EDITORIAL OFFICE neurosci@natureny.com
75 Varick Street, Fl 9, New York, NY 10013-1917
Tel: (212) 726 9319, Fax: (212) 696 0978
Editor: Sandra Aamodt
Senior Editor: Kalyani Narasimhan
Associate Editors: Annette Markus, Charvy Narain
Assistant Editor: Hannah Bayer, Josh McDermott
Copy Editor: Anita Gould, Lavanya Reddy
Senior Production Editor: Jessica Iannuzzi
Production Editor: Elizabeth Jerabek
Illustrators: Kimberly Caesar, Katie Ris
Cover Design: Erin Boyle
Editorial Assistant: Mark Zipkin
NPG New York
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

75 Varick Street, Fl 9, New York, NY 10013-1917
Tel: (212) 726 9200, Fax: (212) 696 9006
Publisher: Beatrice Renault
Executive Editor: Linda Miller
Chief Technology Officer: Howard Ratner
Head of Nature Research & Reviews Marketing: Sara Girard
Assistant Marketing Manager: Amy Maurer
Production Coordinator: Diane Temprano
New Technology Director: Greg Suprock
Head of Web Services: Anthony Barrera
Web Production Manager: Susan Kline

NPG London
The Macmillan Building, 4 Crinan Street, London N1 9XW
Tel: 44 207 833 4000, Fax: 44 207 843 4996
Managing Director: Annette Thomas
Publishing Director: Peter Collins
Editor-in-Chief, Nature Publications: Philip Campbell
Marketing Director: Della Sar
Director of Web Publishing: Timo Hannay

NPG Nature Asia-Pacific
Chiyoda Building, 2-37 Ichigayatamachi, Shinjuku-ku, Tokyo 162-0843
Tel: 81 3 3267 8751, Fax: 81 3 3267 8746
Publishing Director — Asia-Pacific: David Swinbanks
Associate Director: Antoine E. Bocquet
Manager: Koichi Nakamura
Senior Marketing Manager: Peter Yoshihara
Asia-Pacific Sales Director: Kate Yoneyama
Asia-Pacific Sales Manager: Rinoko Asami
DISPLAY ADVERTISING display@natureny.com (US/Canada)
display@nature.com (Europe)
nature@natureasia.com (Asia)
US Head of Display Advertising: Stephen Schwartz, Tel: (212) 726 9256, Fax: (212) 696 9481
Global Head of Display Advertising Sales: John Michael, Tel: 44 207 843 4960, Fax: 44 207 843 4996
Global Head of Online Advertising: Matt Peskett, Tel: 44 207 843 4988, Fax: 44 207 843 4749
Head of Display Advertising—Europe: Gerard Preston, Tel: 44 207 843 4960, Fax: 44 207 843 4996
Business Development Manager: Claire Hines, Tel: 44 207 843 4960, Fax: 44 207 843 4996
Asia-Pacific Sales Manager: Rinoko Asami, Tel: 81 3 3267 8751, Fax: 81 3 3267 8746
Western Region Sales Manager: George Lui, Tel: (415) 781 3804, Fax: (415) 781 3805
Sales Executives:
New England: Sheila Reardon, Tel: (617) 399 4098, Fax: (617) 426 3717
New York, Mid-Atlantic, Southeast: Jim Breault, Tel: (212) 726 9334, Fax: (212) 696 9481
Midwest: Mike Rossi, Tel: (212) 726 9255, Fax: (212) 696 9481
Northwest: Mathieu DesJardins, Tel: (415) 781 6422, Fax: (415) 781 3805
Eastern England/Scotland/Italy/Spain/Israel: Matthew Wilkinson, Tel: 44 207 843 4960, Fax: 44
207 843 4749
South/West United Kingdom/Scandinavia/Holland: Marianne Boulakas, Tel: 44 207 843 4969,
Fax: 44 207 843 4749
Northern Germany: Gerard Preston, Tel: 44 207 843 4960, Fax: 44 207 843 4749
Southern Germany/Austria/Switzerland/France/Belgium: Sabine Hugi-Fürst, Tel: 41 52761 3386,
Fax: 41 52761 3419
NATUREJOBS naturejobs@natureny.com (US/Canada)
naturejobs@nature.com (Europe)
advertising@natureasia.com (Asia)
Publisher: Ben Crowe, Tel: (212) 726 9245, Fax: (212) 696 9482
US Sales Manager: Peter Bless, Tel: (212) 726 9248, Fax: (212) 696 9482
Japan Sales Manager: Rinoko Asami, Tel: 81 3 3267 8751, Fax: 81 3 3267 8746
Americas: Tel: (888) 331 6288 institutions@natureny.com
Asia/Pacific: Tel: 81 3 3267 8751 institutions@natureasia.com
Australia/New Zealand: Tel: 61 3 9825 1160 nature@macmillan.com.au
India: Tel: 00 91 11 324 4186 harpal@nature.com
ROW: Tel: 44 207 843 4759 institutions@nature.com
CUSTOMER SERVICE www.nature.com/help
Senior Global Customer Service Manager: Gerald Coppin
For all print and online assistance, please visit www.nature.com/help
Purchase subscriptions:
Americas: Nature Neuroscience, Subscription Dept., 342 Broadway, PMB 301, New York, NY
10013-3910. Tel: (866) 363 7860, Fax: (212) 689 9108
Europe/ROW: Nature Neuroscience, Subscription Dept., Macmillan Magazines Ltd., Brunel Road,
Houndmills, Basingstoke RG21 6XS, United Kingdom. Tel: 44 1256 329 242, Fax: 44 1256 812 358
Japan: Nature Neuroscience, Nature Japan K.K., MG Ichigaya Bldg. 5F, 19-1 Haraikatamachi,
Shinjuku-ku, Tokyo 162-0841. Tel: 81 3 3267 8751, Fax: 81 3 3267 8746
India: Harpal Singh Gill, Macmillan Magazines Ltd, 5A/12 Ansari Road, Darya Ganj, New Delhi,
110 002 India. Tel: 00 91 11 324 4186, Tel/Fax: 00 91 11 327 2010
REPRINTS reprint@natureny.com
Nature Neuroscience Reprint Department, Nature Publishing Group, 75 Varick Street, Fl 9,
New York, NY 10013-1917, USA.
For commercial reprint orders of 600 or more, please contact:
UK Reprints Sales Executive: Caren Collins, Tel: 44 207 843 4967, Fax: 44 207 843 4839
US Reprints Sales Executive: Sharda Tulsie, Tel: (212) 726 9631, Fax: (212) 679 0843

© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

981 What’s on your mind?

The origins of the neural 983 Neurobiology of Addiction
underpinnings of language are By George F Koob and Michel Le Moal
controversial. A study by Ricardo Gil-
Reviewed by Trevor Robbins
da-Costa and colleagues suggests that
the neural substrate for human speech
evolved from an ancestor common to
humans and macaque monkeys. The NEWS AND VIEWS
authors used PET imaging in macaque
monkeys to find activity specific to 984 Neurons that know when to quit
macaque vocalizations in monkey Jeremy M Wolfe  see also p 1071
brain areas homologous to human
perisylvian language areas. The cover 985 BK channels and circadian output
shows an image of a macaque monkey,
Christopher S Colwell  see also p 1041
which shared a common ancestor with
humans 25–30 million years ago. 987 Choice values
Photo credit: Marc Hauser.
(p 1064)
Yael Niv, Nathaniel D Daw & Peter Dayan  see also p 1057
989 Seeking a function for spontaneous neurotransmission
ChiHye Chung & Ege T Kavalali
991 What’s in control of language?
Angela D Friederici

993 IGF-1 receptor is essential for the establishment of hippocampal neuronal polarity
L Sosa, S Dupraz, L Laurino, F Bollati, M Bisbal, A Cáceres, K H Pfenninger &
S Quiroga
996 UNC5A promotes neuronal apoptosis during spinal cord development
independent of netrin-1
M E Williams, X Lu, W L McKenna, R Washington, A Boyette, P Strickland,
A Dillon, Z Kaprielian, M Tessier-Lavigne & L Hinck
999 Specification of auditory sensitivity by Drosophila TRP channels
M C Göpfert, J T Albert, B Nadrowski & A Kamikouchi
1001 Cross-modal regulation of synaptic AMPA receptors in primary sensory
Insulin-like growth factor essential cortices by visual experience
for hippocampal neuron polarity
(p 993)
A Goel, B Jiang, L W Xu, L Song, A Kirkwood & H-K Lee

Nature Neuroscience (ISSN 1097-6256) is published monthly by Nature Publishing Group, a trading name of Nature America Inc. located at 75 Varick
Street, Fl 9, New York, NY 10013-1917. Periodicals postage paid at New York, NY and additional mailing post offices. Editorial Office: 75 Varick Street,
Fl 9, New York, NY 10013-1917. Tel: (212) 726 9319, Fax: (212) 696 0978. Annual subscription rates: USA/Canada: US$199 (personal), US$1,809
(institution). Canada add 7% GST #104911595RT001; Euro-zone: €271 (personal), €1,558 (institution); Rest of world (excluding China, Japan, Korea):
£175 (personal), £1,005 (institution); Japan: Contact NPG Nature Asia-Pacific, Chiyoda Building, 2-37 Ichigayatamachi, Shinjuku-ku, Tokyo 162-0843.
Tel: 81 (03) 3267 8751, Fax: 81 (03) 3267 8746. POSTMASTER: Send address changes to Nature Neuroscience, Subscriptions Department,
342 Broadway, PMB 301, New York, NY 10013-3910. Authorization to photocopy material for internal or personal use, or internal or personal use of specific
clients, is granted by Nature Publishing Group to libraries and others registered with the Copyright Clearance Center (CCC) Transactional Reporting Service,
provided the relevant copyright fee is paid direct to CCC, 222 Rosewood Drive, Danvers, MA 01923, USA. Identification code for Nature Neuroscience:
1097-6256/04. Back issues: US$45, Canada add 7% for GST. CPC PUB AGREEMENT #40032744. Printed by Publishers Press, Inc., Lebanon Junction,
KY, USA. Copyright © 2006 Nature Publishing Group. Printed in USA.


