -BgTx (Table 1) sug-
gested modulation of this current. In macropatches
pulled from the soma, we found that the voltage
dependence of activation of INa was shifted to more
depolarized potentials in neurons after synaptic block-
ade with -BgTx (measured at 10 mV: p < 0.04; Fig.
8). Whole-cell data obtained with curare blockade of
b the NMJ yielded similar results (T.A.N. and A.B.R.,
unpublished observations).
g/gmax

Fig. 7. GV relation of IK shifted in response to -BgTx in

macropatches pulled from the somata of neurons as in
whole-cell recordings. (a) Sample recordings of IK in
macropatches pulled from motor neuron somata. (b) GV
plots of IK from control and -BgTx-treated motor neurons.
*Significantly different from controls.

146 nature neuroscience volume 3 no 2 february 2000

 2000 Nature America Inc. http://neurosci.nature.com
a pulled from the neuronal soma showed an -
b particularly important during target-
g/gmax
2000 Nature America Inc. http://neurosci.nature.com
DISCUSSION
Our data indicate that motor neuron excitability is functionally
regulated by postsynaptic target cells during development. This is
by no means a passive feedback circuit, as postsynaptic myocytes
must be synaptically activated for modulation of excitability to
occur. In contrast to well-described synapse-specific plasticity610,
the changes in excitability we report are global. Global regula-
tion of excitability that actively depends on feedback cues from
target cells represents a mechanism whereby neural systems may
self-organize and function. This mechanism would allow presy-
naptic neurons to modify all outputs simultaneously in response
to signals from postsynaptic cells.
Interestingly, the process we describe may underlie the spread
of retrograde modulation of presynaptic transmission reported in
the Xenopus NMJ culture system14. However, this group investi-
gated short-term modulation, whereas we investigated long-term
effects. Thus, the studies cannot be directly compared. Further
examination of the temporal aspect of neurotrophin efficacy
should reveal another level of complexity in the neu-
rotrophinexcitability relationship.
Although neurotrophins alter synaptic efficacy and neuronal
excitability3840, the mechanisms through which physiological
control of neurotrophic effects is functionally achieved are not
well understood40. Our data indicate that synaptic activity criti-
cally regulates retrograde signaling between myocyte and neu-
ron, and that the activity-dependent retrograde signal may be a
specific neurotrophin, NT-3.
In the context of neuromuscular development, cell-wide
changes in excitability due to postsynaptic activity may provide
the soma and all synapses of the motor neuron with a measure
of its success at competing in the periphery. Myocytes subject to
polyneuronal innervation may differentially regulate the excitabil-
ity of motor neurons based on synaptic efficacy, as previous stud-
ies suggest that retrograde signaling is localized to the site of
synaptic activation32 and that presynaptic depolarization facili-
tates neurotrophin-induced effects41. A motor neuron that effi-
nature neuroscience volume 3 no 2 february 2000
articles
Fig. 8. The GV relation of INa in macropatches
BgTx-induced shift. (a) Sample recordings of INa
in macropatches pulled from motor neuron
somata. (b) GV plots of INa in macropatches
from control and -BgTx-treated motor neu-
rons. *Significantly different from controls.
ciently stimulates a given myocyte would
become more excitable and, thus, more
successfully activate other contacted
myocytes. This type of regulation may be
dependent motor neuron death, during
which neurons seem to compete for limit-
ed muscle-derived factors42.
Studies suggest that neurons alter their
excitability in response to firing histo-
ry15,16,43. A potential mechanism for this
phenomenon is provided by our data,
which indicate that a neuron changes its
excitability based upon its recent success
at synaptic output, which depends on
action potentials. Cell-wide activity-depen-
dent modulation of presynaptic excitabil-
ity by postsynaptic targets would provide
a feedback mechanism through which the postsynaptic cell can
modulate not only a given presynaptic neurons responsiveness
to inputs, but also its output efficiency across all synapses, thus
altering output to other postsynaptic cells. This type of regula-
tion would have important effects on neural processing. Blocking
transmission at only a subset of synapses would test this hypoth-
esis and might yield further information on potential underly-
ing mechanisms.
METHODS
Animals and cell culture. Xenopus laevis embryos were produced by stan-
dard in-vitro fertilization techniques 44 and staged according to
Nieuwkoop and Faber45. Neuron/myocyte cultures were prepared as
described20. The dorsal-posterior region, which contained the pre-
sumptive spinal cord and surrounding somites, was removed from neur-
al tube stage (stage 1719) embryos and dissociated in divalent cation-free
medium (116.7 mM NaCl, 0.67 mM KCl, 0.4 mM EDTA and 4.6 mM
Tris at pH 7.8). Motor neurons from these embryonic stages have never
contacted muscle46. Co-cultures were plated on plastic dishes in a com-
pletely defined medium (116.7 mM NaCl, 0.67 mM KCl, 1.31 mM
MgSO4, 10 mM CaCl2 and 4.6 mM Tris at pH 7.8) and recorded 1526 h
after plating. Immediately after plating, 50 ng per ml NT-3, BDNF, NGF
(Alomone, Jerusalem, Israel), 0.2 M K252a (Calbiochem, San Diego,
California) and rhodaminated -BgTx (2 g per ml; Molecular Probes,
Eugene, Oregon) were added. Cultures were thus exposed to these
reagents for 1526 h before recording.
Electrophysiology. Current- and voltage-clamp records were obtained
by whole-cell patch clamp47 with 13 M electrodes. In control cultures,
motor neurons with NMJs were identified by their ability to cause
myocyte contraction upon stimulation with 6 voltage steps (60 ms each,
in 10-mV increments from +10 to +60 mV) from a holding potential of
40 mV. Neurons that did not cause contraction were placed in the no-
NMJ group. Cells that did not contact muscle were placed in the solitary
group. Only neurons that clearly had no contact with other neurons were
used. In rhodaminated--BgTx-treated cultures, neurons with NMJs
were identified by their ability to cause postsynaptic clustering of nicotinic
acetylcholine receptors (AChRs), observed as patches of rhodamine stain-
ing under the presynaptic terminal46. Synaptic blockade was confirmed
147
 2000 Nature America Inc. http://neurosci.nature.com
articles
by absence of myocyte contraction under motor neuron stimulation. To
verify that differences in method of identifying motor neurons in con-
trol versus -BgTx-treated cultures did not affect our results, we repeat-
ed experiments with the AChR blocker curare, which could be washed
out, thus allowing us to identify motor neurons by ability to produce
myocyte contraction. Under these conditions, results were similar to
those obtained with -BgTx (T.A.N. & A.B.R., Soc. Neurosci. Abstr. 24,
122.3, 1998).
Although previous studies conflict on whether AChRs on motor neu-
rons are -BgTx-sensitive48,49, both studies used neurons older than those
examined here. In contrast, neurons of the same age showed no acetyl-
choline response50. Because the motor neurons we examined did not express
functional AChRs, they were probably not directly affected by -BgTx.
The pipet solution for current clamp and potassium current records
contained 104 mM KCl, 3 mM MgCl2 and 10 mM HEPES at pH 7.4.
For current-clamp recording, the bath solution contained 125 mM
NaCl, 3 mM KCl, 10 mM CaCl2 and 5 mM HEPES at pH 7.4. Mem-
brane potential was forced to 40 and 80 mV with injected current.
Action potentials were initiated by 2-ms current injections, increased
in 7 increments of 0.25 nA from 0.5 to 2.0 nA, with 5-s recovery periods
2000 Nature America Inc. http://neurosci.nature.com
between stimuli. Only action potentials that overshot the stimulus arti-
fact by at least 2.5 mV were included in analyses. Falling duration was
measured as the time from the peak of the action potential to half-peak.
The minimum current that could reliably evoke action potentials with
a 5-s recovery period was used in a twin-pulse examination of relative
refractory period at 40 mV. The interval between the two pulses was
varied from 2 to 32 ms. Relative refractory period was defined as the
shortest interval between pulses that allowed an action potential in
response to the second pulse. Only depolarizations that overshot the
stimulus artifact by at least 2.5 mV were considered action potentials.
Input resistance was determined by a 500-ms current pulse of 0.01 or
0.1 nA. Capacitance was determined from voltage-clamp recordings.
The membrane potential was held at 40 mV and stepped to 90 mV in
6 steps of 10 mV. The area of the uncompensated capacitative current
(charge) was plotted against voltage and fit with a linear regression.
The slope of this line gave the capacitance. All data were analyzed with
a two-tailed Students t-test and reported as mean standard error.
Data were judged significant if p < 0.05. Numbers of observations are
indicated in parentheses.
For IK records, the bath solution contained 80 mM NaCl, 3 mM
KCl, 5 mM MgCl2, 10 mM CoCl2 and 5 mM HEPES at pH 7.4.
Tetrodotoxin (1 M; Calbiochem) was added to block voltage-gated
sodium currents. Neurons were held at 80 mV and stepped from 50
to +100 mV in 16 steps (10 mV, 60 ms) with a 5-s recovery period
between steps. To minimize the possibility of inadequate space clamp,
cells with capacitances above 35 pF were not used for IK analyses; this
cut-off was selected because currentcapacitance plots show a decline
in measured current density when capacitance is 40 pF (data not
shown). The -BgTx-induced GV shift found in whole-cell record-
ings was also found in macropatches, suggesting that poor control of
membrane voltage in the whole-cell recordings was not responsible
for the altered GV relationship. Records were filtered at 5 kHz and
sampled at 25 kHz with pClamp 6.2 (Axon, Foster City, California).
For current-density measurements, mean steady-state current was
measured 5060 ms after the initiation of the voltage step. Tail current
was measured 0.6 ms after the end of the pulse, which was beyond the
compensated capacitative artifact for all neurons included. Relative
tail current was used as a measure of whole-cell conductance because
this measure obviates considerations of driving force. Calculations of
conductance from steady-state current yielded similar results (data
not shown). For macropatch recordings, the signal-to-noise ratio was
not as high, and thus conductance was calculated from the steady-
state current. Maximum tail-current density was used as a measure of
channel number because the headstage of our amplifier (Axopatch
1C, Axon) was saturated by the large steady-state currents in control
NMJs, and because this measure removes driving-force considera-
tions. Inherent in the use of current density as a measure of channel
number is the assumption that single-channel conductance does not
change. For INa records, the bath solution contained 105 mM NaCl, 20
mM TEA-Cl, 3 mM KCl, 10 mM CoCl2 and 5 mM HEPES at pH 7.4.
148
The pipet solution contained 95 mM CsCl, 5 mM NaCl, 0.64 mM
CaCl2, 2 mM EGTA and 10 mM HEPES at pH 7.4. Neurons were held
at 80 mV and stepped from 40 to +40 mV in 9 steps (10 mV, 20 ms)
with a 5-s recovery period between steps.
ACKNOWLEDGEMENTS
We thank W. J. Betz, J. W. Karpen, J. L. Lubischer, T. C. Rich, K. R. Svoboda and
B. G. Wallace for reading the manuscript; and B. Lu, T. J. Carew, L. K.
Kaczmarek and M.-M. Poo for comments and suggestions. This work was
supported by NIH grants to T.A.N. and A.B.R.
RECEIVED 8 SEPTEMBER; ACCEPTED 19 NOVEMBER 1999
1. Schilling, K., Dickinson, M. H., Connor, J. A. & Morgan, J. I. Electrical
activity in cerebellar cultures determines Purkinje cell dendritic growth
patterns. Neuron 7, 891902 (1991).
2. Spitzer, N. C. A developmental handshake: neuronal control of ionic currents
and their control of neuronal differentiation. J. Neurobiol. 22, 659673 (1991).
3. Ruthazer, E. S. & Stryker, M. P. The role of activity in the development of
long-range horizontal connections in area 17 of the ferret. J. Neurosci. 16,
72537269 (1996).
4. Sherrard, R. M. & Bower, A. J. Role of afferents in the development and cell
survival of the vertebrate nervous system. Clin. Exp. Pharmacol. Physiol. 25,
487495 (1998).
5. Sanes, J. R. & Lichtman, J. W. Development of the vertebrate neuromuscular
junction. Annu. Rev. Neurosci. 22, 389442 (1999).
6. Malinow, R. Transmission between pairs of hippocampal slice neurons:
quantal levels, oscillations, and LTP. Science 252, 722724 (1991).
7. Huang, Y. Y., Colino, A., Selig, D. K. & Malenka, R. C. The influence of prior
synaptic activity on the induction of long-term potentiation. Science 255,
730733 (1992).
8. Castro-Alamancos, M. A., Donoghue, J. P. & Connors, B. W. Different forms
of synaptic plasticity in somatosensory and motor areas of the neocortex. J.
Neurosci. 15, 53245333 (1995).
9. Davis, G. W. & Goodman, C. S. Synapse-specific control of synaptic efficacy
at the terminals of a single neuron. Nature 392, 8286 (1998).
10. Martin, K. C. et al. Synapse-specific, long-term facilitation of Aplysia sensory
to motor synapses: a function for local protein synthesis in memory storage.
Cell 91, 927938 (1997).
11. Vincent, P. & Marty, A. Neighboring cerebellar Purkinje cells communicate
via retrograde inhibition of common presynaptic interneurons. Neuron 11,
885893 (1993).
12. Otani, S., Connor, J. A. & Levy, W. B. Long-term potentiation and evidence
for novel synaptic association in CA1 stratum oriens of rat hippocampus.
Learn. Mem. 2, 101106 (1995).
13. Muller, D., Hefft, S. & Figurov, A. Heterosynaptic interactions between LTP
and LTD in CA1 hippocampal slices. Neuron 14, 599605 (1995).
14. Cash, S., Zucker, R. S. & Poo, M. M. Spread of synaptic depression mediated
by presynaptic cytoplasmic signaling. Science 272, 9981001 (1996).
15. Desai, N. S., Rutherford, L. C. & Turrigiano, G. G. Plasticity in the intrinsic
excitability of cortical pyramidal neurons. Nat. Neurosci. 2, 515520 (1999).
16. Stemmler, M. & Koch, C. How voltage-dependent conductances can adapt to
maximize the information encoded by neuronal firing rate. Nat. Neurosci. 2,
521527 (1999).
17. Spitzer, N. C. New dimensions of neuronal plasticity. Nat. Neurosci. 2,
489491 (1999).
18. Desai, N. S., Rutherford, L. C. & Turrigiano, G. G. BDNF regulates the
intrinsic excitability of cortical neurons. Learn. Mem. 6, 284291 (1999).
19. Henderson, L. P., Smith, M. A. & Spitzer, N. C. The absence of calcium blocks
impulse-evoked release of acetylcholine but not de novo formation of
functional neuromuscular synaptic contacts in culture. J. Neurosci. 4,
31403150 (1984).
20. Spitzer, N. C. & Lamborghini, J. E. The development of the action potential
mechanism of amphibian neurons isolated in culture. Proc. Natl. Acad. Sci.
USA 73, 16411645 (1976).
21. Henderson, L. P. & Spitzer, N. C. Autonomous early differentiation of
neurons and muscle cells in single cell cultures. Dev. Biol. 113, 381387
(1986).
22. Bixby, J. L. & Spitzer, N. C. The appearance and development of
neurotransmitter sensitivity in Xenopus embryonic spinal neurones in vitro. J.
Physiol. (Lond.) 353, 143155 (1984).
23. Lohof, A. M., Ip, N. Y. & Poo, M.-M. Potentiation of developing
neuromuscular synapses by the neurotrophins NT-3 and BDNF. Nature 363,
350353 (1993).
24. Stoop, R. & Poo, M.-M. Potentiation of transmitter release by ciliary
neurotrophic factor requires somatic signaling. Science 267, 695699
(1995).
25. Wang, T., Xie, K. & Lu, B. Neurotrophins promote maturation of developing
neuromuscular synapses. J. Neurosci. 15, 47964805 (1995).
nature neuroscience volume 3 no 2 february 2000
 2000 Nature America Inc. http://neurosci.nature.com
26. Liou, J.-C. & Fu, W.-M. Regulation of quantal secretion from developing
motoneurons by postsynaptic activity-dependent release of NT-3. J. Neurosci.
17, 24592468 (1997).
27. Wang, X. H. & Poo, M. M. Potentiation of developing synapses by
postsynaptic release of neurotrophin-4. Neuron 19, 825835 (1997).
28. Kellerth, J. O., Mellstrom, A. & Skoglund, S. Postnatal excitability changes of
kitten motoneurones. Acta Physiol. Scand. 83, 3141 (1971).
29. Ziskind-Conhaim, L. Electrical properties of motoneurons in the spinal cord
of rat embryos. Dev. Biol. 128, 2129 (1988).
30. Dahm, L. M. & Landmesser, L. T. The regulation of synaptogenesis during
normal development and following activity blockade. J. Neurosci. 11,
238255 (1991).
31. Xie, K., Wang, T., Olafsson, P., Mizuno, K. & Lu, B. Activity-dependent
expression of NT-3 in muscle cells in culture: implications in the
development of neuromuscular junctions. J. Neurosci. 17, 29472958 (1997).
32. Cash, S., Dan, Y., Poo, M.-M. & Zucker, R. Postsynaptic elevation of calcium
induces persistent depression of developing neuromuscular synapses. Neuron
16, 745754 (1996).
33. Fulton, B. P. & Walton, K. Electrophysiological properties of neonatal rat
motoneurones studied in vitro. J. Physiol. (Lond.) 370, 651678 (1986).
34. Xie, H. & Ziskind-Conhaim, L. Blocking Ca2+-dependent synaptic release
delays motoneuron differentiation in the rat spinal cord. J. Neurosci. 15,
59005911 (1995).
35. Berg, M. M., Sternberg, D. W., Parada, L. F. & Chao, M. V. K-252a inhibits
2000 Nature America Inc. http://neurosci.nature.com
nerve growth factor-induced trk proto-oncogene tyrosine phosphorylation
and kinase activity. J. Biol. Chem. 267, 1316 (1992).
36. Marom, S., Goldstein, S. A., Kupper, J. & Levitan, I. B. Mechanism and
modulation of inactivation of the Kv3 potassium channel. Receptors Channels
1, 8188 (1993).
37. Kupper, J., Bowlby, M. R., Marom, S. & Levitan, I. B. Intracellular and
extracellular amino acids that influence C-type inactivation and its
modulation in a voltage-dependent potassium channel. Pflugers Arch. 430,
111 (1995).
nature neuroscience volume 3 no 2 february 2000
articles
38. Lesser, S. S., Sherwood, N. T. & Lo, D. C. Neurotrophins differentially
regulate voltage-gated ion channels. Mol. Cell. Neurosci. 10, 173183
(1997).
39. Sherwood, N. T., Lesser, S. S. & Lo, D. C. Neurotrophin regulation of ionic
currents and cell size depends on cell context. Proc. Natl. Acad. Sci. USA 94,
59175922 (1997).
40. McAllister, A. K., Katz, L. C. & Lo, D. C. Neurotrophins and synaptic
plasticity. Annu. Rev. Neurosci. 22, 295318 (1999).
41. Boulanger, L. & Poo, M.-M. Presynaptic depolarization facilitates
neurotrophin-induced synaptic potentiation. Nat. Neurosci. 2, 346351 (1999).
42. Purves, D. & Lichtman, J. W. in Principles of Neural Development 131154
(Sinauer, Sunderland, Massachusetts, 1985).
43. Golowasch, J., Abbott, L. F. & Marder, E. Activity-dependent regulation of
potassium currents in an identified neuron of the stomatogastric ganglion of
the crab Cancer borealis. J. Neurosci. 19, RC33 (1999).
44. Tabti, N. & Poo, M.-M. in Culturing Nerve Cells (eds. Banker, G. & Goslin, K.)
137154 (MIT Press, Cambridge, Massachusetts, 1991).
45. Nieuwkoop, P. D. & Faber, J. Normal Table of Xenopus laevis (Daudin,
Amsterdam, 1967).
46. Kullberg, R. W., Lentz, T. L. & Cohen, M. W. Development of the myotomal
neuromuscular junction in Xenopus laevis: an electrophysiological and fine-
structural study. Dev. Biol. 60, 101129 (1977).
47. Hamill, O. P., Marty, A., Neher, E., Sakmann, B. & Sigworth, F. J. Improved
patch-clamp techniques for high-resolution current recording from cells and
cell-free membrane patches. Pflugers Arch. 391, 85100 (1981).
48. Perrins, R. & Roberts, A. Nicotinic and muscarinic ACh receptors in
rhythmically active spinal neurones in the Xenopus laevis embryo. J. Physiol.
(Lond.) 478, 221228 (1994).
49. Fu, W. M., Liou, H. C. & Chen, Y. H. Nerve terminal currents induced by
autoreception of acetylcholine release. J. Neurosci. 18, 99549961 (1998).
50. Bixby, J. L. & Spitzer, N. C. The appearance and development of
chemosensitivity in Rohon-Beard neurones of the Xenopus spinal cord. J.
Physiol. (Lond.) 330, 513536 (1982).
149
 2000 Nature America Inc. http://neurosci.nature.com

articles

A new form of long-term

depression in the perirhinal cortex
K. Cho1, N. Kemp1, J. Noel1,2, J. P. Aggleton3, M. W. Brown1 and Z. I. Bashir1

1 MRC Centre for Synaptic Plasticity, Dept. of Anatomy, University of Bristol, Bristol BS8 1TD, UK
2 Present address: Biologie cellulaire des compartiments calciques, Universite de Nice-Sophia Antipolis, Faculte des sciences, 06108 Nice, France
3 School of Psychology, Cardiff University, Cardiff CF10 3YJ, UK
Correspondence should be addressed to Z.I.B. (z.i.bashir@bris.ac.uk)

We demonstrate a form of long-term depression (LTD) in the perirhinal cortex that relies on interac-
tion between different glutamate receptors. Group II metabotropic glutamate (mGlu) receptors
2000 Nature America Inc. http://neurosci.nature.com

facilitated group I mGlu receptor-mediated increases in intracellular calcium. This facilitation plus
NMDA receptor activation may be necessary for induction of LTD at resting membrane potentials.
However, depolarization enhanced NMDA receptor function and removed the requirement of
synergy between group I and group II mGlu receptors: under these conditions, activation of only
NMDA and group I mGlu receptors was required for LTD. Such glutamate receptor interactions
potentially provide new rules for synaptic plasticity. These forms of LTD occur in the perirhinal
cortex, where long-term decreases in neuronal responsiveness may mediate recognition memory.

There are good reasons for believing that decreases in synaptic
efficacy in neurons of the perirhinal cortex, the region of tem-
poral cortex adjacent to the rhinal sulcus, are related to recogni-
tion memory1. Lesions of perirhinal cortex in rats and primates
impair performance of recognition-memory tasks24. Importantly,
perirhinal neuronal responses to novel visual stimuli decrease
markedly once such stimuli become familiar1,5,6. The decrement
in responsiveness can last many hours, and this change may pro-
vide the information required to solve such tasks. Activity-depen-
dent long-term depression of synaptic transmission in perirhinal
cortex provides a potential mechanism for explaining decreases in
neuronal responsiveness.
Induction of homosynaptic LTD depends on postsynaptic
increases in calcium79 brought about by different mechanisms
that include activation of NMDA receptors1012 or mGlu recep-
tors1317. However, there is no evidence for a role of mGlu recep-
tor activation under conditions in which NMDA receptor
activation underlies the induction of activity-dependent LTD.
Furthermore, there seems to be no role for NMDA receptor acti-
vation when mGlu receptors are involved in the induction of
activity-dependent LTD. Group I mGlu receptors and NMDA
receptors cause distinct forms of LTD by coupling to the activa-
tion of protein kinase C and protein phosphatases, respective-
ly1416,18. An alternative potential means of LTD induction by
group I mGlu receptor activation is through phosphoinositide-
induced release of calcium from intracellular stores1922. A presy-
naptic role for group II mGlu receptors in LTD at mossy fiber
synapses23,24 seems to depend on cAMP turnover25, but the mech-
anisms underlying the role of group II mGlu receptors in other
regions are not well understood2628.
Perirhinal cortex is critically involved in a number of different
types of memory. NMDA receptor-dependent LTP29,30 can be
induced in perirhinal cortex in vitro; in addition, investigating
mechanisms of LTD may provide insights into fundamental
mechanisms of learning and memory in this and other regions
of cortex. Here we describe conditions for the induction of activ-
ity-dependent LTD in adult perirhinal cortex. We demonstrate
a new form of LTD in this cortex that relies on the activation of
NMDA and group I and group II mGlu receptors.
RESULTS
In the perirhinal cortex, stimulation delivered to either side of the
rhinal sulcus in layers II/III resulted in synaptic transmission that
was dependent on AMPA/kainate and NMDA receptors30 (Fig. 1).
Low-frequency stimulation (LFS; 200 stimuli, 1 Hz) delivered to
either the entorhinal or the temporal side of the rhinal sulcus
combined with depolarization of the postsynaptic neuron to 40
mV induced robust homosynaptic LTD measured 3035 minutes
after stimulation. No differences were found between the magni-
tude of LTD with stimulation to the entorhinal or the temporal
cortex side (p > 0.05); therefore, data from the two inputs were
pooled (mean depression of 47 8%, n = 10; Fig. 2a) in this and
subsequent experiments. Voltage clamp of neurons at 40 mV for
200 s without LFS did not result in LTD (+1 8%; n = 8, p > 0.05;
data not shown). LTD was also induced by LFS delivered while
the postsynaptic cell was voltage clamped at 70 mV (depression
of 37 8%, n = 8; Fig. 2b). LTD induced by either of the above
protocols was maintained for as long as the recording was con-
tinued (up to 180 min; data not shown).
The NMDA receptor antagonist AP5 (50 M) blocked the
induction of LTD measured 3035 min after LFS delivered at 40
mV (2 5%, n = 7, p > 0.05; Fig. 2c). In three experiments, AP5
was washed out, and subsequent LFS resulted in LTD (52 11%,
n = 3, p < 0.05; Fig. 2c). In a separate series of experiments, AP5
blocked LTD at 70 mV (1 5%, n = 7; p > 0.05; Fig. 2d) in a
reversible manner (48 9%, p < 0.05, n = 3; Fig. 2d). In keep-
ing with the voltage dependence of NMDA receptor-mediated
synaptic transmission, LTD was prevented by hyperpolarizing
neurons to 90 mV (7 6%, n = 3, not shown) or 110 mV
(3 11%, n = 3, not shown). To determine if a postsynaptic rise

150 nature neuroscience volume 3 no 2 february 2000

 2000 Nature America Inc. http://neurosci.nature.com
a required for the induction of LTD. This result was unexpected
b EGLU was washed out, and subsequent LFS induced LTD (46
2000 Nature America Inc. http://neurosci.nature.com
c value close to normal resting potential. With the membrane
Fig. 1. The perirhinal cortex. (a) Lateral view of the rat brain (from ref.
44) indicating the position of the perirhinal cortex (areas 35 and 36) to
either side of the rhinal sulcus. Approximate angles of in vitro slices are
illustrated by diagonal lines. (b) Representation of the perirhinal cortex
slice indicating electrode positions. Whole-cell recordings (Rec) were
obtained from single neurons in layer II/III. Stimulating electrodes (S1
and S2) were positioned in layer II/III either side (temporal and entorhi-
nal) of the recording electrode. (c) Synaptic EPSCs in perirhinal cortex
neurons voltage clamped at 70 mV rely on activation of AMPA/kainate
and NMDA receptors, as demonstrated by the effects of NBQX and
AP5 (ref. 30).
in calcium was necessary for LTD, we used whole-cell filling solu-
tion containing 10 mM EGTA. Under these recording conditions,
induction of LTD was prevented when LFS was delivered at either
40 or 70 mV (Fig. 2e and f).
To determine if metabotropic glutamate (mGlu) receptors
have a role in LTD, we used the broad-spectrum (group I/II)
mGlu receptor antagonist MCPG31,32. MCPG (500 M) prevented
the induction of LTD when LFS was delivered at 40 mV (Fig.
3a; 1 6%, n = 7, p > 0.05). Following MCPG washout in 3
experiments, LFS resulted in LTD (47 7%, n = 3, p < 0.05; Fig.
3a). MCPG also blocked induction of LTD at 70 mV (Fig. 3b;
1 9%, n = 7, p > 0.05) in a reversible manner (46 6%, n = 3,
p < 0.05; Fig. 3b). The group I mGlu receptor antagonist
AIDA33,34 blocked LTD at 40 mV (Fig. 3c; +1 7%, n = 7, p >
0.05) in a reversible manner (35 7%, n = 3, p < 0.05). At 70
mV, AIDA also produced a block of LTD induction (Fig. 3d; 1
11%, n = 6, p > 0.05) that was reversible (35 10%, n = 2).
MAP4, a group III mGlu receptor antagonist35, failed to prevent
LTD at either 40 or 70 mV (43 11%, and 41 10%, respec- onism of either NMDA or group I mGlu receptors and because
tively, n = 4, p < 0.05 for each; Fig. 3e and f). These results suggest
that both group I mGlu receptors and NMDA receptors are
nature neuroscience volume 3 no 2 february 2000
articles
because in previously reported cases of activity-dependent LTD,
induction relies on either NMDA or mGlu receptors but not on
activation of both.
The activation of group II metabotropic glutamate (mGlu)
receptors by application of DCG-IV can result in a long-lasting
depression of field EPSPs in perirhinal cortex36. We investigated
whether synaptic activation of group II mGlu receptors has a role
in LTD. The group II mGlu receptor antagonist EGLU37 did not
block the induction of LTD when LFS was delivered at 40 mV
(depression of 40 9%, n = 4; p < 0.01, Fig. 4a). Surprisingly,
however, EGLU blocked LTD when LFS was delivered at 70 mV
(depression of 3 7%, n = 7; p > 0.05; Fig. 4b). In 3 experiments,
12%, n = 3, p < 0.05; Fig. 4b). LTD was also blocked by the group
II/III mGlu receptor antagonist CPPG (200 M) when LFS was
delivered at 70 mV (2 10% in CPPG and 47 14% following
washout; n = 3; data not shown).
Because the role of group II mGlu receptors in LTD depend-
ed on membrane potential, we investigated whether group II
mGlu receptors were involved in LTD under protocols in which
the membrane potential was not voltage clamped, thus more
closely approximating conditions in vivo. LFS was delivered while
the postsynaptic neuron was held in current clamp at 70 mV, a
potential free to change during LFS, LTD was still reliably induced
(depression of 39 7%, n = 3; p < 0.05, Fig. 4c). Interestingly,
under these conditions LTD still depended on the activation of
group II mGlu receptors, as LTD was blocked by EGLU (3 14%,
n = 3; p > 0.05, Fig. 4d).
The voltage dependence of involvement of group II mGlu
receptors in LTD was unexpected. We reasoned that group II
mGlu receptors might be important during decreased NMDA
receptor activation. Therefore, their involvement might relate to
voltage dependence of NMDA receptor activation, and not to a
voltage dependence of group II mGlu receptors per se. If this were
the case, increasing NMDA receptor activation and calcium influx
at 70 mV should remove the requirement for group II mGlu
receptors in LTD induction. In keeping with this suggestion, when
extracellular calcium was increased (from 2 to 4 mM), LTD
induced at 70 mV was not blocked by EGLU (63 7%, n = 3; p
< 0.01, Fig. 5a). When extracellular magnesium was reduced
(from 1 to 0.01 mM) to enhance NMDA receptor function, LTD
induced at 70 mV was not blocked by EGLU (41 9% depres-
sion, n = 3; p < 0.01, Fig. 5b). Thus enhancing NMDA receptor
activation removes group II mGlu receptor involvement in LTD.
Conversely, decreasing NMDA receptor activation at 40 mV
might produce a requirement for group II mGlu receptor activa-
tion in LTD at depolarized membrane potentials. In a low con-
centration of AP5 (2 M), LTD was induced by pairing LFS with
depolarization of the postsynaptic neuron to 40 mV. LTD
induced at 40 mV under these conditions was blocked by EGLU
(+6 10%, measured 10 min after LFS; n = 3; p > 0.05, Fig. 5c
and d). These results suggest that the levels of NMDA receptor
activation determine the requirement of group II mGlu receptors
in LTD.
One explanation for these results is that cooperativity between
group I and II mGlu receptors and NMDA receptors may be
required to increase calcium levels for induction of LTD at rest-
ing membrane potentials. Because LTD is blocked during antag-
group II mGlu receptors do not directly couple to calcium sig-
naling, it is possible that group II mGlu receptors interact with
151
 2000 Nature America Inc. http://neurosci.nature.com
articles
a b
Normalized peak EPSC
(% of baseline)
Time (min)
c d
Normalized peak EPSC
2000 Nature America Inc. http://neurosci.nature.com
(% of baseline)
Time (min)
e f
Normalized peak EPSC
(% of baseline)
Time (min)
NMDA or group I mGlu receptors. We demonstrated a syner-
gistic interaction between group II and group I mGlu receptors
(Fig. 6). At 70 mV, the inward current produced by the group
I mGlu receptor agonist DHPG37 was significantly (p < 0.05)
increased by the group II mGlu receptor agonist DCG-IV (Fig. 6a;
15 6 without DCG-IV; 59 17 pA with DCG-IV, n = 4).
Because DCG-IV acts as an agonist at NMDA receptors38 (albeit
at higher concentrations than 0.5 M), these experiments were
carried out in the presence of 50 M AP5. In control experiments
without DCG-IV, a second application of DHPG produced
inward currents of similar magnitudes to that produced by a first
application of DHPG (data not shown).
To investigate how interaction between group I and group II
mGlu receptors might facilitate LTD, we monitored intracellular
calcium levels. In cultured perirhinal cortex neurons, DHPG (20
M) increased fluorescence of Fluo-3-AM filled neurons
(21 6% increase; n = 16 neurons). This was significantly (p <
0.05) enhanced (45 10%; n = 16, Fig. 6b and d) by co-applica-
tion of DHPG with the group II mGlu receptor agonist DCG-IV
(1 M). DCG-IV itself had no effect on fluorescence levels
152
Fig. 2. LTD relies on activation of
NMDA receptors and a postsynaptic
increase in calcium. (a) LFS paired
with depolarization to 40 mV results
in LTD in the input receiving LFS but
not in the control input (not shown).
(b) Delivering LFS with the neuron
voltage clamped at 70 mV also results
in homosynaptic LTD, with no effect
on the control input (not shown). (c)
LTD is blocked by the NMDA recep-
tor antagonist AP5 when LFS is paired
with depolarization to 40 mV. LFS fol-
lowing AP5 washout in three of these
Time (min) experiments induced LTD. (d) LTD
was blocked by AP5 when LFS was
delivered at 70 mV. In three experi-
ments, LFS induced LTD after AP5 was
washed out. (e, f) The calcium chela-
tor EGTA (10 mM) blocks the induc-
tion of LTD 2025 min after LFS at
both 40 mV (e) and 70 mV (f). The
control concentration of EGTA (0.5
mM) had no effect on LTD in inter-
leaved control experiments (not
shown). Illustrated synaptic responses
are taken from appropriate time
points, as indicated (1, 2). The control
and depressed responses (and voltage
Time (min)
steps; 5 mV) are also shown superim-
posed (1, 2). The period of LFS (200
stimuli, 1 Hz) is indicated by the hori-
zontal bar joined by the two arrows. In
this and subsequent figures, filled cir-
cles indicate experiments in which the
neuron was depolarized to 40 mV
only for the duration of LFS. Open cir-
cles indicate experiments in which LFS
was delivered while the neuron was
voltage clamped at 70 mV. Breaks in
the x axes (c, d) indicate the time at
which AP5 was washed out.
Time (min)
(Fig. 6c). In separate experiments (Fig. 6c and e), the enhance-
ment by DCG-IV of DHPG-induced calcium mobilization
(44 7%; n = 23 neurons) was reversibly blocked by the group II
mGlu receptor antagonist EGLU (15 4%; n = 23 neurons).
Therefore, group II and group I mGlu receptors can interact to
increase intracellular calcium levels in neurons of the perirhinal
cortex. We postulate that this synergistic increase in calcium may
underlie the observed induction of LTD at resting membrane
potentials.
DISCUSSION
This study describes an activity-dependent LTD in the adult
perirhinal cortex. This finding is significant because this LTD
may reflect mechanisms underlying recognition memory-related
decreases in neuronal responses in perirhinal cortex in vivo1,5,6.
The mechanisms of homosynaptic activity-dependent LTD in
the perirhinal cortex are notable because these depend on synap-
tic activation of both NMDA receptors and mGlu receptors, in
marked contrast to previously described forms of activity-depen-
dent LTD that depend either on the activation of NMDA recep-
nature neuroscience volume 3 no 2 february 2000
 2000 Nature America Inc. http://neurosci.nature.com

articles

Fig. 3. LTD depends on group I but

a b not group III mGlu receptor activa-
tion. (a) LTD is blocked by the mGlu
Normalized peak EPSC

receptor antagonist MCPG when

(% of baseline)