1004 Maternal presence serves as a switch between learning fear and attraction
in infancy
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

BK channels modulate
S Moriceau & R M Sullivan
circadian output 1007 Anterior cingulate cortex responds differentially to expectancy violation
(pp 985 and 1041)
and social rejection
L H Somerville, T F Heatherton & W M Kelley

1009 BDNF-mediated neurotransmission relies upon a myosin VI motor complex
H Yano, I Ninan, H Zhang, T A Milner, O Arancio & M V Chao
1019 Vesicular proteins exocytosed and subsequently retrieved by compensatory
endocytosis are nonidentical
M Wienisch & J Klingauf
1028 Generalization of amygdala LTP and conditioned fear in the absence
of presynaptic inhibition
H Shaban, Y Humeau, C Herry, G Cassasus, R Shigemoto, S Ciocchi, S Barbieri,
H van der Putten, K Kaupmann, B Bettler & A Lüthi
1036 G(o) signaling is required for Drosophila associative learning
J Ferris, H Ge, L Liu & G Roman
1041 BK calcium-activated potassium channels regulate circadian behavioral
rhythms and pacemaker output
A L Meredith, S W Wiler, B H Miller, J S Takahashi, A A Fodor, N F Ruby &
R W Aldrich  see also p 985
Maternal presence switches infant rats
from fear learning to attraction 1050 PET imaging of dopamine D2 receptors during chronic cocaine self-administration
(p 1004) in monkeys
M A Nader, D Morgan, H D Gage, S H Nader, T L Calhoun, N Buchheimer,
R Ehrenkaufer & R H Mach
1057 Midbrain dopamine neurons encode decisions for future action
G Morris, A Nevet, D Arkadir, E Vaadia & H Bergman  see also p 987
1064 Species-specific calls activate homologs of Broca’s and Wernicke’s
areas in the macaque
R Gil-da-Costa, A Martin, M A Lopes, M Muñoz, J B Fritz & A R Braun
1071 LIP responses to a popout stimulus are reduced if it is overtly ignored
A E Ipata, A L Gee, J Gottlieb, J W Bisley & M E Goldberg  see also p 984

See back pages

Lateral intraparietal cortex modulates
salience of pop-out stimuli
(pp 984 and 1071)


What’s on your mind?
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

rom stage magicians to the Vulcan mind meld, mind reading has computer scientists, electrical engineers, scientists who come from an
been the stuff of magic and science fiction. Recent developments applied physics and mathematics background. The fact that we are
in neuroimaging might be bringing it one step closer to reality, finally looking for patterns of activity across the whole population
however, as increasingly sophisticated analysis techniques move toward rather than areas with different function means that we are taking a
the decoding of mental states from functional imaging data in humans. more information-based approach. This is a fundamental shift.”
Two companies even offer fMRI-based lie detectors1. Although legal Pattern classification can also overcome another criticism of
applications are still premature, these pattern classification techniques functional imaging: that it lacks the temporal resolution of other whole-
represent a new way of looking at neuroimaging data2 and may extend brain imaging techniques such as EEG. Pattern classification has been
the power of functional imaging substantially. used to identify distributed patterns of activity associated with different
Conventional neuroimaging analysis correlates external regressors categories of objects6. During later free recall, the patterns of activity
such as task condition with activity in specific brain areas. Pattern clas- associated with the specific object category reappear several seconds
sification inverts this methodology and instead predicts the external before the verbal recall of the object. The greater sensitivity of pattern
stimulus based on neuroimaging data. Unlike conventional analyses, classification techniques compared to conventional imaging analysis
these pattern-based analyses take into account the full spatial pattern of results in a temporal resolution approaching that of EEG.
brain activity rather than concentrating on specific regions. Thus, even As yet, however, these techniques cannot generate the sort of brain
if activity at a particular voxel does not distinguish different cognitive maps that we all know from conventional neuroimaging studies, show-
states, the pattern of activity distributed over many regions can do so, ing spots of localized activity associated with particular functions.
increasing sensitivity. This multivariate approach generates pattern vec- Efforts are underway to develop such maps based on pattern classifi-
tors corresponding to specific cognitive states, and a classifier is trained cation techniques, but conventional image analysis is likely to remain
to discriminate between these states. This classifier can then be used the preferred method for generating activity maps to understand where
to predict the cognitive state on the basis of brain activity alone. Such in the brain a process is occurring. Localization via conventional brain
approaches have been used to predict what percept is dominant in a imaging complements the information about how a process is occur-
binocular rivalry protocol3 or what orientation subjects are viewing4, ring that can be provided by pattern classification techniques, so both
even when they are not consciously aware of the stimulus. approaches will continue to be useful.
These techniques make it easier to evaluate responses to naturalistic Researchers are attempting to use pattern classification techniques to
stimuli, as pattern classification algorithms are designed to analyze activ- predict brain states in real-world applications, such as lie detection, but
ity over the whole brain without attempting to localize function. In a this endeavor seems much less promising. One problem is that activity
recent competition at the University of Pittsburgh (http://www.ebc.pitt. is more likely to be variable but, more importantly, an fMRI lie detector
edu/competition.html), participants were given fMRI data and subjective would rely crucially on the compliance of its subjects. To train a clas-
ratings from observers as they viewed two short film clips. Competitors sifier to categorize lying and truth telling, a suspect would essentially
then had to produce an algorithm that could predict what the subjects be asked to calibrate the instrument for his own conviction, poten-
were seeing based on a third fMRI data set. The winning entries achieved tially violating the fifth amendment to the United States Constitution,
correlations as high as 0.86 for basic features such as the presence of which protects people from being forced to incriminate themselves.
music. Uri Hasson from New York University, one of the researchers who Even though the accuracy of the technology is likely to improve, it is
judged the competition, says, “I am much more optimistic as to the power unclear if such an fMRI lie detector can surpass conventional polygraph
of fMRI to read and predict human experience. Many of the participants and EEG lie detectors, with which it is likely to share drawbacks such as
managed to predict the twelve features chosen for the competition (such noise introduced by arousal or emotional responses.
as language, music, emotion), as well as the specific observers who coded Although the applicability of pattern classification techniques to lie
the movie. Moreover, a few groups managed to predict the identity of the detection is uncertain, their influence on basic research is likely to be
actor being seen or the location the subject is watching.” important. Neuroimaging’s obsession with localization has often led to
The power of this approach extends beyond just predicting cognitive accusations that it is little more than phrenology. By using population
states from brain activity. Pattern classification techniques can provide responses across the whole brain to ask how rather than where informa-
clues about how this information is processed as well. For example, tion is processed, neuroimaging may be starting to come of age. 쐽
this technique shows that object categories with shared image-based 1. Pearson, H. Nature 441, 918–919 (2006).
2. Haynes, J.-D. & Rees, G. Nat. Rev. Neurosci 7, 523–534 (2006).
attributes have shared neural representation, even when multiple views 3. Haynes, J.-D. & Rees, G. Nat. Neurosci. 8, 686–691 (2005).
of objects are included or when line drawings without much detail are 4. Kamitani, Y. & Tong, F. Nat. Neurosci. 8, 679–685 (2005)
used5. This kind of information would be difficult to uncover using a 5. O’Toole, A.J., Jiang, F., Abdi, H. & Haxby, J.V. J. Cogn. Neurosci. 17, 580–590
conventional analysis technique. James Haxby of Princeton University 6. Polyn, S.M., Natu, V.S., Cohen, J.D. & Norman, K.A. Science 310, 1963–1966
says, “I find myself working with a whole new community of people: (2005).


Neurons that know when to quit
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

Jeremy M Wolfe
With sufficient training, monkeys as well as people can be taught to ignore visually salient stimuli. Now Ipata and
colleagues report that activity in monkey lateral intraparietal cortex (LIP) correlates with this ability to ignore
salient stimuli, suggesting that activity in this area represents top-down modulation that adjusts visual salience.

In Euripides’ play Helen, a servant to the king
of Egypt says, “Man’s most valuable trait is a
judicious sense of what not to believe.” He
wasn’t talking about visual attention at the
time, but an ability to ignore misleading
information is vital to the successful
deployment of spatial attention. In this issue,
Ipata and colleagues1 report on activity in
neurons of the lateral interparietal area
(LIP) of macaque cortex that corresponds to
this ability. They taught macaque monkeys
to identify the orientation of a T in a field
containing a number of crosses. The monkey
needed to find the T and then make a response
indicating its orientation. Critically, for this
experiment, the visual displays contained a
uniquely colored, attention-grabbing item
that was never the T. The visual system is
built to respond strongly to salient colored
stimuli. Such color singletons are said to ‘pop
out’ of a display2. All else being equal, they Figure 1 That red A grabs your attention, but does it have to? Cells in macaque lateral interparietal area
will summon attention. However, because the can suppress their normal response to a salient stimulus like this if it is misleading.
uniquely colored item was never the target,
Ipata’s judicious monkeys learned not to
believe that item when it screamed, “Attend to odd item on nearly every trial. When the rules Figure 1. The visual system cannot fully pro-
me,” and the monkeys’ LIP neurons learned were changed so that the singleton was never cess everything everywhere in a single step.
to actively suppress their response to what the target, the number of saccades to that item Visual selective attention is part of the response
would have otherwise been a very stimulating dropped dramatically. Like human observers, to this computational reality. If you want to
stimulus. After training, responses to the monkeys were not perfectly consistent in their identify letters, attention needs to be deployed
pop-out singleton were actually lower than behavior, and this provided an added piece of to one or perhaps a few letters at a time. That
responses to other distractors in the display. evidence that the behavior of LIP neurons was deployment is controlled by a combination of
Ipata et al. measured the tendency of the related to the behavior of the whole organism. top-down and bottom-up factors.
color singleton to grab attention by monitor- Sometimes the monkeys returned to making On first glance, that red A probably grabbed
ing the monkeys’ eye movements. In training, frequent saccades to the singleton even though your attention. Bottom-up, stimulus-driven
when the color oddball could be the target, it was never the target. During those periods, mechanisms guide attention to items that are
monkeys made a saccadic eye movement to that their LIP neurons again showed a stronger dramatically different from their neighbors3. If
response to the singleton than to other stimuli. bottom-up salience were the end of the story, you
This was true even when the monkey made a would be the slave of the stimulus. Instead, you
Jeremy M. Wolfe is at the Brigham and Women’s saccade away from the singleton, showing that are able to impose top-down, user-driven control,
Hospital Visual Attention Laboratory, Harvard the strong response was not related to the plan- allowing you to modify your impression of this
University, 64 Sidney Street, Suite 170, Cambridge, ning or execution of the eye movement. image without changing the stimulus itself. For
Massachusetts 02139, USA. To see why this ability to suppress response example, attend to the blue items. As you do this,
e-mail: wolfe@search.bwh.harvard.edu to a salient stimulus is important, look at you may notice that the red A, while not vanish-


ing, makes less of a demand on your attention. command will give you the entire set of Es in seem to lend support to the nonmandatory side,
The blues, in contrast, now seem more salient. the same, seemingly effortless manner in which to the argument that it is possible to put in place
You may now notice that they form a rough circle. the blue or black sets were delivered to you. a top-down filter that allows you to successful
Moreover, having somehow taken note of the So, salient stimuli will grab your attention ignore a salient stimulus. However, as noted
whole set, you can now direct your attention to in a bottom-up manner, but top-down control above, there are stimuli that may have a greater
individual blue items and notice that they consist allows you to avoid perseverating on the one ability to capture attention than the color
of the letters B through G in clockwise order. or two hot spots in a scene. Are there stimuli singletons used in this study. Both top-down
Notice that there are two steps involved that are so salient that they will grab your and bottom-up components of attentional
here. The top-down command to favor blue attention regardless of any top-down desires or guidance are powerful. Sufficiently strong
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

items changes the effective salience of all blue commands? This is the problem of attentional task demands can produce attentional ‘tunnel
items. Object identification mechanisms can capture. Fortunately, the answer to the strongest vision’ where everything but the attended
then select individual items within the blue form of this question is almost undoubtedly item seems blocked out9,10 but, in the end, the
set. The Ipata results pertain to the first step. “no.” If there were a visual stimulus so salient 19th century writer Sully had a fair point when
Response to the color singleton is suppressed that it would grab attention and never let go, he said, “One would like to know the fortunate
so that it is almost never a candidate for the advertisers would have us glued to it. The (or unfortunate) man who could receive a box
second selective step. The distinction can more reasonable question is whether there are on the ear and not attend to it”11.
be illustrated by returning to the figure to stimuli that attract attention in a mandatory 1. Ipata, A., Gee, A.L., Gottlieb, J., Bisley, J.W. & Goldberg,
search for the S and for the black P. You will fashion even if we are subsequently released to M.E. Nat. Neurosci. 9, 1071–1076 (2006).
2. Neisser, U. Am. J. Psychol. 76, 376–385 (1963).
probably find that you are faster to find the attend elsewhere. A variety of stimuli have been
3. Nothdurft, H.-C. Vision Res. 33, 1937–1958 (1993).
black P because that first, guiding step can be suggested to have the ability to capture attention 4. Egeth, H.E., Virzi, R.A. & Garbart, H. J. Exp. Psychol.
used to eliminate letters that are not black4. in this way. For example, luminance onsets have Hum. Percept. Perform. 10, 32–39 (1984).
5. Wolfe, J.M. & Horowitz, T.S. Nat. Rev. Neurosci. 5,
If the first step were sufficiently powerful, the proven more successful than our color singleton 495–501 (2004).
second step would not be necessary. Thus, in the in forcing attentional deployment. It may not 6. Yantis, S. & Jonides, J. J. Exp. Psychol. Hum. Percept.
Ipata task, if the monkeys could just teach LIP to be simple luminance but the appearance of a Perform. 22, 1505–1513 (1996).
7. Franconeri, S.L., Hollingworth, A. & Simons, D.J.
suppress all of the crosses, then attention would new object that is particularly effective6, though Psychol. Sci. 16, 275–281 (2005).
be deployed to the T, first time, every time. That others disagree7. Moreover, it is possible that 8. Folk, C.L., Remington, R.W. & Wright, J.H. J. Exp.
Psychol. Hum. Percept. Perform. 20, 317–329
will not happen because the abilities of that first attentional capture occurs only if you are in
stage are quite limited5. Returning again to the a state where you are, in a sense, ready to be 9. Williams, L.J. Hum. Factors 27, 221–227 (1985).
figure, are any of the Es mirror-reversed? Of captured (so-called ‘contingent capture’)8. 10. Lavie, N. & Tsal, Y. Percept. Psychophys. 56, 183–197
course, you can find the one mirror-reversed Opinion varies about the mandatory nature 11. Sully, J. The Human Mind: A Text-book of Psychology (D.
E, but you will need to search. No top-down of attentional capture. The Ipata results would Appleton & Co., New York, 1892).

BK channels and circadian output
Christopher S Colwell
The suprachiasmatic nucleus (SCN) controls circadian behavior, and neurons in the SCN are intrinsic oscillators.
Meredith et al. now identify the BK potassium channel as a key modulator of spontaneous firing of the SCN.

You are lying in your bed, staring at the ceiling, few days while we adjust to new time zones? New for generation of the molecular oscillations.
waiting for sleep to come your way and wondering research in this issue by Meredith and colleagues For example, disruption of electrical activity
why it has become so difficult to find. Among into the ionic mechanisms underlying circadian with tetrodotoxin (TTX) damps molecular
other reasons, these sleepless nights are caused by oscillations may well open up the prospect for circadian rhythms of mPer1 levels in SCN
our circadian timing system turning on arousal such manipulations in the future1. tissue2. A similar loss of function at the molec-
centers in our brain at inappropriate times. This Humans and other organisms have daily ular level is observed in mice deficient in
problem may be particularly common during the rhythms in their behavior and physiology. receptors for the neuropeptide transmitter
summer travel season when we are jetting off to In mammals, the part of the nervous system vasoactive intestinal polypeptide3. Thus,
attend a conference or perhaps to catch a World responsible for most circadian behavior is a clarifying the ionic mechanisms responsible
Cup match. Would it not be nice to be able to turn bilaterally paired structure in the hypothalamus, for the generation of rhythms in electrical
down this biological timing signal, if only for a the SCN. Many neurons in the SCN are intrinsic activity in SCN neurons is an important step
oscillators that continue to generate near 24- for understanding the generation and output of
hour rhythms in electrical activity, secretion and circadian oscillations.
The author is at the Laboratory of Circadian gene expression when isolated from the rest of The new study by Meredith and colleagues
Neurobiology, Psychiatry and Biobehavioral Sciences, the organism. Individual SCN neurons contain takes a step in this direction by examining
University of California Los Angeles, 760 Westwood a molecular feedback loop that drives these the role of the large-conductance calcium-
Plaza, Los Angeles, California 90024-1759, USA. rhythms. However, membrane excitability and/ activated potassium (BK) channels in circadian
e-mail: ccolwell@mednet.ucla.edu or synaptic transmission may also be required behavior1. With this work, the authors have


identified a specific channel that seems to be Figure 1 Working model of the bimodal state
responsible for the output of the SCN without of SCN neurons. The dSCN neurons exist in
Day Night
an obvious role in other aspects of the circa- two stable electrical configurations. During the
upstate (day), the cells show spontaneous activity;
dian system. The BK channels seem not to be during the downstate (night), the cells are silent.
important for the generation of daily rhythms A set of currents are responsible for each of
or for their synchronization by light. these two states. At night, the SCN neurons are
To explore the role of this channel in circadian hyperpolarized due to a net increase in potassium
conductance. The BK current may be responsible Prototypical
function, Meredith and colleagues took advan- dSCN neuron
tage of mice carrying a deletion of the Kcnma1 for this nightly silencing of SCN neurons.
(Slo−/−) gene that encodes the core-forming
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

subunit of the BK channel. The authors found Vm Depolarized Hyperpolarized
that BK-deficient mice have lower amplitude different between mutant and wild-type
circadian rhythms in wheel running and home hypothalamus. The only transcript that was Rin

cage activity as well as in core body temperature. consistently different between genotypes was a
The mice still showed circadian rhythms, but the transcript of the Kcnma1 gene itself. Notably, a gK
magnitude of the day/night variation was greatly more detailed analysis of Arntl (Bmal1) expres-