LFS is delivered at 40 mV. In three

experiments, MCPG was washed
out, and LTD was subsequently
induced. (b) LTD was blocked by
MCPG at 70 mV; in 3 experiments,
LFS resulted in LTD following
washout. (c) The group I mGlu
receptor antagonist AIDA prevented
the induction of LTD at 40 mV; in
Time (min) Time (min) three experiments, LFS after AIDA
washout resulted in LTD. (d) LTD
c d was blocked by AIDA at 70 mV. In
two experiments, AIDA was washed
out and LTD subsequently induced
Normalized peak EPSC

by LFS. (e, f) LTD was not blocked

by MAP4 either at 40 mV (e) or at
(% of baseline)
2000 Nature America Inc. http://neurosci.nature.com

70 mV (f).

ever, at resting membrane poten-

tials, when calcium influx
through NMDA receptor chan-
nels is restricted, the additional
Time (min) Time (min)
activation of group I mGlu
receptors is insufficient for LTD.
Under these conditions, induc-
e tion of LTD requires activation of
f group II mGlu receptors. It is not
known if this mechanism confers
Normalized peak EPSC

unique properties on perirhinal

(% of baseline)

cortical plasticity or if it also

Time (min)
tors1012 or on the activation of mGlu receptors13,14,16,17, but not
on the conjoint activation of both. Indeed, distinct NMDA and
mGlu receptor-dependent forms of LTD coexist in the CA1 region
of the hippocampus15. Although we have not investigated bio-
chemical mechanisms of perirhinal LTD, NMDA receptor-depen-
dent LTD relies on protein phosphatase activation18, whereas
mGlu receptor-dependent LTD depends on protein kinase C acti-
vation14,16 in the hippocampus. The conjoint activation of NMDA
and mGlu receptors provides an additional rule for the induction
of synaptic plasticity. The cerebral cortex can potentially exploit
the variety of such rules, either to effect different types of memo-
ry or to increase the conditions for memory storage.
The most intriguing finding is that Group II mGlu receptors
are involved in LTD at resting membrane potentials but not at
depolarized potentials. Our results indicate that the role for group
II mGlu receptors depends on the level of NMDA receptor acti-
vation, rather than membrane potential per se. We propose that
at depolarized potentials, when NMDA receptor activation is
increased, co-activation of NMDA and group I mGlu receptors
is sufficient and necessary to trigger the induction of LTD. How-
occurs in widespread cortical
regions. Thus, although group II
mGlu receptors have a role in
LTD in other brain regions2628,
it is not known if this also occurs
at resting membrane potentials,
as reported here. Although acti-
vation of postsynaptic group II
Time (min) mGlu receptors probably
explains our results, the possibil-
ity that presynaptic group II mGlu receptors also have some role
in LTD cannot be ruled out23,24. It is interesting to note that the
mGlu receptor antagonists MCPG and EGLU reduced the depres-
sion of transmission during LFS, an effect more evident when
LFS was delivered at 70 mV. This suggests that synaptic activa-
tion of mGlu receptors may transiently decrease synaptic trans-
mission, possibly via a presynaptic mechanism, as has been
demonstrated at mossy fiber synapses39. At 40 mV, the decreased
driving force also results in EPSP depression, making it difficult
to determine the effects of mGlu receptor antagonists on depres-
sion during LFS.
Group II mGlu receptors are located both pre- and postsy-
naptically and are known to decrease forskolin-stimulated cAMP
turnover23,24. However, we provide evidence that the activation
of group II mGlu receptors can increase group I mGlu receptor-
mediated calcium mobilization. This result extends previous
work40,41 showing that group II mGlu receptors increase group
I mGlu receptor-dependent phosphoinositide turnover. The pre-
cise mechanisms underlying the synergy between these groups
of mGlu receptors remain to be identified. However, this inter-

nature neuroscience volume 3 no 2 february 2000 153

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

articles

Fig. 4. The role of group II

mGlu receptors in LTD is a b
voltage dependent. (a) The
group II mGlu receptor

Normalized peak EPSC

antagonist EGLU did not

(% of baseline)
block the induction of LTD
when LFS was delivered at
40 mV. (b) However, when
LFS was delivered at 70
mV, LTD was blocked by
EGLU but could be induced
in three experiments fol-
lowing washout. (c)
Delivering LFS to the neu-
ron in current clamp (at 70 Time (min) Time (min)
mV) results in LTD induc- c d
tion that is also blocked by
the group II mGlu receptor
antagonist EGLU (d).
Normalized peak EPSC

current clamp
(% of baseline)
2000 Nature America Inc. http://neurosci.nature.com

current clamp

action may occur directly

between G subunits
(activated by group II
mGlu receptors) and
Gq/11 (activated by group
I mGlu receptors)40 rather Time (min) Time (min)
than involve group II
mGlu receptor-mediated
changes in cAMP turnover41. Whatever the underlying mecha- Therefore, these results suggest a mechanism for the requirement
nism, our results demonstrate that this interaction increases intra- of synaptic activation of group II mGlu receptors in LTD.
cellular calcium. Given the essential role of calcium in the Our results demonstrate that LTD in the perirhinal cortex can
induction of synaptic plasticity, this synergy may be required for depend on several different types of glutamate receptor, thus rais-
LTD in the perirhinal cortex at resting membrane potentials. ing the question as to the physiological relevance of various mech-

a c
Normalized peak EPSC

Normalized peak EPSC

(% of baseline)

(% of baseline)

b Time (min) d Time (min)

Normalized peak EPSC

Normalized peak EPSC

(% of baseline)
(% of baseline)

Time (min) Time (min)

Fig. 5. The role of group II mGlu receptors in LTD is influenced by the level of NMDA receptor activation. (a, b) EGLU does not block LTD at 70
mV in 4 mM extracellular Ca2+ (a) or in 0.01 mM extracellular Mg2+ (b). (c, d) Single example (c) and pooled data (d) show that EGLU prevents the
induction of LTD at 40 mV when NMDA receptor activation is decreased by 2 M AP5. Induction of LTD was not blocked by this concentration of
AP5 alone (c).

154 nature neuroscience volume 3 no 2 february 2000

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

articles

Fig. 6. Group II mGlu receptor activation by DCG-IV

enhances group I mGlu receptor (DHPG)-induced inward a

Change in holding
current and calcium mobilization. (a) DHPG induced an

current (pA)
inward current that was significantly enhanced by DCG-IV. To
prevent any agonist effect of DCG-IV at NMDA receptors, we
used a low concentration of DCG-IV (0.5 M) in the presence
of 50 M AP5. (b) Single example of fluorescence increase in a
cultured perirhinal cortex neuron in response to application
of 20 M DHPG. The calcium rise in response to DHPG is
synergistically increased in a reversible manner by the co-
application of DCG-IV. (c) Single example illustrates reversible Time (min)
block by EGLU of DCG-IVs enhancement of DHPG-induced
fluorescence increase. (d) Pooled data illustrate that DHPG- b c
induced calcium mobilization was significantly enhanced by
DCG-IV. (e) Pooled data show that DCG-IVs enhancement of

( % of baseline)

(% of baseline)
Fluorescence
the DHPG response is blocked by EGLU. Neither EGLU nor

Fluorescence
DCG-IV alone had any effect on fluorescence.
2000 Nature America Inc. http://neurosci.nature.com

anisms. One possibility is that the decreased neuronal

activity with stimulus repetition in perirhinal cortex1 aris- Time (s) Time (s)
es from a variety of afferent-activation patterns appro- d e
priately encoded by the various synaptic and molecular
mechanisms we identified. It is also possible that the dif-
ferent mechanisms underlying synaptic plasticity are nec-

(% of baseline)
Fluorescence
(% of baseline)
Fluorescence

essary for the variety of forms of learning and memory

involving neurons in perirhinal cortex14. In conclusion,
the results of this study identify new mechanisms of activ-
ity-dependent synaptic plasticity that rely on a function-
al interaction between different classes of synaptically
activated glutamate receptors.
METHODS
Electrophysiology. Slices of perirhinal cortex were prepared
from adult male DA rats (150270 g, 712 weeks, Bantin and
Kingman, UK). All efforts were made to minimize numbers
of animals used. Animals were anesthetized with halothane,
decapitated in accordance with the UK Animals (Scientific Procedures)
Act 1986, the brain rapidly removed and placed in ice-cold artificial
cerebrospinal fluid (aCSF; bubbled with 95% 02/5% CO2), which com-
prised 124 mM NaCl, 3 mM KCl, 26 mM NaHCO3, 1.25 mM NaH2PO4,
2 mM CaCl2, 1 mM MgSO4 and 10 mM D-glucose. A midsagittal section
of the brain was made, the rostral and caudal parts were removed by sin-
gle scalpel cuts approximately 45 to the dorsoventral axis, and each half
was glued by its caudal end to a vibroslice stage (Campden Instruments,
Sileby, UK). Slices (400 m) that included perirhinal, entorhinal and
temporal cortices were stored submerged in aCSF (2025C). A single
slice was placed in a submerged recording chamber (2830C; flow rate,
2 ml per min) when required. Picrotoxin (5 M) was present through-
out the experiment. Blind whole-cell recordings were obtained from
neurons in layer II/III. Pipet (47 M) solutions (280 mOsm, pH 7.2)
comprised 130 mM cesium methyl sulfate, 8 mM NaCl, 4 mM Mg-ATP,
0.3 mM Na-GTP, 0.5 mM EGTA, 10 mM HEPES and 6 mM QX-314. In
some experiments, potassium methyl sulfate was substituted for cesium
methyl sulfate, and QX-314 was omitted. High-EGTA filling solution
contained 10 mM EGTA, substituted for equimolar cesium methyl sul-
fate to maintain osmolarity. One stimulating electrode was placed dor-
sorostrally on the temporal cortex side (area 36) and one ventrocaudally
on the entorhinal cortex side (area 35/entorhinal cortex) of the rhinal
sulcus. Stimuli (constant voltage) were delivered alternately to the two
electrodes (each electrode 0.033 Hz). Neurons were voltage clamped at
70 mV unless otherwise indicated. To induce LTD in each slice, low-
frequency stimulation (LFS; 200 stimuli, 1 Hz) was delivered to only one
input. LFS was delivered at 70 mV, paired with depolarization to 40
mV or delivered under current clamp at 70 mV, as appropriate for dif-
ferent experiments. Alterations of membrane potential were made only
during the LFS. Amplitudes of the evoked EPSCs were measured and
expressed relative to the normalized pre-conditioning baseline. Effects
of LFS were measured at appropriate time points (averaged over five
min) after delivering LFS. Data were analyzed from only one slice per rat
unless otherwise indicated. Data pooled across slices are expressed as
means s.e. and significance (p < 0.05) tested using either paired or
unpaired t-tests as appropriate. Only experiments in which there was lit-
tle baseline drift (<10%) were included in the pooled data. Data were
recorded using an Axopatch 200 amplifier (Axon Instruments, Foster
City, California), monitored and analyzed on line and re-analyzed off
line (W.W. Anderson, G. L. Collingridge, Soc. Neurosci. Abstr. 23, 665,
1997). Agonists and antagonists were applied by addition to the per-
fusate.
Cell culture. Perirhinal cultures were prepared according to described
methods42 for preparation of hippocampal cultures. Briefly, the perirhi-
nal cortex was removed from rats (three to five days old) and neurons
recovered by enzymatic digestion and mechanical dissociation. Cells were
plated onto coverslips in 35-mm petri dishes. Cultures were maintained
at 37C in a 95% O2/5% CO2-humidified incubator. The culture media
was composed of minimal essential media (Gibco; 30 mM glucose, 2 mM
glutamine; 15 mM HEPES, 100 g per ml bovine transferrin and 30 g
per ml insulin, complemented with 510% fetal calf serum). From the
second day in culture, the media were supplemented with cytosine--D-
arabinofuranoside (2.5 M). Neurons were used for calcium-imaging
studies 1430 days after plating.
Calcium imaging. Imaging techniques were used as described previous-
ly32,43. Briefly, cells were washed three times in HBS buffer (119 mM NaCl,
5 mM KCl, 25 mM HEPES, 33 mM glucose, 2 mM CaCl2, 2 mM MgCl2,
500 nM TTX, 1 M glycine, 100 M picrotoxin, pH 7.4; osmolarity,

nature neuroscience volume 3 no 2 february 2000 155

 2000 Nature America Inc. http://neurosci.nature.com
articles
300310 mOsm) and loaded with 5 M of the membrane-permeable Ca2+
indicator fluo-3-AM made up in 1 mg per ml bovine serum albumin/HBS
at 37C for 20 min. Cells were then washed 3 times in HBS and incubated for
30 min in a 5% CO2 atmosphere at 22C to allow for the de-esterification of
the flurophore. Cells were then viewed on a BioRad (Hemel Hempstead,
UK) MRC600 confocal microscope equipped with an argon ion laser using
standard green filter sets and perfused continuously at 2 ml per min with
HBS buffer to which AP5 (50 M) was added. Agonists were bath applied.
Integrations of five individual images were obtained every 10 s before, dur-
ing and after agonist application. In six of seven cover slips used in this
study, DHPG-induced fluorescence was significantly enhanced by DCG-
IV. All neurons from these six cover slips that initially responded to DHPG
with an increase in fluorescence were used in subsequent analysis. The flu-
orescence of individual cells in each preparation was measured using a pub-
lic-domain NIH program (http://rsb.info.nih.gov/nihimage/) and expressed
relative to baseline. The mean peak fluorescence was then calculated.
Drugs (all from Tocris, Bristol, UK) used were the N-methyl-D-aspar-
tate (NMDA) receptor antagonist D-2-amino-5-phosphonopentanoate
(AP5), the metabotropic glutamate (mGlu) receptor antagonists (S)--
methyl-4-carboxyphenylglycine (MCPG) (R,S)-1-aminoindan-1,5-dicar-
2000 Nature America Inc. http://neurosci.nature.com
boxylic acid (AIDA) (S)-2-amino-2-methyl-4-phosphonobutanoic acid
(MAP4) and (2S)--ethylglutamic acid (EGLU) and the mGlu receptor
agonists (R,S)-3,5-dihydroxyphenylglycine (DHPG) and (2S,2R,3R)-
2-(2,3-dicarboxycyclopropyl)glycine (DCG-IV).
ACKNOWLEDGEMENTS
We thank A. Doherty, L. Pickard and V. Collett for help with cell culture and
calcium imaging and G.L. Collingridge for discussions. This work was supported
by the BBSRC, MRC and Wellcome Trust.
RECEIVED 29 OCTOBER, ACCEPTED 23 NOVEMBER 1999
1. Brown, M. W. & Xiang, J.-Z. Recognition memory: neuronal substrates of the
judgement of prior occurrence. Prog. Neurobiol. 55, 149189 (1998).
2. Meunier, M., Bachevalier, J., Mishkin, M. & Murray, B. Effects on visual
recognition of combined and separate ablations of the entorhinal and
perirhinal cortex in rhesus monkeys J. Neurosci. 13, 54185432 (1993).
3. Sakai, K. & Miyashita, Y. Neural organization for the long-term memory of
paired associates. Nature 354, 152155 (1999).
4. Corodimas, K. P. & Ledoux, J. E. Disruptive effects of posttraining perirhinal
cortex lesions on conditioned fear: contributions of contextual cues Behav.
Neurosci. 109, 613619 (1995).
5. Zhu, X. O. & Brown, M. W. Changes in neuronal activity related to the
repetition and relative familiarity of visual stimuli in rhinal and adjacent
cortex of the anesthetized rat. Brain Res. 689, 101110 (1995).
6. Zhu, X. O., McCabe, B. J., Aggleton, J. P. & Brown, M. W. Mapping visual
recognition memory through expression of the immediate early gene c-fos.
Neuroreport 7, 18711875 (1996).
7. Bear, M. F. & Malenka, R. C. Synaptic plasticity: LTP and LTD. Curr. Opin.
Neurobiol. 4, 389399 (1994).
8. Kerr, D. S. & Abraham, W. C. LTD: Many means to how many ends?
Hippocampus 6, 3034 (1996).
9. Bear, M. F. & Abraham, W. C. Long-term depression in hippocampus. Annu.
Rev. Neurosci. 19, 437462 (1996).
10. Fujii, S., Saito, K., Miyakawa, H., Ito, K. & Kato, H. Reversal of long-term
potentiation (depotentiation) induced by tetanus stimulation of the input to
CA1 neurons of guinea pig hippocampal slices. Brain Res. 555, 112122 (1991).
11. Mulkey, R. M. & Malenka, R. C. Mechanisms underlying induction of
homosynaptic long-term depression in area CA1 of the hippocampus.
Neuron 9, 967975 (1992).
12. Dudek, S. M. & Bear, M. F. Homosynaptic long-term depression in area CA1
of hippocampus and effects of N-methyl-D-aspartate receptor blockade. Proc.
Natl. Acad. Sci. USA 89, 43634367 (1992).
13. Bashir, Z. I. & Collingridge, G. L. An investigation of depotentiation of long-
term potentiation in the CA1 region of the hippocampus. Exp. Brain Res. 100,
437443 (1994).
14. Bolshakov, V. Y. & Siegelbaum, S. A. Postsynaptic induction and presynaptic
expression of hippocampal long-term depression. Science 264, 11481152
(1994).
15. Oliet, S. H. R., Malenka, R. C. & Nicoll, R. A. Two distinct form of long-term
depression coexist in CA1 hippocampal pyramidal cells. Neuron 18, 969982
(1997).
16. Otani, S. & Connor, J. A. Requirement of rapid Ca2+ entry and synaptic
activation of metabotropic glutamate receptors for the induction of long-term
156
depression in adult rat hippocampus. J. Physiol. (Lond.) 511, 761770 (1998).
17. Fitzjohn, S. M. et al. The potent mGlu receptor antagonist LY341495
identifies roles for both cloned and novel mGlu receptors in hippocampal
synaptic plasticity. Neuropharmacology 37, 14451458 (1998).
18. Mulkey, R. M., Herron, C. E. & Malenka, R. C. An essential role for protein
phosphatases in hippocampal long-term depression. Science 261, 10511055
(1993).
19. Pin, J.-P. & Duvoisin, R. The metabotropic glutamate receptors: Structure
and functions. Neuropharmacology 34, 126 (1995).
20. Conn, P. J. & Pin, J. P. Pharmacology and functions of metabotropic
glutamate receptors. Annu. Rev. Pharmacol. Toxicol. 37, 205237 (1997).
21. Anwyl, R. Metabotropic glutamate receptors: electrophysiological properties
and role in plasticity. Brain Res. Rev. 29, 83120 (1999).
22. Reyes, M. & Stanton, P. K. Induction of hippocampal long-term depression
requires release of Ca2+ from separate presynaptic and postsynaptic
intracellular stores. J. Neurosci. 16, 59515960 (1996).
23. Yokoi, M. et al. Impairment of hippocampal mossy fibre LTD in mice lacking
mGluR2. Science 273, 645647 (1996).
24. Domenici, M. R., Berretta, N. & Cherubini, E. Two distinct forms of long-
term depression coexist at the mossy fiber-CA3 synapse in the hippocampus
during development. Proc. Natl. Acad. Sci. USA 95, 83108315 (1998).
25. Tzounopoulos, T., Janz, R., Sdhof, T. C., Nicoll, R. A. & Malenka, R. C. A
role for cAMP in long-term depression at hippocampal mossy fiber synapses.
Neuron 21, 837845 (1998).
26. Huang, L. Q., Rowan, M. J. & Anwyl, R. mGluR II agonist inhibition of LTP
induction, and mGluR II antagonist inhibition of LTD induction, in the
dentate gyrus in vitro. Neuroreport 8, 687693 (1997).
27. Manahan-Vaughan, D. Group 1 and 2 metabotropic glutamate receptors play
differential roles in hippocampal long-term depression and long-term
potentiation in freely moving rats. J. Neurosci. 17, 33033311 (1997).
28. Manahan-Vaughan, D. Priming of group 2 metabotropic glutamate receptors
facilitates induction of long-term depression in the dentate gyrus of freely
moving rats. Neuropharmacology 37, 14591464 (1998).
29. Bilkey, D. K. Long-term potentiation in the in vitro perirhinal cortex displays
associative properties. Brain Res. 733, 297300 (1996).
30. Ziakopoulos, Z., Tillet, C. W., Brown, M. W. & Bashir, Z. I. Input- and layer-
dependent synaptic plasticity in the rat perirhinal cortex in vitro.
Neuroscience 92, 459472 (1999).
31. Eaton, S. A. et al. Competitive antagonism at metabotropic glutamate
receptors by (S)-4-carboxyphenylglycine and (R,S)--methyl-4-
carboxyphenylglycine. Eur. J. Pharmacol. 244, 195197 (1993).
32. Bashir, Z. I. et al. Induction of LTP in the hippocampus needs synaptic
activation of glutamate metabotropic receptors. Nature 363, 347350 (1993).
33. Pellicciari, R. et al. 1-aminoindan-1,5-dicarboxylic acid: a novel antagonist at
phosholipase C-linked metabotropic glutamate receptors. J. Med. Chem. 38,
37173719 (1995).
34. Moroni, F. et al. Pharmacological characterization of 1-aminoindan-1,5-
dicarboxylic acid, a potent mGluR1 antagonist. J. Pharmacol. Exp. Ther. 281,
721729 (1997).
35. Salt, T. E. & Eaton, S. A. Distinct presynaptic metabotropic receptors for L-
AP4 and CCG1 on GABAergic terminals: pharmacological evidence using
novel -methyl derivative mGluR antagonists, MAP4 and MCCG, in the rat
thalamus in vivo. Neuroscience 65, 513 (1995).
36. McCaffery, B. et al. Synaptic depression induced by pharmacological
activation of metabotropic glutamate receptors in the perirhinal cortex in
vitro. Neuroscience 93, 977984 (1999).
37. Ito, I. et al. 3,5-Dihydroxyphenylglycine: a potent agonist of metabotropic
glutamate receptors. Neuroreport 3, 10131016 (1992).
38. Wilsch, V. W., Pidoplichko, V. I., Opitz, T., Shinozaki, H. & Reymann, K. G.
Metabotropic glutamate receptor agonist DCG-IV as NMDA receptor
agonist in immature rat hippocampal neurones. Eur. J. Pharmacol. 262,
287291 (1994).
39. Scanziani, M., Salin, P. A., Vogt, K. E., Malenka, R. C. & Nicoll, R. A. Use-
dependent increases in glutamate concentration activate presynaptic
metabotropic glutamate receptors. Nature 385 630634 (1997).
40. Schoepp, D. D. et al. The novel metabotropic glutamate receptor agonist
2R,4R- APDC potentiates stimulation of phosphoinositide hydrolysis in the
rat hippocampus by 3,5dihydroxyphenylglycine: Evidence for a synergistic
interaction between group 1 and group 2 receptors. Neuropharmacology 35,
16611672 (1996).
41. Mistry, R., Golding, N. & Challiss, R. A. J. Regulation of phosphoinositide
turnover in neonatal rat cerebral cortex by group I- and II-selective
metabotropic glutamate receptor agonists. Br. J. Pharmacol. 123, 581589
(1998).
42. Noel, J. et al. Surface expression of AMPA receptors in hippocampal neurones
is regulated by an NSF-dependent mechanisms. Neuron 23, 365376 (1999).
43. Doherty, A. J., Collingridge, G. L. & Jane, D. E. Antagonist activity of -
substituted 4-carboxyphenylglycine analogues at group I metabotropic
glutamate receptors expressed in CHO cells. Br. J. Pharmacol. 126, 205210
(1999).
44. Burwell, R. D., Witter, M. P. & Amaral, D. G. Perirhinal and postrhinal
cortices of the rat: A review of the neuroanatomical literature and
comparison with findings from the monkey brain. Hippocampus 5, 390408
(1995).
nature neuroscience volume 3 no 2 february 2000
 2000 Nature America Inc. http://neurosci.nature.com

articles

Polyglutamine expansion down-

regulates specific neuronal genes
before pathologic changes in SCA1
Xi Lin1,2, Barbara Antalffy3, Dongcheul Kang1,2, Harry T. Orr4 and Huda Y. Zoghbi1,2

1 Howard Hughes Medical Institute, Baylor College of Medicine, One Baylor Plaza, Houston, Texas 77030, USA
2 Department of Molecular and Human Genetics, Baylor College of Medicine, One Baylor Plaza, Houston, Texas 77030, USA
3 Department of Pathology, Baylor College of Medicine, One Baylor Plaza, Houston, Texas 77030, USA
4 Department of Laboratory Medicine and Pathology, Institute of Human Genetics, University of Minnesota, Harvard Street at East River Rd., Minneapolis,
Minnesota 55455, USA
2000 Nature America Inc. http://neurosci.nature.com

Correspondence should be addressed to H.Y.Z. (hzoghbi@bcm.tmc.edu)

The expansion of an unstable CAG repeat causes spinocerebellar ataxia type 1 (SCA1) and several
other neurodegenerative diseases. How polyglutamine expansions render the resulting proteins
toxic to neurons, however, remains elusive. Hypothesizing that long polyglutamine tracts alter gene
expression, we found certain neuronal genes involved in signal transduction and calcium homeosta-
sis sequentially downregulated in SCA1 mice. These genes were abundant in Purkinje cells, the
primary site of SCA1 pathogenesis; moreover, their downregulation was mediated by expanded
ataxin-1 and occured before detectable pathology. Similar downregulation occurred in SCA1 human
tissues. Altered gene expression may be the earliest mediator of polyglutamine toxicity.

Spinocerebellar ataxia type 1 and several other autosomal domi-
nant neurodegenerative diseases are caused by expansions of a CAG
trinucleotide repeat tract that result in an abnormally long polyg-
lutamine tract within the mutated proteins. In SCA1, the degener-
ation afflicts primarily the cerebellar Purkinje cells and brainstem
neurons, leading to the characteristic ataxic phenotype and bulbar
dysfunction. There is persuasive evidence that the mutant proteins
gain some toxic function1,2, but precisely what this new function
might be and how a ubiquitous protein can cause the degeneration
of only highly circumscribed neuronal populations remain some
of the most puzzling questions in SCA1 pathogenesis.
Mouse models yield important clues to this question. The B05
line, the primary SCA1 mouse model, expresses a full-length SCA1
cDNA with 82 CAG repeats (82Q) under the Purkinje cell-spe-
cific promoter of the Pcp2 gene and develops the ataxia and loss of
motor coordination typical of the human disease phenotype3,4.
These mice duplicate the Purkinje cell pathology seen in human
patients, including the localization of the mutant ataxin-1 pro-
tein into a single nuclear aggregate5. A line of transgenic mice that
expresses expanded ataxin-1 with a point mutation in the nuclear
localization signal (so that the protein cannot be transported into
the nucleus) fails to develop disease6. Another mouse line (77)
expressing expanded ataxin-1 without a self-association domain
fails to develop nuclear aggregates despite nuclear localization of
the expanded proteinbut does develop the SCA1 phenotype.
These data clearly demonstrate that nuclear events, though not
nuclear aggregation, mediate SCA1 pathogenesis. Furthermore,
studies in transfected COS-1 cells show that mutant ataxin-1 asso-
ciates with nuclear matrix and disrupts specific nuclear domains
(such as promyelocytic oncogenic domains, PODs)5.
Disruption of nuclear structures and functions might alter gene
expression. In testing this hypothesis, we found several genes with
altered expression in the B05 mice3,4 using a PCR-based cDNA
subtractive-hybridization strategy. Six neuronal genes, all highly
abundant in Purkinje cells, were downregulated at a surprisingly
early stage in pathogenesis, before any known behavioral or patho-
logical changes. This downregulation required the nuclear local-
ization of mutant ataxin-1 and occurred in two independent SCA1
mouse models. We also found upregulation of one gene, a
homolog of human 1-antichymotrypsin (1-ACT), at five weeks
of age; this is most likely a secondary event in pathogenesis. Final-
ly, all the alterations noted in SCA1 transgenic mice were also
found in human SCA1 tissues. These data suggest that polygluta-
mine expansion in SCA1 may mediate early pathogenic events by
downregulating the expression of specific neuronal genes.
RESULTS
In B05 mice, the transgene expression was first detectable at post-
natal day 10 (P10). The mice developed normally with no neu-
robehavioral abnormalities until five weeks of age, when they first
showed mild impairment on the accelerating rotating rod. Purkinje
cells begin to lose their proximal dendrites at six weeks3,4. To cap-
ture the full spectrum of genes that might be altered in SCA1
pathogenesis, we performed PCR-based cDNA subtraction using
cerebellar mRNA from two-month-old B05 mice and wild-type
littermates. The subtractions were done in two different directions,
using B05 mRNA either as the tester or the driver, to identify genes
that were up- or downregulated, respectively. Six hundred clones
from the subtracted cDNA library were screened by dot-blot
hybridization. Twenty clones that differentially hybridized to the
subtracted probes were further examined with northern blots,
from which nine clones representing seven different genes had
altered expression patterns in B05 cerebellum. Differentially
expressed clones were identified by sequence analysis.

nature neuroscience volume 3 no 2 february 2000 157

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

articles

Fig. 1. PCCMT expression was specifically

reduced in SCA1 mice. (a) Northern blots of a
cerebellar RNA from four different lines of
SCA1 transgenic mice and wild-type (WT) lit-
termates using PCCMT cDNA as a probe. B05
and A02 mice express human ataxin-1 with 82
and 30 glutamines, respectively. K772T mice
express expanded ataxin-1 with 82 glutamines
and a point mutation in the nuclear localization
signal (K772T). 77 mice express ataxin-1 with
77 glutamines and no self-association domain.
Substantially reduced PCCMT mRNA expres-
sion was seen in B05 and 77 mice, but not in
A02 and K772T mice. The ages of animals used
in this experiment were eight weeks for B05
and A02 and five weeks for 77 and K772T. b
The -actin hybridization signal demonstrates
equal loading of RNA for littermate pairs at
various time points. (b) Northern blots of
cerebellar RNA from B05 mice and WT litter-
2000 Nature America Inc. http://neurosci.nature.com

mates at different ages using PCCMT cDNA as

a probe. Expression of PCCMT mRNA was
reduced by postnatal day 11. The similar
amount of RNA between B05 and WT litter-
mates was shown by the equal -actin
hybridization signal.