Katie Ris
reduced. Notably, the frequency (or period) of sion by in situ hybridization indicates that the BK
the biological rhythm did not seem to be altered rhythm in expression of this clock gene was
by the loss of the BK channels. Furthermore, the not altered by the loss of the BK channel. So
light response of the circadian system as mea- the BK channel does not seem to be important this channel on the daily rhythms in membrane
sured by the onset of the activity rhythms was in the rhythmic expression of clock genes that potential and conductance that characterize
not altered in the mutants. This analysis suggests lie at the heart of the generation of circadian SCN neurons. Future studies will need to go
that the output of the circadian system is selec- oscillations. Overall, the authors conclude that beyond demonstrating a correlation between the
tively damped by the loss of the BK channels. BK channels are essential for the amplitude or decrease in amplitude in the electrical rhythms
Previous work used oligonucleotide arrays ‘robustness’ of circadian rhythms. of SCN neurons and the downstream behavior.
to show that transcription of the Kcnma1 gene This set of experiments adds to our growing There also remain some tricky questions about
cycles in SCN tissue4. Here the authors deter- understanding of the ionic mechanisms respon- the mechanisms underlying how the molecular
mined that the BK protein is rhythmically sible for the generation of rhythms in electrical clock drives the rhythms in the BK protein and
expressed in the SCN of mice maintained in con- activity in SCN neurons. It is likely that a set of current. Another puzzle is that SCN neurons
stant conditions—that is, the levels of the protein intrinsic voltage-sensitive currents are respon- have circadian rhythms in calcium levels that
vary with the circadian cycle. The peak expres- sible for the circadian variation in firing rate in peak during the day13,14, whereas the calcium-
sion of the BK channel is during the night, when SCN neurons (Fig. 1). One possibility is that sensitive BK currents peak at night. Nonetheless,
the magnitude of the BK current also peaks5. these cells fluctuate between two subthresh- these studies are beginning to provide the infor-
Functionally, the pharmacological blockage of old values, a hyperpolarized downstate and a mation needed to understand the physiological
BK channels decreases the current flow that depolarized upstate. During the upstate (day), underpinnings of the generation of spontaneous
occurs between action potentials6 and decreases neurons are spontaneously active and can fire firing in SCN neurons. From a circadian perspec-
the amplitude of the day/night difference in action potentials. The largest source of the exci- tive, it is important not only to identify those
firing rate in SCN neurons5. Meredith et al. tation driving the cell to generate spontaneous currents that regulate spontaneous firing but also
explored the electrical activity of the SCN in the action potentials seems to be TTX-sensitive Na+ to understand how those currents change from
mutant mice. One of the characteristic features currents that open at subthreshold voltages9. day to night to drive the neural activity rhythm.
of neurons in the SCN is that they show circa- The amplitude of the L-type Ca2+ current also With some luck, this work will ultimately allow
dian rhythms in electrical discharge with more shows a diurnal rhythm in the SCN and may the manipulation of our circadian timing system,
action potentials generated in the day than in the well contribute to the excitatory drive during and fewer sleepless nights.
night. The peak rate of action potential genera- the day10. In addition, SCN neurons express 1. Meredith, A.L. et al. Nat. Neurosci. 9, 1041–1049
tion during the day was not altered in the BK- a K+ current known as the fast delayed recti- (2006).
2. Yamaguchi, S. et al. Science 302, 1408–1412
deficient mice; however, the SCN neurons from fier (fDR). This current is active in the times
the mutant mice showed more electrical activ- between the generation of action potentials dur- 3. Maywood, E.S. et al. Curr. Biol. 16, 599–605 (2006).
ity during the night. Given the close association ing the day. In the SCN, the fDR is thought to be 4. Panda, S. et al. Cell 109, 307–320 (2002).
5. Pitts, G.R., Ohta, H. & McMahon, D.G. Brain Res. 1071,
between SCN electrical activity and rhythmic critical for translating membrane depolarization 54–62 (2006).
behavior, the lower-amplitude rhythms in the into a regular pattern of action potential firing11. 6. Cloues, R.K. & Sather, W.A. J. Neurosci. 23, 1593–1604
SCN may be responsible for the reduction in the During the downstate (night), the SCN neurons (2003).
7. Nitabach, M.N., Blau, J. & Holmes, T.C. Cell 109,
robustness of the behavioral rhythmicity. are electively inactive due to a membrane hyper- 485–495 (2002).
Finally, the authors looked at gene expres- polarization associated with an increase in total 8. Lundkvist, G.B., Kwak, Y., Davis, E.K., Tei, H. & Block,
G.D. J. Neurosci. 25, 762–766 (2005).
sion in the SCN of the mutant mice. A few years membrane conductance12. The nightly upregu-
9. Jackson, A.C., Yao, G.L. & Bean, B.P. J. Neurosci. 24,
ago, this would have seemed like a superfluous lation of the BK current may be essential for the 7985–7998 (2004).
experiment. However, work in Drosophila7 and inactivity of SCN neurons during the night. 10. Pennartz, C.M., de Jeu, M.T., Bos, N.P., Schaap, J. &
Geurtsen, A.M. Nature 416, 286–290 (2002).
mice8 suggests that treatments that alter the Of course, like many good scientific studies, 11. Itri, J.N., Michel, S., Vansteensel, M.J., Meijer, J.H. &
membrane potential can disrupt the rhythms in this new work1 raises more questions than it Colwell, C.S. Nat. Neurosci. 8, 650–656 (2005).
the expression of clock genes such as Period. In answers. It would have been good to see actual 12. Kuhlman, S.J. & McMahon, D.G. Eur. J. Neurosci. 20,
1113–1117 (2004).
this case, microarray data indicates that overall measurements of the BK currents in SCN neu- 13. Colwell, C.S. Eur. J. Neurosci. 12, 571–576 (2000).
gene expression profiles were not systemically rons and to determine the impact of the loss of 14. Ikeda, M. et al. Neuron 38, 253–263 (2003).


Choice values
Yael Niv, Nathaniel D Daw & Peter Dayan
Dopaminergic neurons are thought to inform decisions by reporting errors in reward prediction. A new study reports
dopaminergic responses as monkeys make choices, supporting one computational theory of appetitive learning.
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

A group of neurophysiologists, computer
scientists, psychologists and economists
has found common ground in trumpeting
a detailed hypothesis about the involvement
of the brain’s dopaminergic system in rein-
forcement learning1, which is the process of
learning by trial and error to predict rewards
(and punishments) and to make good deci-
sions. This hypothesis is grounded in an
impressive body of experimental evidence,
notably recordings showing that dopami-
nergic neurons in behaving primates seem
to carry an error signal that could be useful
for learning to predict rewards and also to
choose rewarding actions2,3. This presumed
central role in appetitive choice aligns well
with dopamine’s involvement in synaptic
plasticity, learned habits, drug addiction and
various pathologies4.

Katie Ris
Maddeningly, however, the detailed
neurophysiological evidence concerns pre-
diction learning alone, leaving the story
about decision making bereft of empirical
guidance. This is because dopaminergic
Figure 1 A monkey is presented with a choice between two options (decision trial). Each option is
neurons have rarely been recorded when
associated with a different probability of reward, so the overall worth of this choice can be evaluated in
animals are making nontrivial choices several different ways (thought balloons; see text). These different evaluations would lead to different
between multiple rewarded options (that prediction error signals associated with different reinforcement learning rules. The recordings of Morris et
is, but for a few exceptions5, doing much al. from dopamine neurons, putatively reporting these errors, show that the neural responses quickly reflect
more than following instructions or pas- the distal future choice of the monkey (left and right traces; data from ref. 6). This supports the SARSA
sively receiving rewards). An experiment algorithm for prediction learning and action selection (bottom) against value learning and Q learning.
reported by Morris and colleagues in this
issue6 fills this gap and provides surprising
and detailed information about the compu- as the decision trials were embedded sparsely richer and poorer options in rough propor-
tations underlying decision making. among a set of ‘reference trials’, in which the tion to their relative worth. This behavior
In the experiment, four different images monkeys were presented with single images was fortuitous because it allowed Morris
were associated with fixed probabilities and responded for the associated chance of et al. to record the activity of dopamine
(0.25, 0.5, 0.75 and 1) of obtaining juice (or receiving juice. As in most previous dopa- neurons on decision trials in which the mon-
water) reward. In ‘decision trials’, monkeys mine recording studies, reference trials did key ultimately chose either the richer or the
saw pairs of these images and, after a couple not present the monkeys with a meaningful poorer option. The comparison of one with
of seconds, had to choose between them. choice, and the particular prediction error the other, and with the firing patterns from
Reward was given, or not, according to the reported by dopamine firing on presenta- the single-image reference trials, provides a
probability attached to the chosen image. tion of an image simply corresponds to its window onto the decision process.
The monkeys knew these probabilities well, associated reward probability7. The central finding is that a burst of
Faced with a decision trial, one might dopaminergic responding at the outset of
expect the monkeys to choose the richer a decision trial nearly instantly reflects the
The authors are at the Gatsby Computational option after pondering the pair for a while. average reward associated with the option
Neuroscience Unit, 17 Queen Square, Neither of these expectations was satisfied. that will ultimately be chosen, even though
London WC1N 3AR, UK. Yael Niv is also at the Fortunately for science, but unfortunately the monkey cannot actually submit its deci-
Interdisciplinary Center for Neural Computation, for themselves, the monkeys adopted the sion until some seconds later. Indeed, the
Hebrew University, Jerusalem 91904, Israel. (surprisingly common) suboptimal choice neural response to the presentation of a pair
e-mail: dayan@gatsby.ucl.ac.uk strategy of ‘probability matching’, pursuing of images is nearly the same in (average)


time course and magnitude as the response The value of taking a particular action at a reinforcement learning perspective, it is not
on reference trials to the presentation of only state is called a Q value11, and the two remain- straightforward to expect that the monkey
the image the animal will choose. Within just ing ways to evaluate a decision trial can both will represent the state of a pair of images
200 ms, the monkey has evidently made a be used to learn Q values. The more popular as simply the conjunction of the two single-
decision and communicated it to the dopa- approach is Q learning11, in which the predic- image states. Fourth, it is now pressing to
minergic cells. tion error associated with a decision (which is work out a SARSA-like algorithm that also
The beauty of this result is that it lays what the dopamine cells report) is determined respects the anatomical data on the dual
waste to a crowded field of computational by the Q value of the better option rather than dopamine and striatal systems that helped
ideas about appetitive choice; under a the one actually chosen (Fig. 1, bottom). This motivate the actor-critic model. A relative of
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

friendly interpretation, it leaves just one is a very clever idea, as it decouples learning the actor-critic algorithm called ‘advantage
survivor. An immediate casualty is the idea from the actual choice and allows optimal learning’, which has found some support in
that the dopamine signal might be directly behavior to be acquired while exploring sub- human functional magnetic resonance imag-
involved in selecting actions8—instead, the optimal alternatives. However, going by the ing (fMRI) studies of learned choice10, seems
firing apparently reflects a choice already data, this is evidently too clever for the dopa- not to do the trick, but a variant might.
made. Most other accounts of reinforce- mine cells, whose activity follows the reference Finally, the monkeys in this study were vastly
ment learning assume that dopaminergic activity for the action actually chosen. The overtrained, which allowed for careful study
responses affect decisions only indirectly, by remaining option is the class of algorithms under uniform conditions, but at the cost of
controlling learning. All the accounts agree that acquire Q values using a prediction error ensuring that no learning occurred during
that the dopaminergic prediction error at the that reflects the value of the chosen option. the experiment. The recorded reinforcement
start of a decision trial reports an evaluation It is these so-called SARSA (state-action- learning error signals were therefore appar-
of the overall value (predicted reward) of the reward-state-action) algorithms12 (Fig. 1) ently epiphenomenal. In studies of similar
trial. However, their substantially different that this study favors. decision-making tasks, animals have been
approaches to learned choice are reflected in In sum, the most natural conclusion from exposed to more complex and changing
subtly different ways of assessing this value, the neural data is that dopamine signals reward contingencies and continually updated
which the results of Morris et al. exactly test. report prediction errors based on Q values their behavior in light of received rewards14,15.
There are three main possibilities (Fig. 1) for for SARSA learning. A choice can be made An obvious future direction is to understand
what the value of a decision trial is, in terms between actions by favoring (perhaps still how such behavioral changes relate, trial by
of the values of the two images it comprises. subject to randomness) the one with a larger trial, to recorded dopamine responses. The
It could be (i) the average of the values of the Q value. This would account for the animal’s experiment of Morris et al. sets the stage for a
two options, weighted according to the prob- fortuitous but flawed probability matching new multidisciplinary enterprise of such stud-
ability that each would be chosen, (ii) the behavior. The behavior itself is both infor- ies of dopamine and decisions.
value of the better option, or (iii) the value mative and surprising. Under methods such
of the image that is actually chosen on that as the actor-critic, persistent performance of 1. Sutton, R.S. & Barto, A.G. Reinforcement
Learning: An Introduction (MIT Press, Cambridge,
trial. The data favor the last possibility. a suboptimal action is not possible. That it Massachusetts, 1998).
The first option—that the values are aver- happens is additional evidence that choice 2. Houk, J.C., Adams, J.L. & Barto, A.G. in Models of
aged over the choices (V in Fig. 1)—would is based on Q values. However, it is odd that Information Processing in the Basal Ganglia (eds.
Houk, J.C., Davis, J.L. & Beiser, D.G.) Ch. 13,
have been expected under the so-called the monkeys seemed never to adjust their 249–270 (MIT Press, Cambridge, Massachusetts,
actor-critic algorithm, which posits that a behavior toward exclusive choice of the 1995).
3. Schultz, W., Dayan, P. & Montague, P.R. Science
‘critic’ with no knowledge of the actions can richer option. The common rationale for 275, 1593–1599 (1997).
track the average value of situations (called occasional suboptimal choices is to allow for 4. Wise, R.A. Nat. Rev. Neurosci. 5, 483–495
‘states’); these values can, separately, be used exploration of unfamiliar alternatives, but no (2004).
5. Bayer, H.M. & Glimcher, P.W. Neuron 47, 129–141
as rewards to train an ‘actor’ that makes such exploration was necessary here because (2005).
choices. This notion mirrors venerable the images’ values were stable over weeks of 6. Morris, G., Nevet, A., Arkadir, E. & Bergman, H. Nat.
ideas from psychology about the interac- recording and anyway sampled extensively Neurosci. 9, 1057–1063 (2006).
7. Morris, G., Arkadir, D., Nevet, A., Vaadia, E. &
tion between reward prediction (Pavlovian during reference trials. Bergman, H. Neuron 43, 133–143 (2004).
conditioning in the critic) and action choice As with any illuminating result, many 8. McClure, S.M., Daw, N.D. & Montague, P.R. Trends
Neurosci. 26, 423–428 (2003).
(instrumental conditioning in the actor)9, open issues and interesting implications 9. Dickinson, A. Contemporary Animal Learning Theory
and seems nicely to parallel the anatomi- remain. First, even though dopamine seems (Cambridge Univ. Press, New York, 1980).
cal division of the dopamine system and its not to be involved directly in the choice 10. O’Doherty, J. et al. Science 304, 452–454 (2004).
11. Watkins, C. Learning from Delayed Rewards. Thesis,
targets into ventral (evaluation) and dorsal between options, it may influence other Cambridge Univ. (1989).
(action) components10. The central trick of aspects of the selected action, such as the 12. Rummery, G.A. & Niranjan, M. in Technical Report
the actor-critic algorithm is how it learns vigor with which it is executed13. Second, CUED/F-INENG/TR 166 (Engineering Department,
Cambridge Univ., 1994).
to choose actions using reward predictions dopaminergic responses during decision 13. Niv, Y., Daw, N.D. & Dayan, P. in Advances in Neural
that ignore actions altogether, instead aver- trials, though similar on average to those Information Processing Systems (NIPS) Vol. 18 (eds.
Weiss, Y., Schölkopf, B. & Platt, J.) 1019–1026 (MIT
aging over them. However, the data of Morris on reference trials, were nevertheless much Press, Cambridge, Massachusetts, 2005).
et al. rule out this trick and show that the more variable. Structure in the signal may 14. Lau, B. & Glimcher, P.W. J. Exp. Anal. Behav. 84,
dopamine signal instead incorporates richer still remain undiscovered, perhaps including 555–579 (2005).
15. Corrado, G.S., Sugrue, L.P., Seung, H.S. &
information, separately reporting the value evidence of the monkey changing its mind Newsome, W.T. J. Exp. Anal. Behav. 84, 581–617
of choosing either action at a state. during the waiting period. Third, from a (2005).


Seeking a function for spontaneous neurotransmission
ChiHye Chung & Ege T Kavalali
A recent study proposes that the random and spontaneous, NMDA receptor–dependent miniature postsynaptic currents
at hippocampal synapses serve to regulate local postsynaptic protein synthesis, thereby stabilizing synaptic function.
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

Spontaneous neurotransmitter release is
a common feature of synapses through-
out the nervous system, but does it serve
a purpose? Since Fatt and Katz discovered
neurotransmission in the absence of nerve
impulses1, this question has been in the
minds of neurophysiologists. Spontaneous
neurotransmission is widely studied, though
most often as a simpler proxy for the more
complicated action potential–driven syn-
chronized release of neurotransmitters. A
few studies have examined spontaneous
neurotransmission for its own sake and have
shown that spontaneous release events may
trigger action potential firing in cells with
high membrane resistance and may also be
required for the maturation of synapses2.

Katie Ris
Another mysterious phenomenon that has
puzzled neurophysiologists for over a cen-
tury is the hyperexcitability of target mem-
Figure 1 Block of resting NMDA receptor activity activates mRNA translation machinery and
branes that follows the denervation or other
increases synaptic efficacy. (a) Activation of postsynaptic NMDA receptors suppresses dendritic
disruption of their nervous input3. In the mRNA translation. (b) Inhibition of postsynaptic NMDA receptors at rest overcomes the suppression
neuromuscular junction, the increase in sen- of dendritic mRNA translation and triggers recruitment of newly synthesized GluR1 subtype of
sitivity of muscle tissue to acetylcholine seen AMPA receptors to the postsynaptic sites. This leads to an increase in unitary postsynaptic response
after denervation is due to the upregulation amplitudes as well as a transient augmentation in the number of GluR2-lacking AMPA receptors.
of acetycholine receptors4. A similar recep-
tor upregulation occurs at central synapses,
revealing a powerful mechanism for the
maintenance of homeostatic stability of the dendritic protein translation machinery tetrodotoxin leads to slow rescaling of uni-
CNS synaptic networks 5,6. Furthermore, locally and thereby maintaining receptor tary synaptic efficacy6. Moreover, the authors
chronic blockade of action potential firing composition at synapses. show that this rapid effect of NMDA receptor
in neuronal cultures increases trafficking The initial set of experiments of Sutton blockade on unitary transmission is strictly
of the AMPA receptor subunits GluR1 and et al. documented that, unlike the blockade dependent on protein synthesis, consistent
GluR2 to postsynaptic sites, thus increasing of action potentials, inhibition of either with recent findings from the same group9.
sensitivity to released glutamate7. NMDA receptors or AMPA receptors can To test whether the increase in the
A new study in Cell by Sutton and col- increase the amplitude of miniature excit- amplitudes of mEPSCs is due to increased
leagues8 bridges these two persistent questions atory postsynaptic currents (‘minis’ or mEP- cell surface expression of AMPA receptors,
of neurophysiology through a comprehensive SCs) within mere hours. The frequency of the authors used a biotinylation assay, and
set of experiments and provides a causal link minis remains unchanged, and the increased detected an increase in GluR1 but not GluR2
between the two phenomena. The authors mini amplitude is seen even when action subunits at the cell surface. To visualize the
show that spontaneous neurotransmitter potentials are allowed during receptor block- spatial distribution of these newly recruited
release, rather than evoked neurotransmission, ade. This finding has two surprising aspects. GluR1 subunits, they labeled surface recep-
is a specific regulator of postsynaptic sensi- First, it strongly suggests that NMDA recep- tors in live cultures with specific antibod-
tivity to neurotransmitters—by suppressing tors are active at rest during spontaneous ies and also immunostained for synapse
neurotransmission, despite their reduced markers. These experiments showed that
ion conductance due to Mg2+ block. Second, the increase in GluR1 subunits takes place
The authors are at the Center for Basic the observation that amplitudes of unitary at synapses. Furthermore, whereas blockade
Neuroscience and the Department of Physiology, synaptic responses increase rapidly within of NMDA receptors increased both the size
University of Texas Southwestern Medical Center, an hour after NMDA receptor blockade of individual synaptic GluR1 clusters and the
Dallas, Texas 75390-9111, USA. stands in striking contrast to earlier reports numbers of synapses expressing GluR1, it did
e-mail: ege.kavalali@utsouthwestern.edu that chronic blockade of neuronal firing by not affect the expression levels of other syn-