To determine whether these altered expression levels might
mediate early pathogenesis, we examined the expression of these
genes in younger mice, before any histologic or neurobehavioral
abnormalities develop. Because mutant animals do not begin to
display Purkinje cell pathology until six weeks4, we reasoned that
changes of gene expression before three weeks would be less likely to
represent secondary events in pathogenesis. We ascertained the ear-
liest date of altered expression for each gene by northern blot analy-
ses on total-RNA samples collected at different time points from
B05 mice or normal littermates. The characterization of these genes
and their alterations in SCA1 is detailed in the following sections.
Cloning and characterization of mouse PCCMT
One of the clones identified by cDNA subtractive hybridization was
isolated repeatedly and found to contain a 550-bp insert. Sequenc-
ing and BLAST searches revealed no significant homology to any
known sequences in the databases. Northern blots using the 550-
bp insert as a probe identified a 4.8-kb transcript, which was dra-
matically downregulated in B05 cerebellum (Fig. 1a, B05).
Subsequent screening of a mouse brain cDNA library (Stratagene,
La Jolla, California) allowed the isolation of a clone that contained
a 4,730-bp insert (GenBank accession number, AF209926). Assem-
bly of this 4,730-bp sequence with that of an EST clone (AA022288)
resulted in a cDNA of 4,846 bp and predicted an open reading frame
of 852 bp that was highly homologous to the coding regions of
human (88% identity) and Xenopus (71% identity) prenylcysteine
carboxylmethyltransferase (PCCMT)7,8. This clone was therefore
identified as a mouse homolog of PCCMT. Immunohistochem-
istry using a rabbit polyclonal antibody against a peptide (amino
acids 185200) from human PCCMT protein, equivalent to the
predicted mouse protein amino acids 184199, showed abundant
staining in the Purkinje cells of wild-type mice and sparse staining
in B05 Purkinje cells (data not shown).
To evaluate the potential roles of PCCMT in early SCA1 patho-
genesis, we examined PCCMT mRNA expression at different time
points in B05 and wild-type littermate cerebella. Remarkably, the
reduction in PCCMT mRNA level was already detectable in B05
cerebella at P11 (Fig. 1b), about one day after the initiation of SCA1
transgene expression and two weeks before the earliest morpho-
logic changes3,4. Moreover, this reduction was consistently observed
in multiple experiments using RNA from different litters of mice.
Phosphoimaging analysis confirmed that the reduction of PCCMT
mRNA expression was statistically significant, with an 11 2.36%
decrease at day 11 and a 19.2 7.5% decrease at day 15 (n = 3, p <
0.05). Thus PCCMT downregulation occurred very early.
We examined PCCMT mRNA expression in other SCA1 trans-
genic mouse lines that show no obvious Purkinje cell dysfunction
or degeneration (A02 mice and ataxin-1[82Q]K772T mice)4,6. A02
mice express high levels of wild-type human ataxin-1 with 30 glu-
tamines (30Q); mice expressing ataxin-1[82Q]K772T with a point
mutation in the nuclear-localization signal retain mutant ataxin-
1 in the cytoplasm, where it is nonpathogenic. Neither line showed
altered PCCMT mRNA levels (Fig. 1a, A02 and K772T). In con-
trast, PCCMT mRNA was dramatically decreased in another SCA1
mouse model (77) that manifests behavioral and pathological
changes similar to those of B05 mice (Fig. 1a, 77).
To study the function of PCCMT in mice, we examined its
expression during development and in various peripheral tissues in
northern blots. PCCMT mRNA was highly enriched in adult cere-
bellum, with low levels of expression in other brain regions (Fig.
2a) and very low but detectable (after prolonged exposure) levels in
peripheral tissues such as lung, heart, kidney and skeletal muscles
(Fig. 2b). Within the cerebellum, PCCMT mRNA was expressed
postnatally, with a dramatic increase after P12 (Fig. 2c). The tem-
poral and spatial expression pattern of PCCMT suggests that it
might have an important role in Purkinje cells.
To determine whether PCCMT was expressed in neurons vul-
nerable to SCA1, we examined PCCMT mRNA expression in the
human brain (Fig. 2d). Northern blots using mRNA isolated from
different brain regions showed that PCCMT was expressed in the
cerebellum and putamen at higher levels than in other brain
regions and revealed three different transcripts of 9.0 kb, 5.5 kb

158 nature neuroscience volume 3 no 2 february 2000

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

articles

Fig. 2. PCCMT mRNA

expression in mouse and a d
human brain. (ac) Northern
blots using PCCMT cDNA as
a probe and total RNA (25
g) from different brain
regions of adult WT mice (a),
thalamus and several periph-
eral tissues of B05 and WT
littermates at two months of
age (b), and cerebella of WT
mice at different developmen-
tal stages (c). Ethidium bro-
mide-stained RNA gel images b e
are shown for loading con-
trol, which was confirmed by
hybridization with -actin
cDNA (not shown). PCCMT
was predominantly expressed
in cerebellum, with much
2000 Nature America Inc. http://neurosci.nature.com

lower levels of expression in

other regions of the brain (a,
one-day exposure). In the
peripheral tissues, PCCMT
was detectable only after long
exposure (b, three-day expo-
sure). During cerebellar c
development, PCCMT
expression was low before P6
and dramatically increased
after P12. (d) Northern blot
of polyA RNA (2 g) from
the indicated regions of
human brain probed with
PCCMT cDNA shows the
relatively high levels of
expression in the cerebellum
and putamen. Equal loading in
each lane was confirmed by hybridization with -actin cDNA as a probe (not shown). (e) Immunohistochemical staining of PCCMT in the cerebella and pons
of a SCA1 patient and an age-matched control demonstrates much-reduced expression in Purkinje cells and pontine neurons in the patient.

and 3.6 kb, of which the 5.5 kb was most abundant (Fig. 2d). It
remains unknown whether these transcripts represent alterna-
tively spliced isoforms or distinct homologs. Immunohisto-
chemical analysis using an antibody against human PCCMT
found abundant expression in Purkinje cells and pontine neurons
(Fig. 2e, control). We also examined PCCMT expression in human
brain tissue from a juvenile-onset SCA1 patient, focusing on the
cerebellum and pons, two regions typically affected in SCA1. Lit-
tle PCCMT immunostaining was found in the cerebellum (Fig.
2e, Cerebellum, SCA1). This might be due in part to the paucity
of remaining Purkinje cells in this patient. The pons, however,
which retained many pontine neurons, showed greatly reduced
PCCMT immunostaining (Fig. 2e, Pons, SCA1). Although it is
impossible to investigate the role of PCCMT in early SCA1 patho-
genesis in humans, these snapshots of end-stage disease are in
accordance with the alterations observed in transgenic mice.
IP3R1 and SERCA2 are downregulated at P14
Sequence analysis of two other subtracted clones showed 100% iden-
tity to mouse inositol triphosphate receptor type 1 (IP3R1) and sar-
coplasmic endoplasmic reticulum calcium ATPase type 2 (SERCA2),
two Purkinje cell-abundant proteins. IP3R1 and SERCA2, an intra-
cellular Ca2+-release channel and a Ca2+ pump, respectively, are both
present on the membrane of the endoplasmic reticulum (ER). The
mRNAs of IP3R1 and SERCA2 are downregulated at two weeks of
age in B05 mouse cerebellum, again, preceding known behavioral
or morphological changes (Fig. 3a). Immunohistochemistry using
either an IP3R1- or SERCA2-specific polyclonal antibody on cere-
bellar sections from six-week-old mice revealed dramatic reductions
in IP3R1 and SERCA2 proteins in B05 cerebellar Purkinje cells (data
not shown). Similar downregulations in IP3R1 and SERCA2 mRNA
occurred in mice expressing ataxin-1[77Q], but not in A02 mice or
mice expressing ataxin-1[82Q]K772T (data not shown). Therefore,
downregulation of IP3R1 and SERCA2 specifically occurs in the
transgenic mice that show Purkinje cell degeneration caused by
mutant ataxin-1. Immunohistochemistry using IP3R1 (not shown)
and SERCA2 (Fig. 3b) antibodies revealed profoundly reduced
immunostaining in cerebellar Purkinje cells and pontine neurons
from SCA1 patient tissues.
Downregulation of 5-phosphatase, mTRP3 and EAAT4
Three genes were downregulated by three to four weeks of age in
B05 mouse cerebellum: type 1 inositol polyphosphate 5-phosphatase
(5-phosphatase; Fig. 4a), transient receptor potential type 3 (TRP3;
Fig. 4b) and excitatory amino acid transporter type 4 (EAAT4; Fig.
4c). Type 1 inositol polyphosphate 5-phosphatase is a major inac-
tivating enzyme of IP3 and IP4, which mobilize Ca2+ through inter-
actions with the intracellular calcium release channels, IP3Rs9. TRP3

nature neuroscience volume 3 no 2 february 2000 159

 2000 Nature America Inc. http://neurosci.nature.com
articles
Fig. 3. IP3R1 and SERCA2 expression
is reduced in SCA1 mice and patient a b
brain tissues. (a) Northern blots of
total cerebellar RNA (25 g) from
B05 mice and WT littermates at dif-
ferent ages using IP3R1 or SERCA2
cDNA as probe. IP3R1 and SERCA2
mRNAs are reduced in B05 cerebel-
lum at two weeks of age. Equal loading
of RNA in each lane is shown by
similar intensity of ethidium bro
mide-stained 28S and 18S rRNAs.
(Hybridization signal of -actin is not
shown.) (b) Immunohistochemical
staining of SERCA2 in the cerebellum
and pons of a SCA1 patient and an
age-matched control.
2000 Nature America Inc. http://neurosci.nature.com
is a putative plasmalemmal calcium channel controlled by the con-
tent or filling state of calcium stores through some unknown mech-
anism10. Transcripts of these two genes were dramatically reduced
at four weeks of age in B05 mouse cerebellum (Fig. 4a and b). These
downregulations probably reflected reduced expression in Purkinje
cells, as type 1 inositol 5-phosphatase and TRP3 are predominantly
expressed in Purkinje cells in the cerebellum.
EAAT4 is a Purkinje cell-specific glutamate transporter local-
ized on dendritic spines11. In B05 cerebellum, EAAT4 mRNA down-
regulation was obvious at three weeks of age. By six weeks, EAAT4
protein was hardly detectable in Purkinje cells by immunohisto-
chemistry using an EAAT4-specific polyclonal antibody (Fig. 4c).
Analysis of the expression patterns of these three genes in the
other SCA1 transgenic mice demonstrated that their downregu-
lation was associated specifically with pathogenic mutant ataxin-
a clone was identical to EB22/4, a serine protease
b was upregulated by five weeks (Fig. 6b), whereas no
c
160
1, as downregulation occurred in ataxin-1[77Q] mice but not
A02 or ataxin-1[82Q]K772T mice (data not shown).
Purkinje cell gene expression was not globally altered
To rule out the possibility that downregulation might be a global
effect, we looked at the expression patterns of eight other genes
known to be abundant in Purkinje cells in three-month-old B05
mice and wild-type littermates. Among those examined were
cGMP binding-phosphodiesterase (cGB-PDE), phospholipase C4
(PLC4) and IP3 3-kinase (IP3-3K)1214. There were no apparent
changes in the expression of any of the eight genes in B05 mouse
cerebellum (three examples shown in Fig. 5).
Mouse 1-antichymotrypsin upregulated at five weeks
Not all the genes identified with our subtractive hybridization
approach were downregulated in B05 mice. One clone
was upregulated. Sequence analysis revealed that this
inhibitor isolated from a mouse teratocarcinoma cell
line and presumably the physiological homolog of
human 1-antichymotrypsin (61% amino acid iden-
tity)15. Northern blots showed a single transcript of
2 kb (Fig. 6a). In B05 cerebellum, EB22/4 mRNA
change was detected in A02 mice (Fig. 6a). EB22/4
mRNA was also increased in ataxin-1[77Q] mouse
cerebellum (Fig. 6a), consistent with the pathology
of Purkinje cell degeneration. Surprisingly, we
observed an increase in EB22/4 in ataxin-1[82Q]K772T
mice (Fig. 6a), which have no obvious Purkinje cell
Fig. 4. Downregulation of type 1 inositol polyphosphate 5-
phosphatase, TRP3 and EAAT4 in SCA1 mice. Northern
blots of cerebellar RNA (25 g) from B05 mice and WT lit-
termates at different ages were probed with 5-phosphatase 1
(a), TRP3 (b) and EAAT4 (c) cDNAs. Equal amount of total
RNA in each lane was confirmed by similar -actin hybridiza-
tion signals and ethidium bromide staining of the gel (not
shown). (a, b) At four weeks of age, inositol polyphosphate
5-phosphatase type 1 and TRP3 showed reduced expression
in B05 cerebellum. (c) At three weeks, EAAT4 mRNA is
reduced. Immunohistochemical staining of EAAT4 in B05 and
WT littermate mouse cerebella at six weeks of age showed
greatly reduced EAAT4 protein in Purkinje cell dendrites.
nature neuroscience volume 3 no 2 february 2000
 2000 Nature America Inc. http://neurosci.nature.com
Fig. 5. Mutant ataxin-1 does not affect expression of all Purkinje cell-
specific genes. Northern blots of total cerebellar RNA (25 g) from
three-month-old B05 mice and WT littermates using the indicated
cDNAs as probes. The mRNA levels of cGMP binding-phosphodi-
esterase (cGB-PDE), phospholipase C4 (PLC4) and IP3 3-kinase (IP3-
3K) are similar between B05 and WT.
2000 Nature America Inc. http://neurosci.nature.com
degeneration but do express mutant ataxin-1 in the cytoplasm.
Unlike the other genes characterized above, EB22/4 mRNA upreg-
ulation thus seemed to depend on the polyglutamine expansion in
ataxin-1, but not on nuclear localization of its product.
Because EB22/4 is highly homologous to human 1-ACT and
anti-human 1-ACT polyclonal antibodies recognize a single band
on western blots of mouse tissues16, we examined 1-ACT-like
protein (EB22/4) expression by immunohistochemistry. The white
matter of B05 mouse cerebellum showed intense 1-ACT-anti-
body staining, whereas wild-type mouse cerebellum showed a very
low level of staining (Fig. 6c). This staining pattern suggests that
upregulation of 1-ACT-like protein in SCA1 mouse cerebellum
was a response secondary to mutant ataxin-1 expression in the
Purkinje cells. It is interesting that, in SCA1 patient tissue, 1-
ACT stains intensely in neuronal cytoplasm (Fig. 6d).
DISCUSSION
We identified and characterized several specific genes whose
expression was altered in SCA1 transgenic mice at unexpectedly
early stages in pathogenesis. This downregulation is the earliest
known effect of mutant ataxin-1, occurring before any detectable
pathological changes. This effect of mutant ataxin-1 is selective:
a number of other Purkinje cell-specific genes were not altered.
Moreover, the downregulation seemed to occur sequentially, with
PCCMT affected first, followed by IP3R1, SERCA2, EAAT4, TRP3
and 5-phosphatase 1. The finding that downregulation of the same
genes occurred in two independent SCA1 mouse models (B05
and [77Q]) and not in mice overexpressing non-pathogenic atax-
in-1 (A02 and mice expressing ataxin-1[82Q]K772T) makes it plau-
sible that reduced expression of these genes is involved in SCA1
pathogenesis. Substantially reduced immunostaining of PCCMT,
SERCA2 and IP3R1 in human SCA1 brain tissues corroborated
the mouse data and suggested that expression of these genes is
altered in human SCA1 pathogenesis as well. To rule out the
remote possibility that this downregulation was secondary to
Purkinje cell distress, we evaluated the expression pattern of
PCCMT, SERCA2 and TRP3 in two mouse mutants that have dys-
functional Purkinje cells, Ube3a and Sca1 knockouts2,17. No alter-
ations in gene expression were detected (data not shown).
EB22/4 was the only gene we found to be upregulated. Avail-
able evidence suggests that mouse EB22/4 is the physiological
homolog of human 1-antichymotrypsin15, a serine protease
nature neuroscience volume 3 no 2 february 2000
articles
inhibitor and an acute response protein involved in a variety of
pathological conditions. In the brain, 1-antichymotrypsin is
expressed in glia; its expression is increased in Alzheimer disease,
Huntington disease and normal aging1820, and it may well be
involved in similar molecular events in these different pathophys-
iological conditions. The observation that EB22/4 was also upreg-
ulated in K772T mice indicates its dependence on the expression,
but not the nuclear localization, of mutant ataxin-1. Although
K772T mice do not exhibit abnormalities, their Purkinje cells
might respond in subtle ways to the presence of expanded ataxin-
1 in the cytoplasm. Excessive quantities of misfolded ataxin-1 in
the cytoplasm could elicit enough of a stress response to affect cel-
lular functions, if not cause actual degeneration.
EAAT4, TRP3 and inositol polyphosphate 5-phosphatase type
1 were downregulated in B05 cerebellum by three to four weeks of
age. EAAT4 is a Purkinje cell-specific glutamate transporter local-
ized to the extra-junctional dendritic membrane. It presumably
restricts glutamate spillover to adjacent synapses and enhances
spatial and temporal resolution of glutamate signaling11,21. TRP3,
a calcium channel abundant in Purkinje cell membranes, is
involved in cellular calcium homeostasis10,22,23. Inositol polyphos-
phate 5-phosphatase type 1 is the major inactivating enzyme of
calcium-mobilizing molecules IP3 and IP4 and is involved in the
regulation of calcium signaling9,24,25. In both human and rat
brains, inositol 5-phosphatase type 1 is highly abundant in the
Purkinje cells12,26. Specific downregulation of inositol 5-phos-
phatase type 1 and TRP3 suggests that calcium homeostasis may
be perturbed in SCA1 pathogenesis.
At two weeks, IP3R1 expression and SERCA2 expression were
reduced in B05 mice. Both of these proteins localize to the endo-
plasmic reticulum and are involved in calcium store regulation2729.
IP3R1 is a calcium-release channel and increases free cytoplasmic
calcium concentration ([Ca2+]cyt) by releasing Ca2+ from the ER
in response to IP3 generated by extracellular signals. Interestingly,
lack of IP3R1 due to either spontaneous mutation or targeted dele-
tion leads to severe ataxia, seizures and premature death in
mice30,31. SERCA2, an ATPase, balances IP3R1 by pumping calci-
um into the ER, effectively reducing [Ca2+]cyt. The abundance of
IP3R1 and SERCA2 in cerebellar Purkinje cells reflects their impor-
tance in regulating calcium levels3234. The net effect of IP3R1 and
SERCA2 downregulation on [Ca2+]cyt in Purkinje cells may not be
straightforward, however. Depending on the relative contributions
of IP3R1 and SERCA2, average [Ca2+]cyt could increase, decrease or
not change. Even subtle changes in calcium transients might lead
to Purkinje cell dysfunction and progressive degeneration in
SCA135,36.
The cellular calcium signaling system is markedly flexible and
adaptable, and expression of different components are interde-
pendent37. It thus seems probable that downregulating IP3R1
and/or SERCA2 leads to changes in the expression of the other
genes by a cellular compensatory mechanism that then resets the
cellular calcium-signaling system. Because the expression of
expanded ataxin-1 in SCA1 mouse Purkinje cells is not transient (it
is driven by the potent Pcp2/L7 promoter), the constant, high level
of mutant ataxin-1 might pose a sustained stress signal to the cal-
cium signaling system.
PCCMT, whose expression was altered first, is an abundant
Purkinje cell protein in both mouse and human cerebella; it is
involved in the posttranslational modification of a number of very
important cellular proteins, including Ras-like small GTPases, het-
erotrimeric G protein subunits and nuclear lamins3840. Inhibit-
ing PCCMT activity impairs membrane targeting of Ras in
transfected cells41. Downregulation of PCCMT in cerebellar Purk-
161
 2000 Nature America Inc. http://neurosci.nature.com
articles
a inje cells could strain the normal functions of these molecules and
b tion6, we favor a model of transcriptional downregulation. Expand-
2000 Nature America Inc. http://neurosci.nature.com
c on how the proteasomal activity is modulated by expanded ataxin-
d expression is altered by mutant ataxin-1 are regulated by nuclear
Fig. 6. Upregulation of EB22/4, a mouse homolog of human 1-ACT. (a,
b) Northern blots of total cerebellar RNA (25 g) from different lines of
SCA1 transgenic mice and WT littermates at five to eight weeks of age (a)
and B05 mice and WT littermates at different ages (b) were probed with
EB22/4 cDNA. EB22/4 expression is increased not only in B05 and 77
mice, but also in K772T mice. The increase of EB22/4 expression in B05
cerebellum was obvious by five weeks of age. (c, d) Immunohistochemical
staining for 1-ACT in B05 and WT littermate mouse cerebella at three
months of age (c) as well as the pons of a SCA1 patient and age-matched
control (d). Staining for 1-ACT is increased in the white matter of B05
cerebellum (c) and in pontine neurons of the SCA1 patient (d).
162
have adverse effects on many aspects of cellular physiology. Alter-
natively, the reduced expression of PCCMT (as well as IP3R1 and
SERCA2) in early developmental stages might perturb Purkinje
cell differentiation and/or render the cells more susceptible to stress.
It is striking that all the genes whose expression was altered
early in the pathogenesis in SCA1 mouse models were downreg-
ulated; perhaps a common mechanism operated at transcription-
al and/or posttranscriptional levels. Because expanded ataxin-1
has to localize to Purkinje cell nuclei to cause neuronal degenera-
ed ataxin-1 might interact directly with certain transcription factors
and coactivators, thus interfering specifically with the transcrip-
tion of their target genes. Alternatively, expanded ataxin-1 might
affect transcription indirectly, possibly by perturbing nuclear pro-
teasomal activity42. Some transcription factors, such as NFB, p53,
nuclear receptors and their cofactors, are regulated by proteaso-
mal degradation4346. Ataxin-1 colocalizes with proteasomal com-
ponents in nuclear aggregates in the Purkinje cells of both SCA1
transgenic mice and patients42. Proteasomal perturbation by
mutant ataxin-1 in the nucleus could lead to downregulation of
certain genes through either increased degradation of their acti-
vators and/or decreased degradation of their repressors, depending
1. Regardless, identification of genes downregulated in SCA1 makes
it possible for future studies to evaluate this hypothesis.
Mutant ataxin-1 might also downregulate gene expression by
disrupting nuclear structures and nuclear protein functions. For
example, mutant ataxin-1 redistributes PML5, which interacts
with CBP/p300 and facilitates CBP/p300 localization to PODs47.
PML enhances the activity of nuclear receptors as transactivators
in mammalian cells, suggesting that it could be a transcriptional
co-activator. It is possible that, by redistributing PML, mutant
ataxin-1 interferes with its normal function as a transcriptional
co-activator, resulting in reduced expression of specific groups of
genes. It will be worthwhile to determine if the genes whose
receptors and PML. Alternatively, mutant ataxin-1 might alter
gene expression by interfering with mRNA metabolism through
various nuclear events, such as mRNA processing and mRNA
nucleocytoplasmic transport. These possibilities can be explored
in the SCA1 mouse models by studying the regulation of the genes
identified in this study.
Identifying genes whose expression is altered by mutant ataxin-
1 provides a platform for further studies aimed at understanding
the earliest molecular events disrupted in SCA1 pathogenesis.
Characterization of the transcriptional regulation of these genes
should yield greater insight into not only the molecular mecha-
nisms by which expanded ataxin-1 affects gene expression, but the
pathogenesis of other polyglutamine diseases as well.
METHODS
Northern blots. Total RNA was isolated from different mouse tissues
with TRIZOL reagent following manufacturers instructions (Gibco-BRL,
Gaithersburg, Maryland). RNA (25 g) was fractionated and transferred
to nylon membrane according to standard protocol. Various probes were
labeled using Megaprime kit (Amersham, Cleveland, Ohio) and
hybridized in ExpressHyb hybridization solution (Clontech, Palo Alto,
California) according to manufacturers instruction. The relative amounts
of RNA on the blots were evaluated by both ethidium bromide-stained
RNA gel images and hybridization using -actin cDNA as a probe.
Subtractive cDNA cloning. To identify genes with altered expression pat-
terns in B05 mice, we performed subtractive cDNA cloning using Clontech
nature neuroscience volume 3 no 2 february 2000
 2000 Nature America Inc. http://neurosci.nature.com
PCR-Select cDNA Subtraction Kit (Clontech). Briefly, total RNA was iso-
lated from the cerebella of two-month-old B05 mice and wild-type, sex-
matched littermates using TRIZOL as described above. mRNA was then
selected using Dynabeads Oligo(dT)25 (Dynal A. S., Oslo, Norway). We
then precisely followed the protocol of PCR-Select cDNA Subtraction Kit
provided by the manufacturer.
Potential, differentially expressed clones were verified by northern blots
of wild-type and B05 cerebellar RNA using the cDNA insert as a probe.
Sequences of the differentially expressed clones were obtained by ABI
PrismTM DNA Sequencer and BigDyeTM Terminator Cycle Sequencing
Ready Reaction Kit (PE Applied Biosystems, Foster City, California). The
sequences were analyzed by DNASTAR program and BLAST search.
Immunohistochemistry. The immunohistochemical staining on mouse
brain sections was performed as described previously5. The rabbit poly-
clonal anti-PCCMT antibody, a gift from Mark Philips7, was used at 1:800
dilution. The rabbit anti-IP3R1 antibody (sc6093, Santa Cruz, Santa
Cruz, California) was used at 1:100 dilution. The anti-SERCA2 mono-
clonal antibody (Novocastra Laboratories, Newcastle, UK) was used at
1:50 dilution. The rabbit anti-1-ACT antibody (Zymed Laboratories,
2000 Nature America Inc. http://neurosci.nature.com
San Francisco, California) was used at 1:50 dilution. The rabbit anti-
EAAT4 antibody, provided by Kohichi Tanaka and Kei Watase11,21, was
used at 0.1 mg per ml.
ACKNOWLEDGEMENTS
The authors thank A. Beaudet for providing the Ube3a null mice, M. Philips at
New York University School of Medicine and K. Tanaka at the National Institute
of Neuroscience in Japan for providing the anti-PCCMT and anti-EAAT4
antibodies, H. J. Bellen for reading the manuscript and V. Brandt for comments.
This work was supported by grants NS27699 and NS22920 from the National
Institutes of Health (to H.Y.Z. and H.T.O.) and by the core facilities of the Baylor
College of Medicine Mental Retardation Research Center. X. Lin is an Associate
and H. Zoghbi an Investigator with the Howard Hughes Medical Institute.
RECEIVED 20 SEPTEMBER; ACCEPTED 6 DECEMBER 1999
1. Zoghbi, H. Y. & Orr, H. T. Spinocerebellar ataxia type 1. Seminars in Cell Biology:
Unstable Repeat Diseases Vol. 6, 2935 (Saunders, Philadelphia, 1995).
2. Matilla, A. et al. Mice lacking ataxin-1 display learning deficits and decreased
hippocampal paired-pulse facilitation. J. Neurosci. 18, 55085516 (1998).
3. Burright, E. N. et al. SCA1 transgenic mice: a model for neurodegeneration
caused by an expanded CAG trinucleotide repeat. Cell 82, 937948 (1995).
4. Clark, H. B. et al. Purkinje cell expression of a mutant allele of SCA1 in transgenic
mice leads to disparate effects on motor behaviors, followed by a progressive
cerebellar dysfunction and histological alterations. J. Neurosci. 17, 73857395
(1997).
5. Skinner, P. J. et al. Ataxin-1 with extra glutamines induces alterations in nuclear
matrix-associated structures. Nature 389, 971974 (1997).
6. Klement, I. A. et al. Ataxin-1 nuclear localization and aggregation: Role in
polyglutamine-induced disease in SCA1 transgenic mice. Cell 95, 4153 (1998).
7. Dai, Q. et al. Mammalian prenylcysteine carboxyl methyltransferase is in the
endoplasmic reticulum. J. Biol. Chem. 273, 1503015034 (1998).
8. Imai, Y., Davey, J., Kawagishi-Kobayashi, M. & Yamamoto, M. Genes encoding
farnesyl cysteine carboxyl methyltransferase in Schizosaccharomyces pombe and
Xenopus laevis. Mol. Cell. Biol. 17, 15431551 (1997).
9. Mitchell, C. A., Brown, S., Campbell, J. K., Munday, A. D. & Speed, C. J.
Regulation of second messengers by the inositol polyphosphate 5-phosphatases.
Biochem. Soc. Trans. 24, 9941000 (1996).
10. Mori, Y. et al. Differential distribution of TRP Ca2+ channel isoforms in mouse
brain. Neuroreport 9, 507515 (1998).
11. Yamada, K. et al. EAAT4 is a post-synaptic glutamate transporter at Purkinje cell
synapses. Neuroreport 7, 20132017 (1996).
12. Mailleux, P., Takazawa, K., Erneux, C. & Vanderhaeghen, J. J. Comparison of
neuronal inositol 1,4,5-trisphosphate 3-kinase and receptor mRNA distributions
in the adult rat brain using in situ hybridization histochemistry. Neuroscience 49,
577590 (1992).
13. Kotera, J. et al. Expression of rat cGMP-binding cGMP-specific
phosphodiesterase mRNA in Purkinje cell layers during postnatal neuronal
development. Eur. J. Biochem. 249, 434442 (1997).
14. Roustan, P. et al. The rat phospholipase C beta 4 gene is expressed at high
abundance in cerebellar Purkinje cells. Neuroreport 6, 18371841 (1995).
15. Inglis, J. D., Lee, M., Davidson, D. R. & Hill, R. E. Isolation of two cDNAs
encoding novel alpha 1-antichymotrypsin-like proteins in a murine
chondrocytic cell line. Gene 106, 213220 (1991).
nature neuroscience volume 3 no 2 february 2000
articles
16. Akaaboune, M., Ma, J., Festoff, B. W., Greenberg, B. D. & Hantai, D. Neurotrophic
regulation of mouse muscle beta-amyloid protein precursor and alpha 1-
antichymotrypsin as revealed by axotomy. J. Neurobiol. 25, 503514 (1994).
17. Jiang, Y. H. et al. Mutation of the Angelman ubiquitin ligase in mice causes
increased cytoplasmic p53 and deficits of contextual learning and long-term
potentiation. Neuron 21, 799811 (1998).
18. Abraham, C. R., Selkoe, D. J. & Potter, H. Immunochemical identification of the
serine protease inhibitor alpha 1-antichymotrypsin in the brain amyloid deposits
of Alzheimers disease. Cell 52, 487501 (1988).
19. Das, S. & Potter, H. Expression of the Alzheimer amyloid-promoting factor
antichymotrypsin is induced in human astrocytes by IL-1. Neuron 14, 447456
(1995).
20. Koo, E. H., Abraham, C. R., Potter, H., Cork, L. C. & Price, D. L. Developmental
expression of alpha 1-antichymotrypsin in brain may be related to astrogliosis.
Neurobiol. Aging 12, 495501 (1991).
21. Tanaka, J., Ichikawa, R., Watanabe, M., Tanaka, K. & Inoue, Y. Extra-junctional
localization of glutamate transporter EAAT4 at excitatory Purkinje cell synapses.
Neuroreport 8, 24612464 (1997).
22. Birnbaumer, L. et al. On the molecular basis and regulation of cellular
capacitative calcium entry: roles for Trp proteins. Proc. Natl. Acad. Sci. USA 93,
1519515202 (1996).
23. Zhu, X. et al. trp, a novel mammalian gene family essential for agonist-activated
capacitative Ca2+ entry. Cell 85, 661671 (1996).
24. Drayer, A. L. et al. The family of inositol and phosphatidylinositol polyphosphate
5-phosphatases. Biochem. Soc. Trans. 24, 10011005 (1996).
25. Woscholski, R. & Parker, P. J. Inositol lipid 5-phosphatasestraffic signals and
signal traffic. Trends Biochem. Sci. 22, 427431 (1997).
26. De Smedt, F., Verjans, B., Mailleux, P. & Erneux, C. Cloning and expression of
human brain type I inositol 1,4,5-trisphosphate 5-phosphatase. High levels of
mRNA in cerebellar Purkinje cells. FEBS Lett. 347, 6972 (1994).
27. Takei, K. et al. Ca2+ stores in Purkinje neurons: endoplasmic reticulum
subcompartments demonstrated by the heterogeneous distribution of the InsP3
receptor, Ca2+-ATPase, and calsequestrin. J. Neurosci. 12, 489505 (1992).
28. Villa, A. et al. The endoplasmic reticulum of Purkinje neuron body and
dendrites: molecular identity and specializations for Ca2+ transport.
Neuroscience 49, 467477 (1992).
29. Volpe, P., Nori, A., Martini, A., Sacchetto, R. & Villa, A. Multiple/heterogeneous
Ca2+ stores in cerebellum Purkinje neurons. Comp. Biochem. Physiol. Comp.
Physiol. 105, 205211 (1993).
30. Matsumoto, M. et al. Ataxia and epileptic seizures in mice lacking type 1 inositol
1,4,5-trisphosphate receptor. Nature 379, 168171 (1996).
31. Street, V. A. et al. The type 1 inositol 1,4,5-trisphosphate receptor gene is altered
in the opisthotonos mouse. J. Neurosci. 17, 635645 (1997).
32. Miller, K. K., Verma, A., Snyder, S. H. & Ross, C. A. Localization of an
endoplasmic reticulum calcium ATPase mRNA in rat brain by in situ
hybridization. Neuroscience 43, 19 (1991).
33. Nakanishi, S., Maeda, N. & Mikoshiba, K. Immunohistochemical localization of
an inositol 1,4,5-trisphosphate receptor, P400, in neural tissue: studies in
developing and adult mouse brain. J. Neurosci. 11, 20752086 (1991).
34. Ross, C. A., Danoff, S. K., Schell, M. J., Snyder, S. H. & Ullrich, A. Three
additional inositol 1,4,5-trisphosphate receptors: molecular cloning and
differential localization in brain and peripheral tissues. Proc. Natl. Acad. Sci. USA
89, 42654269 (1992).
35. Berridge, M. J. Neuronal calcium signaling. Neuron 21, 1326 (1998).
36. Verkhratsky, A. & Toescu, E. C. Calcium and neuronal aging. Trends Neurosci .21,
27 (1998).
37. Liu, B. F., Xu, X., Fridman, R., Muallem, S. & Kuo, T. H. Consequences of
functional expression of the plasma membrane Ca2+ pump isoform 1a. J. Biol.
Chem. 271, 55365544 (1996).
38. Rando, R. R. Chemical biology of protein isoprenylation/methylation. Biochim.
Biophys. Acta 1300, 516 (1996).
39. Clarke, S. Protein isoprenylation and methylation at carboxyl-terminal cysteine
residues. Annu. Rev. Biochem. 61, 355386 (1992).
40. Hrycyna, C. A. & Clarke, S. Modification of eukaryotic signaling proteins by C-
terminal methylation reactions. Pharmacol. Ther. 59, 281300 (1993).
41. Choy, E. et al. Endomembrane trafficking of ras: the CAAX motif targets proteins
to the ER and Golgi. Cell 98, 6980 (1999).
42. Cummings, C. J. et al. Chaperone suppression of aggregation and altered
subcellular proteasome localization imply protein misfolding in SCA1. Nat.
Genet. 19, 148154 (1998).
43. Ciechanover, A. The ubiquitin-proteasome proteolytic pathway. Cell 79, 1321
(1994).
44. Masuyama, H. & MacDonald, P. N. Proteasome-mediated degradation of the
vitamin D receptor (VDR) and a putative role for SUG1 interaction with the AF-
2 domain of VDR. J. Cell. Biochem. 71, 429440 (1998).
45. Nawaz, Z., Lonard, D. M., Dennis, A. P., Smith, C. L. & OMalley, B. W.
Proteasome-dependent degradation of the human estrogen receptor. Proc. Natl.
Acad. Sci. USA 96, 18581862 (1999).
46. Zhang, J., Guenther, M. G., Carthew, R. W. & Lazar, M. A. Proteasomal
regulation of nuclear receptor corepressor-mediated repression. Genes Dev. 12,
17751780 (1998).
47. Doucas, V., Tini, M., Egan, D. A. & Evans, R. M. Modulation of CREB binding
protein function by the promyelocytic (PML) oncoprotein suggests a role for
nuclear bodies in hormone signaling. Proc. Natl. Acad . Sci. USA 96, 26272632
(1999).
163
 2000 Nature America Inc. http://neurosci.nature.com

articles

High-resolution mapping of iso-

orientation columns by fMRI
Dae-Shik Kim, Timothy Q. Duong and Seong-Gi Kim

Center for Magnetic Resonance Research, University of Minnesota Medical School, 2021 6th Street S.E., Minneapolis, Minnesota 55455, USA
Correspondence should be addressed to S.-G.K.(kim@cmrr.umn.edu)

Blood-oxygenation level-dependent (BOLD) functional magnetic resonance imaging (fMRI) is an

important tool for localizing brain functions in vivo. However, the ability of BOLD fMRI to map corti-
cal columnar structures is highly controversial, as the ultimate functional specificity of BOLD remains
unknown. Here we report a biphasic BOLD response to visual stimulation in the primary visual cortex
of cats. In functional imaging, the initial BOLD signal decrease accurately labeled individual iso-
2000 Nature America Inc. http://neurosci.nature.com

orientation columns. In contrast, the delayed positive BOLD changes indicated the pattern of overall
activation in the visual cortex, but were less suited to discriminate active from inactive columns.

Localization is a principle that is widely used in brain: cytoar-
chitectonically distinct areas form the basis for functional spe-
cialization1. Such parcellation of the cortical tissue into functional
subunits is especially prominent at the level of individual cortical
columns. In visual cortex of mammals, neurons with similar
response properties such as ocular dominance or orientation
preference are clustered into columns, spanning the entire cor-
tical plate from the pia to white matter24. Studies of the struc-
ture, function and plasticity of cortical columns using a variety
of traditional mapping techniques, however, suffer from funda-
mental limitations. For example, intra- and extracellular record-
ings yield insufficient field of view, and the 2-deoxyglucose5
technique is not viable for mapping in vivo. The optical imaging
of intrinsic signals allows simultaneous recording of neuronal
activity over large areas of cortex69. However, this technique can-
not be considered to be noninvasive, and furthermore, its appli-
cation is limited to the exposed cortical surface9,10.
The progress of blood-oxygenation level-dependent func-
tional magnetic resonance imaging11,12 raises hope that the func-
tional architecture of the living brain can be visualized
noninvasively, avoiding the limitations of the aforementioned
techniques. Using the paramagnetic deoxyhemoglobin as an
endogenous contrast agent11,12, BOLD-based functional images
can be obtained in vivo (in contrast to the 2-deoxyglucose (2-
DG) technique5,13), do not require extrinsic contrast agents (in
contrast to the positron emission technique14,15) and can access
activation signals from the entire brain (in contrast to optical
imaging69). Most importantly, the noninvasiveness of MRI ide-
ally suits this technique for studying the human brain during
cognitive and perceptual tasks1621.
Numerous BOLD studies during cognitive16, motor17 and per-
ceptual1821 tasks indicate good spatial correlation between neu-
ronal and hemodynamic responses at a coarse scale (several
millimeter to centimeter), and the BOLD signal pointspread is
comparable to that of optical imaging21. The ability of BOLD
fMRI to map the columnar architecture of the brain, however, is
controversial, as the ultimate functional specificity of BOLD is
undetermined. Because optical spectroscopy data predicts a
biphasic BOLD response following neuronal stimulation22,23
(with each BOLD phase potentially yielding different mapping
resolutions), it becomes imperative to determine the limits of
the functional specificity that can be achieved with BOLD. The
exact temporal kinetics of the BOLD responses in mammalian
brains, however, remain vigorously debated (compare refs. 2426
with refs. 27, 28); thus it remains to be seen whether the func-
tional specificity of BOLD is sufficient to map the basic compu-
tational units of the brains functional architecture, namely, that
of cortical columns.
RESULTS
To resolve the question of whether and to what extent the colum-
nar architecture of the brain can be labeled using the BOLD-
fMRI technique, we used ultra-high field magnets to obtain MR
signals originating from individual orientation columns in cat
visual cortex (area 18). Visual stimuli were optimized to drive
orientation-selective, complex-type area 18 neurons29. In this
study, area 18 was used because the distance between two iso-
orientation columns is greater in this area than in area 17 (ref.
30). Furthermore, area 18 on the lateral gyrus is essentially flat
in the cat and can be covered by a single imaging slice. We carried
out a total of ten semi-chronic experiments in ten hemispheres of
five different animals. Unless otherwise mentioned, similar results
were obtained in all ten experiments. Statistical data for all ten
studies are given in parentheses.
Figure 1a shows an anatomical MR image of cat visual cor-
tex on the lateral gyrus. All activation maps were derived from a
plane tangential to area 18 on the lateral gyrus (green box,
Fig. 1a). Colored pixels indicating regions of increased BOLD-
signal change (Fig. 1c; see Methods) reveal the pattern of cortical
activation in response to a moving grating oriented at 45. As
indicated in this panel, robust and homogenous activities were
observed in the lateral gyri of both hemispheres. The region of
activity extended several millimeters in anteriorposterior and
mediallateral directions.
In cat area 18, the average spacing between two neighboring
iso-orientation columns is 1.21.4 mm (ref. 30). Therefore, the
nominal in-plane resolution of 156 156 m2 per pixel achieved
in this study (see Methods) should have been sufficient to resolve
individual orientation columns. As evident in Fig. 1c, however, a
columnar layout was not obtained. All four activation maps

164 nature neuroscience volume 3 no 2 february 2000

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

articles

a b expected to greatly enhance the func-

tional specificity of fMRI. Such an
improvement of functional speci-
ficity was achieved in our study (Fig.
1d). Here, only the negative BOLD
percent changes originating within
the first two seconds after the stim-
ulus onset were used to generate
functional maps (see Methods). The
colored pixels indicate regions of sig-
nificant decrease in the MR signals
following visual stimulation. In
c d e sharp contrast to the conventional
fMRI maps based on the positive
BOLD-signal changes (Fig. 1c), the
foci of the negative signal changes are
confined to patchy clusters (Fig. 1d).
In line with the iso-orientation
2000 Nature America Inc. http://neurosci.nature.com

columns in cat detected with 2-DG30

and optical imaging32,33 techniques,
such semi-ellipsoidal and irregular-
ly shaped clusters were broadly dis-
tributed over the approximated area
18 with an average periodicity of
Fig. 1. Improvement of spatial specificity of BOLD using the initial negative signal changes. about 1.34 0.23 mm. Note also that
(a) Anatomical image of cat area 18 on the lateral gyrus (image size, 2 2 cm2). The green box indicates the areas defined by negative signal
the position of the imaging slice used to obtain the activation maps displayed in this study. (b) The bipha-
changes were located largely in tis-
sic MR signal time course following visual stimulation. Stimulus duration of 10 s is marked by the gray
box behind the time course. Scale bar (b), 10 s. (c) Pattern of increased BOLD activity in response to
sue areas, excluding regions of large
moving 45 gratings. The positive BOLD percent changes are marked with colors as indicated by the blood vessels such as that surround-
color scale below. (d) Functional map for which only negative BOLD percent changes occurring within ing the sagittal sinus.
the first two seconds after stimulus onset were used. The color scale below indicates the negative per- Given the large difference in
cent changes. (e) The pattern of positive BOLD responses after the threshold for functional image con- absolute amplitude between the neg-
struction was raised to match the number of the activated pixels to that in panel (d). See text for details. ative and positive signal changes, it
Scale bar (ce), 1 mm. A, anterior; P, posterior. is conceivable that the differences in
spatial pattern between the positive
(Fig. 1c) and negative (Fig. 1d) maps
could be simply due to the difficulty
obtained in response to moving gratings of four different orienta- in equating signal-threshold levels. Therefore, as a control, we
tions (0, 45, 90, 135) yielded homogeneous spatial distributions raised the threshold for positive responses to match the num-
that were hardly distinguishable (data not shown; see also ref. 31). ber of activated pixels to that in the negative map (Fig. 1e). The
This result is consistent with optical spectroscopy data22,23, resulting positive map largely reflects the pattern of surface vas-
suggesting that the stimulation of cortical neurons gives rise to culature in visual cortex.
a biphasic hemodynamic response of which only the early Patterns of negative activity obtained from the same patch of
increase of local deoxyhemoglobin (and hence a decrease in MR cortex while the animal viewed four different orientation stimuli
signal) reflects increased neuronal activity and oxygen con- further corroborate functional significance of the negative BOLD
sumption22,23. Subsequent decrease of deoxyhemoglobin (and signals (Fig. 2ad). Analogous to 2-DG30 and optical imaging32,33,
hence an increase of MR signal), which forms the basis for con- we show regions of high activity (large negative signals) as dark
ventional BOLD11,12, is hypothesized to originate mainly from pixels. Each pattern of negative activity was highly specific to the
the widespread increase in blood flow, thus making it less suit- respective stimulus orientation. To support this notion, outlines
able to discriminate electrically active from inactive columns. of the orientation patches for 90 and 135 maps were overlaid
In cat visual cortex, the time evolution of MR signals was indeed on the maps obtained for the respective orthogonal orientations
biphasic (Fig. 1b). Following visual stimulation (as indicated by the (Fig. 2ad).
gray box behind the time course), the MR signal decreased, reach- It is clear from these panels that the patches for orthogonal
ing a minimum of about 0.2% to 0.4% (0.28 0.1%, n = 10) orientations occupied cortical territories that were mostly com-
around 3 seconds (2.9 0.7 s, n = 10) after the stimulus onset. The plementary to each other. The complementarity between maps
signal then reversed, yielding a maximum positive signal change of of orthogonal orientations was highly significant (linear correla-
1.02.0% (1.3 0.4%, n = 10) approximately 810 seconds after tion between 0 and 90 maps, r = 0.4, p < 106; between 45 and
stimulus onset (8.0 1.3 s, n = 10). Biphasic BOLD responses fol- 135, r = 0.55, p < 106; see Methods). Stimulus-selective respons-
lowing visual stimulation were observed in all ten experiments at es based on negative BOLD signals were reflected at the colum-
two different magnetic fields. nar scale (Fig. 2e and f); these panels present the signal time
As the early negative signal changes are likely to reflect the courses obtained from 45 and 135 orientation columns (in green
transient increase of local deoxyhemoglobin in parenchymal tis- and in red, respectively; see Methods). MR signals in pixels rep-
sue 22,23, their use in generation of a functional map can be resenting 45 columns (green trace) decreased transiently during

nature neuroscience volume 3 no 2 february 2000 165

 2000 Nature America Inc. http://neurosci.nature.com
articles
a b
Negative BOLD activity
c d
e f
2000 Nature America Inc. http://neurosci.nature.com
Signal change (%)
Fig. 2. Representation of orthogonal orientations in complementary
cortical domains. (ad) Patterns of negative activity in response to four
orientations (0, 45, 90, 135). Regions of negative signal changes dur-
ing the first two seconds following visual stimulation are displayed as
dark pixels. For visual inspection of the complementarity between the
orthogonal orientation maps, the patches in panels (b) and (d) are out-
lined (in blue and red, respectively) and overlaid on the maps in panels
(a) and (c), respectively. With few exceptions (marked by yellow arrow-
heads), the maps for orthogonal orientations are complementary to
each other. Scale bar, 1 mm. (e, f) The signal time courses for 45
(green) and 135 (red) columns during the stimulation with 45 (c) and
135 (d) orientation stimuli. The stimulation duration of ten seconds is
marked by the gray boxes behind the time courses. Scale bar (e, f), 10 s.
45 stimulation (Fig. 2e), whereas little or no such decrease was
observed during 135 stimulation (Fig. 2f). Likewise, MR signals
from 135 columns (red trace) decreased transiently in response to
the 135 stimulus (Fig. 2f), but not to the complementary 45
stimulus (Fig. 2e). The difference between 45 and 135 columns
in magnitude of the early negative signals during the 45 (Fig. 2e)
and 135 (Fig. 2f) stimulation was statistically significant (p <
0.005 and p < 0.003, respectively; paired t-test).
The positive BOLD signals (Fig. 2e and f), on the other
hand, were much less suited to discriminate between 45 and
135 columns, as they yielded largely overlapping time cours-
es for the two orthogonal stimuli. The differences in positive
signals originating from the orthogonal columns were statis-
tically insignificant (p = 0.19, paired t-test). To allow more
direct comparison of the map quality, we also calculated the
45 and 135 maps obtained during the positive BOLD period
(Fig. 3a and b). All functional maps depicted in Figs. 2 and 3
were acquired in the same fMRI studies, and images were
processed with analogous methods (see Methods). In contrast
166
to the patchy, interdigitized columns of the negative maps, the
domains of the positive BOLD responses were larger, with no
apparent periodicity. The outlines of the 135 map from Fig.
3b were then overlaid on the 45 map (Fig. 3a). The regions of
positive activity for the orthogonal orientations were mostly
overlapping (linear correlation between the two maps, r = 0.69,
p < 106). The results in Figs. 2 and 3 indicate, therefore, that
although the amplitudes of the negative signals were far small-
er than those of the positive signals, the stimulus-specific con-
trast of the negative signals was superior to that of the positive
signals. Based on Fig. 2e and f, the stimulus-specific contrast
of the negative signals was calculated to be about 6.5 times larg-
er than that of the positive signals.
The ultimate validation of the veracity of the iso-orientation
columns based on negative BOLD changes would require simul-
taneous single-unit recording. Such simultaneous recording,
however, would require placement of the recording electrodes
in the iso-center of the magnetic bore, resulting in devastating
magnetic field interferences. Furthermore, it would be almost
impossible to change the electrode track or introduce new elec-
trodes without repositioning the animal. Alternatively, the speci-
ficity of the negative BOLD signals at the columnar scale can be
tested by comparing their detailed spatial pattern with that
obtained by using more traditional brain-mapping techniques.
A characteristic and invariant feature of the mammalian orien-
tation system is the presence of topological singularities3234,
which are observed in many species using both multi-elec-
trode34,35 and optical-imaging32,33,3638 techniques. Therefore, to
validate the functional specificity of the negative BOLD maps,
we calculated the composite-angle maps through vector sum-
mation of the underlying activation maps obtained for the four
different orientations (see Methods; Fig. 4).
In the composite map based on only the early negative
BOLD signals (Fig. 4a), the colors (preferred orientations)
change smoothly, forming a map of orientation selectivity. The
continuity of orientation preferences is interrupted only at the
singular points (termed orientation pinwheels) where the
domains for all orientations converge, as observed using multi-
electrode34,35 and optical imaging32,33,3638 techniques. Each ori-
entation is represented only once around such a pinwheel;
therefore each of these topological singularities takes on one of
two rotational chiralities, namely clockwise or counter-clock-
wise pinwheels. Two such pinwheels are enlarged (Fig. 4a, right).
The pinwheel density and the ratio between clockwise and
counter-clockwise singularities found in negative BOLD com-
posite maps were in excellent agreement with those obtained
with optical imaging. We found a pinwheel density of
1.46 0.17 pinwheels per mm2 (n = 4); in comparison, optical
imaging studies 32,37 yielded average pinwheel densities of
1.21.95 pinwheels per mm2. Likewise, the ratio of clockwise
to counterclockwise pinwheels was found to be roughly 1:1 in
both negative-BOLD (1:0.89, n = 4) and optical imaging 37
(1:0.9) data.
Although the above data are in excellent agreement with data
from multi-electrode34,35 and optical imaging studies32,33,3638,
it is theoretically possible but unlikely that such topological sin-
gularities could arise from a randomly distributed pattern of
activity39. Therefore, as a control, composite maps based on sig-
nals obtained before stimulus onset (Fig. 4b) and those during
the delayed positive BOLD changes (Fig. 4c) were also calcu-
lated. All steps for composite-map construction were performed
exactly as for Fig. 4a (see Methods). Unlike the composite map
based on negative BOLD signals (Fig. 4a), the maps based on
nature neuroscience volume 3 no 2 february 2000
 2000 Nature America Inc. http://neurosci.nature.com
Fig. 3. Representation of orthogonal orienta-
tions of positive BOLD maps. (a, b) Functional a b
maps based on positive signal changes obtained
with 45 and 135 stimuli, respectively. High
positive signals are represented by dark pixels
(see text). Maps in Figs. 2 and 3 were obtained
from the same experiment. The domains of
high activity in 135 map (b) are outlined and
overlaid on the 45 map (a). Most of the active
domains in both maps overlapped extensively,
except a few areas as marked with green
arrowheads. Scale bar, 1 mm.
2000 Nature America Inc. http://neurosci.nature.com
control (Fig. 4b) or delayed positive BOLD signals (Fig. 4c) dis-
played none of the characteristic features of the mammalian
angle maps.
DISCUSSION
Our results indicate that the functional specificity of BOLD at
the columnar scale depends highly on the temporal dynamics
of the underlying signals. If the early negative signals are used,
functional maps at columnar resolution can be obtained without
differential imaging (see below). The delayed positive BOLD
changes, on the other hand, clearly indicated the pattern of the
overall activation per se in the visual cortex, but were less suit-
ed to discriminate between active and inactive columns, as they
were more diffuse2123 and less specific to the individual stimu-
lus properties.
It is, of course, conceivable that, in principle, even such dif-
fuse positive BOLD signals might yield columnar patterns of
activity if maps of orthogonal conditions were subtracted from
each other. Such differential imaging had been suggested for
analyzing optical imaging40 and conventional (positive) BOLD
data41. Although the use of a differential method for optical-
imaging data might be defendable given the well established ver-
ification of intrinsic signals with extensive single-unit studies37,38,
a commensurate increase in blood flow22,23.
b c
nature neuroscience volume 3 no 2 february 2000
articles
Positive BOLD activity
the use of a differential method for BOLD fMRI data poses
severe difficulties. In differential imaging, one activation map
(for example, of the left eye) is subtracted from the other acti-
vation map (the right eye), with the assumption that the two
maps yield complementary activation patterns. If this assump-
tion is truesuch as between left- and right-eye domainsthen
the result of the subtraction is tautological, because it was known
in advance. However, if the assumption is not true or simply
unknown, as for most receptive-field properties, then the sub-
traction method will probably give the wrong answer. Without
direct validation of BOLD using simultaneous single-unit
recordings, the applicability of the differential method for BOLD
data remains questionable. The main significance of the nega-
tive BOLD imaging is therefore that a columnar pattern of high
functional specificity can be obtained directly, without the need
for such differential methods.
As the T2 and T2*-weighted BOLD contrast predominantly
originates from the regional changes in paramagnetic deoxyhe-
moglobin concentration10,11,42, the early decrease of BOLD signals
can most likely be attributed to the regional increase of deoxyhe-
moglobin following neuronal stimulation. Although alternative
explanations exist43,44, the most parsimonious explanation for such
transient BOLD signal decrease is caused by elevated oxygen con-
sumption44 in the active orientation column without a
Recently, similar decreases in BOLD signals have been
observed in the visual cortices of awake human2426,45 and
anesthetized nonhuman primates46 during perceptual
Fig. 4. fMRI-based composite angle maps. (a) The composite
angle map obtained through pixel-by-pixel vector addition of
the four single iso-orientation maps based on negative signal
changes. The color key next to (a) was used for color coding
the resulting orientation preferences. Overall continuity of the
orientation preferences is interrupted at the orientation pin-
wheels where the cortical columns for different orientations
are circularly arranged. The white and black circles in (a) depict
clockwise and counterclockwise pinwheels, respectively. Two
of such pinwheels are enlarged right (a). Scale bar for the
enlarged pinwheels, 200 m. As a control, the composite maps
based on MR signals obtained before stimulus onset (b) and
during positive BOLD signals (c) are also displayed. Maps in
(b) and (c) were obtained from the same cortical region as in
(a). The control maps are devoid of topological structures
characteristic of genuine composite angle maps. Scale bar
(ac), 1 mm. A, anterior; P, posterior; M, medial; L, lateral.
167
 2000 Nature America Inc. http://neurosci.nature.com
articles
tasks, indicating that the capability of BOLD-fMRI to label func-
tional columns in vivo should also be applicable in humans and
monkeys. We conclude from our study that noninvasive fMRI of
brain functions can be performed at the columnar levels. Thus
it now becomes possible to study the precise three-dimensional
layouts of columnar structures in mature and developing brains,
regardless of their location in the brain.
METHODS
Animal preparation. All animal experiments were performed with insti-
tutional approval. Animals were initially anesthetized with a ketamine
(1025 mg per kg, i.v.) and xylazine (2.5 mg per kg) cocktail. The ani-
mals were intubated and artificially ventilated (2535 strokes per min,
1530 ml per stroke) under isoflurane anesthesia (0.81.5%) in a N2O:O2
mixture of 70:30 throughout the experiment. The animals eyes were
refracted and corrective contact lenses used if necessary. The animal was
placed in a cradle and restrained in normal postural position using a cus-
tom-designed stereotaxic frame. Endtidal CO2, respiration rate and rec-
tal temperature were monitored throughout the experiment. Following
2000 Nature America Inc. http://neurosci.nature.com
the three-to-five-hour MR experiments, gas anesthesia was discontin-
ued, and animals were kept on the respirator until they could breathe
independently. The animals were observed for 0.52.0 h before being
returned to their littermates.
Stimulation protocol. Animals were stimulated binocularly. Visual stim-
uli consisted of high-contrast square-wave moving gratings (0.15 cycles
per degree, 2 cycles per s) of four different orientations (0, 45, 90,
135), optimized to elicit responses from neurons in area 18 of the cat
visual cortex29. During the resting period, a stationary grating pattern
of identical spatial frequency and orientation was presented. A video pro-
jector (Resonance Technology, Northridge, California; resolution,
640 480 pixels) was used to project the visual stimuli onto a screen
positioned 15 cm from the animals eyes. The screen covered about 37 of
the animals visual field.
MR methods. After placing the animal in a cradle, we placed a small sur-
face coil (diameter, 1.2 cm) on top of the animals head corresponding
to the Horsley-Clark A3. All MR experiments were performed on an
Oxford (4.7 T, 40 cm in diameter; Oxford, UK) or Magnex (9.4 T, 31 cm
in diameter; Abingdon, UK) horizontal MRI scanner equipped with a
magnetic field gradient of 15 or 30 gauss per cm, respectively. Data
obtained at two magnetic field strengths were analyzed separately. The
amplitude and phase of the biphasic time courses obtained at the two
field strengths however, were averaged together, as they yielded compa-
rable results. BOLD measurements on a single image slice was made using
gradient-echo echo-planar imaging (EPI). The imaging slice was posi-
tioned 500 m below the pia to avoid superficial draining vessels. The
MR imaging parameters were data matrix, 64 64; single-shot (4.7 T)
or 2-shot (9.4 T) EPI; FOV, 2 cm 2 cm; slice thickness, 2 mm; TE, 31
ms (4.7 T) or 12 (9.4 T) ms; TR, 0.5 s. A total of 160 images were acquired
during each epoch (60 images before stimulation, 20 images during stim-
ulation and 80 images after stimulation). Images with a 64 64 matrix
were zero-filled to a 128 128 matrix, resulting in nominal in-plane res-
olution of 156 156 m2 per pixel. Images obtained for the same orien-
tations within a single fMRI session were averaged for signal-to-noise
ratio improvement (usually five to ten epochs).
Functional image construction. To determine the existence of BOLD
activity per se, the MR signals were first cross-correlated with the stimu-
lation protocol47. The cross-correlation coefficient threshold was set at
> 98% confidence level. The percent-change maps within the negative
and positive portions of BOLD time course were averaged into respec-
tive time-binned maps (positive and negative map, respectively). For the
map of the negative response, individual pixels showing negative percent
change 0.51.0 s.d. from the mean baseline percent change were taken
as statistically significant. Because we used clustered pixel analysis with a
cluster size of 4 pixels, the nominal p-value for our threshold was p < 0.2
(ref. 47). The same threshold was used for calculating the four different
168
iso-orientation maps for each experiment. For the positive response (Fig.
1c), the threshold was raised in proportion to ratio of the maximum pos-
itive to the minimum negative BOLD response for each animal (approx-
imately fivefold). For the negative and positive BOLD maps depicted in
Fig. 1, no subtraction or image filtering method was applied.
Analysis of iso-orientation maps. Similar image-processing methods
were used to construct the iso-orientation maps based on negative or
positive BOLD data. Analogous to the standard analysis of optical-imag-
ing data32,33,37, responses to the four different orientations were obtained
by dividing each single orientation map (raw percent-change maps) by
the normalized sum of four orientations (cocktail blank). For display-
ing the maps, the dynamic range of the negative (Fig. 2ad) and posi-
tive (Fig. 3a and b) maps were clipped at the upper and lower three
percent of the distribution and linearly mapped on an eight-bit gray
scale. In both the negative and positive maps, high-frequency noise was
removed using a 3 3 pixel Gaussian kernel. The complementarity of
the orthogonal maps was tested by calculating the linear correlation coef-
ficient between the two maps48. Correlation coefficient p-values were
calculated using standard algorithms48. For the MR time courses in Fig.
2, the ROI for the 45 columns was selected on the basis of negative
responses during only the 45 stimulation. The same ROI was then used
to plot the time course of those pixels during the 135 stimulation. Like-
wise, the ROI for the 135 columns was selected on the basis of negative
responses during only the 135 stimulation. This ROI was then used to
plot the time course of the pixels during the 45 stimulation. The differ-
ences in positive signal contrasts during the 45 and 135 stimulations
were therefore independent of the initial pixel selection process.
Analysis of angle maps. Composite angle maps for the negative BOLD
changes (Fig. 4a) were obtained through a pixel-by-pixel vector addition
of the four iso-orientation maps (see above) with the negative percent
changes as vector amplitudes and the respective stimulus orientations as
vector angles7,32,33. Such composite angle maps were obtained for four
different experiments that yielded the best contrast-to-noise ratio. The
resulting angle at each pixel was color-coded using a circular color table.
The positions and densities of the pinwheel centers were detected by cal-
culating the spatial gradients of the composite angle map32,37 and checked
visually. As a control, composite maps were also calculated based on
MR signals obtained before visual stimulation (Fig. 4b) and those
obtained during delayed positive BOLD responses (Fig. 4c; 820 s after
stimulus onset). For the sake of consistency, the composite maps in Fig.
4ac were generated using identical methods for the region of interest,
threshold and vector-summation.
ACKNOWLEDGEMENTS
We thank K. Ugurbil for continuing support of our project and S. Ogawa, A.
Grinvald, R.B. Tootell, J. Ashe and A. P. Georgopoulos for suggestions and com-
ments. H. Merkle and J. Strupp provided support in hardware and software.
This work was supported by the NIH (NS38295, MH57180, NS10930,
RR08079), the Minnesota Medical Foundation and the Keck Foundation.
RECEIVED 20 AUGUST; ACCEPTED 29 NOVEMBER 1999
1. Brodmann, K. Vergleichende Lokalisationslehre der Grosshirnrinde (Johann
Ambrosius Barth, Leipzig, 1909).
2. Hubel, D. H. & Wiesel, T. N. Receptve field, binocular interactions and
functional architecture in the cats visual cortex. J. Physiol. (Lond.) 160,
106154 (1962).
3. LeVay, S., Stryker, M. P. & Shatz, C. J. Ocular dominance columns and
their development in layer IV of cats visual cortex. J. Comp. Neurol. 179,
223244 (1978).
4. LeVay, S. & Nelson, S. B. Vision and Visual Dysfunction 266315
(Macmillan, Houndsmill, 1991).
5. Sokoloff, L. et al. The 14C-deoxyglucose method for the measurement of
local cerebral glucose utilization: theory, procedure, and normal values in
the conscious and anesthetized albino rat. J. Neurochem. 28, 897916
(1977).
nature neuroscience volume 3 no 2 february 2000
 2000 Nature America Inc. http://neurosci.nature.com
6. Grinvald, A., Lieke, E., Frostig, R. D., Gilbert, C. D. & Wiesel, T. N. Functional
architecture of cortex revealed by optical imaging of intrinsic signals. Nature
324, 361364 (1986).
7. Tso, D. Y., Frostig, R. D., Lieke, E. E. & Grinvald, A. Functional organization
of primate visual cortex revealed by high resolution optical imaging. Science
249, 417420 (1990).
8. Frostig, R. D., Lieke, E. E., Tso, D. Y. & Grinvald, A. Cortical functional
architecture and local coupling between neuronal activity and the
microcirculation revealed by in vivo high-resolution optical imaging of
intrinsic signals. Proc. Natl. Acad. Sci. USA 87, 60826086 (1990).
9. Bonhoeffer, T. & Grinvald, A. The layout of iso-orientation domains in area
18 of cat visual cortex: optical imaging reveals a pinwheel-like organization. J.
Neurosci. 13, 41574180 (1993).
10. Stetter, M. & Obemayer K. Simulation of scanning laser technique for optical
imaging of blood-related intrinsic signals. J. Opt. Soc. Am. A 16, 5870
(1999).
11. Ogawa, S. et al. Intrinsic signal changes accompanying sensory stimulation:
functional brain mapping with magnetic resonance imaging. Proc. Natl.
Acad. Sci. USA 89, 59515955 (1992).
12. Kwong, K. K. et al. Dynamic magnetic resonance imaging of human brain
activity during primary sensory stimulation. Proc. Natl. Acad. Sci. USA 89,
56755679 (1992).
13. Loewel, S. & Singer, W. Tangential intracortical pathways and the
development of iso-orientation bands in cat striate cortex. Dev. Brain Res. 56,
2000 Nature America Inc. http://neurosci.nature.com
99115 (1990).
14. Fox, P. T. & Raichle, M. E. Focal physiological uncoupling of cerebral blood
flow and oxidative metabolism during somatosensory stimulation in human
subjects. Proc. Natl. Acad. Sci. USA 83, 11401144 (1986).
15. Posner, M. I. & Raichle, M. E. Images of Mind (Freeman, New York, 1994).
16. Wagner, A. D. et al. Building memories: remembering and forgetting of verbal
experiences as predicted by brain activity. Science 281, 11881191 (1998).
17. Kim, S.-G. et al. Functional magnetic resonance imaging of motor cortex:
hemispheric asymmetry and handedness. Science 261, 615617 (1993).
18. DeYoe, E. A. et al. Mapping striate and extrastriate visual areas in human
cerebral cortex. Proc. Natl. Acad. Sci. USA 93, 23822386 (1996).
19. Sereno, M. I. et al. Borders of multiple visual areas in humans revealed by
functional magnetic resonance imaging. Science 268, 889893 (1995).
20. Tootell, R. B. et al. Functional analysis of V3A and related areas in monkey
visual cortex. J. Neurosci. 17, 70707078 (1997).
21. Engel, A. S., Glover, G. H. & Wandell, B. A. Retinotopic organization in
human visual cortex and the spatial precision of functional MRI. Cereb.
Cortex 7, 181192 (1997).
22. Malonek, D. & Grinvald, A. Interactions between electrical activity and
cortical microcirculation revealed by imaging spectroscopy: Implication for
functional brain mapping. Science 272, 551554 (1996).
23. Malonek, D. et al. Vascular imprints of neuronal activity: relationships
between the dynamics of cortical blood flow, oxygenation, and volume
changes following sensory stimulation. Proc. Natl. Acad. Sci. USA 94,
1482614831 (1997).
24. Ernst, T. & Henning, J. Observation of a fast response in functional MR.
Magn. Reson. Med. 32, 146149 (1994).
25. Menon, R. S. et al. BOLD based functional MRI at 4 Tesla includes a capillary
bed contribution: Echo planar imaging correlates with previous optical
imaging using intrinsic signals. Magn. Reson. Med. 33, 453459 (1995).
26. Hu, X., Le, T. H. & Ugurbil, K. Evaluation of the early response in fMRI in
individual subjects using short stimulus duration. Magn. Reson. Med. 37,
877884 (1997).
nature neuroscience volume 3 no 2 february 2000
articles
27. Marota, J. J. A. et al. Investigation of the early response to rat forepaw
stimulation. Magn. Reson. Med. 41, 247252 (1999).
28. Silva, A. C., Lee, S.-P., Iadecola, C. & Kim, S.-G. Early characteristics of CBF
and deoxyhemoglobin changes during somatosensory stimulation. J. Cereb.
Blood Flow Metab. 20, 201206 (2000).
29. Movshon, J. A., Thompson, I. D. & Tolhurst, D. J. Spatial and temporal
contrast sensitivity of neurons in areas 17 and 18 of the cats visual cortex. J.
Physiol. (Lond.) 283, 101120 (1978).
30. Loewel, S., Freeman, B. & Singer, W. Topographic organization of the
orientation column system in large, flat-mounts of the cat visual cortex: A 2-
deoxyglucose study. J. Comp. Neurol. 255, 401415 (1987).
31. Jezzard, P., Rauschecker, J. P. & Malonek, D. An in vivo model for functional
MRI in cat visual cortex. Magn. Reson. Med. 38, 699705 (1997).
32. Bonhoeffer, T. & Grinvald, A. Iso-orientation domains in cat visual cortex are
arranged in pinwheel-like pattern. Nature 353, 429432 (1991).
33. Kim, D.-S. & Bonhoeffer, T. Reverse occlusion leads to a precise restoration of
orientation preference maps in visual cortex. Nature 370, 370372 (1994).
34. Swindale, N. W, Matsubara, J. A. & Cynader, M. S. Surface organization of
orientation and direction selectivity in cat area 18. J. Neurosci. 7, 14141427
(1987).
35. Diao, Y. C., Jia, W. G., Swindale, N. V. & Cynader, M. S. Functional
organization of the cortical 17 per 18 border region in the cat. Exp. Brain Res.
79, 271282 (1990).
36. Maldonado, P. E., Goedecke, I., Gray, C. M. & Bonhoeffer, T. Orientation
selectivity in pinwheel centers in cat visual cortex. Science 276, 15511555
(1997).
37. Shmuel, A. & Grinvald, A. Functional organization for direction of motion
and its relationship to orientation maps in cat area 18. J. Neurosci. 16,
69456964 (1996).
38. Crair, M. C., Gillespie, D. C. & Stryker, M. P. The role of visual experience in
the development of columns in cat visual cortex. Science 279, 566570 (1998).
39. Rojer, A. S. & Schwartz, E. L. Cat and monkey cortical columnar patterns
modeled by bandpass-filtered 2D white noise. Biol. Cybern. 62, 381391
(1990).
40. Blasdel, G. G. Differential imaging of ocular dominance and orientation
selectivity in monkey striate cortex. J. Neurosci. 12, 31153138 (1992).
41. Menon, R. S., Ogawa, S. Strupp, J. P. & Ugurbil, K. Ocular dominance in
human V1 demonstrated by functional magnetic resonance imaging. J.
Neurophysiol. 77, 27802787 (1997).
42. Ogawa, S., Menon, R. S., Kim, S.-G. & Ugurbil, K. On the characterizatics of
functional magnetic resonance imaging of the brain. Annu. Rev. Biophys.
Biomol. Struct. 27, 447474 (1998).
43. Janz, C., Speck, O. & Henning, J. Time-resolved measurements of brain
activation after a short visual stimulus: new results on the physiological
mechanisms of the cortical response. NMR Biomed. 10, 222229 (1997).
44. Vanzetta, I. & Grinvald, A. Increased cortical oxidative metabolism due to
sensory stimulation: implications for functional brain imaging. Science 286,
15551558 (1999).
45. Yacoub, E. & Hu, X. Detection of the early negative response in fMRI at 1.5
Tesla. Magn. Reson. Med. 41, 10881092 (1999).
46. Logothetis, N. K., Guggenberger, H., Peled, S. & Pauls, J. Functional imaging
of the monkey brain. Nat. Neurosci. 2, 555562 (1999).
47. Forman, S. D. et al. Improved assessment of significant activation in
functional magnetic resonance imaging (fMRI): use of a cluster-size
threshold. Magn. Reson. Med. 33, 636647 (1995).
48. Press, W. H., Teukolsky, S. A., Vetterling, W. T. & Flannery, B.P. Numerical
Recipes in C (Cambridge, UK, 1992).
169
 2000 Nature America Inc. http://neurosci.nature.com

articles

Dynamics of perceptual oscillations

in form vision
Hugh R. Wilson1, Boris Krupa1 and Frances Wilkinson2

1 Visual Science Center, University of Chicago, 939 E. 57th Street, Chicago, Illinois 60637, USA
2 Department of Psychology, 1205 Docteur Penfield Avenue, McGill University, Montreal, Quebec, H3A 1B1, Canada
Correspondence should be addressed to H.R.W. (hrw6@midway.uchicago.edu)

Certain periodic dot patterns (Marroquin patterns) generate a percept of dynamically oscillating
circles, and analogous effects were explored by op artists in the 1960s. Here we show psychophysically
that circles are perceived in these patterns only around specific points that are quantitatively predicted
2000 Nature America Inc. http://neurosci.nature.com

by a neural model of configural units hypothesized to reside in cortical area V4. Circles superimposed
on the pattern mask perception of illusory circles. A neural model of lateral inhibitory interactions
among V4 configural units showing spike-frequency adaptation quantitatively accounts for the human
data. The model is consistent with ideas on the neural basis of attention in V4, and it suggests that
attention may be biased via neuromodulation of slow hyperpolarizing potentials in cortical neurons.