aptic markers. The authors also did a techni- might function as a ‘marker’ or a ‘tag’ for in AMPA receptor activity. One possibility
cally demanding set of experiments using a synaptic modification. After induction of long- is that synaptic scaling in response to
dual micropipette local perfusion system to term potentiation as well as during homeostatic chronic action potential blockade does
inhibit NMDA receptors focally; these exper- plasticity, GluR2-deficient receptors are not involve the transient expression of
iments revealed selective surface expression gradually replaced with GluR2-containing Ca 2+-permeable AMPA receptors 7. The
of GluR1 subunits near the inhibition site ones, albeit with significantly different time exact mechanism mediating synaptic scal-
but not in dendritic regions away from this courses. In a previous study, GluR2-lacking ing induced by chronic tetrodotoxin treat-
site. Thus, the increase of GluR1 expression AMPA receptors were detectable only for ment remains unknown. It is thought to
after NMDA receptor block is under tight 25 minutes11, whereas in the study of Sutton act more globally14, which is consistent
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

temporal and spatial regulation. et al., the replacement of GluR2-lacking with the idea of large-scale homeostatic
An overall increase in surface expression receptors by GluR2-containing ones takes place maintenance of synaptic circuits. If chronic
in GluR1 subunits but not GluR2 subunits over 24 hours8. The mechanisms underlying blockade of action potential firing also medi-
suggests a concomitant increase of synaptic this differential rate of receptor replacement ates its effect synaptically but by activating
GluR2- lacking receptors, which are after distinct forms of plasticity will be the a different signaling cascade, then how do
incidentally permeable to Ca 2+ (ref. 10). interesting subject of future study. neurons distinguish evoked and spontaneous
To characterize the subunit composition How do AMPA and NMDA receptors neurotransmission? Are the differences in
of AMPA receptors electrophysiologically, cross-talk at individual synapses? The study temporal characteristics of the two forms of
the authors used a selective inhibitor by Sutton and colleagues, as well as earlier neurotransmission sufficient? Or is there a
of GluR2- lacking AMPA receptors, work12, showed that blocking AMPA recep- distinct postsynaptic detection machinery for
1- napthylacetylspermine (Naspm). tors also induces synaptic rescaling. This spontaneously released neurotransmitters?
Once synapses were scaled up after the finding raises the question of whether block- Spontaneous fusion events may not origi-
application of the NMDA receptor blocker ade of AMPA and NMDA receptors share the nate from the same pool of vesicles as evoked
D (-)-2-amino-5-phosphonovaleric acid same downstream mechanisms. Does spon- neurotransmission15. Better understanding
(AP5), the increased amplitude of mEPSCs taneous release rate per synapse determine of the presynaptic machinery and its postsyn-
was decreased upon application of Naspm, the unitary response amplitudes in a linear aptic counterparts that underlie spontaneous
suggesting a rise in the contribution of way? If so, does the receptor composition of and evoked neurotransmission will provide
GluR2-lacking receptors after AP5 treatment. a particular synapse depend on its history us with molecular and pharmacological
The authors also used fluorescence imaging of spontaneous transmission? Or is sponta- tools that can selectively manipulate the two
to visualize the increase in GluR2-lacking, neous transmission merely permissive, such forms of neurotransmission. Independent
and thus Ca2+-permeable, AMPA receptors that above a certain threshold of sponta- analysis of spontaneous and evoked
on dendrites. For this purpose, they took neous release, it can suppress translational neurotransmission may uncover more sur-
advantage of the Co 2+ permeability of machinery? Future identification of the sig- prises in the intricacies of communication
Ca2+-permeable AMPA receptors, a unique naling pathway connecting NMDA receptors within individual synapses.
feature among Ca2+-permeable channels. to protein synthesis machinery and recruit-
Co2+ influx can be detected as a decrease ment of GluR1 subunits may provide some 1. Fatt, P. & Katz, B. J. Physiol. (Lond.) 117, 109–128
in calcein fluorescence, as Co2+ acts as a answers to these questions. In addition to the 2. Zucker, R.S. Neuron 45, 482–484 (2005).
potent quencher. These measurements cross-talk in downstream signaling, it is pos- 3. Cannon, W.B. & Rosenblueth, A. The Supersensitivity
confirmed the increase of GluR2-lacking sible that AMPA receptor activity may aug- of Denervated Structures: A Law of Denervation
(Macmillan, New York, 1949).
AMPA receptors after prolonged focal ment NMDA receptor activity at rest through 4. Axelsson, J. & Thesleff, S. J. Physiol. (Lond.) 147,
AP5 treatment and helped the authors electrical means. Such interaction between 178–193 (1959).
bridge their electrophysiology and the activation of the two types of receptors 5. O’Brien, R.J. et al. Neuron 21, 1067–1078
immunocytochemistry results. Application may occur if at least some dendritic spines 6. Turrigiano, G.G., Leslie, K.R., Desai, N.S.,
of the protein synthesis inhibitor anisomycin comprise electrically isolated compartments Rutherford, L.C. & Nelson, S.B. Nature 391, 892–
896 (1998).
together with AP5 also impaired the due to high spine neck resistance13. In such 7. Wierenga, C.J., Ibata, K. & Turrigiano, G.G. J.
increase in GluR2-lacking AMPA receptors, circumstances, activation of AMPA receptors Neurosci. 25, 2895–2905 (2005).
corroborating the authors’ earlier results may result in sufficient local depolarization 8. Sutton, M.A. et al. Cell 125, 785–799 (2006).
9. Sutton, M.A., Wall, N.R., Aakalu, G.N. & Schuman,
on local protein synthesis dependence of to facilitate the relief of adjacent NMDA E.M. Science 304, 1979–1983 (2004).
synaptic scaling8 (Fig. 1). receptors from Mg2+ block. 10. Hollmann, M., Hartley, M. & Heinemann, S. Science
Transient recruitment of GluR2-lacking An intriguing implication of this work is 252, 851–853 (1991).
11. Plant, K. et al. Nat. Neurosci. 9, 602–604
AMPA receptors also occurs shortly after the the divergence of mechanisms underlying (2006).
induction of long-term potentiation11. GluR2- synaptic scaling after chronic inhibition 12. Thiagarajan, T.C., Lindskog, M. & Tsien, R.W. Neuron
47, 725–737 (2005).
lacking AMPA receptors seem to serve as of NMDA receptor–mediated miniature 13. Bloodgood, B.L. & Sabatini, B.L. Science 310,
pioneers for changes in postsynaptic efficacy currents versus chronic inhibition of action 866–869 (2005).
and may initiate the synthesis or recruitment potentials. The difference in the time course 14. Turrigiano, G.G. & Nelson, S.B. Nat. Rev. Neurosci.
5, 97–107 (2004).
of additional factors for maintenance of the of action of the two maneuvers argues for 15. Sara, Y., Virmani, T., Deak, F., Liu, X. & Kavalali, E.T.
plasticity. Thus GluR2-lacking AMPA receptors separate mechanisms mediating the increase Neuron 45, 563–573 (2005).


What’s in control of language?
Angela D Friederici
Language functions are thought to be controlled largely by cortical areas. A study now finds that the subcortical caudate
nucleus is sensitive to language change in bilingual speakers, suggesting a role for this area in control processes.
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

Anyone who has learned a second language
late in life knows how frustrating it is to search
for a particular word in still-foreign language,
especially when that word is easily remem-
bered in one’s native language. Bilinguals,
those lucky people who grew up with two lan-
guages or learned their second language early
in life, do not have this problem. Bilinguals
have easy access to words in both their lan-
guages, and they can switch between languages
without difficulty.
In a recent study in Science1, Crinion and
colleagues used functional neuroimaging to
investigate how bilinguals accomplish this lin-
guistic feat, and found that the left caudate
is critical for monitoring and controlling the
language in use. The study involved a semantic
priming task in which two written nouns were
presented sequentially, with a short interval
between the first and second words of the
Figure 1 The caudate receives input from the prefrontal premotor, temporal and parietal cortex,
pair. Words were from the same language
and connects reciprocally to the cortex via the thalamus. Left, neuroanatomical topography in
(for example, English trout – SALMON or relation to other brain structures. Right, a principled but simplified view of the connections
German forelle – LACHS) or from different between the different brain structures.
languages (for German-English bilinguals,
trout – LACHS or forelle – SALMON). Some
of these word pairs were related, and others The authors found that a common set of pair), but only when both words were in the
were not. Participants were required to make a frontal, temporal and parietal regions were same language. There was no semantic prim-
semantic decision on the perceptual properties activated by semantic decisions in both lan- ing across languages for the left caudate. This
of the object only on the second word of the guages. However, there were large differences pattern was independent of whether the second
pair. In monolinguals, the one-language among the three groups, and it would have word was in the native or the second language.
version of this task usually reveals a decrease been interesting to learn whether the increased The left caudate is therefore sensitive to language
in reaction time and neuronal activation for activation in the fMRI German group was change across word pairs, suggesting that it is
semantically related compared to semantically correlated with the increased response times critical for language control.
unrelated second words in a pair. Are there observed in this group. This is a surprising finding because subcorti-
similar effects in a cross-lingual setting? To Most relevant for the argument put forward cal structures such as the caudate are not viewed
test this, the authors examined three groups: in the present study, however, is the semantic as being primarily involved in conscious cogni-
two groups of German-English bilinguals priming effect: the reduction in brain activation tive control. In contrast, frontal regions located
(one in a positron emission tomography to semantically related second words compared in the lateral prefrontal cortex support conscious
(PET) study and one in a functional magnetic to unrelated second words. The study dissoci- cognitive control2,3 and language switching4,5,
resonance imaging (fMRI) study) and a group ated two such effects in all three groups tested. but were not activated by the priming task used
of Japanese-English bilinguals. All participants First, activity in the left anterior temporal in this study.
had learned English as a second language and pole decreased in proportion to the degree of This may be because priming is a highly
had mastered English at different proficiency semantic relatedness, regardless of whether the automatic process and may therefore not
levels, according to the behavioral tests two words were in the same language or not. involve conscious control even when a lan-
reported in the online material. This clearly demonstrates that semantic prim- guage change is involved. This, however, also
ing of nouns can take place across languages suggests that the left caudate is involved, first,
and suggests overlapping neural representa- in automatic semantic priming for words
The author is at the Max Planck Institute tions for nouns in different languages. within the same language, as reflected in the
for Human Cognitive and Brain Sciences, In contrast, the left caudate showed the same activation reduction, and, second, in a percep-
Stephanstr. 1a, 04103 Leipzig, Germany. semantic priming effect (reduced activation for tual change between two languages, as reflected
e-mail: angelafr@cbs.mpg.de the second word of a semantically related word in an activation increase. This twofold


observation is condensed in the formula that explain the results in the study by Crinion structures and their possible interplay with
the left caudate responses are highest when et al. as well, as more conscious processes are well-known cortical language regions. Though
there is a change in meaning (semantically likely to be recruited across languages com- the data on this topic are still sparse, available
unrelated words in a pair) or a change in lan- pared to processing within the same language. studies suggest that the left caudate’s function
guage (different languages in a pair). It is compatible with the finding that, com- is not to control the language in use, but rather
In an attempt to further specify the role pared to real words, it takes longer to decide to recruit controlled processes when language
of the left caudate, Crinion et al. refer to two whether a pseudoword (a nonsense letter processing cannot rely primarily on auto-
other imaging studies on word processing in string similar to a real word) is a meaning- matic processes. Such a description of the left
monolinguals, comparing semantic decisions to ful word or not8. Such lexical decision times caudate’s function, which stresses the system’s
a phonological task6,7, which also found activa-
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

are also slower for words learned later com- adaptation to the processing demands rather
tion in the left caudate. On the combined basis of pared to those learned early on9. Moreover, than its control over processes, is in accordance
these studies, they conclude that the left caudate it is in line with the observation that second with the functional neuroanatomical view that
responds in general when the pattern of neural language processing is more demanding than control processes are primarily supported by
input changes. This conclusion, however, is not native language processing10–12. cortical regions. If this more general functional
entirely clear, as these two last studies differ from This more general functional description of description of the left caudate is valid, the
the present study in that identical words were the left caudate fits with the suggestion that it degree to which the language system recruits
presented during semantic and phonological is crucial in controlling and selecting motor controlled rather than automatic processes
task blocks, thus eliminating any direct percep- sequences necessary for articulation1,12. This will determine the degree to which the left
tual changes within a short time window. argument is supported by neuroanatomical caudate is activated during language process-
Moreover, this functional description of data showing that the left caudate receives pro- ing. This should hold not only for monolingual
the left caudate cannot easily be generalized jections from prefrontal, temporal and parietal and multilingual adults, but also for children
to a number of other fMRI studies that also cortex, and connects reciprocally with the cortex during language development. Thus the left
found activation in the left caudate. One via the thalamus (Fig. 1). The anterior portion caudate may be more involved during language
study compared monolingual reading of late- of the body of the left caudate, the area activated acquisition compared to adult processing.
learned words to those learned early in life and in the study by Crinion et al., receives dense
observed increased activation in the left cau- projections from the lateral and medial premo-
date8. Another word processing study found tor cortex, which is functionally related to the 1. Crinion, J. et al. Science 312, 1537–1540 (2006).
2. Koechlin, E., Ody, C. & Kouneiher, F. Science 302,
an increase in the left caudate activation for sequencing and scheduling of compiled motor 1181–1185 (2003).
words compared to pseudowords in monolin- routines13. For their own findings, Crinion et al. 3. Miller, E.K. & Cohen, J.D. Annu. Rev. Neurosci. 24,
guals9 and for translating words from native to suggest that the left caudate is active because of 167–202 (2001).
4. Hernandez, A.E., Dapretto, M., Mazziotta, J. &
second language compared to reading aloud the differing motor patterns across languages. A Bookheimer, S. Neuroimage 14, 510–520 (2001).
words in the native language5. At the sentence more general account could be that the motor- 5. Price, C.J., Green, D.W. & von Studnitz, R. Brain 122,
level, increased activation of left caudate has related patterns necessary for articulatory 2221–2235 (1999).
6. Mummery, C.J., Patterson, K., Hodges, J.R. & Price, C.J.
been observed for processing the second lan- processes are activated during language com- J. Cogn. Neurosci 10, 766–777 (1998).
guage as compared to the native language prehension whenever comprehension requires 7. Price, C.J., Moore, C.J., Humphreys, G.W. & Wise, R.J.S.
J. Cogn. Neurosci. 9, 727–733 (1997).
during reading and auditory comprehension less automatic and more controlled processes. A 8. Kotz, S.A., Cappa, S.F., von Cramon, D.Y. & Friederici,
for correct, semantically incorrect and syn- number of event-related brain potential studies A.D. Neuroimage 17, 1761–1772 (2002).
tactically incorrect sentences10–12. Thus, these support this interpretation by demonstrating a 9. Fiebach, C.J., Friederici, A.D., Müller, K.,
von Cramon, D.Y. & Hernandez, A.E. Neuroimage 19,
studies together suggest that the left caudate is selective deficit of controlled syntactic processes 1627–1637 (2003).
part of a language network, be it monolingual as a result of focal lesions of the basal ganglia, 10. Wartenburger, I. et al. Neuron 37, 159–170 (2003).
or multilingual. At this stage, it is still unclear including the left caudate nucleus14,15. The 11. Rüschemeyer, S.-A., Zysset, S. & Friederici, A.D.
Neuroimage 31, 354–365 (2006).
whether its specific function is that of control- extent to which controlled processes at the word 12. Rüschemeyer, S.-A., Fiebach, C.J., Kempe, V. &
ling the language in use. and sentence level depend on the articulatory Friederici, A.D. Hum. Brain Mapp. 25, 266–286
Instead, the reviewed studies suggest that aspects of language use needs to be tested using 13. Schubotz, R.I. & von Cramon, D.Y. Cereb. Cortex 11,
the left caudate activates when the language articulatory suppression tasks. 210–222 (2001).
processing system cannot rely entirely on The new data open an exciting view of 14. Friederici, A.D. & Kotz, S.A. Neuroimage 20, Suppl.,
S8–S17 (2003).
automatic mechanisms but has to recruit the neural basis of language processing by 15. Kotz, S.A., Frisch, S., von Cramon, D.Y. & Friederici, A.D.
controlled processes as well. This notion can highlighting the crucial role of subcortical J. Int. Neuropsychol. Soc. 9, 1053–1060 (2003).


IGF-1 receptor is essential for the these cells, accumulation of active phosphatidylinositol 3-kinase (PI3k)
and its product PIP3 at the growth cone of an undifferentiated neurite
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

establishment of hippocampal is crucial for the outgrowth of the future axon and the polarized
distribution of mPar3 and mPar6, two proteins required for axon
neuronal polarity specification2–4. Indeed, selection of the future axon requires activation
of PI3k by a (so far unspecified) growth factor receptor tyrosine
Lucas Sosa1, Sebastian Dupraz1, Lisandro Laurino1, Flavia Bollati2, kinase2. Insulin-like growth factor-1 receptors (IGF-1R) of developing
Mariano Bisbal2, Alfredo Cáceres2, Karl H Pfenninger3 & neurons contain an immunochemically distinct b subunit termed bgc
Santiago Quiroga1 (ref. 5). The bgc subunit is highly enriched in the axonal growth cone5,6.
Our results show that the sequence encoding the bgc subunit is
How a neuron becomes polarized remains largely unknown. contained in that of the IGF-1R (Supplementary Methods online).
Results obtained with a function-blocking antibody and an Thus, the bgc variant must be the result of alternative splicing of the sole
siRNA targeting the insulin-like growth factor-1 (IGF-1) Igf1r gene known to exist in rats. We found that the distribution of
receptor suggest that an essential step in the establishment
of hippocampal neuronal polarity and the initiation of axonal
outgrowth is the activation of the phosphatidylinositol 3-kinase a tub P-IGF-1 r Overlay b Co

(PI3k)-Cdc42 pathway by the IGF-1 receptor, but not by the
12 h

TrkA or TrkB receptors.