Certain static patterns generate dynamic visual percepts, an
effect on which op artists capitalized frequently during the
1960s1,2. One salient example is the pattern (Fig. 1a) created in
1976 by Marroquin3. This pattern is produced by superimpos-
ing three copies of a square dot grid, each copy being rotated
by 60 relative to the others. This static pattern generates a per-
cept of circular shapes that appear and vanish at various loca-
tions in an oscillatory fashion. Vision scientists have
characterized this and related oscillating patterns as products
of dynamic grouping4,5. Such explanations, however, raise ques-
tions concerning the nature of this grouping and the underlying
dynamical processes producing oscillations. Here we apply psy-
chophysical techniques to measure the frequency and duration
of circle visibility throughout the Marroquin pattern, and we
demonstrate that a plausible neural model can quantitatively
account for the data.
In searching for a possible neural basis for the oscillating cir-
cles in the Marroquin pattern, a connection with the perception
of circular structure in Glass patterns6,7 suggested itself. (See refs.
8, 9 for examples of Glass patterns.) Circular structure in these
random-dot patterns is considerably more salient than radial,
hyperbolic or translational structure8,9. The visual system extracts
concentric structure from Glass patterns by linear pooling of
quasi-concentric orientations, and a neural model derived from
these observations provides a quantitative explanation of the
data8. Comparison with primate single-unit physiology1012 sug-
gests that such concentric units might arise in V4, an intermedi-
ate area in the ventral form-vision system13,14. Evidence for units
optimized for analysis of quasi-concentric structure in human
V4 is supported by converging evidence from fMRI (James et al.,
Invest. Ophthalmol. Vis. Sci. 40, S819, Abstr. 4313, 1999), event-
related potentials recorded directly from the human cortical sur-
face15 and studies of a patient with unilateral V4 damage (Mazer
et al., Soc. Neurosci. Abstr. 25, 212.14, 1999). Neural simulations
have demonstrated that model V4 units can provide a popula-
tion code for the location, shape and bilateral symmetry of
human heads16. For example, model V4 concentric units can
extract the center and mean radius of a face transparently added
to a house (Fig. 1b), a stimulus that has been used to provide
evidence for object-based attention17. Thus, V4 concentric units
may be important in face and other ellipsoidal object analysis as
well as in selective attention.
Selective attention effects are evident in V4 at the single neu-
ron level1821 and are believed to result from inhibitory compe-
tition in parallel networks22. This suggested that the Marroquin
illusion might arise from competitive interactions among V4
concentric units, and that the illusion might therefore provide
further insight into the mechanisms of selective attention in V4.
The results reported here confirm that inhibitory interactions
among model V4 concentric units account quantitatively for vis-
ibility of Marroquin circles throughout the pattern. The resul-
tant neural network may therefore form a basis for spatial or
object based selective attention, with neuromodulation of spike-
frequency adaptation providing a mechanism for top-down bias-
ing of attentional effects.
RESULTS
Psychophysics
To understand the dynamics underlying the Marroquin illusion,
we first asked whether circles were more likely to appear at some
locations than at others. The visibility of circles centered at dif-
ferent points in the Marroquin pattern was quantified using a
procedure analogous to that generally used in binocular rivalry
measurements23. In each experiment, subjects fixated a red dot at
a designated position within the Marroquin pattern and depressed
a button whenever an illusory circle was perceived to be centered
on that dot (see Methods). Because of the hexagonal symmetry
of these patterns, measurements were restricted to ten dot posi-
tions within one sextant of the pattern. Visibility was defined as
the fraction of a two-minute viewing period during which an illu-
sory circle of specified diameter (line in Fig. 1a) was perceived to
surround the designated center dot. Data for one subject are plot-
ted at all ten positions for runs on three different days (Fig. 2a).
Although data at different positions were collected in pseudo-

170 nature neuroscience volume 3 no 2 february 2000

 2000 Nature America Inc. http://neurosci.nature.com
a (a) Marroquin pattern3 produced by superimposing three square grids
2000 Nature America Inc. http://neurosci.nature.com
b of extracting ellipsoidal object boundaries from complex images.
random order on different days, performance was highly repro-
ducible from day to day. Results averaged over three days of exper-
iments for each of four subjects show a similar pattern (Fig. 2b).
In comparing subjects, visibilities were normalized relative to that
at position 1. (Absolute visibilities are listed in Table 1.) Although
there was some variability among subjects, illusory-circle visibil-
ity was high for all subjects at positions 1, 7 and 10, whereas vis-
ibility was low at positions 5 and 8. One-way ANOVA showed a
significant effect of position (F9,110 = 8.42, p < 0.0001). A post-hoc
Fischers PLSD showed significant differences between each of
positions 1, 7 and 10 and positions 5 and 8 at p < 0.0001.) Clear-
ly, therefore, Marroquin patterns generate illusory circles prefer-
entially surrounding particular hot spots for all subjects, whereas
circles are rarely perceived around other cold spots.
Mean durations of illusory circle visibility are indicated for
position 1 (the center of the pattern) in Table 1. Two subjects
(FW and HRW) showed mean durations in the 23.5 s range,
similar to binocular rivalry23,24, and both produced standard devi-
ations comparable to the mean. The remaining two subjects had
considerably longer mean visibilities.
As the Marroquin oscillations are suggestive of instabilities in
a competitive network, we wondered whether pattern masking
might be effective in altering the mean visibility of illusory cir-
cles. To test this, further data were gathered at position 1 with a
superimposed mask comprising four surrounding gray circles
nature neuroscience volume 3 no 2 february 2000
articles
Fig. 1. Patterns generating percepts of circular or ellipsoidal structure.
of dots, each rotated by 60 relative to the others. In experiments, dots
were presented against a gray background of mean luminance 50.5 cd
per m2. While subjects fixate this pattern, illusory circles appear and
vanish centered at various locations throughout the pattern. Although
illusory circles of several different diameters are sometimes seen, with
the dot separations and viewing distances used in our experiments, only
illusory circles of one size were apparent (diameter indicated by hori-
zontal line in lower right). (b) Simulated response of model V4 concen-
tric units to a transparent face/house image of the type used in a study of
object-based attention17. Processing of the face alone produced a maxi-
mum model response at the point indicated by the upper black circle.
The mean radius of the stimulus was encoded by the maximum model
response and corresponded to the distance indicated by the arrow. The
model population response also encoded the stimulus as vertically elon-
gated16. Superposition of the house onto the face only shifted the maxi-
mum V4 model response to the lower black circle but left both the
estimated radius and axis of elongation essentially unchanged. Model V4
concentric units only responded to the face shape and thus are capable
(Fig. 3a). These masking circles were centered at varying dis-
tances from the illusory center circle in different experiments.
Both large circles (diameter 1.80) and smaller circles (diameter
0.90) produced significant reductions in the visibility of the illu-
sory circle centered about position 1 (Fig. 3b). Significantly,
masking data from both circle sizes fell off as the same function
F of increasing radius R from position 1:
F(R) = 1 A exp ( R5/5) (1)
Only the amplitude A increased with increasing circle diam-
eter, and s = 1.97 for both diameters. This suggests lateral
inhibitory interactions among units optimized to extract quasi-
circular structure from stimuli. To test this idea further, the cir-
cular masks were replaced with four parallel vertical lines having
the same separations as the circle diameter and the same total
length as the circle circumferences. These lines produced almost
no masking of illusory circle visibility (Fig. 3b). This provides
further evidence that Marroquin patterns evoke lateral inhibi-
tion among units selective for concentric structure.
Neural model
The Marroquin oscillations can be explained by extending a neur-
al model developed to account for the visibility of concentric
structure in Glass patterns8. This model uses a sequence of ori-
ented filtering, full-wave rectification and orthogonal second-
stage oriented filtering (Fig. 5) to extract local curvature
information from the stimulus25,26. There is direct psychophysi-
cal evidence that the final stage incorporates linear concentric
pooling of curvature information8,9. Responses of such model
units agree with a subset of neurons recorded in V4 in several
studies1012, and we will therefore refer to them as model V4 con-
centric units. Processing the Marroquin stimulus by an array of
these units produced a response showing strong activation of
model V4 units by certain pattern locations, but little or no acti-
vation by other locations (Fig. 4a). The three hot spots and two
cold spots from the psychophysical data are marked, demon-
strating qualitative agreement of the model responses with the
data. A quantitative comparison among all ten positions demon-
strates that the model responses agree well with relative visibili-
ties averaged across subjects (Fig. 2c). Thus, the V4 concentric
model accurately predicts relative visibilities of illusory circles at
different locations throughout the Marroquin pattern.
171
 2000 Nature America Inc. http://neurosci.nature.com
articles
a Marroquin pattern. Because of hexagonal symmetry, these points were
FW #1
FW #2
FW #3
Visibility
Position
b
2000 Nature America Inc. http://neurosci.nature.com
Relative visibility
c Position
Mean data
Relative visibility or model response
V4 model
Position
The oscillations of illusory circles in the Marroquin pattern
must now be explained. The psychophysical masking data pro-
vide evidence for lateral inhibitory interactions that decline with
the distance between concentric units, and this suggests the exis-
tence of a spatially regional winner-take-all (WTA) network
among V4 concentric units. WTA networks in general, however,
cannot generate oscillations unless the winners adapt or fatigue
with time27. Here a property of virtually all neocortical excitatory
(but not inhibitory) neurons in humans and other mammals sug-
gests itself, namely, spike-frequency adaptation2830. When stim-
ulated by a constant-current input, neocortical excitatory neurons
initiate firing at a high rate that typically drops by a factor of about
3.0 within a few hundred ms. This spike-frequency adaptation is
mediated by a slow Ca2+-mediated K+ afterhyperpolarizing cur-
rent (AHP). Inclusion of spike-frequency adaptation in a region-
al WTA network of model V4 concentric units produces a
quantitative account of Marroquin pattern oscillations.
172
Fig. 2. Visibility of illusory circles centered at ten different points in the
all chosen from one sextant of the pattern. Position 1 was the center of
the pattern, and successively higher numbered points progressed out-
ward toward the upper left. (a) Visibility (fraction of time an illusory cir-
cle was seen during a two-minute run) is plotted for all positions from
three different days of experiments on FW. Despite pseudorandom
ordering of data collection from day to day, a high degree of repeatabil-
ity is evident. (b) Mean data for all four subjects are plotted normalized
to the relative visibility at position 1. (Absolute visibilities are listed in
Table 1.) Illusory circles were highly visible at positions 1, 7 and 10 for
all subjects, whereas circles were seldom visible around positions 5 and
8. These five positions have been marked (Fig. 4a). This dependence of
illusory circle visibility was highly statistically significant (see text). (c) V4
model predictions of visibility variations across positions are compared
with data averaged across all subjects.
Designating excitatory and inhibitory neuron spike rates by
E and I, and the AHP current by H, a suitable set of equations to
describe competitive interactions within a two-dimensional array
of model V4 concentric cells is
2
dEn 100 P+
E = En + 2 2
dt (10 + Hn) + P+
where P = SMarroquin 0.6n=k Ik exp ( Rnk5/5)
dIn
I = In + En
dt
dHn
H
dt
= Hn + gEn
(2)
The first equation here indicates that the excitatory firing rates
follow a Naka-Rushton31 function of the net postsynaptic input P
produced by stimulus SMarroquin, extracted from the Marroquin
pattern by the model V4 circuit (Fig. 5a) minus network inhibi-
tion. Naka-Rushton functions provide good fits to firing rates of
visual cortical neurons32,33. Each excitatory neuron En activates
an inhibitory neuron In, which in turn provides subtractive inhi-
bition to a range of neighboring E neurons (excluding En) using
the connectivity function of relative distance Rnk derived from
equation (1). Finally, adapting effects of the AHP current are sim-
ulated by an increase in the semi-saturation constant of the Naka-
Rushton function mediated by Hn, a formulation that provides
a reasonable approximation to cortical neuron adaptation27. In
general, g = 2.9 (but it is assumed to be under neuromodulatory
control; see Discussion), and the time constants were assigned
the plausible values of E = 15 ms, I = 8 ms and H = 400 ms.
These constants reflect the faster spiking of I than of E neurons28,
and H falls within the range reported for AHP currents in human
cortical neurons30,34.
The temporal response from a model E neuron at position 1
of the Marroquin pattern shows the effects of spike-frequency
adaptation in the rapid drop of spike rate to about one-third of
the initial transient rate (Fig. 6a). As this neuron continues to
fatigue further during its firing plateau, the effects of time-varying
inhibitory inputs become evident. Ultimately, inhibition termi-
nates spiking in this neuron, and neighboring neurons become
active. Repeated simulations of the two-minute duration of our
experiments with slightly different initial conditions produced
the model responses summarized in Table 1. Similar to FW and
HRW, the model predicted a visibility at position 1 of 0.62 and a
mean visibility duration of 2.24 1.93 s. Durations for both the
model and human subjects were plotted in a histogram and fit-
nature neuroscience volume 3 no 2 february 2000
 2000 Nature America Inc. http://neurosci.nature.com

articles

a b

Relative visibility
1.8 circle
0.9 circle
lines
2000 Nature America Inc. http://neurosci.nature.com

Mask displacement

Fig. 3. Masking of illusory circle visibility. (a) The masking stimulus comprised four circular contours superimposed on the Marroquin pattern. These mask-
ing circles were always spaced at equal distances from position 1, with distance being varied from experiment to experiment. As shown by mean data for
four subjects (b), the masking circles greatly reduced the relative visibility of illusory circles centered on position 1, with larger effects being produced by
larger diameter masking circles. Data for both circle sizes were well fit by the function in equation (1), with only amplitude A depending on size. For com-
parison, four vertical parallel lines with the same total length as the masking circle circumferences and spaced apart by a distance equal to the circle diame-
ters produced almost no noticeable masking. These data thus reflect a form of pattern-specific masking rather than simple contour interactions per se.

ted with a gamma distribution function (Fig. 6b and c). The radial gratings than by conventional sinusoidal gratings,
model and subjects both show comparably good fits to the gamma although these studies did not test for the possibility of com-
distribution (parameters summarized in Table 1). To illustrate the petitive interactions.
overall spatial percept implied by the model, neural responses The complex oscillations in this regional winner-take-all net-
30 seconds into one simulation were convolved with the activating work are driven by spike-frequency adaptation in excitatory neu-
input extracted from the Marroquin pattern, which produced a rons, which is known to be caused by AHP currents 2830 .
circular configuration resembling the percept (Fig. 4b). Physiological studies of excitatory neurons from human neo-
As the model responses were sufficiently complex to suggest cortex have shown that AHP currents are modulated by hista-
chaos, further simulations were conducted to compute the max- mine, acetylcholine, norepinephrine and serotonin, all of which
imum Lyapunov exponent using standard techniques27. The reduce the magnitude of spike-frequency adaptation34. Effects of
extreme sensitivity to initial conditions of a chaotic system is such neuromodulation in the model V4 competitive network can
manifested by a positive . We found that = 0.10/ms, so the be incorporated simply by reducing the parameter g in equation
neural model is not chaotic but rather represents a highly com- (2). Simulations have shown that a 14% reduction to g = 2.5 caus-
plex, asymptotically stable limit cycle oscillation. Rather than es the model to produce a 0.92 visibility as demonstrated by sub-
representing a simple alternation between two states as in binoc- ject EL (Table 1). This provides a plausible explanation for
ular rivalry, this oscillation cycles through a very large ensemble individual differences in our study.
of states before repeating. As discussed elsewhere8,9, the filterrectifyfilter sequence at
the earlier stages of the model is incorporated into many models
DISCUSSION for texture segregation. The unique aspect of our model is the
The foregoing analysis demonstrates that the Marroquin illu- final concentric, linear summation stage, which was derived from
sion can be explained quantitatively by inhibitory interactions psychophysics8. To date, proponents of other approaches to texture
giving rise to a regional winner-take-all
computation. Input to this network is Table 1: Summary of data for four subjects and V4 model simulation.
provided by model units that explain
Subject Visibility Duration (s) Gamma n Gamma (1/s)
the perception of structure in circular
FW 0.57 2.02 2.57 1.59 2.45
Glass patterns 8 . As these units have
properties similar to a subset of primate HRW 0.62 3.56 2.92 2.01 1.00
V4 neurons1012, it is hypothesized that BK 0.77 6.33 4.24 1.33 0.46
the competitive neural network devel- EL 0.91 32.67 46.78 n/a n/a
oped here reflects human V4 activity8,9. V4 Model 0.62 2.24 1.93 1.72 1.56
Results from fMRI (James et al., Invest.
Mean visibilities and mean durations are reported for position 1 (center of pattern). The two gamma para-
Ophthalmol. Vis. Sci. 40, S819, Abstr. meters are from least mean squares fits of K (t)n exp (t) to histograms of the duration data (see Fig.
4313, 1999) and event-related poten- 6). No gamma parameters could be calculated for EL because of the very long durations and consequent
tials15 indicate that human V4v is more small number of duration data. Model results fall within the data range for the first three subjects, and
agreement with EL can be obtained by a small shift in one model parameter (see text).
strongly activated by concentric and

nature neuroscience volume 3 no 2 february 2000 173

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

articles

Fig. 4. Pseudocolor representa-

tion of model V4 concentric unit8
responses to the Marroquin pat- a b
tern. (a) From maximum to zero,
responses are plotted in white,
yellow, red, dark red and black. 10
Model responses at the three high 8
points (positions 1, 7 and 10) and
two low points (positions 5 and 8) 1
of the human data are marked.
Relative model responses at all 7
ten positions agree well with 5
mean human data (Fig. 2c).
(b) One frame of the neural net-
work simulation in equation (2).
The small cross-shaped patches
are active local islands of E neu-
rons that have suppressed all
other neural activity in their vicin-
ity. Illusory circles have been
2000 Nature America Inc. http://neurosci.nature.com

added by convolving the network

activity pattern with a circle representing the effective input to these neurons from the Marroquin patterns. The circles thus represent the percept sig-
naled by network activity. (An animation of V4 network dynamics may be viewed on the web at http://neurosci.nature.com/web_specials/)

grouping have not attempted to explain the Marroquin illusion. It
is not clear how such a model might operate, but it will be inter-
esting to see whether alternative approaches can provide a com-
parable fit to our psychophysical data. In any event, it will still be
necessary to incorporate regional inhibitory interactions and adap-
tation to account for the observed perceptual oscillations.
Both the data and model reported here show similarities to
binocular rivalry. The Marroquin illusion and rivalry both gen-
erate visibility intervals well approximated by a gamma distri-
bution 23,24,35 . In addition, rivalry has been modeled by
competitive interactions plus fatigue36, and there is evidence
that rivalry is not chaotic37,38. Finally, it has been shown that
disappearances such as those of the Marroquin circles are sim-
ilar to binocular rivalry and are not caused by eye movements39.
Modifications of the current model may therefore be useful in
explaining the patchwork percept and spreading waves in
binocular rivalry.
There have been several theoretical proposals that spatial atten-
tion may reflect the activity of winner-take-all networks40,41. Fur-
thermore, studies of primate V4 demonstrate prominent
attentional effects1821,4244. This work has led to the proposal that
selective attention results from biasing of competitive interactions
in parallel networks22. It is therefore reasonable to hypothesize that
the parallel competitive network embodied in equation (2) reflects
one substrate for selective attention operating among V4 units. If
so, a novel form of network biasing suggests itself: localized reduc-
tion of spike-frequency adaptation controlled by modulatory neu-
rotransmitters. Changes in the magnitude of spike-frequency

3D model Model cross section Fig. 5. Schematic of the complete V4 concen-

tric unit dynamical model illustrated both in
perspective (left) and in cross-section (right)
for clarity. V1 filtering was performed at 12
different simple cell orientations spaced at
15 intervals. The filtered response at each
orientation was then full-wave rectified and
filtered by a larger second-stage oriented fil-
ter (hypothesized to reside in V2) at an
orthogonal orientation, a sequence of opera-
tions that extracts curvature information27,28.
Finally, responses of 24 concentrically
arranged second-stage filters are summed lin-
early (S in the model cross section) to pro-
duce the input to model V4 concentric units.
Further details of this feedforward model for
concentric V4 units were reported previ-
ously8,9. At the V4 stage, these units engage in
competitive lateral inhibition (lines with solid
circles at the top of the diagram) over spatial
regions defined by the masking function in
equation (1). Given the dimensions of the
connectivity function, each neuron inhibited a
circular region comprising approximately 12%
(at 1/e level) of the 64 64 network. There
was no inhibition between a unit and itself or
its two nearest neighbors in each direction.

174 nature neuroscience volume 3 no 2 february 2000

 2000 Nature America Inc. http://neurosci.nature.com
Fig. 6. Comparison of model net
work responses and human data. a
(a) Temporal response of the model E
neuron in position 1 (center of the
network) is plotted. Following a tran-
Spike rate
sient overshoot when the neuron
escapes from inhibition and switches
on, the rapid drop of the spike rate to
about 1/3 of the initial peak value
results from spike-frequency adapta-
tion mediated by H in equation (2).
Small fluctuations in firing level during
the subsequent plateau result from
time-varying inhibition arising from
oscillations elsewhere in the network.
Ultimately, competition and spike-fre-
quency adaptation cause the neuron b
to cease firing. Histograms of visibility
durations at position 1 are shown for
both subject FW (b) and the V4 model
2000 Nature America Inc. http://neurosci.nature.com
(c). Both are well fit by a distribution
Interval count
(parameters in Table 1).
adaptation would have a small
effect on the initial transient
response of cortical neurons, but
would significantly enhance the
subsequent asymptotic firing level; Interval duration (s)
this time course of adaptational
effects is observed in primate V4
neurons21. Aspects of selective attention in visual areas V1 and V2
might be explained by incorporating recurrent connections from
the V4 model to these areas, thus generating a competitive atten-
tional hierarchy41. Delayed figureground enhancements observed adaptation, the dots were counterphase flickered between black and
in V1 might also be explained by such feedback to V1 from V4
concentric units45.
Despite their intrinsic fascination, illusions attain their pri-
mary significance as windows for glimpsing underlying visual
processes. From this perspective, the Marroquin illusion may be
viewed as the result of presenting a dense set of competing stim-
uli to an attentional network. Similarly, the oscillation triggered
by this set of stimuli may be construed as a form of neural search
sequence incorporating inhibition of return46,47 mediated by spike-
frequency adaptation. We have presented evidence that model V4
concentric units can detect ellipsoidal objects in complex scenes
(Fig. 1b) and that such units interact via spatially regional win-
ner-take-all competition. Given evidence for object-based atten-
17
tion , this raises the exciting possibility that further study of V4
concentric units may help to elucidate the physiological bases of
such attentional processes.
METHODS
Psychophysics. Marroquin patterns were generated by superimposing
three square dot grids rotated by 60 with respect to each other. At the
viewing distance of 1.5 m, each dot subtended 40.6 arc s, and the spac-
ing between the dots in the square grids was 10.8 arc min. Overall, the
Marroquin pattern subtended 8.12 8.12. The dots were at 100% con-
trast and superimposed on a background of mean luminance 50.5 cd
per m2. In each experiment, three collinear dots were colored red: a cen-
tral one that was fixated by the subject and two others located on the
diameter of an illusory circle (see line in Fig. 1a) that the subject was
instructed to monitor. The subjects task was to fixate the center red dot
and depress the mouse button whenever a concentric, illusory circle was
perceived to pass through the two flanking red dots. By recording the
times at which the mouse button was depressed and released, a time
nature neuroscience volume 3 no 2 february 2000
articles
Transient
Inhibition
Time (s)
FW Model
Gamma Gamma
n = 103
n = 478
Interval count
Interval duration (s)
series of visibility durations was obtained. Each experiment was cen-
tered at one of ten points in the Marroquin pattern, and each experi-
ment lasted two min. To minimize fading of dots due to retinal
white at 1.5 Hz, which made the task considerably easier for subjects.
Data were collected at all fixation points on each of three days, with
points measured in pseudo-random order each day.
Modeling. Simulations of equation (2) were conducted in MatLab
software on a Macintosh G3 computer using a Runge-Kutta routine
with constant step size. Before solving the differential equations, we
applied the concentric V4 unit model8 consisting of oriented V1 fil-
tering, full-wave rectification, orthogonal V2 filtering and final con-
centric linear pooling to a 512 512 portion of the Marroquin pattern.
To reduce the size of the simulation, this filtered input to V4 concen-
tric units was subsampled down to 64 64. This was carefully done
in order to retain the symmetries in the V4 processed image
(see Fig. 4a). Equation (2) therefore represents a system of 3 642 or
12,288 coupled nonlinear differential equations. The first-stage filters
were derived from earlier masking studies with parameters for 16.0
cycles per degree48. The second-stage filters were scaled for the first-
stage filters (16.0 cycles per degree) from values previously reported8.
FFTs were used to evaluate the summation in the definition of p in
equation (2), so boundary conditions were periodic. As p represents
the postsynaptic input to excitatory neurons, which are described by
typical spike rate equations27, p provides an input to the Naka-Rush-
ton function31 that is positive or zero: p+ = max (p, 0). This thresh-
olding ensured that negative postsynaptic potentials produced a zero
firing rate. The maximum rate (100) and semi-saturation (10) of the
Naka-Rushton function were chosen to produce oscillations in the net-
work given the strength of the filtered V4 model inputs. The equation
for the I neurons is linear for mathematical convenience; an I equa-
tion with a Naka-Rushton rate nonlinearity and suitable parameters
produces very similar results. Because of the wide range of time con-
stants (8400 ms), these are stiff equations, so a series of preliminary
simulations were run to determine an appropriate time step for the
simulations, which turned out to be 0.05 ms. The maximum Lyapunov
175
 2000 Nature America Inc. http://neurosci.nature.com
articles
exponent was computed by comparing the main trajectory at each
time step with a second that started the interval displaced a distance
of 106 in the state space of the system. This was repeated throughout
the entire 2.0 min of each simulation, and the maximum Lyapunov
exponent was then estimated using standard formulas27.
Note: An animation of V4 network dynamics can be found on the Nature
Neuroscience web site (http://neurosci.nature.com/web_specials/).
ACKNOWLEDGEMENTS
This research was supported in part by NIH grant #EY02158 to H.R.W., by a
grant from Research to Prevent Blindness to the University of Chicago and by
NSERC grant #OGP0007551 (Canada) to F.W.
RECEIVED 23 AUGUST; ACCEPTED 3 DECEMBER 1999
1. Barrett, C. Op Art (Studio Vista, London, 1970).
2. Wade, N. J. Op art and visual perception. Perception 7, 2146 (1977).
3. Marroquin, J. L. Human Visual Perception of Structure. Masters thesis, MIT
2000 Nature America Inc. http://neurosci.nature.com
(1976).
4. Marr, D. Vision: A Computational Investigation into the Human
Representation and Processing of Visual Information 4950 (Freeman, San
Francisco, 1982).
5. Stevens, K. A. & Brookes, A. Detecting structure by symbolic constructions
on tokens. Comput. Graph. Image Process. 37, 238260 (1987).
6. Glass, L. Moir effect from random dots. Nature 223, 578580 (1969).
7. Glass, L. & Prez, R. Perception of random dot interference patterns. Nature
246, 360362 (1973).
8. Wilson, H. R., Wilkinson, F. & Asaad, W. Concentric orientation summation
in human form vision. Vision Res. 37, 23252330 (1997).
9. Wilson, H. R. & Wilkinson, F. Detection of global structure in Glass patterns:
implications for form vision. Vision Res. 38, 29332947 (1998).
10. Gallant, J. L., Braun, J. & Van Essen, D. C. Selectivity for polar, hyperbolic,
and Cartesian gratings in macaque visual cortex. Science 259, 100103
(1993).
11. Gallant, J. L., Connor, C. E., Rakshit, S., Lewis, J. W. & Van Essen, D. C. Neural
responses to polar, hyperbolic, and Cartesian gratings in area V4 of the macaque
monkey. J. Neurophysiol. 76, 27182739 (1996).
12. Kobatake, E. & Tanaka, K. Neuronal selectivities to complex object features in
the ventral visual pathway of the macaque cerebral cortex. J. Neurophysiol. 71,
856867 (1994).
13. Mishkin, M., Ungerleider, L. G. & Macko, K. A. Object vision and spatial vision:
two cortical pathways. Trends Neurosci. 6, 414417 (1983).
14. Van Essen, D. C., Anderson, C. H. & Felleman, D. J. Information processing in
the primate visual system: an integrated systems perspective. Science 255,
419423 (1992).
15. Allison, T., Puce, A., Spencer, D. D. & McCarthy, G. Electrophysiological studies
of human face perception. I: Potentials generated in ocippitotemporal cortex by
face and non-face stimuli. Cereb. Cortex 9, 415430 (1999).
16. Wilson, H. R., Wilkinson, F., Lin, L. M. & Castillo, M. Perception of head
orientation. Vision Res. (in press).
17. OCraven, K. M., Downing, P. E. & Kanwisher, N. fMRI evidence for objects as
the units of attentional selection. Nature 401, 584587 (1999).
18. Motter, B. C. Neural correlates of feature selective memory and pop-out in
extrastriate area V4. J. Neurosci. 14, 21902199 (1994).
19. Connor, C. E., Preddie, D. C., Gallant, J. L. & Van Essen, D. C. Spatial attention
effects in macaque area V4. J. Neurosci. 17, 32013214 (1997).
20. 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, 2442 (1997).
176
21. McAdams, C. J. & Maunsell, J. H. R. Effects of attention on orientation tuning
functions of single neurons in macaque cortical area V4. J. Neurosci. 19, 431441
(1999).
22. Desimone, R. & Duncan, J. Neural mechanisms of selective visual attention.
Annu. Rev. Neurosci. 18, 193222 (1995).
23. Fox, R. & Herrmann, J. Stochastic properties of binocular rivalry alternations.
Percept. Psychophys. 2, 432436 (1967).
24. Blake, R. A neural theory of binocular rivalry. Psychol. Rev. 96, 145167 (1989).
25. Dobbins, A., Zucker, S. W. & Cynader, M. S. Endstopped neurons in the visual
cortex as a substrate for calculating curvature. Nature 329, 438441 (1987).
26. Wilson, H. R. & Richards, W. A. Curvature and separation discrimination at
texture boundaries. J. Opt. Soc. Am. A 9, 16531662 (1992).
27. Wilson, H. R. Spikes, Decisions, and Actions: Dynamical Foundations of
Neuroscience (Oxford Univ. Press, Oxford, 1999).
28. Connors, B. W. & Gutnick, M. J. Intrinsic firing patterns of diverse
neocortical neurons. Trends Neurosci. 13, 99104 (1990).
29. Foehring, R. C., Lorenzon, N. M., Herron, P. & Wilson, C. J. Correlation of
physiologically and morphologically identified neuronal types in human
association cortex in vitro. J. Neurophysiol. 66, 18251837 (1991).
30. Lorenzon, N. M. & Foehring, R. C. Relationship between repetitive firing and
afterhyperpolarizations in human neocortical neurons. J. Neurophysiol. 67,
350363 (1992).
31. Naka, K. I. & Rushton, W. A. S-potentials from colour units in the retina of
fish. J. Physiol. (Lond.) 185, 584599 (1966).
32. Albrecht, D. G. & Hamilton, D. B. Striate cortex of monkey and cat: contrast
response function. J. Neurophysiol. 48, 217237 (1982).
33. Sclar, G., Maunsell, J. H. R. & Lennie, P. Coding of image contrast in central
visual pathways of the macaque monkey. Vision Res. 30, 110 (1990).
34. McCormick, D. A. & Williamson, A. Convergence and divergence of
neurotransmitter action in human cerebral cortex. Proc. Natl. Acad. Sci. USA
86, 80988102 (1989).
35. Borsellino, A., De Marco, A., Allazetta, A., Rinesi, S. & Bartolini, B. Reversal
time distribution in the perception of visual ambiguous stimuli. Kybernetik
10, 139144 (1972).
36. Lehky, S. An astable multivibrator model of binocular rivalry. Perception 17,
215228 (1988).
37. Richards, W., Wilson, H. R. & Sommer, M. A. Chaos in percepts? Biol.
Cybern. 70, 345349 (1994).
38. Lehky, S. R. Binocular rivalry is not chaotic. Proc. R. Soc. Lond. B Biol. Sci.
259, 7176 (1995).
39. Wade, N. J. Distortions and disappearances of geometrical patterns.
Perception 6, 407433 (1977).
40. Koch, C. & Ullman, S. Shifts in selective visual attention: towards the
underlying neural circuitry. Hum. Neurobiol. 4, 219227 (1985).
41. Tsotsos, J. K. Analyzing vision at the complexity level. Behav. Brain Sci. 13,
423469 (1990).
42. Motter, B. C. Focal attention produces spatially selective processing in visual
cortical areas V1, V2, and V4 in the presence of competing stimuli.
J. Neurophysiol. 70, 909919 (1993).
43. De Weerd, P., Peralta, M. R., Desimone, R. & Ungerleider, L. G. Loss of
attentional stimulus selection after extrastriate cortical lesions in macaques.
Nat. Neurosci. 2, 753758 (1999).
44. Reynolds, J. H., Chelazzi, L. & Desimone, R. Competitive mechanisms
subserve attention in macaque areas V2 and V4. J. Neurosci. 19, 17361753
(1999).
45. Zipser, K., Lamme, V. A. F. & Schiller, P. H. Contextual modulation in
primary visual cortex. J. Neurosci. 16, 73767389 (1996).
46. Posner, M. I., Cohen, Y., Choate, L. S., Hockey, R. & Maylor, E. in Preparatory
States and Processes (eds. Kornblum, S. & Requin, J.) 4965 (Erlbaum,
Hillsdale, New Jersey, 1984).
47. Maylor, E. A. & Hockey, R. J. Inhibitory component of externally controlled
covert orienting in visual space. J. Exp. Psychol. Hum. Percept. Perform. 11,
777787 (1985).
48. Wilson, H. R. in Spatial Vision (ed. Regan, D.) 6486 (MacMillan, London,
1991).
nature neuroscience volume 3 no 2 february 2000
 2000 Nature America Inc. http://neurosci.nature.com

articles

Motion perception during saccadic

eye movements
Eric Castet and Guillaume S. Masson

Centre de Recherche en Neurosciences Cognitives (CRNC), UPR 9012 du CNRS31, chemin J. Aiguier13402, Marseille cedex 20, France
Correspondence should be addressed to E.C. (castet@lnf.cnrs-mrs.fr)

During rapid eye movements, motion of the stationary world is generally not perceived despite
displacement of the whole image on the retina. Here we report that during saccades, human
observers sensed visual motion of patterns with low spatial frequency. The effect was greatest when
the stimulus was spatiotemporally optimal for motion detection by the magnocellular pathway.
Adaptation experiments demonstrated dependence of this intrasaccadic motion percept on
2000 Nature America Inc. http://neurosci.nature.com

activation of direction-selective mechanisms. Even two-dimensional complex motion percepts

requiring spatial integration of early motion signals were observed during saccades. These results
indicate that the magnocellular pathway functions during saccades, and that only spatiotemporal
limitations of visual motion perception are important in suppressing awareness of intrasaccadic
motion signals.

During saccadic eye movements, we do not perceive the world
as moving. This observation is paradoxical because high-speed
retinal motion can be detected with static eyes, provided it is pro-
duced by patterns with low spatial frequency1. Therefore, it is
widely assumed that motion perception is switched off during
saccades. Based on psychophysical results reporting decreased
intrasaccadic sensitivity (for instance, to contrast), two radically
different theories account for the absence of conscious motion
perception during saccades. The first approach postulates that
extraretinal signals triggered by saccades actively suppress
intrasaccadic motion processing2,3. Alternatively, some researchers
propose that visual factors alone might be involved. For exam-
ple, intrasaccadic information can be effectively masked by pre-
and postsaccadic visual activations4,5.
The general issue of intrasaccadic motion processing remains
unclear. Notably, proposals concerning the physiological imple-
mentation of extraretinal suppression raise a number of prob-
lems. The visual system comprises two relatively independent
pathways from the retina to the cortex. One pathway, which
extends from the magnocellular subdivision in the lateral genic-
ulate nucleus (LGN) to the parietal cortex, is thought to be
important for assessing motion6,7. This magnocellular path-
way contains neurons responding optimally to gratings of low
spatial and high temporal frequencies8,9. It is proposed, based
on psychophysical results, that the magnocellular pathway is
inactivated during the saccade by selective suppression in the
LGN 10,11. However, no physiological evidence supports this
claim. Moreover, neurons in the middle temporal area (MT) of
awake monkeys can be transiently activated by the visual flow
created by fixational saccades12, suggesting that precortical sup-
pression of the motion pathway is unlikely. Furthermore, such
suppression would be surprising, as there is reason to believe
that the essential functions of the magnocellular stream are still
needed during saccades, whether or not they contribute to con-
scious awareness13. Therefore, it was important to investigate
how much intrasaccadic information is available in the mag-
nocellular stream.
The ability to detect a two-frame displacement briefly pre-
sented during the saccade is well studied14,15. It is unclear if dis-
placement perception involves activation of low-level motion
detectors within the magnocellular pathway or a different process,
like attentional tracking of a moving objects successive posi-
tions16,17. Attempts to assess intrasaccadic motion perception
using stimuli more specifically linked to early cortical motion
processing are scarce. Intrasaccadic retinal stimuli used in most
studies do not optimally activate magnocellular motion detec-
tors. As a result, human capacity for intrasaccadic motion pro-
cessing is probably largely underestimated. In brief, visual
functions that might be carried out by the magnocellular stream
during saccades, including low-level motion processing, are cur-
rently poorly understood.
We present direct evidence for intrasaccadic motion percep-
tion mediated by low-level motion signals transmitted through
the magnocellular pathway. Our experiments pinpoint the rele-
vant visual factors that must be optimized to produce a clear,
conscious sensation of visual motion during saccades.
RESULTS
Motion detection
The basic stimulus consisted of a vertical grating with low spatial
frequency (0.17 cycles per degree) moving at a constant high speed
(360 per s). With static eyes, the low-contrast grating (10%) was
above the critical fusion frequency and was therefore invisible18.
While the grating was continuously drifting, observers were
required to make horizontal saccades between two fixation points
whose spatial separation was varied across trials to alter eye veloc-
ity. The eye speed was concurrently measured in three observers
who had to report whether or not grating motion was perceived
during the saccade (yes/no task). Whenever saccades of moder-
ate amplitudes (6) were made in the direction of the grating, a
compelling perception of bars drifting in the direction of the grat-
ing occurred during the saccade. Mean velocity profiles of sac-
cades of different amplitudes for one observer were plotted
(Fig. 1a). The highest speed reached during a saccade increased

nature neuroscience volume 3 no 2 february 2000 177

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

articles

b
a b
second)

per second)
(/s)
per

(degrees(/s)
velocity
velocity (degrees

Peak velocity
Horizontal

Peak velocity
Horizontal

Saccade
Saccade amplitude
amplitude ()
(degrees)
Time (ms)
Time (ms)

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

of perceiving
perceiving motion
Probability of
motion
Probability

Peak velocity
Peak velocity (/s)
(degrees per second)

Fig. 1. Motion perception during a saccade in the same direction as a high-speed vertical grating (10% contrast). (a) Eye speed during leftward sac-
cades (observer EC) as a function of time and saccade amplitude. Dashed line shows constant speed of a leftward-moving grating. For each saccade
amplitude (two indicated by arrows), the mean velocity profile was obtained from aligning onsets of individual saccades and averaging their velocity
profiles. (b) Dependence of peak velocity on saccade amplitude. For each combination of saccade amplitude and grating speed (solid symbols, 360
per s; open symbols, 300 per s), we measured the mean saccadic peak velocity by averaging individual peak velocities. (c) Probability of perceiving
motion as a function of saccadic peak velocity for three observers. Gratings moving at either 360 per s (filled symbols) or 300 per s (open symbols)
were presented in successive blocks. Error bars to right show average s.e. For all observers, the two inverted U-shaped curves were laterally shifted
by 60 per s (horizontal arrows).

with saccade amplitude19,20 (Fig. 1b). The probability of perceiv-
ing motion depended on this saccadic peak velocity (Fig. 1c). With
gratings either at 360 per s (solid symbols) or 300 per s (open
symbols), the data curves resembled an inverted U and were lat-
erally shifted by a peak-velocity difference of about 60 per s (hor-
izontal arrows). This value corresponded to the difference between
the two grating speeds. These results strongly suggest that motion
perception depends on the difference between the peak eye veloc-
ity and grating speed; the relevant factor is the speed (or tempo-
ral frequency) of the grating relative to the retina. This is best seen
when data from Fig. 1c are replotted as a function of the retinal
temporal frequency at the peak of the saccade (Fig. 2a). Here the
two curves are superimposed, peaking between 10 and 25 Hz,
depending on the observer.
The phenomenon can be understood if we look at the effec-
tive spatiotemporal stimulus at the retinal level. The retinal speed
(and temporal frequency) of the grating was high at the begin-
ning of the saccade and decreased until the peak was reached.
After the peak, retinal speed increased again (Fig. 1a). We pro-
pose that this constantly changing retinal temporal frequency
partly explains the usual impairment of intrasaccadic motion
processing. This derives from two fundamental spatiotemporal
characteristics of motion detection. First, individual low-level
motion detectors only respond to a restricted range of spa-
tiotemporal frequencies2123, and second, they need a minimum
amount of time to integrate motion energy24. Thus, for any spa-
tial-frequency channel, the temporal frequency at the retinal
level changes too rapidly to provide a reliable signal, at least dur-
ing the early acceleration and late deceleration phases of the
velocity profile. However, within a brief period during which
the velocity peaks, retinal temporal frequency is relatively con-
stant. It is during this critical period that motion detectors could
be effectively activated, provided the spatiotemporal combina-
tion were optimal. This hypothesis is fully consistent with our
motion effect. With small saccades, the temporal frequency of
the grating relative to the retina was too high even when the peak
velocity was reached (Fig. 2a). For instance, with a 2 saccade,
the retinal temporal frequency at the peak was about 40 Hz (240
per s), much too close to the fusion frequency to produce a clear
motion percept. With medium saccades, motion perception was

178 nature neuroscience volume 3 no 2 february 2000

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

articles

optimal, because these sac- a

cades yielded retinal tempo-

perceivingmotion
motion
ral frequencies around the

of
Probability of
peak that fell within the opti-

Probability
mal temporal range for

perceiving
motion detection 25 . With
large saccades (when the
peak velocity approaches the
grating speednull relative
speed), static visual channels
are optimally activated. Peak retinal temporal frequency (Hz)
Observers report that the
sense of motion is lost and
replaced by the perception of b
Fig. 2. Temporal tuning of motion perception. (a) Data in Fig.
a flashed grating. A previ-
1c replotted as a function of retinal temporal frequency at the
Response (impulses/sec)

ous study has shown that

(impulses/s)

peak of the saccade (circles). Positive and negative abscissa val-

stabilizing the retinal image ues denote retinal motion in the same direction as the saccade
of a rapidly moving grating
2000 Nature America Inc. http://neurosci.nature.com

and in the opposite direction, respectively. Triangles, data

with a saccade produces a obtained for observer EC during fixation. The stimulus was a
static flash percept26. How- 25-ms movie (4 frames) crudely approximating the different
Response

ever, observers did not
report motion perception,
presumably because the sin-
gle spatial frequency used
(1 cycle per degree) was too
Temporal frequency (Hz)
high. Beside making stimuli
non-optimal for the magno-
cellular pathway, high spatial
frequencies present another
problem: during a saccade, the range of retinal temporal fre-
quencies swept through over a given period increases propor-
tionally with spatial frequency. As a result, the restricted range of
temporal frequencies required for motion processing within a
single temporal-frequency channel is encountered for shorter
periods as spatial frequency is increased. Thus, this period is
approximately sixfold shorter in the previous study26 than in
ours, possibly accounting for the failure to perceive intrasac-
cadic motion with high spatial frequencies.
Processing intrasaccadic visual motion under our protocol
requires integrating different speeds across time, presumably
by some form of averaging, within a brief period of 25 ms.
We tested the ability to process this kind of visual stimulus with
static eyes. For each saccade amplitude previously tested, a 25-
ms stimulus with 3 different grating speeds was created to
approximate the variation in speeds experienced near the sac-
cadic peak velocity (see Methods). Again, subjects indicated
whether or not motion was perceived. The results were quali-
tatively similar to that obtained during saccades: for low con-
trasts, probability of perceiving motion as a function of peak
retinal temporal frequency was an inverted U-shaped curve.
Quantitative differences were demonstrated by observer EC
(triangles, Fig. 2a). The leftward shift of the curve probably
reflects the different perceptual criterion used for this task.
Contrast sensitivity was also higher in the static eye condi-
tion: with a 10% contrast grating (as in the intrasaccadic task),
high temporal frequencies were not above fusion frequency
and, therefore, produced motion perception (solid triangles).
By halving contrast, motion perception was decreased for these
high frequencies, yielding an inverted U-curve (open trian-
gles). It is impossible to determine whether this sensitivity dif-
ference results from extraretinal or visual factors, mainly
because the effective retinal stimulus differs between experi-
mental conditions.
retinal temporal frequencies experienced around the peak
velocity of the saccade. The grating contrast was either 10%
(solid triangles) or 5% (open triangles). (b) Temporal tuning
curve of a V1 magnocellular cell measured with a 0.9-cycle-per-
degree grating (replotted from ref. 8). This cell is direction
selective and projects to area MT.
We showed that a saccade toward a high-speed grating of low
spatial frequency could produce a clear motion percept: the grat-
ing briefly appeared to move in the direction of the saccade. This
percept was optimal when the retinal stimulus was within the range
for motion detection during the period including peak velocity.
The optimal retinal temporal frequencies for motion detection
(1025 Hz; Fig. 2b) were consistent with the broad and high-cutoff
tuning curves of direction-selective cells in primate striate cortex8,9.
Direction discrimination
Possible inconsistency in criteria for motion percepts across
observers and runs may be a weakness of the previous study. This
may explain why peak values of the inverted U-curves differed
for different observers (Fig. 2a). To overcome this, we also mea-
sured thresholds for direction discrimination. Because direction
selectivity is an essential property of early motion detectors, a
direction discrimination task further probed the involvement of
motion pathways in intrasaccadic motion perception.
We predicted that perceived direction would oppose the direc-
tion of the saccade when the peak of the velocity profile exceed-
ed the grating speed for a sufficiently long duration. To test this,
a grating whose speed (or equivalent temporal frequency) was
controlled by an adaptive-staircase procedure was presented on
each trial as before. Observers made large saccades of constant
amplitude (14) and reported whether perceived motion was in
the same direction as the saccade (forward response) or its oppo-
site (backward response). Using 2 staircases, we could assess the
grating speeds producing forward responses with probabilities
of 0.29 and 0.71. Backward percepts appeared less vivid than for-
ward percepts, presumably because of concurrent activation of
static channels during the two brief periods when eye and grating
speeds matched. The point of subjective stationarity (PSS) was
determined by tracking the speed of a grating that produced a
static percept (equiprobable forward and backward responses).

nature neuroscience volume 3 no 2 february 2000 179

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

articles

Threshold (degrees per second)

PSS = 355/s PSS = 358/s PSS = 347/s

Direction-specific adaptation
Although the spatiotemporal selectivity of
these motion effects suggests the involvement
of early motion detectors, one could argue
that the percept is merely due to static mech-
anisms recruited to detect a temporal change
of position16,17. To rule out this possibility, we
used a classic protocol for direction-specific
adaptation27. A vertical high-contrast adapt-
Forward

Backward

Forward

Backward

Forward

Backward
ing grating (12 Hz, 0.17 cycles per degree, 72
Forward

Forward
Forward
Backward

Backward
Backward
per s) covering the whole screen was present-
ed while observers stared at a fixation point.
Its temporal frequency was chosen to opti-
mally activate motion detectors25. During the
Fig. 3. Direction discrimination experiment. Thresholds for perceiving motion either in the
direction of the saccade (forward) or in the opposite direction (backward). Forward and back-
test period, a vertical grating (50 Hz, 0.17
ward thresholds respectively correspond to grating speeds larger and smaller than the average cycles per degree, 300 per s) of variable con-
saccadic peak velocity. trast presented in the top or bottom half of
the screen was moved either in or opposite to
2000 Nature America Inc. http://neurosci.nature.com

the adapting direction. Observers were

Thresholds for all observers lay within a 2535 per s range
(46 Hz): forward differential thresholds correspond to the dif-
ference between 0.71 and 0.50 levels, and backward thresholds
correspond to the 0.500.29 difference (Fig. 3).
For comparison, we also measured these direction-discrimi-
nation thresholds during fixation. The stimulus was a 25-ms movie
of 4 frames simulating 3 successive speeds around the peak of a
14 saccade. We varied the average speed from a few degrees per
second to the left to a few degrees per second to the right. Mea-
sured thresholds (observers E.C., G.S.M., data not shown) were
four- to sixfold lower than those during saccades (Fig. 3). Howev-
er, it is impossible to assess whether this decrease was related to
extraretinal signals or to visual factors alone, especially as retinal
stimuli differed between conditions. One clear difference was that,
because of the large saccade, the effective stimulus width for the
intrasaccadic condition was narrower (about 12) than the full
screen available during the fixation condition (27). As a result,
spatial integration mechanisms could not be as efficient as during
fixation. Moreover, the screen edges sweeping over peripheral reti-
na may produce lateral masking. Intrasaccadic thresholds may also
be inflated by variability in peak saccade speed across trials, as we
used average, rather than trial-by-trial, eye speed. Therefore, inves-
tigating the relative influence of visual versus
extraretinal factors on direction discrimina-
tion during saccades requires further research.
required to make saccades in the same direc-
tion as the test grating and to indicate the part of the screen in
which the intrasaccadic motion was perceived. The saccade
amplitude for each observer was chosen to optimize motion
perception (as assessed in the first experiment). Contrast
thresholds were dramatically increased when the adapting and
test gratings were in the same direction, whereas adaptation in
the opposite direction had only a slightly detrimental effect
(Fig. 4; comparison of across-observer means, t 2 = 8;
p = 0.015). This direction-specific threshold elevation is a clas-
sic signature of an early direction-sensitive mechanism27. In
primates, direction selectivity is a visual property that starts at
the level of the primary visual cortex28. Therefore, this finding
is at odds with suggestions that the activity of the magnocel-
lular pathway (constituting the dominant input to motion
detectors 6,7) is suppressed at a precortical level during sac-
cades10,11.
Two-dimensional motion perception
In the experiments above, the perceived direction of the
intrasaccadic percept always matched the axis of the saccade
itself. This coincidence raises the possibility that the percept
was actually determined by an extraretinal signal associated
threshold

Here, the relevant difference between fix-

Contrast threshold

ation and intrasaccadic thresholds concerned

elevation

the PSS. During saccades, the PSS (above each

elevation

graph in Fig. 3) was smaller than the average

Contrast

saccadic peak velocity (370 per s), so the cor-

responding retinal speed averaged over 25 ms
was close to zero (mean, 6 per s). This sug-
gests that stationarity is perceived when reti-
nal speeds within 2530 ms are of opposite
directions and thus approximately average to
zero, consistent with our measurement of the Opposite
PSS during fixation at nearly 0 per s. Same
In summary, the intrasaccadic ability to
Fig. 4. Direction-selective adaptation. Observers made horizontal saccades and detected
discriminate between forward and back- whether a horizontally moving test grating was located in the top or bottom half of the screen
ward directions supports our hypothesis that (spatial 2AFC). Contrast sensitivity was measured with or without adaptation to a moving grat-
direction signals relevant to the task are ing. The data clearly show direction-selective adaptation: contrast-threshold elevation (relative
indeed extracted during a short period in to the unadapted condition) was larger when the adapting grating was in the same rather than in
which eye velocity peaks. the opposite direction.

180 nature neuroscience volume 3 no 2 february 2000

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

articles

a b
2000 Nature America Inc. http://neurosci.nature.com

Fig. 5. Perceived direction of two-dimensional motion. (a) Stimuli were oblique (45) moving gratings (contrast 30%; temporal frequency as before)
presented within square (20 20) or rectangular (20 10) apertures. As before, observers made horizontal saccades of optimal amplitude and
then rotated an arrow in the middle of the screen to indicate the direction of perceived intrasaccadic motion. (b) Distributions of perceived direc-
tions in polar coordinates. The origin (0) of the angular axis represents the motion component perpendicular to the bars of the grating. The radial
axis plots, in log coordinates, the number of observations collected for each direction. Mean direction is given by . With a square aperture, per-
ceived directions cluster around the perpendicular component. With a rectangular aperture, perceived directions are parallel to the long axis (barber-
pole illusion).

with the saccade14. To avoid this confounding effect, we used DISCUSSION

the display from the first experiment but rotated the moving
grating by 45 and presented it within a large square aperture
(Fig. 5a, top). Observers made horizontal saccades of optimal
amplitudes, and then adjusted the direction of an arrow pre-
sented on the screen to report the perceived direction of the
intrasaccadic percept. The distribution of the perceived direc-
tions was centered around the grating direction (Fig. 5b, top),
as would be expected if motion detectors with the preferred
oblique orientation matching the grating orientation were acti-
vated in this condition.
Furthermore, motion perception of real complex objects
requires the spatial integration of different local motion sig-
nals that are not in the direction of the eye movements29,30. If
early motion signals are not suppressed during saccades, as
suggested by previous experiments, one could argue instead
that integration of these signals into a two-dimensional veloc-
ity signal is actively suppressed. A classic stimulus used to study
motion integration is the barber-pole illusion 31 : when an
oblique grating moves within a rectangular, rather than square,
aperture, its perceived motion is parallel to the long side of the
aperture. This phenomenon, which indicates the importance
of the motion signals elicited near aperture borders32,33, was
also tested. The height of the aperture used before was simply
halved to obtain a horizontal barber-pole (Fig. 5a, bottom).
All observers perceived the classic barber-pole illusionhere,
in a horizontal direction (Fig. 5b, bottom). Intrasaccadic
motion perception based on activation of low-level detectors
tuned to all possible directions followed by integration of the
resulting signals, presumably at a second stage29,34,35, is therefore
possible during saccades.
We show that making saccades in the direction of a rapidly mov-
ing grating produced clear motion percepts when retinal image
motion (averaged over a critical period around the saccadic peak
velocity) was within the motion-detection range. Observers could
distinguish between retinal motion in the direction of the sac-
cade and retinal motion in the opposite direction, and perceived
a static grating when retinal speed averaged over 2530 ms was
close to null. Adapting during fixation to a low spatial and high
temporal frequency grating produced a direction-specific reduc-
tion in contrast sensitivity to intrasaccadic motion perception.
Additionally, spatial integration of early motion signals leading to
two-dimensional motion perception was still observed during
saccades. We conclude that these intrasaccadic motion percepts
revealed the activation of direction-selective mechanisms tuned
to low spatial and high temporal frequencies within the magno-
cellular pathway.
We designed the stimulus to optimize intrasaccadic motion
perception. First, as the grating was invisible at the beginning
(and at the end) of each saccade, there was no pre- or post-sac-
cadic visual information that might have produced forward or
backward masking4,5,36,37. Moreover, in contrast to studies using
a two-frame sequence, the grating was not abruptly presented
during the saccade. Instead, it was continuously displayed on
each trial. This point was crucial: the gradual transitions between
a retinal temporal range above fusion frequency and an optimal
range for motion detection minimized the temporal energy
spread and the intrusion of masking transient signals. This opti-
mization allowed a study of motion detection per se rather than
displacement detection. The involvement of early motion pro-
cessing is unclear in numerous studies of intrasaccadic displace-

nature neuroscience volume 3 no 2 february 2000 181

 2000 Nature America Inc. http://neurosci.nature.com
articles
ment perception14,15. Two-frame presentations used in displace-
ment studies are weak stimuli for motion processing. As a con-
sequence, it is possible that performance in these studies might
rely exclusively on visual processes that detect changes in posi-
tion over time16,17.
In our experiments, intrasaccadic motion perception was
easily experienced when magnocellular motion detection was
optimized. In primates, the first stage of motion processing
takes place in the striate cortex (V1). The one-dimensional
edge-motion signals extracted in V1 are thought to be inte-
grated at a later stage into a two-dimensional velocity signal
within the middle temporal area34,35. There is ample evidence
that motion perception depends on the activity of MT cells3840.
In addition, MT cells of alert monkeys consistently respond to
directional signals induced by randomly directed small-ampli-
tude (about 0.8) saccades while monkeys fixate12. MT neurons
transiently increase their firing rates when these fixational sac-
cades induce a retinal flow in the neurons preferred direction,
2000 Nature America Inc. http://neurosci.nature.com
whereas neuronal responses are mildly suppressed when sac-
cades induce null retinal flow. In this framework, the spa-
tiotemporal characteristics of the effects in experiments 13
suggest the involvement of direction-selective V1 neurons8,9,
whereas motion integration observed in experiment 4 was prob-
ably related to MT responses.
Our findings suggest that visual factors are the essential
determinants of suppression of consciousness of motion dur-
ing saccades4,5,36,37. Clearly, extraretinal signals do not switch
off motion processing. Studies suggesting such suppression
actually measured the visibility of static gratings (0 Hz)10,11, and
therefore are not inconsistent with our study, which used spa-
tiotemporal stimuli optimized for motion detection by the mag-
nocellular pathway (with low spatial and high temporal
frequencies).
Our results also pertain to the issue of perceptual stability
during eye movements in relation to extraretinal signals14,15. In
our experiments, it was striking that a static flash percept
occurred when the eye peak velocity equaled the speed of the
moving grating. Such a static appearance was inconsistent with
the idea that extraretinal signals are used to compensate for the
effect of eye movements, as generally assumed with smooth eye
movements 4144. This hypothesis would predict high-speed
motion perception of the grating. Therefore, it seems that sac-
cadic eye movements invoke other mechanisms to achieve per-
ceptual stability 13,45 . Alternatively, it is possible that a
recalibration process due to extraretinal signals acts more slow-
ly than the actual saccadic eye movements46, or that extraretinal
signals carry a damped representation of eye position47.
Here we showed that the magnocellular pathway functions
during saccades. Indeed, suppression of the magnocellular-
dominated stream would have important adverse effects48. For
instance, the role of the parietal cortex in visual attention49
makes it well suited to alert and orient the visual system toward
a peripheral transient signal detected during the saccade. Given
the high frequency of saccades in normal vision, future research
should examine the behavioral importance of processing
intrasaccadic visual information.
Our findings do not rule out the possibility that an extrareti-
nal signal alters intrasaccadic visual motion processing. Howev-
er, we clearly showed that such a putative extraretinal signal does
not suppress motion perception during saccades. Therefore, it is
more parsimonious to interpret the usual absence of intrasac-
cadic motion perception as the result of visual and/or attention-
al factors alone.
182
METHODS
Psychophysics. Stimuli were displayed on a 21-inch Sony color monitor
(GDM-4011P) driven by a display controller (Cambridge Research Sys-
tem VSG 2/3F, Rochester, UK) with a 160-Hz frame rate. At a viewing
distance of 68 cm, the average horizontal separation between adjacent
pixels subtended 0.035 of visual angle. The screen subtended 27 21.
A lookup table in the software was used to linearize the intensity response
of the screen phosphors at an 8-bit luminance resolution.
Stimuli displayed during saccades. On each trial, a moving grating of
low spatial frequency (0.17 cycles per degree) was displayed for 2 s. Dur-
ing the first 500 ms, the contrast increased using a raised cosine func-
tion from 0% to the desired level, which was kept constant for 1000 ms.
The contrast then decayed to 0% using the same function during the last
500 ms. A low-pitch sound indicated that the contrast had reached the
desired level, prompting saccades. A high-pitch sound, presented when
the grating disappeared, prompted observers to respond. The temporal
frequency was either 50 Hz (300 per s) or 60 Hz (360 per s) except in the
second experiment (direction discrimination), where it was varied with
an adaptive procedure. The mean luminance of the grating was 22 cd per
m2 (4.5 cd per m2 in the second experiment). Observers AL and YR were
naive as to the purpose of the experiments.
Stimuli displayed with static eyes. A 25-ms movie was presented at 3 dif-
ferent speeds during fixation by displaying 4 frames of a grating whose
spatial phase was appropriately shifted. For each saccade amplitude (test-
ed in the first experiment), 3 retinal speeds centered on the saccadic peak
velocity were calculated from the saccadic velocity profiles over 25 ms.
For instance, for a 2 amplitude, the 3 successive retinal speeds of the
grating were S1, 270 per s (6 ms before the peak), S2, 240 per s (at the
peak) and S1 again (6 ms after the peak); average speed, 260 per s.
Measurement of eye movements. Horizontal movements of the right eye
were recorded using a high-resolution infrared scleral-reflectance system
(IRIS Skalar, Delft, Netherlands). Analog signals were digitized at 500
Hz with a 12-bit resolution ADC. A bite bar was used to restrain head
movements. Two points of different colors indicated the size and the
direction of the saccade to be made. On each trial, a sound instructed
observers to make the saccade. Data were analyzed off line. Inappropri-
ate saccades were discarded from the analysis.
Direction discrimination. Across trials of the same block, the direction of
the saccade (14 amplitude) was constant (leftward or rightward).
Observers were required to indicate whether perceived motion of the
grating was in the same direction as the saccade (forward) or in the oppo-
site direction (backward). The grating (20% contrast) covered the whole
screen and had its temporal frequency controlled by one of 3 staircases
with steps of 0.5 Hz50. One staircase (starting value, 360 per s; 60 Hz)
converged on a 0.50 threshold indicating the speed at which the grating
appeared static. The two other staircases tracked the 0.29 and 0.71 thresh-
olds. Their starting values, which were 300 per s (50 Hz) and 420 per s
(70 Hz) in the training period, were respectively increased and lowered as
performance improved across blocks. The actual starting values were
randomly chosen within a range of 2 Hz centered on the above values.
The end of a block occurred when a minimum of nine reversals had been
completed in each staircase. The three thresholds were then estimated
by averaging the six last reversals in each staircase. The same direction
differential threshold (DT) was defined as the difference between the
0.71 and 0.50 levels, and the opposite direction DT as the difference
between the 0.50 and 0.29 levels.
Direction-specific adaptation. Required saccade direction (leftward or
rightward, in the direction of the test grating moving at 300 per s) was
randomly alternated across trials. For each observer, a constant saccade
amplitude was chosen from results of the first experiment to optimize
motion perception (4 for EC and 3 for AL and YR). The contrast of the
test grating was adjusted with an adaptive staircase procedure converging
on a 0.71 threshold (steps of 0.4% contrast)50. Two staircases, each
assigned to a grating direction, were randomly interleaved within a block.
nature neuroscience volume 3 no 2 february 2000
 2000 Nature America Inc. http://neurosci.nature.com
Contrast thresholds were measured with or without adaptation to a mov-
ing grating (80% contrast), which covered the whole screen and had the
same spatial frequency as the test grating and a temporal frequency of
12 Hz (optimal for motion detection of low spatial frequencies25). The
adapting grating, whose direction was constant within any block, was
presented for one minute at the beginning of each block. Observers were
then exposed to cycles of test periods (during which one saccade had to
be made within 1500 ms) and 10-s adapting periods. Test gratings were
presented either in the upper or lower part of the screen (spatial 2AFC).
A block ended when a minimum of eight reversals had been completed in
each staircase. The two thresholds were then estimated by averaging the
last six reversals in each staircase.
Two-dimensional motion perception. For a given saccade direction (left-
ward or rightward), the orientation of the grating was randomized so
that the motion component perpendicular to the bars was either above or
below horizontal. The two different saccade directions were presented
in distinct blocks. The perceived directions plotted in Fig. 5b were aver-
aged over the four possible quadrants (two for each saccade direction).
The square and rectangular aperture conditions were randomly inter-
2000 Nature America Inc. http://neurosci.nature.com
leaved within each block.
ACKNOWLEDGEMENTS
We thank L. S. Stone, M. J. Morgan, J. K. ORegan and A. Riehle for comments
on the manuscript.
RECEIVED 19 AUGUST; ACCEPTED 8 DECEMBER 1999
1. Burr, D. C. & Ross, J. Contrast sensitivity at high velocities. Vision Res. 22,
479484 (1982).
2. Shioiri, S. & Cavanagh, P. Saccadic suppression of low-level motion. Vision
Res. 29, 915928 (1989).
3. Ilg, U. J. & Hoffmann, K. P. Motion perception during saccades. Vision Res.
33, 211220 (1993).
4. Judge, S. J., Wurtz, R. H. & Richmond, B. J. Vision during saccadic eye
movements. I. Visual interactions in striate cortex. J. Neurophysiol. 43,
11331155 (1980).
5. Campbell, F. W. & Wurtz, R. H. Saccadic omission: why we do not see a grey-
out during a saccadic eye movement. Vision Res. 18, 12971303 (1978).
6. Maunsell, J. H., Nealey, T. A. & DePriest, D. D. Magnocellular and
parvocellular contributions to responses in the middle temporal visual area
(MT) of the macaque monkey. J. Neurosci. 10, 33233334 (1990).
7. Merigan, W. H., Byrne, C. E. & Maunsell, J. H. Does primate motion
perception depend on the magnocellular pathway? J. Neurosci. 11, 34223429
(1991).
8. Movshon, J. A. & Newsome, W. T. Visual response properties of striate
cortical neurons projecting to area MT in macaque monkeys. J. Neurosci. 16,
77337741 (1996).
9. Hawken, M. J., Shapley, R. M. & Grosof, D. H. Temporal-frequency selectivity
in monkey visual cortex. Vis. Neurosci. 13, 477492 (1996).
10. Ross, J., Burr, D. & Morrone, C. Suppression of the magnocellular pathway
during saccades. Behav. Brain Res. 80, 18 (1996).
11. Burr, D. C., Morrone, M. C. & Ross, J. Selective suppression of the
magnocellular visual pathway during saccadic eye movements. Nature 371,
511513 (1994).
12. Bair, W. & OKeefe, L. P. The influence of fixational eye movements on the
response of neurons in area MT of the macaque. Vis. Neurosci. 15, 779786
(1998).
13. ORegan, J. K. Solving the real mysteries of visual perception: the world as an
outside memory. Can. J. Psychol. 46, 461488 (1992).
14. Bridgeman, B., Van der Heijden, A. H. C. & Velichkovsky, B. M. A theory of
visual stability across saccadic eye movements. Behav. Brain Sci. 17, 247292
(1994).
15. Li, W. & Matin, L. Saccadic suppression of displacement: separate influences of
saccade size and of target retinal eccentricity. Vision Res. 37, 17791797 (1997).
16. Braddick, O. J. Low-level and high-level processes in apparent motion.
Phil. Trans. R. Soc. Lond. B Biol.Sci. 290, 137151 (1980).
nature neuroscience volume 3 no 2 february 2000
articles
17. Cavanagh, P. Attention-based motion perception. Science 257, 15631565
(1992).
18. Robson, J. G. Spatial and temporal contrast sensitivity functions of the visual
system. J. Opt. Soc. Am. 56, 11411142 (1966).
19. Bahill, A. T., Clark, M. R. & Stark, L. The main sequence, a tool for studying
human eye movements. Math. Biosci. 24, 191204 (1975).
20. Collewijn, H., Erkelens, C. J. & Steinman, R. M. Binocular co-ordination of
human horizontal saccadic eye movements. J. Physiol. (Lond.) 404, 157182
(1988).
21. Morgan, M. J. Analogue models of motion perception. Phil. Trans. R. Soc.
Lond. B Biol. Sci. 290, 117135 (1980).
22. van Santen, J. P. & Sperling, G. Elaborated Reichardt detectors. J. Opt. Soc.
Am. A Opt. Image Sci. Vis. 2, 300321 (1985).
23. Watson, A. B. & Ahumada, A. J., Jr. Model of human visual-motion sensing.
J. Opt. Soc. Am. A Opt. Image Sci. Vis. 2, 322341 (1985).
24. Burr, D. C. Temporal summation of moving images by the human visual
system. Proc. R. Soc. Lond. B Biol. Sci. 211, 321339 (1981).
25. Anderson, S. J. & Burr, D. C. Spatial and temporal selectivity of the human
motion detection system. Vision Res. 25, 11471154 (1985).
26. Deubel, H., Elsner, T. & Hauske, G. Saccadic eye movements and the
detection of fast-moving gratings. Biol. Cybern. 57, 3745 (1987).
27. Levinson, E. & Sekuler, R. The independence of channels in human vision
selective for direction of movement. J. Physiol. (Lond.) 250, 347366 (1975).
28. Hubel, D. H. & Wiesel, T. N. Functional architecture of macaque monkey
visual cortex. Proc. R. Soc. Lond. B Biol. Sci. 198, 159 (1977).
29. Adelson, E. H. & Movshon, J. A. Phenomenal coherence of moving visual
patterns. Nature 300, 523525 (1982).
30. Yuille, A. L. & Grzywacz, N. M. A computational theory for the perception of
coherent visual motion. Nature 333, 7174 (1988).
31. Wuerger, S., Shapley, R. & Rubin, N. On the visually perceived direction of
motion by Hans Wallach: 60 years later. Perception 11, 13171367 (1996).
32. Kooi, F. L. Local direction of edge motion causes and abolishes the barber
pole illusion. Vision Res. 33, 23472351 (1993).
33. Castet, E., Charton, V. & Dufour, A. The extrinsic/intrinsic classification of
2D motion signals with barber-pole stimuli. Vision Res. 39, 915932 (1999).
34. Albright, T. D. & Stoner, G. R. Visual motion perception. Proc. Natl. Acad. Sci. USA
92, 24332440 (1995).
35. Movshon, A. J., Adelson, E. H., Gizzi, M. S. & Newsome, W. T. in Pattern
Recognition Mechanisms (eds. Chagas, C., Gattas, R. & Gross, C. G.) 117151
(Vatican Press, Rome, 1985).
36. Breitmeyer, B. G. & Ganz, L. Implications of sustained and transient channels
for theories of visual pattern masking, saccadic suppression, and information
processing. Psychol. Rev. 83, 136 (1976).
37. Volkmann, F. C. Human visual suppression. Vision Res. 26, 14011416
(1986).
38. Salzman, C. D., Britten, K. H. & Newsome, W. T. Cortical microstimulation
influences perceptual judgements of motion direction. Nature 346, 174177
(1990).
39. Britten, K. H., Newsome, W. T., Shadlen, M. N., Celebrini, S. & Movshon,
J. A. A relationship between behavioral choice and the visual responses of
neurons in macaque MT. Vis. Neurosci. 13, 87100 (1996).
40. Pasternak, T. & Merigan, W. H. Motion perception following lesions of the
superior temporal sulcus in the monkey. Cereb. Cortex 4, 247259 (1994).
41. Newsome, W. T. & Wurtz, R. H. Probing visual cortical function with discrete
chemical lesions. Trends Neurosci. 11, 394400 (1988).
42. Erickson, R. G. & Thier, P. A neuronal correlate of spatial stability during
periods of self-induced visual motion. Exp. Brain Res. 86, 608616 (1991).
43. Haarmeier, T., Thier, P., Repnow, M. & Petersen, D. False perception of
motion in a patient who cannot compensate for eye movements. Nature 389,
849852 (1997).
44. Matin, L. et al. Oculoparalytic illusion: visual-field dependent spatial
mislocalizations by humans partially paralyzed with curare. Science 216,
198201 (1982).
45. Deubel, H., Bridgeman, B. & Schneider, W. X. Immediate post-saccadic
information mediates space constancy. Vision Res. 38, 31473159 (1998).
46. Grsser, O. J., Krizic, A. & Weiss, L. R. Afterimage movement during saccades
in the dark. Vision Res. 27, 215226 (1987).
47. Bockisch, C. J. & Miller, J. M. Different motor systems use similar damped
extraretinal eye position information. Vision Res. 39, 10251038 (1999).
48. Van Essen, D. C. & DeYoe, E. A. in The Cognitive Neurosciences (ed.
Gazzaniga, M. S.) 383400 (MIT Press, Cambridge, Massachusetts, 1995).
49. Corbetta, M., Miezin, F. M., Shulman, G. L. & Petersen, S. E. A PET study of
visuospatial attention. J. Neurosci. 13, 12021226 (1993).
50. Levitt, H. Transformed up-down methods in psychoacoustics. J. Acoust. Soc. Am.
49, 467477 (1971).
183
 2000 Nature America Inc. http://neurosci.nature.com

articles

Thermosensory activation of insular

cortex
A. D. Craig1, K. Chen2,3, D. Bandy2 and E. M. Reiman2,4

1 Division of Neurosurgery, Barrow Neurological Institute, 350 West Thomas Rd., Phoenix, Arizona 85013, USA
2 Positron Emission Tomography Center, Good Samaritan Regional Medical Center, Phoenix, Arizona 85006, USA
3 Department of Mathematics, Arizona State University, Tempe, Arizona 85280, USA
4 Department of Psychiatry, University of Arizona, Tucson, Arizona 85721, USA
Correspondence should be addressed to A.D.C. (bcraig@mha.chw.edu)

Temperature sensation is regarded as a submodality of touch, but evidence suggests involvement of

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

insular cortex rather than parietal somatosensory cortices. Using positron emission tomography
(PET), we found contralateral activity correlated with graded cooling stimuli only in the dorsal
margin of the middle/posterior insula in humans. This corresponds to the thermoreceptive- and
nociceptive-specific lamina I spinothalamocortical pathway in monkeys, and can be considered an
enteroceptive area within limbic sensory cortex. Because lesions at this site can produce the post-
stroke central pain syndrome, this finding supports the proposal that central pain results from loss
of the normal inhibition of pain by cold. Notably, perceived thermal intensity was well correlated
with activation in the right (ipsilateral) anterior insular and orbitofrontal cortices.

The cutaneous sense of temperature is a distinct form of somat-
ic sensibility mediated by specific primary afferent receptors, yet
its representation in the brain is unknown1,2. Thermoreception
has two aspects, an exteroceptive, discriminative aspect impor-
tant for object recognition and environmental exploration and
an enteroceptive, affective aspect important for autonomic activ-
ity, homeostasis and thermoregulatory behavior. Whereas the
former is more obvious and traditionally emphasized, the latter
is an essential feature of mammalian physiology. Conceptual
recognition that thermal sensibility is an emergent aspect of ente-
roception (the sense of the physiological condition of the body
itself) would accomodate a variety of considerations and have
strong implications for other sensations from the body typically
regarded as exteroceptive, for instance, the sensation of pain.
Temperature and pain sensations are intimately associated
functionally and anatomically in the central nervous system
consistent with their common importance for maintenance of
body integrityand they are accordingly integrated. A funda-
mental interaction, the cold inhibition of pain, is a well estab-
lished therapeutic phenomenon with a demonstrable central
basis3,4. Loss of this interaction, which may result in the disinhi-
bition of pain, is proposed as a possible cause of the post-stroke
central pain syndrome5. This syndrome is generally character-
ized by intractable burning or aching pain referred to deep and
cutaneous tissues within a region of paradoxical hypalgesia and
thermal hypesthesia. (It occurs in 28% of stroke patients and
2540% of spinal cord injury and multiple sclerosis patients; it
is unresponsive to opiates.) Brain lesions that produce this syn-
drome interrupt the ascending spinothalamic pathway respon-
sible for pain and temperature sensations6,7, and the critical
cortical locus lies in a parieto-insular region8.
Other findings directly suggest that thermal sensation is rep-
resented in insular cortex, an area usually associated with auto-
nomic and limbic function911. Functional anatomy in monkeys
indicates that a dedicated thalamic nucleus relays topographic, dis-
criminative thermoreceptive-specific and nociceptive-specific lam-
ina I spino- and trigeminothalamic projections to the dorsal
margin of middle/posterior insular cortex2,12,13. Microstimulation
seemingly localized to the homologous thalamic region in awake
humans can produce discrete, graded sensations of cold14. Data
from a PET imaging study of the thermal grill illusion of pain4 and
an fMRI study14 indicate that strong innocuous cool stimulation
(20C) activates contralateral insular cortex, but not parietal
somatosensory cortical areas.
We used PET imaging of regional cerebral blood flow (rCBF)
to examine changes in forebrain activity directly related to the
intensity of graded cooling stimuli. Our results demonstrate that
discriminative thermosensory cortex lies in the dorsal margin of
the middle/posterior insula of humans.
RESULTS
We addressed the hypothesis that thermal stimuli activate the con-
tralateral insular cortex by examining rCBF changes with graded
cooling of the right hand. Tonic thermal stimuli were used,
because the tonic responses of thermoreceptive-specific lamina I
neurons linearly encode innocuous cool temperatures, whereas
their dynamic responses saturate more quickly1,15,16. Thus, the
stimuli ramped slowly downward and achieved a stable target
temperature 15 seconds before scan initiation. Preliminary testing
verified the discriminability of these stimuli and the stability of
the evoked sensations over the 4560-second stimulation peri-
od1. Using a verbal, open-ended zero to ten scale for magnitude
estimation of cold intensity, all volunteers were able to differentiate
stimuli that differed by 2C in ascending or descending sequences,
and their ratings were correlated with stimulus intensity (Fig. 1a).
We first explored changes in regional activity over the entire
forebrain by examining differences (subtractions) between the var-
ious thermal stimulus conditions (30C, 28C, 26C, 24C, 20C)

184 nature neuroscience volume 3 no 2 february 2000

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

articles

a c

Activity
Ratings

Temperature (C) Temperature (C)

b Fig. 1. Regression plots. (a) Average ratings given by the volunteers for the
thermal stimuli during the PET scans (r = 0.83; p < 106). (b) Average rCBF
2000 Nature America Inc. http://neurosci.nature.com

activity in left (contralateral) insular cortex (coordinates = 36, 22, 24; r =

0.55; p < 0.00005). (c) Step-like shift of average rCBF activity in right poste-
Activity

rior parietal cortex (coordinates = 36, 40, 36). Bars indicate standard error.

sion, but there were obvious shifts in magnitudes and spatial extents.
The largest correlations with ratings were located in the right
orbitofrontal and anterior insular regions (Table 2; Fig. 2). A sig-
nificant correlation was seen in the left anterior insula (Fig. 2, 24
mm), where there had been little correlation with stimulus tem-
perature. In contrast, the correlations in posterior parietal cortex
and the baseline condition (33C) and averaging across subjects. were greatly reduced. Surprisingly, the extent of the correlated acti-
For example, we found that the 20C stimulus produced signifi- vation in the upper brainstem increased (Fig. 2, 16 mm).
cant rCBF increases on the left (contralateral) side in the mid- Upon closer examination, we noted that the correlations with
dle/posterior insular region, the adjacent lentiform nuclei, posterior stimulus intensity in left middle/posterior and right anterior insular
parietal cortex and middle cingulate cortex and on the right (ipsi- cortex had equal strength. However, the correlation with ratings was
lateral) side in the region of the anterior insula, lentiform nuclei, significantly stronger in right anterior insula than in left middle/pos-
posterior parietal cortex and orbitofrontal cortex (Fig. 2). There terior insula (F-test, F1,116 = 4.61, p < 0.034; Fig. 3). The correlation
was slight activation in the middle/upper brainstem, probably rep- in right anterior insula with ratings was marginally better than with
resenting the periaqueductal gray and the parabrachial nucleus5. temperature (F = 2.1887, p < 0.07, one-tailed). Similarly, activation
The extent of activation in several regions seemed smaller with less in right orbitofrontal cortex was better correlated with ratings than
intense cooling stimuli, so we proceeded to correlational analyses. with temperature (F = 40.3478, p < 108). Given the close corre-
We used a voxel-by-voxel regression analysis to explore cor- spondence of the ratings with stimulus intensity, these are substan-
relations between the intensity of thermal stimulation and tial differences. The significance of these correlations with ratings
regional activity over the entire forebrain, testing the hypotheses was confirmed by ANCOVA with effects due to temperature, inter-
that there would be significant correlations in the left insular subject variability and scan-session factored out (right anterior insu-
region and the other regions noted above. In the left forebrain, la, z = 2.95, p < 0.003; right orbitofrontal, z = 3.59, p < 0.0002). Thus
rCBF was strongly correlated with stimulus intensity in the mid- activation on the right (nondominant) side in the anterior insula
dle/posterior insular region (Fig. 2). There was a weak correla- and orbitofrontal regions (and also left anterior insula) was related
tion in left posterior parietal cortex (but see below). Significant to subjective evaluation of the thermal stimuli.
correlations were observed on the right side in the anterior insu- In contrast, the activation in right and left posterior parietal
lar/lentiform, posterior parietal and orbitofrontal regions and cortices was significantly greater during all thermal stimuli than
middle/upper brainstem. The rCBF correla-
tion with stimulus intensity in the left mid- Table 1. Regression with stimulus temperature.
dle/posterior insula was highly significant and Site Coordinates (x, y, z) Peak Z r Slope p
was the largest in the entire forebrain L. post. insula 36 22 24 4.26 0.55 0.11 0.00005
(p < 0.00005; Fig. 1b; Table 1). We interpret
R. ant. insula 32 12 20 4.12 0.51 0.16 0.00004
this region as the contralateral cortical repre-
sentation of discriminative thermal sensation. brainstem 10 32 16 2.93 0.38 0.10 0.002
We next examined correlations between the R. post. parietal 36 40 36 4.12 0.47 0.12 0.00008
volunteers subjective ratings of thermal stimu- R. orbitofrontal 24 38 0 3.72 0.43 0.12 0.0004
lus intensity and rCBF activity in these regions, Results for the regression against stimulus temperature in the identified regions. Stereotaxic
again using a voxel-by-voxel regression analysis. (Talairach) coordinates are given as x, mediolateral; y, anteroposterior; z, horizontal. The peak Z
As expected, correlations were found in most of value is given for the single voxel at the focus, and the regression values encompass a region extend-
the same regions as with the temperature regres- ing 2 voxels in each direction around the focus (1 voxel for the brainstem site). L., left; R., right.

nature neuroscience volume 3 no 2 february 2000 185

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

articles

in the baseline condition (t-test, p < 105, right Table 2. Regression with raw ratings.
and left), but nearly uniform (ANOVA; right,
F 4,45 = 0.9767, p > 0.42; left, F = 0.4584, Site Coordinates (x, y, z) Peak Z r Slope p
p > 0.77; Fig. 1c). Thus, we interpret these L. post. insula 36 22 24 3.54 0.52 0.13 0.0006
step-like increases in parietal activation as R. ant. insula 32 12 20 4.69 0.60 0.24 0.000004
related to the attention directed to the right Brainstem 10 32 16 2.90 0.41 0.18 0.002
hand during the thermal stimuli. R. post. parietal 36 40 36 2.69 0.47 0.16 0.0005
Frontal projection showed that graded ther- R. orbitofrontal 24 38 0 4.80 0.57 0.21 0.000005
mal activation in the left (contralateral) mid-
dle/posterior insula was focused at the dorsal Results for the regression against subjects ratings in the identified regions. Conventions as in Table 1.
margin (Fig. 4). Notably, because it was centered
in the fundus of the superior limiting (circular) sulcus, the associ- gin of the middle/posterior insula. The activation was linearly
ation of this site with insular cortex might not be appreciated if it correlated with stimulus intensity, indicating that this site repre-
were viewed only in an axial (horizontal) projection. The localiza- sents discriminative thermal sensation. The demonstration that
tion of the foci of activation correlated with the subjects ratings in human thermosensory cortex is located in insular cortex, rather
the right (ipsilateral) anterior insular and orbitofrontal regions is than in parietal somatosensory areas, is significant for several
shown in Fig. 5. reasons. This directly confirms the prediction of several com-
parative and clinical observations, it substantiates the key per-
2000 Nature America Inc. http://neurosci.nature.com

DISCUSSION spective that thermal sensation is a specific aspect of

We found highly significant activation with innocuous cooling enteroception, and it provides strong evidence for the proposal
stimuli at one site in the contralateral forebrain, the dorsal mar- that central pain is an integrative thermosensory disorder.
Innocuous primary afferent (A
cool and C warm) thermoreceptors
innervating skin and deep tissues ter-
20 33 minate in the superficial spinal (and
trigeminal) dorsal horn, where a unique
population of lamina I spinothalamic
neurons is thermoreceptive-specific and
morphologically distinct1619. These neu-
Temp. rons have thresholds near normal skin
regr. temperature, evince linearly graded
responses to innocuous cooling and
have ongoing discharge that is inhibited
by warming and activity that is corre-
lated with discriminative thermal sen-
Rating sation15,16. (Neurons selectively excited
regr.
by warm are also present, though rela-
tively rare.) They project to thalamus in
the crossed lateral spinothalamic tract,
40 mm 32 mm 28 mm 24 mm 20 mm 16 mm
which is clinically and behaviorally crit-
ical for temperature sensibility, as well
as for pain, itch and sexual sensa-
tions2022. In primates, lamina I neurons
20 33
terminate in a dedicated thalamic relay
nucleus (VMpo, the posterior part of
the ventral medial nucleus), which con-
tains thermoreceptive- and nociceptive-
specific neurons2,12. In awake humans,
Temp. such neurons seem to be located in the
regr.
same region, where microstimulation
produces sensations of graded cold or
pain14,23,24. The homologous pathway
exists in cats and rats in primordial
Rating form; VMpo is visible only in primates,
regr. and it is proportionately greatly
enlarged in humans12. Anatomic data in
monkeys indicate that VMpo projects
12 mm 8 mm 0 mm 8 mm 16 mm 20 mm topographically to a cytoarchitectoni-
Fig. 2. Comparison of the statistical maps of rCBF activation for three different analyses at 12 of the 18 axial
cally distinct field in the fundus of the
levels examined. Abbreviations: 20 33, subtraction comparing activation in the 20C stimulus with the superior limiting sulcus at the dorsal
baseline 33C condition; Temp. regr., correlation with stimulus temperature; Rating regr., correlation with margin of middle/posterior insular cor-
subjects ratings. The yellow area indicates activation significant at the z > 2.58 level (p < 0.005), and the red tex5,13. This location matches the site of
area indicates activation significant at the z > 3.90 level (p < 0.0001). Numbers indicate axial (horizontal) level. the human thermosensory cortex

186 nature neuroscience volume 3 no 2 february 2000

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

articles

ceptors1,22. On the premise that temperature is an exteroceptive

Left sensation, prior investigators sought thermoreceptive-specific
insula neurons within the somatosensory system, and the homeostatic
aspect of thermoreception was relegated to the brainstem and
Percent activ-

hypothalamus1. Our findings substantiate a different perspective.

ity

Our findings fit the concept that temperature sensation is a

specific, emergent aspect of enteroception. Like all feelings
from the body, but in contrast to exteroceptive (touch) and
teloreceptive (vision) modalities, thermal sensibility is inher-
ently endowed with a characteristic affect that motivates behav-
ior, and it reflexively generates autonomic responses that signal
its primary role in thermoregulation and its integration with
homeostasis. The functional anatomical characteristics of lam-
ina I and insular cortex manifest this perspective (reviewed in
Right ref. 5). Small-diameter afferents that terminate in lamina I orig-
insula inate from nearly all tissues of the body and are sensitive to
changes in physiological conditions that require systemic
Percent activity

responses (for example, temperature, strong mechanical strain,

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

exercise, low pH, hypoglycemia, hypoxia, hypo-osmolarity, his-

tamine or inflammatory mediators). Consonantly, lamina I has
major projections to spinal autonomic and brainstem homeo-
static integration sites, including regions that receive conver-
gent vagal afferent activity and are heavily interconnected with
the hypothalamus and amygdala, namely, the parabrachial
Ratings nucleus (PB), periaqueductal gray (PAG) and the A1 cell group.
Lamina I also receives descending modulation directly from
Fig. 3. Regression plots comparing the proportional rCBF increases brainstem pre-autonomic sources (A5, A7, raphe) and from the
with ratings in left posterior insula (slope against absolute rCBF = 0.13)
hypothalamus. These and other considerations indicate that the
and right anterior insula (slope against absolute rCBF = 0.24) for each
subject at each temperature.
fundamental role of lamina I is to distribute modality-selective
sensory information on the ongoing physiological status of the
tissues of the body to a hierarchical network involved in the
maintenance of the integrity (or well-being) of the organism5.
demonstrated by the present observations. We conclude that this The termination of the ascending lamina I pathway in insu-
site is the terminus of the lamina I spinothalamocortical path- lar cortex is thus consistent with the view of the insula as limbic
way in humans. Together, these findings demonstrate the central sensory cortex, which is based on its association with visceral
pathway for discriminative thermal sensation.
This observation explains available clinical
findings. Contralateral thermosensory deficits
are caused by lesions involving the lateral
spinothalamic tract, posterolateral thalamus
or parieto-insular cortex6,8,21,25. The critical cor-
tical region is focused at the site identified by
our observations. In contrast, lesions of pari-
etal somatosensory areas have little or no effect
on thermal (or pain) sensation1,25,26. Similarly,
nearly all cooling-sensitive cells recorded in
parietal somatosensory areas show nonlinear
sensitivity associated with the cross-modal
responsiveness of slowly adapting mechanore-

Fig. 4. Localization of thermosensory cortex in

the dorsal margin of left (contralateral) insular cor-
tex (coordinates = 36, 22, 24), identified by
regression analysis of rCBF activation with stimulus
temperature, in frontal, axial and sagittal views. The
low cutoff (purple) of the spectral display is at
z > 2.58 (p < 0.005), and the white indicates z > 3.90
(p < 0.0001). Note that in the axial view, this site
seems parietal simply because the insula cannot be
seen. The activation in right anterior insula is also Temperature regression,
visible in the axial view. These color-coded images left middle posterior insula
were produced with the program Register (David (36, 22, 24)
MacDonald, Montreal Neurological Institute).

nature neuroscience volume 3 no 2 february 2000 187

 2000 Nature America Inc. http://neurosci.nature.com
articles
Rating regression,
right anterior insula
(32, 12, 20)
2000 Nature America Inc. http://neurosci.nature.com
Fig. 5. Localization of the evaluative regions in right anterior insular and orbitofrontal cortices, identified by regression analysis of rCBF activation
with subjects ratings, in frontal, axial and sagittal views. Conventions as in Fig. 3.
and autonomic function, its multimodal features and, partic-
ularly, its strong interconnections with hypothalamus, amyg-
dala and cingulate and orbitofrontal cortices911. Descending
projections from insular cortex terminate in lamina I as well as
in the same brainstem pre-autonomic and homeostatic sites
noted above27. Stimulation or lesions of insular cortex affect
cardiorespiratory, gastrointestinal, sympathetic and ther-
moregulatory activity11. In primates, the VMpo projection to
the dorsal margin of the insula is contiguous anteriorly with
the region that receives general (vagal) and special (gustatory)
visceral input by way of the thalamic VMb nucleus (which is
contiguous with VMpo)27,28. The common source of ascending
input to insular cortex (by way of VMb) in all mammals is the
parabrachial nucleus, the brainstem homeostatic site that inte-
grates both vagal and lamina I inputs; accordingly, the primor-
dial role of insular cortex can be regarded as modulation of PB
and as a source of multimodal input to goal-directed, homeo-
static motor processing in the hypothalamus, amygdala and
other sites10,11,29. Consonant with the enormous encephalization
in primates, especially humans, primate enteroceptive sensory
inputs bypass PB, with a direct gustatory projection from the
solitary nucleus to VMb30 and a topographic, dedicated lami-
na I projection to VMpo. These pathways seem to provide a
highly resolved enteroceptive representation of the bodys con-
dition in humans, including the specific sensations of temper-
ature, pain and other feelings from the body.
Many observations from functional imaging reinforce these
considerations. Middle/posterior insular activation is observed
with gustatory stimulation (taste), fasting (hunger), hyperton-
ic saline injection (thirst and, perhaps, induced hyperthermia),
lactate injection (tachycardia and panic in inducible patients),
cholecystokinin injection (tachycardia and induced anxiety),
inspiration, isometric exercise and various types of cutaneous
and deep pain4,3138. All of these can be considered aspects of pri-
mary enteroceptive sensation.
A broader association of anterior insula with internally gen-
erated emotion is suggested by its activation with recall-gener-
ated sadness, anticipatory anxiety, panic, disgust and visually
188
Rating regression,
right orbitofrontal
(24, 38, 0)
evoked sexual arousal35,3941. Consistent with our association of
right anterior insular activity with thermosensory evaluation,
similar activation is observed with the signaled anticipation of
heat pain42 and with tonically evoked heat pain38, in addition to
contralateral middle/posterior insular activation that can be
ascribed to the lamina I spinothalamocortical projection from
VMpo. Thus, these findings are consistent with the neurological
hypothesis that the right (nondominant) anterior insula is inte-
gral to mentally generating the image of ones physical state that
underlies basic emotions43, an essential part of the somatic mark-
er hypothesis of consciousness, or in different words, a limbic
sensory substrate involved in the evaluation that invests internal
feelings with emotional significance35. Our findings suggest that
perceptual dissociation of the affect generated by a thermal stim-
ulus, which depends on thermoregulatory integration, from the
discriminative aspect, which does not44, could occur upon con-
textual evaluation in the right anterior insula, and further in the
orbitofrontal cortex2,45. Nonetheless, whereas thermal sensation
(like ticklish or sensual feelings) can be either pleasant or unpleas-
ant, thermal distress is selectively associated with concomitant
activation of the anterior cingulate, that is, limbic motor cortex4.
The location of human thermosensory cortex corresponds
unmistakably with the critical cortical region damaged in certain
post-stroke central pain patients8, consistent with the proposal
that disruption of thermal sensation may cause central pain5.
Head and Holmes recognized that central pain is released by loss
of a specific pain and temperature pathway25, but inferred that
discriminative pain processing normally inhibits the emotional
aspect of pain. Others suggest that loss of the lateral pain path-
way releases a hypothetical spinoreticulothalamic pathway or that
deafferentation causes bursting and hyperactivity in somatosen-
sory pathways7. However, clinical observers report high correla-
tion of ongoing, burning central pain with the loss of thermal
sensibility6. Analyses of peripheral nerve block3 and of the ther-
mal grill illusion of pain reveal that reduced thermosensory (cool)
activity disinhibits a painful burning sensation. In particular, the
thermal grill, in which a sensation of burning, ice-like pain is
generated by innocuous cool (20C) and warm (40C) stimuli
nature neuroscience volume 3 no 2 february 2000
 2000 Nature America Inc. http://neurosci.nature.com
presented together in a spatially unusual fashion, demonstrates
that reduced activity in the cooling-sensitive lamina I spinothal-
amocortical pathway unmasks polymodal nociceptive activation
of the anterior cingulate cortex that is selectively associated with
thermal pain4,15. The intense burning experienced when warm
water is applied to a foot numbed by cold is directly comparable
to the thermal grill sensation and is an exigent thermoregulato-
ry distress signal. Cutaneous cooling inhibits various enterocep-
tive feelings; one normal function of the cold inhibition of pain 1. Darian-Smith, I. in Handbook of Physiology, Section 1, The Nervous System,
may be to produce graded (integrated) differentiation of nox-
ious (stressful) cold from innocuous cool, thereby motivating
appropriate homeostatic responses to thermal distress that
requires protective behavior.
Thus, the thermosensory disinhibition hypothesis proposes
that disruption of the thermosensory (enteroceptive) area in insu-
lar cortex disinhibits a limbic network, involving brainstem
homeostatic sites (PB, PAG) and anterior cingulate cortex, that
engenders affect associated with thermoregulatory motivation5.
2000 Nature America Inc. http://neurosci.nature.com
The affective feeling associated with a thermal stimulus reflects
the needs of thermoregulatory integration44 and seems to persist
following thalamocortical thermosensory lesions22. Vague auto-
nomic disturbances are repeatedly noted in central pain patients,
and many experience relief simply by going into a warm room,
exercising lightly or sitting in a spa6,25. The present results pro-
vide strong evidence supporting the idea that central pain is an
enteroceptive, thermoregulatory dysfunction.
METHODS
Our protocol was approved by local institutional review boards. We mea-
sured rCBF during 12 scans (60 s each) in each of 10 normal right-hand-
ed volunteers (3 female, 7 male; ages 2349, mean 33) using an ECAT 951/31
scanner (CTI, Knoxville, Tennessee). After a 30-min transmission scan,
bolus injections of 40 mCi of H215O were made at 15-min intervals4. Ther-
mal stimuli were applied to the palmar surface of the right hand with a 20
14 cm feedback-controlled Peltier device4,15. The volunteers had received
similar, but not identical, stimuli before the scans. The cooling stimuli
ramped at 1 per s from a baseline of 33C and attained a stable plateau
temperature of 30, 28, 26, 24 or 20C before each bolus injection, whereas
the neutral condition remained at 33C. Each volunteer was instructed to
remain still with eyes closed. Scanning was initiated upon the sharp increase
in measured radioactivity (1518 seconds after the injection). The temper-
ature was stable through the first 30 s of the scan and then returned to base-
line (ramp rate, 10 per s), to maximize the contrast between
state-dependent rCBF increases and noise46. The volunteers gave a magni-
tude estimate of the perceived intensity of cold following each scan using
an open-ended scale of zero to ten. The six different stimulus conditions
were presented two times each, first in a descending sequence of tempera-
tures and then in an ascending sequence, to counterbalance order effects.
The scanner produced 31 horizontal 128 128 slices with a separa-
tion of 3.375 mm, a 10.8-cm axial field of view, an in-plane resolution
of 9.5 mm, full width at half maximum (FWHM) and an axial resolu-
tion of 5.07.1 mm FWHM. The data were analyzed with Statistical Para-
metric Mapping (SPM96) implemented in Matlab (Mathworks, Sherborn
Massachusetts)47,48. The PET images for each subject were aligned with
the initial scan49, transformed into a standard stereotaxic coordinate
space50, smoothed with an isotropic Gaussian kernel (16 mm FWHM)
and normalized for global variations in absolute measurements. Statistical
parametric maps were constructed using data from all ten volunteers and
superimposed onto a spatially standardized MRI volume. The statistical
comparisons between different stimulus conditions and the regression
analyses were performed at each voxel using a multi-subject design with
replications. Normalized t-values (z-scores) were computed to compare
differences between each active thermal condition and the neutral con-
dition and for regressions against temperature and magnitude ratings.
Significant rCBF activation in the regions of interest occurred at z > 2.58,
yielding p < 0.005 uncorrected for multiple comparisons35,38.
nature neuroscience volume 3 no 2 february 2000
articles
ACKNOWLEDGEMENTS
We thank S. Goodwin, J. Frost and D. Andrew for technical assistance. This
study was supported by the Robert S. Flinn Foundation and the Atkinson Pain
Research Fund administered by the Barrow Neurological Foundation.
RECEIVED 8 OCTOBER; ACCEPTED 6 DECEMBER 1999
Vol. III, Sensory Processes (ed. Darian-Smith, I.) 879913 (American
Physiological Society, Bethesda, 1984).
2. Craig, A. D., Zhang, E. T. & Blomqvist, A. A distinct thermoreceptive
subregion of lamina I in nucleus caudalis of the owl monkey. J. Comp. Neurol.
404, 221234 (1999).
3. Yarnitsky, D. & Ochoa, J. L. Release of cold-induced burning pain by block of
cold-specific afferent input. Brain 113, 893902 (1990).
4. Craig, A. D., Reiman, E. M., Evans, A. & Bushnell, M. C. Functional imaging
of an illusion of pain. Nature 384, 258260 (1996).
5. Craig, A. D. A new version of the thalamic disinhibition hypothesis of central
pain. Pain Forum 7, 114 (1998).
6. Boivie, J., Leijon, G. & Johansson, I. Central post-stroke painA study of the
mechanisms through analyses of the sensory abnormalities. Pain 37, 173185
(1989).
7. Pagni, C. A. Central Pain: A Neurosurgical Challenge. (Ediziona Minerva
Medica S. P. A., Turin, 1998).
8. Schmahmann, J. D. & Leifer, D. Parietal pseudothalamic pain syndrome:
Clinical features and anatomic correlates. Arch. Neurol. 49, 10321037
(1992).
9. Mesulam, M.-M. & Mufson, E. J. Insula of the old world monkey. III: Efferent
cortical output and comments on function. J. Comp. Neurol. 212, 3852
(1982).
10. Allen, G. V., Saper, C. B., Hurley, K. M. & Cechetto, D. F. Organization of
visceral and limbic connections in the insular cortex of the rat. J. Comp.
Neurol. 311, 116 (1991).
11. Augustine, J. R. Circuitry and functional aspects of the insular lobe in
primates including humans. Brain Res. Rev. 22, 229244 (1996).
12. Craig, A. D., Bushnell, M. C., Zhang, E.-T. & Blomqvist, A. A thalamic
nucleus specific for pain and temperature sensation. Nature 372, 770773
(1994).
13. Craig, A. D. in Forebrain Areas Involved in Pain Processing. (eds. Besson, J.-M.,
Guilbaud, G. & Ollat, H.) 1326 (John Libbey, Paris, 1995).
14. Davis, K. D. et al. Thalamic relay site for cold perception in humans.
J. Neurophysiol. 81, 19701973 (1999).
15. Craig, A. D. & Bushnell, M. C. The thermal grill illusion: unmasking the burn
of cold pain. Science 265, 252255 (1994).
16. Dostrovsky, J. O. & Craig, A. D. Cooling-specific spinothalamic neurons in
the monkey. J. Neurophysiol. 76, 36563665 (1996).
17. Christensen, B. N. & Perl, E. R. Spinal neurons specifically excited by noxious
or thermal stimuli: marginal zone of the dorsal horn. J. Neurophysiol. 33,
293307 (1970).
18. Dostrovsky, J. O. & Hellon, R. F. The representation of facial temperature in
the caudal trigeminal nucleus of the cat. J. Physiol. (Lond.) 277, 2947
(1978).
19. Han, Z.-S., Zhang, E.-T. & Craig, A. D. Nociceptive and thermoreceptive
lamina I neurons are anatomically distinct. Nat. Neurosci. 1, 218225 (1998).
20. White, J. C. & Sweet, W. H. Pain and the Neurosurgeon: A Forty-Year
Experience 1000 (Thomas, Springfield, Massachusetts, 1969).
21. Norrsell, U. Behavioural thermosensitivity after unilateral, partial lesions of
the lateral funiculus in the cervical spinal cord of the cat. Exp. Brain Res. 78,
369373 (1989).
22. Norrsell, U. & Craig, A. D. Behavioral thermosensitivity after lesions of
thalamic target areas of a thermosensory spinothalamic pathway in the cat.
J. Neurophysiol. 82, 611625 (1999).
23. Dostrovsky, J. O., Wells, F. E. B. & Tasker, R. R. in Processing and Inhibition of
Nociceptive Information, International Congress Series 989 (eds. Inoka, R.,
Shigenaga, Y. & Tohyama, M.) 115120 (Excerpta Medica, Amsterdam,
1992).
24. Lenz, F. A. et al. Neurons in the area of human thalamic nucleus ventralis
caudalis respond to painful heat stimuli. Brain Res. 623, 235240 (1993).
25. Head, H. & Holmes, G. Sensory disturbances from cerebral lesions. Brain 34,
102254 (1911).
26. Adams, R. W. & Burke, D. Deficits of thermal sensation in patients with
unilateral cerebral lesions. Electroencephalogr. Clin. Neurophysiol. 73,
443452 (1989).
27. Yasui, Y., Breder, C. D., Saper, C. B. & Cechetto, D. F. Autonomic responses
and efferent pathways from the insular cortex in the rat. J. Comp. Neurol. 303,
355374 (1991).
28. Pritchard, T. C., Hamilton, R. B., Morse, J. R. & Norgren, R. Projections of
thalamic gustatory and lingual areas in the monkey, Macaca fascicularis.
J. Comp. Neurol. 244, 213228 (1986).
29. Shi, C. J. & Cassell, M. D. Cortical, thalamic, and amygdaloid connections of the
anterior and posterior insular cortices. J. Comp. Neurol. 399, 440468 (1998).
189
 2000 Nature America Inc. http://neurosci.nature.com
articles
30. Beckstead, R .M., Morse, J. R. & Norgren, R. The nucleus of the solitary tract
in the monkey: projections to the thalamus and brain stem nuclei. J. Comp.
Neurol. 190, 259282 (1980).
31. Kinomura, S. et al. Functional anatomy of taste perception in the human brain
studied with positron emission tomography. Brain Res. 659, 263266 (1994).
32. Benkelfat, C. et al. Functional neuroanatomy of CCK4-induced anxiety in normal
healthy volunteers. Am. J. Psychiatry 152, 11801184 (1995).
33. Casey, K. L., Minoshima, S., Morrow, T. J. & Koeppe, R. A. Comparison of
human cerebral activation patterns during cutaneous warmth, heat pain, and
deep cold pain. J. Neurophysiol. 76, 571581 (1996).
34. King, A. B., Menon, R. S., Hachinski, V. & Cechetto, D. F. Human forebrain
activation by visceral stimuli. J. Comp. Neurol. 413, 572582 (1999).
35. Reiman, E. M. The application of positron emission tomography to the study
of normal and pathological emotions. J. Clin. Psychiatry 58, 412 (1997).
36. Denton, D. et al. Neuroimaging of genesis and satiation of thirst and an
interoceptor- driven theory of origins of primary consciousness. Proc. Natl.
Acad. Sci. USA 96, 53045309 (1999).
37. Tataranni, P. A. et al. Neuroanatomical correlates of hunger and satiation in
humans using positron emission tomography. Proc. Natl. Acad. Sci. USA 96,
45694574 (1999).
38. Derbyshire, S. W. G. & Jones, A. K .P. Cerebral responses to a continual tonic
pain stimulus measured using positron emission tomography. Pain 76,
127135 (1998).
39. Stoleru, S. et al. Neuroanatomical correlates of visually evoked sexual arousal
2000 Nature America Inc. http://neurosci.nature.com
in human males. Arch. Sex. Behav. 28, 121 (1999).
40. Phillips, M. L. et al. A specific neural substrate for perceiving facial
expressions of disgust. Nature 389, 495498 (1997).
190
41. Mayberg, H. S. et al. Reciprocal limbic-cortical function and negative mood:
converging PET findings in depression and normal sadness. Am. J. Psychiatry
156, 675682 (1999).
42. Ploghaus, A. et al. Dissociating pain from its anticipation in the human brain.
Science 284, 19791981 (1999).
43. Damasio, A. R. Descartes Error: Emotion, Reason, and the Human Brain 312
(Putnam, New York, 1993).
44. Mower, G. Perceived intensity of peripheral thermal stimuli is independent
of internal body temperature. J. Comp. Physiol. Psychol. 90, 11521155
(1976).
45. Francis, S. et al. The representation of pleasant touch in the brain and its
relationship with taste and olfactory areas. Neuroreport 10, 453459 (1999).
46. Volkow, N. D., Mullani, N., Gould, L. K., Adler, S. S. & Gatley, S. J. Sensitivity
of measurements of regional brain activation with oxygen-15-water and PET
to time of stimulation and period of image reconstruction. J. Nucl. Med. 32,
5861 (1991).
47. Friston, K. J., Frith, C. D., Liddle, P. F. & Frackowiak, R. S. Comparing
functional (PET) images: the assessment of significant change. J. Cereb. Blood
Flow Metab. 11, 690699 (1991).
48. Worsley, K. J., Evans, A. C., Marrett, S. & Neelin, P. A three-dimensional
statistical analysis for CBF activation studies in human brain. J. Cereb. Blood
Flow Metab. 12, 900 918 (1992).
49. Woods, R. P., Cherry, S. R. & Mazziotta, J. C. Rapid automated algorithm for
aligning and reslicing PET images. J. Comput. Assist. Tomogr. 16, 620633
(1992).
50. Talairach, J. & Tournoux, P. Co-Planar Stereotaxic Atlas of the Human Brain
(Thieme, New York, 1988).
nature neuroscience volume 3 no 2 february 2000
 2000 Nature America Inc. http://neurosci.nature.com

articles

Expertise for cars and birds recruits

brain areas involved in face
recognition
Isabel Gauthier1,2, Pawel Skudlarski2, John C. Gore2 and Adam W. Anderson2

1 Present Address: Department of Psychology, Vanderbilt University, Wilson Hall, Nashville, Tennessee 37240, USA
2 Department of Diagnostic Radiology, Yale University Medical School, Fitkin Basement, 333 Cedar Street, New Haven, Connecticut 06510, USA
Correspondence should be addressed to I.G. (isabel.gauthier@vanderbilt.edu)

Expertise with unfamiliar objects (greebles) recruits face-selective areas in the fusiform gyrus (FFA)
2000 Nature America Inc. http://neurosci.nature.com

and occipital lobe (OFA). Here we extend this finding to other homogeneous categories. Bird and
car experts were tested with functional magnetic resonance imaging during tasks with faces,
familiar objects, cars and birds. Homogeneous categories activated the FFA more than familiar
objects. Moreover, the right FFA and OFA showed significant expertise effects. An independent
behavioral test of expertise predicted relative activation in the right FFA for birds versus cars within
each group. The results suggest that level of categorization and expertise, rather than superficial
properties of objects, determine the specialization of the FFA.

Face and object recognition differ in at least two ways. First, faces
are recognized at a more specific level of categorization (for exam-
ple, Adam) than most objects (for example, chair or car). Sec-
ond, although we are experts with faces, we have much less
experience discriminating among members of other categories.
Level of categorization and expertise are relevant even for unfa-
miliar faces and objects. A person passed on the street may be
encoded at the individual level and recognized the next day,
whereas a mug may be replaced by another mug without our
noticing. Processing biases for different categories depend on our
experience with levels of categorization and our expertise in
extracting diagnostic features1.
Viewing faces activates a small extrastriate region called the
fusiform face area (FFA)210. Neuropsychological studies suggest
that the brain areas responsible for face and object processing
can be dissociated1114. According to one view, extrastriate cor-
tex contains a map of visual features1516, suggesting that the same
region should not be recruited for processing different object cat-
egories when the relevant features differ.
On the other hand, prosopagnosia is often associated with
deficits discriminating among nonface objects within categories.
For example, a bird watcher became unable to identify birds17,
whereas another patient could no longer identify car makes18.
Thus, one hypothesis holds that prosopagnosia is a deficit in
evoking a specific context from a stimulus belonging to a class
of visually similar objects19. At least some prosopagnosic patients
have difficulty with classes in which objects are both visually and
semantically homogeneous20,21. Evidence from brain-lesion stud-
ies is still under debate13,22; however, additional data from brain
imaging may help resolve these questions.
Several lines of research converge to suggest that level of cate-
gorization and expertise account for a large part of the activation
difference between faces and objects. First, behavioral effects2325
once thought unique to faces have been obtained with objects,
often with expert subjects2629. Second, nonface objects elicit more
activation in the FFA when matched to specific labels as compared
to more categorical ones (for example, ketchup bottle versus bot-
tle)3,30. Third, expertise with animal-like unfamiliar objects (gree-
bles) recruits the right FFA4. However, it remains unclear whether
expertise with any homogeneous category is capable of recruiting
the neural substrate of face recognition.
This experiment had three purposes. First, we tested whether
long-term expertise with birds and cars would recruit face-selec-
tive areas. Second, the interaction between level of categoriza-
tion and expertise was investigated. Third, we tested how these
two factors depend on attention to stimulus identity. The FFA
typically activates more for faces than objects, even during passive
viewing7. This suggests that faces are processed automatically at
the subordinate level. Here we asked whether this is also true for
other expertise domains.
RESULTS
We tested 11 car experts and 8 bird experts with many years of
experience recognizing car models or bird species (Table 1). The
right and left FFA and right occipital face area (OFA) were
defined in passive-viewing localizer scans (see Methods). The
OFA is also face selective31 and active in greeble experts4. A right
FFA was found in all subjects (median size, 6 voxels), a left FFA
was found in 13 subjects (4 bird experts, 9 car experts; median
size, 5 voxels) and a right OFA in 15 subjects (7 bird experts, 8
car experts; median size, 7 voxels).
Subjects also underwent identity and location scans. Stimulus
presentation was identical in both conditions, and subjects detect-
ed immediate (1-back) repetitions in either the identity of the pic-
ture or its location while ignoring the other dimension. Blocks of
16 grayscale faces, objects, cars or birds shown sequentially were
alternated with periods of fixation (Fig. 1). Pilot experiments indi-
cated that the absence of color cues did not eliminate the advantage
of experts over novices. Behavioral data in the scanner was available
for 16 of the 19 subjects. Performance was better in the identity

nature neuroscience volume 3 no 2 february 2000 191

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

articles

Inspection of Fig. 2 suggests baseline differences between cate-

gories and between groups. First, responses to birds were larger than
to cars in FFAs of both hemispheres (right, F1,17 = 11.13, p < 0.05;
left, F1,11 = 5.47, p < 0.05). Again, although animal faces may acti-
vate this area more than objects10, here we found no difference
between cars and birds in novices. Bird experts showed more acti-
vation for any object category than car experts, although not signif-
icantly in any ROI. Given these baseline trends, it was crucial to
measure the effect of expertise by comparing activation for cars and
birds in the two groups. The predicted expertise effect was signifi-
cant (a group category interaction) in the right FFA (Figs. 2 and 3;
F1,17 = 19.22, p < 0.0005) and in the right OFA (F1,13 = 4.86, p < 0.05).
There was no expertise effect in the left FFA (F < 1).
One important question is whether this expertise effect aris-
es from the same area as face expertise. To test this, we used a set
a of criteria (see Methods) based on a definition of face cells in neu-
rophysiology32. This defines a smaller FFA than any other defini-
tion to our knowledge (also eliminating the majority of OFA and
2000 Nature America Inc. http://neurosci.nature.com

left FFA ROIs, which were not analyzed further). We call this ROI
the center of the FFA (median = 3 voxels), in which each voxel is
highly face selective. Even in this ROI, both the level of catego-
rization effect in novices (F1,17 = 6.37, p < 0.02) and the expertise
effect were present (F1,17 = 10.25, p < 0.006; Fig. 3). These effects
also held when analyzed in a subset of subjects whose FFA could
Location Location Identity be defined using described criteria6,10 (see supplemental material
or identity
at http://neurosci.nature.com/web_specials/). To assess the mag-
Time
b nitude of the expertise effect, we plotted the main effects and inter-
action separately for the center of the right FFA33 (Fig. 4). The
Fig. 1. Examples of stimuli and tasks for the fMRI protocol. (a) Images statistically significant expertise effect contributed a difference of
(256 256 pixels in size, 256 grays) from each of 4 categories about 0.4% signal change between groups, whereas the group and
(Caucasian faces without hair, passerine birds from New England, car category main effects contributed about 0.3% and 0.1% signal
models for the years 1995 and 1998 and various familiar objects) were change, respectively, and were not significant (F < 1 for both).
used in the fMRI study. (b) Example of stimulus presentation during the Corresponding values in the larger right FFA ROI were 0.2, 0.1
fMRI runs. Subjects made 1-back repetition judgments regarding either
and 0.3, respectively. The expertise effect alone accounted for 32%
location or identity (an identity repeat would show identical images,
although sometimes in different locationssee Methods for details). of the difference between faces and objects in the right FFA defined
at t = 2 and for 36% in the center of the right FFA.
We measured the center of mass of the signal change for acti-
vated voxels for birds, cars and faces (relative to objects). This
than the location runs (identity performance s.e., 89.4 2.1; was done in a ROI of 25 25 voxels (each 1.3 mm by 1.7 mm,
location, 86.0 2.3; F1,14 = 9.98, p < 0.01) and this effect was larg- y x over 3 slices in Talairach space, centered on the right and
er for birds and objects than for cars and faces (task category left FFA from the localizer). The only significant differences for
interaction, F1,14 = 5.86, p < 0.01). These categories varied more in cars or birds relative to faces were obtained in novice subjects
shape, making location judgments more difficult.
The percent signal changes in the three regions of
interest (ROIs) were assessed using a fixation baseline. Table 1. Subject information and behavioral results.
First, we describe all significant effects pooled across Bird experts Car experts
task, coming back to this factor later. The level of cat- Mean age s.e. 34.4 2.0 31 2.5
egorization effect was measured by comparing activa- Mean years experience se 18 3.3 20.6 3.8
tion in novices to cars or birds versus objects. The
effect of level of categorization was significant in the
right FFA (F1,17 = 14.36, p < 0.02) and in the left FFA Behavioral data during fMRI
(F1,11 = 8.76, p = 0.02.). This effect was marginal in the (% correct identity s.e.; location s.e.)
right OFA (F1,13 = 3.67, p < 0.08). The interaction Objects 86 3; 81 4 93 3; 88 3
between level and group was significant in both the Faces 85 3; 82 3 92 3; 91 3
right FFA (F 1,17 = 6.61, p < 0.02) and the left FFA Cars 84 3; 81 3 92 2; 91 2
(F1,11 = 6.47, p < 0.05). Post-hoc tests (p < 0.05) indi- Birds 87 3; 81 4 92 3; 89 3
cated that the level effect was only significant for car
experts viewing birds. It may be tempting to believe Behavioral data pre-test (d s.e.)
that birds activate the FFA because of their faces10. Birds upright 2.53 0.10 1.06 0.07
However, the difference between birds and cars for
Cars upright 1.41 0.12 2.42 0.14
novices was not significant in either area (p > 0.5 for
Birds inverted 2.23 0.20 1.01 0.09
both), and the group effect arises from a difference in
activity for common objects (larger in birders). Cars inverted 0.84 0.13 1.58 0.20

192 nature neuroscience volume 3 no 2 february 2000

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

articles

bird experts
3.125 7 mm3) window in Fig. 5.
Experts (and to some extent bird-
car experts ers viewing cars) showed a distri-
bution of activation for birds and
Right FFA Center of right FFA
cars that was relatively limited to
the localizer peak of the FFA. The
mean percent signal change in the
Mean % signal change

Mean % signal change

center voxel was compared to that
in the 8 voxels surrounding it, and
to the surrounding outside 16
voxels. An ANOVA (3 regions 3
categories 2 groups) revealed a
main effect of region (F2,22 = 7.18,
p < 0.004) with more activity in
the center than in both outer
regions. Because of greater activ-
Objects Cars Birds Faces Objects Cars Birds Faces
ity for faces than birds and cars
only in the center, the category
2000 Nature America Inc. http://neurosci.nature.com

region interaction was marginal

Left FFA Right OFA
(F4,44 = 2.18, p < 0.09), consistent
with our other analyses. Crucial-
ly, there was no significant differ-
ence among the three categories
Mean % signal change

Mean % signal change

in activity for each of the two sur-

rounding regions, suggesting that
activation is as focused for objects
as for faces.
As a more direct way of assess-
ing the expertise effect, we mea-
sured the correlation between
behavioral performance outside
Objects Cars Birds Faces Objects Cars Birds Faces the scanner with the signal
change in the three ROIs during
Fig. 2. Mean percent signal change for each object category in the two expert groups in three face-specific ROIs the location and identity tasks. In
and in the center of the right FFA. The average percent signal increase from fixation for each object category in the
the behavioral test, subjects
different ROIs was averaged across subjects in each ROI for each expert group. Error bars indicate standard error
of the mean. The Talairach coordinates for the center of each ROI standard error were right FFA, x = 38 2, judged whether sequentially pre-
y = 50 1, z = 7 1; left FFA, x = 38 2, y = 56 4, z = 6 2; right OFA, x = 40 2, y = 75 3, z = 3 1. sented pairs of birds and cars
(upright or inverted) belonged to
the same species or car model.
The expertise effect was signifi-
(see Table 2). Center of mass for expert categories was indistin- cant (group category interaction; Table 1), bird experts being
guishable from that obtained for faces (even using a lenient sta- more sensitive for birds than cars and vice versa for car experts
tistical test). We also compared the activation distribution in the (F1,17 = 59.40, p < 0.0001). The effect of orientation and interac-
right FFA for the three categories by averaging nonspatially tion of category with orientation were significant, with the inver-
smoothed individual maps, centered on the most face-selective sion effect stronger for cars than for birds (F1,17 = 14.27, p <
voxel in the localizer. This is shown in a 5 5 voxel (each 3.125 0.002). Both groups were poorer with inverted than upright cars,

Fig. 3. The right FFA shows an expertise

Faces objects Cars objects Birds objects
effect for birds and cars. One axial oblique
slice through the FFA for one expert for each
category shows the t-maps obtained when
Car expert

comparing the activation for faces, cars and

birds with the activation elicited by objects t = 4.5
during the location 1-back runs. The voxels
marked by white crosses indicate the right
FFA and OFA as defined in the passive viewing
runs for these two subjects. (In this car
expert, the OFA was actually in the slice
immediately below and is shown on the same
Bird expert

slice as the FFA only to illustrate its in-plane

location.) Note that the center of the right
t=1
FFA may be slightly different depending on the
task (here passive viewing versus 1-back loca-
tion) and that its size varies between subjects.

nature neuroscience volume 3 no 2 february 2000 193

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

articles

whereas the inversion effect for birds only bird experts

approached statistical significance in bird- Main effects car experts Interaction
ers (p = 0.068).
A group analysis was performed on
the fMRI data during all 1-back tasks

Percent signal change

Percent signal change

with cars and birds (see Methods). This
less precise analysis allowed us to seek
other regions showing an expertise effect
regardless of the category, beyond the
ones we could define functionally. This
showed an area in right ventral temporal
cortex that was more activated in experts
than novices (Fig. 6). In addition to this
stream of activation, going from the right
OFA toward the right FFA, a bilateral
region in the parahippocampal gyrus was Cars Birds Cars Birds
also more active in experts. This area
overlaps with the parahippocampal place Fig. 4. Main effects with grand mean and interaction partialed out and interaction effect with the
2000 Nature America Inc. http://neurosci.nature.com

area (PPA)34, functionally defined as the grand mean and main effects partialed out, in the center of the right FFA. Each observed condition
region responding more to scenes than mean can be reconstructed by adding the value for the main effects and the interaction effect to the
objects. (It also responds more to objects grand mean (in this case, 0.845).
than faces.) Further work will be required
to identify the role of this area in percep-
tual expertise. The only region more acti-
vated in novices than experts was a small bilateral area of the small (1-back identity versus passive viewing; but see ref. 36). We
lateral occipital gyrus, superior to the OFA. This area has been also found no effect of task on the advantage of faces over objects in
found to activate more for letter strings than faces35, and its selec- the right FFA (p > 0.28), nor on the expertise effect. Expertise may
tive activation for novices could reflect a switch from a featural influence how objects are automatically processed, an idea that we
to a more configural strategy. come back to in our discussion.
In each ROI, we correlated the percent signal change for birds
minus cars with relative expertise, the difference in sensitivity DISCUSSION
(d) for upright birds minus upright cars. As the 2 groups com- Previous studies suggest that level of categorization and expertise
bined would produce a bimodal distribution, the correlation coef- contribute to the specialization of the FFA. The present results show
ficients were calculated for each group separately using our largest how their contributions add up to account for a considerable part
homogeneous sample (the 12 subjects scanned with axial slices.) of the difference typically found between objects and faces.
For both groups, relative expertise was positively correlated with In our experiment, experts would know more names for the
relative percent signal change for birds versus cars in the right birds or cars than novices would. However, naming is not likely
FFA and only for the location task (car experts, r = 0.75; bird to account for the effects in the FFA because unfamiliar faces
experts, r = 0.82; p < 0.05 for both; Fig. 7). activated this area the most, whereas common objects that are
We also considered task effects beyond that found in the corre- easily named elicited the least activation. In addition, expertise
lation analyses. The only ROI showing a significant influence of effects for novel objects can be obtained in the FFA for unfa-
task was the right FFA, where this factor interacted with level of cat- miliar exemplars of a trained category4.
egorization (F1,17 = 6.58, p < 0.02): the subordinate-level advantage Why would faces recruit the FFA more than expert recog-
was larger when novices attended to the identity than to the location nition of objects? There are many possibilities. First, the FFA
of the stimuli. In prior studies6,10, the effect of task in the FFA was may be dedicated to face recognition (innately or through expe-
rience), although it may mediate the processing of
Table 2. Center of mass coordinates in the middle temporal lobe for other objects to some extent. At the least, our study
category-selective areas, given in Talairach coordinates. demonstrates that an innate bias is unnecessary for
Left hemisphere Right hemisphere objects to recruit this area with expertise. Second,
x y z x y z we cannot claim to have equated objects with faces
on level of categorization and expertise22. The faces
Bird experts
may constitute a more visually homogeneous set
Faces 31.3 49.8 7.6 40.8 48.2 8.5
than our bird or car images. Faces are recognized
Cars 29.3* 47.3* 7.8 39.9 47.5 9.0 as individual exemplars, whereas even experts
Birds 30.8 49.6 7.9 40.3 48.1 8.3 mainly recognize cars and birds at the
model/species level. Although our subjects had
Car experts years of experience with cars or birds, they still had
Faces 29.0 48.7 9.1 38.6 48.1 9.1 been practicing face recognition for many more
Cars 28.9 49.2 8.1 38.3 47.2 8.5 years. Thus, face recognition being in a sense more
Birds 30.5* 49.8 7.0 41.2* 47.8 8.9 subordinate and relying on greater expertise may
be what make it seem special, leaving little contri-
*Value significantly different from the coordinate for faces in the same expert group accord- bution for a component of object category per se.
ing to a least significant difference test; p < 0.05.
Additionally, categorization level and expertise may

194 nature neuroscience volume 3 no 2 february 2000

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

articles

Fig. 5. Spatial distribution of percent signal Mean % signal change

change for faces, birds and cars (relative to an
0 0.9 1.8
objects baseline) in a 5 5 voxel window in
the right FFA, centered on the most strongly
activated voxel in the localizer. Dashed lines
indicate the three regions within which acti- Faces objects Birds objects Cars objects
vation was averaged for analyses. Note that
the highest activation during experimental
runs for faces may not be identical to the
highest peak in the localizer (consider car

Bird experts
experts in the faces objects condition).

be only two of several factors that

determine the specialization of this
area. (Other factors may include sym-
metry, properties of associated seman-
tic knowledge, number of exemplars,
2000 Nature America Inc. http://neurosci.nature.com

value to the perceiver37.)

Car experts

The effect obtained in the right

OFA suggests that expertise may be
responsible for specialization of a large
part of the face recognition system (at
least in the right hemisphere). In the
left FFA, we found an effect of level of
categorization, with no detectable con-
tribution of expertise. Whereas subordinate-level processing
may recruit both hemispheres, here visual expert recognition
of homogeneous categories seems to be mainly a right hemi-
sphere process.
Our most striking result may be a very strong correlation
between a behavioral test of object expertise and the relative acti-
vation in the right FFA for birds and cars. It is remarkable that
the expertise of a subject was so accurately predicted from the
activation in a small part (six voxels) of the brain, especially as
the behavioral and fMRI experiments shared neither a common
task nor stimulus set. In addition, this analysis suggests that acti-
vation of the right FFA was more directly correlated with expert
performance than the right OFA.
There seems to be an important interaction between auto-
maticity of processing at the subordinate level and expertise. It is
argued that the preference of the FFA for faces does not depend on
the task6,10 (but see ref. 35), a claim also supported by our results.
However, we found an interaction between level of categorization
and task for novices, indicating that, for most people, simply seeing
an object among similar exemplars may not prompt complete sub-
ordinate-level processing. Automatic subordinate-level processing
for experts could also explain a surprising finding: the correlation
between behavior and activation in the right FFA was significant
only when subjects attended to the location of the objects. During
the identity task, subjects had to perform subordinate-level recog-
nition with both categories, regardless of expertise. Novices may

Fig. 6. Expertise effect in the temporal cortex. The t-maps

for all subjects with axial slices (14) were transformed into a
common standard space. Voxels showing a significant exper-
tise effect across subjects (p < 0.01) are displayed on the
transformed anatomical images for slices 2 to 5 for a single
subject. The red to yellow voxels were more active for
experts than novices across the identity and location tasks,
whereas the blue voxels were more active for novices. The
right hemisphere is shown on the left. The FFA is typically
found in slice three.

nature neuroscience volume 3 no 2 february 2000 195

 2000 Nature America Inc. http://neurosci.nature.com
articles
Location task
bird advantage
car experts, r = 0.75*
bird experts, r = 0.82*
Relative % signal change
car advantage
Relative expertise (d)
2000 Nature America Inc. http://neurosci.nature.com
car advantage
Identity task
bird advantage
car experts, r = 0.10
bird experts, r = 0.004
Relative % signal change
car advantage
Relative expertise (d)
car advantage
Fig. 7. Relationship between a behavioral measure of expertise and
activation in the right FFA. Relative expertise is the sensitivity (d) for
bird minus car matching. The dashed and full lines respectively indicate
the best linear fits for car and bird experts. Significant correlation coef-
ficients are marked with an asterisk (p < 0.05).
then use a featural strategy, whereas experts may use a more con-
figural strategy2628. Perhaps only configural processing is a good
predictor of behavioral expertise. In contrast, during the location
task novices may not access the subordinate level, whereas experts
did so automatically.
Birds and cars differ in many aspects. (Birds are small ani-
mals with moveable parts, covered with feathers that have spe-
cific markings; cars are large man-made objects made of metal
and typically uniform in texture.) Combined with a previous
study showing an expertise effect in the FFA with greebles4,
our results suggest very few constraints on the structure of the
objects for which expertise can recruit this small area. This is
important for any theory of visual representation in the ven-
196
tral pathway, because it suggests that responses of neurons in
extrastriate cortex may not be organized according to the visu-
al features that they detect15,16; rather, their functional organi-
zation may depend on the different processes important for
object recognition. For instance, some areas may be more suit-
ed for featural processing, whereas other areas may support
configural and holistic processing, hallmarks of subordinate-
level expertise. Our results suggest that expert subordinate-level
recognition for any category may be mediated in the same
regions, either by virtue of activating common cells or through
selectively activating different populations that are intermin-
gled. Other techniques, such as single-cell recording, will be
necessary to distinguish between these two alternatives.
METHODS
Subjects. Subjects, all male, included 11 car experts and 8 bird experts.
Informed consent was obtained from each subject, and the study was
approved by the Human Investigation Committee at the School of Med-
bird advantage icine, Yale University. Eight subjects were left handed. Handedness did
not correlate with any effect reported here.
Stimuli. One hundred and seventy six images each of passerine birds and
cars were obtained from public sources on the world-wide web. Images
were converted to 8-bit grayscale 256 256-pixel format, and objects
were isolated and placed on a 50% gray background. Objects were select-
ed to be familiar to our expert population. For each category, 112 images
were used in the behavioral test, whereas the remaining 64 objects from
each category were used for experimental scans. Faces without hair
(n = 64, scanned in a 3-D laser scanner, courtesy of Niko Troje and Hein-
rich Blthoff, Max Planck Institute, Tbingen, Germany) and 64 images
of non-living familiar objects were prepared in the same way as the cars
and birds and also used in experimental scans. Localizer scans used 90
grayscale photographs of faces and 90 pictures of familiar objects.
Behavioral task. Each subject performed 10 blocks of 56 sequential
matching trials, alternating between blocks showing birds or cars. There
were four conditions (upright and inverted cars and birds). Each trial
showed two images from the same category and orientation. Upright and
inverted trials were randomly intermixed. On each trial, a fixation cross
appeared for 500 ms, followed by stimulus 1 for 1000 ms and a pattern
mask for 500 ms before stimulus 2 appeared and remained on the screen
until a response was made. Subjects judged whether the two images
showed birds from the same species or whether cars were from the same
model but different years (mostly 1995 versus 1998). No difference was
bird advantage found in mean sensitivity between categories for novices. However,
responses to cars were slower than responses to birds for all subjects (RTs
for hits with cars, 1138 ms; birds, 1046 ms; p < 0.05, suggesting that the
cars were more difficult).
fMRI task. Experimental scans consisted of three runs of a one-back loca-
tion task alternated with three runs of a one-back identity task. The only
difference between identity and location runs was instructions to detect
immediate repetitions in either location or identity (Fig. 1). Each run
lasted 5 min, 36 s and consisted of 16 epochs (16 s each) with 5 fixation
periods (16 s each) interleaved at regular intervals. During each epoch, 16
objects appeared, each shown for 725 ms followed by a 275 ms blank.
Objects (each 12 12) appeared in one of 8 locations within an overall
area subtending 18 18 of visual angle. The order of the four categories
was counterbalanced across runs.
ROI selection. Regions of interests were functionally defined using two
localizer scans, which included 16 epochs (16 s each) of passive viewing of
faces or common objects centered on the screen (25 pictures per epoch).
Each run began with 16 s of fixation, and an 8-s fixation period was includ-
ed after every 2 passive viewing epochs. The right and left FFA and right
OFA were defined as contiguous voxels activated at an arbitrary threshold of
t = 2 in the middle fusiform gyri (cd, F-G, 9-10 in Talairach space), and
the same threshold was applied in the right ventral occipital lobe (c-d, H-I,
nature neuroscience volume 3 no 2 february 2000
 2000 Nature America Inc. http://neurosci.nature.com
9-10) for the right OFA . The raw data were noisier than when using a high-
er field scanner and a surface coil6,10 so the same level of significance for
ROI definitions could not be applied. However, the magnitude and spatial
extent of the effect for a given functional area should be similar regardless of
statistical power, and we used a criterion leading to ROIs comparable in
size to those in published studies6,10 . Our FFAs show at least twice as much
percent signal change for faces as for objects (each compared to fixation.) To
eliminate the influence of less face-selective voxels, we defined the center
of the right FFA using criteria more stringent than in any published study.
The only voxels selected were found within the cluster of contiguous voxels
selected using the less stringent criterion, showed twofold greater percent-
age signal change for faces as objects (compared with fixation) and, in each
subject, did not have less than half the percentage signal change for faces of
the voxel showing the maximum signal change for faces.
fMRI imaging parameters and analyses. Most (16) subjects were scanned
at the Yale School of Medicine on a 1.5 T GE Signa scanner equipped with
resonant gradients (Advanced NMR, Wilmington, Massachusetts) using
echo-planar imaging (gradient echo single shot sequence, 168 images per
slice, FOV = 40 20 cm, matrix = 128 64, NEX = 1, TR = 2000 ms,
2000 Nature America Inc. http://neurosci.nature.com
TE = 60 ms, flip angle = 60 ms). Six contiguous 7-mm-thick axial-oblique
slices aligned along the longitudinal extent of the fusiform gyrus covered
most of the temporal lobe. Some subjects (two car experts and one bird
expert) were scanned using coronal-oblique slices. Three more subjects
(two car experts and 1 bird expert, I. G. et al., Soc. Neurosci. Abstr. 25,
212.9, 1999) were scanned using coronal slices on the 3 T GE scanner at
the MGH-NMR Center in Charlestown, Massachusetts. In this case, a
custom bilateral surface coil was used to collect 168 images per slice in 12
near-coronal slices, 6 mm-thick. The imaging parameters were TR = 2000
ms, TE = 70 ms, flip angle = 90, 180 degrees and offset = 25 ms.
Before statistical analysis, images from the 1.5 T scanner were motion
corrected for three translation directions and the three possible rotations
using SPM-96 software (Wellcome Department of Cognitive Neurolo-
gy, London, UK). On the 3 T scanner, a bite bar was used to minimize
head motion. Maps of t-values and percent signal change, both correct-
ed for a linear drift in the signal38, were created. Maps were spatially
smoothed using a Gaussian filter with a full-width half-maximum value
of two voxels, except for analyses in the center of the right FFA, where
regions of interests were very small and no smoothing was performed.
In group composite maps, the percent signal change relative to fixation
baseline for both birds and cars was multiplied with contrast weights for
each subject (1 and 1 for bird experts and 1 and 1 for car experts). Under
the null hypothesis of no expertise effect, the expected value for this contrast
was equal to zero. We used a randomization test to asses the statistical sig-
nificance of percent signal changes. A population distribution for each voxel
was generated by calculating randomized mean values (1000 times) of the
contrast in which randomly chosen subsets of half the subjects got reversed
weights. The observed contrast, calculated without sign reversal, was assigned
a p value or proportion of times that the observed contrast was more extreme
than the randomized contrast). To show the anatomy clearly, the p values
were overlaid on the normalized anatomical images for a single subject
(threshold at p < 0.01; Fig. 6).
Note: Additional analysis can be found on the Nature Neuroscience web site
(http://neurosci.nature.com/web_specials/).
ACKNOWLEDGEMENTS
We thank Nancy Kanwisher and Ren Marois for discussions and Jill Moylan,
Terry Hickey and Hedy Sarofin for technical assistance. This work was supported
by NINDS grant NS33332 to J.C.G. and NIMH grant 56037 to N. Kanwisher.
I.G. was supported by NSERC.
RECEIVED 11 AUGUST; ACCEPTED 7 DECEMBER 1999
1. Archambault, A., ODonnell, C. & Schyns, P. G. Blind to object changes: When
learning the same object at different levels of categorization modifies its perception.
Psychol. Sci. 10, 249255 (1999).
2. Allison, T. et al. Face recognition in human extrastriate cortex. J. Neurophysiol. 71,
821825 (1994).
nature neuroscience volume 3 no 2 february 2000
articles
3. Gauthier, I., Tarr, M. J., Moylan, J., Anderson, A. W. & Gore, J. C. The functionally-
defined face area is engaged by subordinate-level recognition. Cognit.
Neuropsychol. (in press).
4. Gauthier, I., Tarr, M. J., Anderson, A. W., Skudlarski, P. & Gore, J. C. Activation of
the middle fusiform face area increases with expertise in recognizing novel objects,
Nat. Neurosci. 2, 568573 (1999).
5. Haxby, J. V. et al. The functional organization of human extrastriate cortex: A PET-
RCBF study of selective attention to faces and locations. J. Neurosci. 14, 63366353
(1994).
6. Kanwisher, N., McDermott, J. & Chun, M. M. The fusiform face area: A module in
human extrastriate cortex specialized for face perception. J. Neurosci. 17,
43024311 (1997).
7. McCarthy, G., Puce, A., Gore, J. C. & Allison, T. Face-specific processing in the
human fusiform gyrus. J. Cogn. Neurosci. 9, 605610 (1997).
8. Puce, A., Allison, T, Asgari, M, Gore, J. C. & McCarthy, G. Face-sensitive regions in
extrastriate cortex studied by functional MRI. Neurophysiology 74, 11921199 (1996).
9. Sergent, J., Otha, S. & MacDonald, B. Functional neuroanatomy of face and object
processing. Brain, 115, 1536 (1992).
10. Kanwisher, N., Stanley, D. & Harris, A. The fusiform face area is selective for faces
not animals. Neuroreport 10, 183187 (1999).
11. Sergent, J. & Signoret, J. L. Varieties of functional deficits in prosopagnosia. Cereb.
Cortex 2, 375388 (1992).
12. McNeil, J. E. & Warrington, E. K. Prosopagnosia: A face-specific disorder. Q. J. Exp.
Psychol. A 46, 110 (1993).
13. Moscovitch, M., Winocur, G. & Behrmann, M. What is special in face recognition?
Nineteen experiments on a person with visual object agnosia and dyslexia but
normal face recognition. J. Cogn. Neurosci. 9, 555604 (1997).
14. Assal, G., Favre, C. & Anderes, J. P. Non-reconnaissance danimaux familiers chez
un paysan. Rev. Neurol. 140, 580584 (1984).
15. Ishai, A, Ungerleider, L. G., Martin, A., Schouten, J. L. & Haxby, J. Distributed
representation of objects in the human ventral visual pathway. Proc. Natl. Acad. Sci.
USA 96, 93799384 (1999).
16. Tanaka, K. Inferotemporal cortex and object vision. Annu. Rev. Neurosci. 19,
109139 (1996).
17. Bornstein, B. in Problems of Dynamic Neurology (ed. Halpern, L.) 283318
(Hadassah Medical Organization, Jerusalem, 1963).
18. Lhermitte, J., Chain, F, Escouroole, R. Ducarne, B & Pillon, B. tude anatomo-
clinique dun cas de prosopagnosie. Rev. Neurol. (Paris) 126, 329346 (1972).
19. Damasio, A. R. & Damasio, H. & Van Hoesen, G. W. Prosopagnosia: anatomic
basis and behavioral mechanisms. Neurology 32, 331341(1982).
20. Dixon, M. J., Bub, D. N. & Arguin, M. Semantic and visual determinants of face
recognition in a prosopagnosic patient. J. Cogn. Neurosci. 10, 362376 (1998).
21. Riddoch, J. M. & Humphreys, G. W. Visual processing in optic aphasia: A case of
semantic access agnosia. Cognit. Neuropsychol. 4, 131186 (1987).
22. Gauthier, I., Behrmann, M. & Tarr, M. J. Can face recognition really be dissociated
from object recognition? J. Cogn. Neurosci. 11, 349370 (1999).
23. Yin, R. K. Looking at upside-down faces. J. Exp. Psychol. 81, 141145 (1969).
24. Tanaka, J. W. & Farah, M. J. Parts and wholes in face recognition. Q. J. Exp. Psychol.
A 46, 225245 (1993).
25. Farah, M. J., Wilson, K. D., Drain, H. M. & Tanaka, J. W. The inverted face inversion
effect in prosopagnosia: Evidene for mandatory, face-specific perceptual
mechanisms. Neuropsychologia 33, 661674 (1995).
26. Diamond, R. & Carey, S. Why faces are and are not special: An effect of expertise. J.
Exp. Psychol. Gen. 115, 107117 (1986).
27. Gauthier, I. & Tarr, M. J. Becoming a Greeble expert: Exploring the face
recognition mechanisms. Vision Res. 37, 16731682 (1997).
28. Gauthier, I., Williams, P., Tarr, M. J. & Tanaka, J. W. Training Greeble experts: A
framework for studying expert object recognition processes. Vision Res. 38,
24012428 (1998).
29. deGelder, B., Bachoud-Lvi, A.-C. & Degos, J.-D. Inversion superiority in visual
agnosia may be common to a variety of orientation-polarised objects besides faces.
Vision Res. 38, 28552861 (1998).
30. Gauthier, I., Anderson, A. W., Tarr, M. J., Skudlarski, P. & Gore, J. C. Levels of
categorization in visual object recognition studied with functional MRI. Curr. Biol.
7, 645651 (1997).
31. Halgren E., Dale, A. M., Sereno, M. I., Tootell, R. B., Marinkovic, K. & Rosen,
B. R. Location of human face-selective cortex with respect to retinotopic areas.
Hum. Brain Mapp. 7, 2937 (1999).
32. Rolls, E. T. & Baylis, G. C. Size and contrast have only small effects on the response
to faces of neurons in the cortex of the superior temporal sulcus of the monkey.
Exp. Brain Res. 65, 3848 (1986).
33. Rosnow, R. L. & Rosenthal, R. Some things you learn arent so: Cohens paradox,
Aschs Paradigm, and the interpretation of interaction. Psychol. Sci. 6, 39 (1995).
34. Epstein, R., Harris, A., Stanley, D. & Kanwisher, N. The parahippocampal place
area: Recognition, navigation, or encoding? Neuron 23, 115125 (1999).
35. Puce, A., Allison, T, Asgari, M, Gore, J. C. & McCarthy, G. Differential sensitivity of
human visual cortex to faces, letterstrings, and textures: A functional magnetic
resonance imaging study, J. Neurosci. 16, 52055215 (1996).
36. Wojciulik, E., Kanwisher, N. & Driver, J. Modulation of activity in the fusiform face
area by covert attention: an fMRI study. J. Neurophysiol. 79, 15741578 (1998).
37. Tranel, D., Logan, C. G., Frank, R. J. & Damasio, A. R. Explaining category-related
effects in the retrieval of conceptual and lexical knowledge for concrete entities:
operationalization and analysis of factors. Neuropsychologia 35, 13291339 (1997).
38. Skudlarski, P., Constable, R. T. & Gore, J. C. ROC analysis of statistical methods
used in functional MRI: Individual subjects. Neuroimage 9, 311329 (1999).
197