During differentiation, neurons become polarized by growing a single
long axon and several branching dendrites. The initial signals and
pathways that determine polarity are largely unknown. A widely used
18 h

model to elucidate the underlying molecular mechanisms is the
primary dissociated hippocampal neuron of the rat in culture1. In

Figure 1 Reducing IGF-1R function by incubation with blocking antibody c 80 Co 24 h
Ab 24 h
or by siRNA silencing prevents axon formation in rat hippocampal neurons. 70 Co 48 h
(a) Double immunofluorescence micrographs showing the distribution of Ab 48 h
Percentage of neurons

tyrosinated tubulin and active IGF-1 receptor in hippocampal pyramidal
neurons after 12 h (top) or 18 h (bottom) in culture (Supplementary 50

Methods). (b) Micrographs of hippocampal pyramidal neurons 40
immunostained with antibody to tyrosinated tubulin in control cultures (Co)
d C SS Si
30 βgc
and cultures grown in the presence of an antibody blocking the IGF-1R (Ab).
(c) Percentage (± s.e.m.) of neurons at specific stages of differentiation, cdc42
grown for 24 h or 48 h in culture, in control conditions (Co) or in the 10
presence of an IGF-1R–blocking antibody (Ab). n ¼ 3 independent 0 trkB
Stage 1 Stage 2 Stage 3
experiments. At least 150 cells were scored for each condition. (d) Western
blots showing protein levels of bgc, TrkB and Cdc42 in control cells (Co) and e f 80
cells transfected with a scrambled RNA sequence (ss) or IGF-1R siRNA (si). siRNA + NGF
Transfection efficiency was B60–70%. A substantial decrease of bgc 70
Percentage of neurons

expression in the siRNA-treated cells is evident. (e) Triple immuno- 60
fluorescence micrographs showing the distribution of bgc, tyrosinated 50
a-tubulin and IGF-1R–targeted, biotinylated siRNA in hippocampal
βgc siRNA 40
pyramidal neurons cultured for 24 h in the presence of 50 ng ml–1 BDNF.
The transfected neuron (arrow) did not develop an axon. (f) Percentage
(± s.e.m.) of control neurons (ssRNA) or neurons containing IGF-1R–targeted 20
siRNA at specific stages of differentiation after 24 h in culture in the 10
presence of 50 ng ml–1 BDNF or 50 ng ml–1 BDNF + 50 ng ml–1 NGF. 0
n ¼ 3 independent experiments. At least 150 cells were scored for each







condition. Scale bars: 20 mm in a and e; 60 mm in b. tub Overlay



1Departamento de Quı́mica Biológica, Facultad de Ciencias Quı́micas, Universidad Nacional de Córdoba y Centro de Investigaciones en Quı́mica Biológica de Córdoba
(CIQUIBIC), Consejo de Investigaciones Cientı́ficas y Técnicas (CONICET), Córdoba, Argentina. 2Instituto de Investigaciones Mercedes y Martin Ferreyra (INIMEC-
CONICET), Córdoba, Argentina. 3Department of Pediatrics, University of Colorado School of Medicine and University of Colorado Cancer Center, Aurora, Colorado 80045,
USA. Correspondence should be addressed to S.Q. (squiroga@dqbfcq.uncor.edu).
Received 17 May; accepted 26 June; published online 16 July 2006; doi:10.1038/nn1742


active (phosphorylated) IGF-1R was polarized to one neurite even in not related to failing neuronal health because (i) there was no
stage-2 neurons that did not yet exhibit a discernible axon (Fig. 1a, top; significant reduction in the mean number of neurites per cell at stages
specificity of the antibody is shown in Supplementary Fig. 1 online). 2 and 3 for the siRNA-treated neurons (4.3 ± 1.1%) compared to
As differentiation proceeded to early stage 3, these receptors were control neurons (4.9 ± 0.6%), and (ii) a detailed time course analysis of
completely segregated to the axon (Fig. 1a, bottom). This distribution apoptosis index, cell viability, total number of neurons per mm–2 and
pattern matched that of bgc, which recognized both the active (phos- differentiation of control versus siRNA-treated neurons showed no
phorylated) and the nonphosphorylated forms of the IGF-1R (Supple- noticeable differences in these parameters, whereas substantial differ-
mentary Fig. 1). Addition of an IGF-1R–blocking antibody to the ences were evident in neuronal polarization (Supplementary Fig. 4
culture medium prevented polarization (Fig. 1b) so that most neurons online). Thus, most neurons were unable to form axons even though
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

remained at stage 2 of differentiation for up to 48 h in culture (Fig. 1c). they were grown in the presence of 50 ng ml–1 brain-derived neuro-
Antibody-induced arrest of polarization was also reflected in the failure trophic factor (BDNF) and expressed near-normal levels of the
of tau to distribute to a single neurite, as in the case of control neurons activatable BDNF receptor TrkB (Fig. 1d and Supplementary Fig. 5
(Supplementary Fig. 2 online). Transfection of cultured neurons online). The neurons also expressed insulin receptor (not affected by
with siRNA derived from the IGF-1R sequence uniformly silenced IGF-1R–targeted siRNA; Supplementary Fig. 6 online) but, in contrast
the expression of bgc-containing IGF-1R to nondetectable levels to the distribution of bgc-containing IGF-1R, it was absent from the
(Fig. 1d,e and Supplementary Fig. 3 online). Failing to form axons, distal axon.
these IGF-1R–silenced cells generated only short, minor neurites As stated above, PI3k activation and PIP3 accumulation at the
(Fig. 1e and Supplementary Fig. 3). To analyze this observation growth cone of a minor neurite are involved in axon specification3.
quantitatively, we scored the differentiation stages of neurons trans- Our results show that IGF-1R–silenced neurons (cultured in the
fected with control RNA (scrambled sequence, ss) or with IGF-1R– presence of 50 ng ml–1 BDNF and challenged with 20 nM IGF-1)
targeted siRNA, after 24 h in culture. Only about 10% of siRNA- had active PI3k (p-tyr-p85), but only in the cell perikaryon (probably
containing neurons formed a discernible axon, in contrast to controls stimulated by the insulin receptor), and they remained at stage 2
expressing ssRNA (about 70%; Fig. 1f). The siRNA-treated cultures (Fig. 2a and Supplementary Figs. 5 and 6). In contrast, control
also included neurons that did not contain siRNA and thus expressed neurons showed substantial accumulation of active PI3k primarily in
bgc; these control neurons were predominantly at stage 3 (72 ± 4%). the distal axon and axonal growth cone (Fig. 2a), where bgc was
Reduced differentiation and polarization after siRNA transfection were enriched (Supplementary Fig. 6). mPar3, a component of the

mPar3 siRNA tub Overlay
a ptyr-p85 siRNA tub Overlay


c cdc42 βgc e fc-cdc42 siRNA tau-1 Overlay


tub Overlay ss si
84.2 ± 7.1

37.3 ± 2.5

cdc42 βgc tub Overlay

Figure 2 Cotransfection of neurons with IGF-1R–targeted siRNA and cDNA encoding fc- Cdc-42 rescues the phenotype. (a,b) Triple immunofluorescence
micrographs showing the distribution of tyrosinated a-tubulin, IGF-1R siRNA and either (a) active PI3k (p-tyr-p85) or (b) mPar3 in hippocampal pyramidal
neurons cultured for 24 h in the presence of 50 ng ml–1 BDNF. (c) Triple immunofluorescence micrographs showing the distribution of Cdc42, tyrosinated
a-tubulin and bgc in hippocampal pyramidal neurons (24 h in culture in the presence of 50 ng ml–1 BDNF). The transfection efficiency of the cells shown in
panels a–c was 60–70%. All cells cultured in the presence of ssRNA in stage 3 (72 ± 4%) showed polarized distributions of active PI3k, mPar3 and Cdc42.
In contrast, no cells transfected with the siRNA in stage 2 showed polarization of these antigens. (d) Western blots of GTP-bound Cdc42 from neurons
transfected with ssRNA or siRNA. Numbers below are the average optical densities ± s.e.m. of three independent experiments. (e) Triple immunofluorescence
micrographs of hippocampal pyramidal neurons cotransfected with the DNA encoding fc-Cdc42 and IGF-1R siRNA (top) or transfected only with siRNA
(bottom), and grown for 24 h in vitro. Scale bar, 15 mm.


Par3/Par6/aPKC complex, is enriched in the axonal growth cones of possible, and there are further differences: we have shown that PI3k
stage-3 neurons and is important for the specification of neuronal activation by IGF-1 (but not by BDNF) is necessary for plasmalemmal
polarity. It was essentially undetectable in the neurites of the IGF-1R– expansion at the growth cone and thus for axonal outgrowth in
silenced stage-2 neurons (Fig. 2b). hippocampal neurons13,14. Notably, TrkB-null mice show no obvious
Cdc42 is a Rho-family GTPase operating downstream of PI3k and is structural abnormalities in the CNS (ref. 15). Taken together, our
implicated in neuronal polarization7. We found that Cdc42 was results suggest that activation of the IGF-1R/PI3k/Cdc42 pathway
enriched at the distal axon and growth cone of control stage-3 neurons (perhaps together with transactivation of the EGFR) is essential for
(Fig. 2c). In contrast, knockdown of bgc-containing IGF-1R caused the determination of polarity and that IGF-1 is the growth factor
uniform distribution of Cdc42 in the neurons (Fig. 2c). Suppressed initiating axon specification in hippocampal neurons.
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

neurons contained near-normal amounts of total Cdc42 protein
Note: Supplementary information is available on the Nature Neuroscience website.
(Fig. 1d) but showed significantly (P o 0.02, Student’s t-test) reduced
Cdc42 activity as revealed by decreased amounts of GTP-bound Cdc42 ACKNOWLEDGMENTS
(Fig. 2d). Cotransfection of neurons with IGF-1R siRNA and a cDNA This work was supported by grants from Consejo de Investigaciones Cientı́ficas
encoding an myc-tagged, fast-cycling form of Cdc42 rescued the y Técnicas (CONICET), Agencia Córdoba Ciencia, Secretarı́a de Ciencia y
Tecnologı́a de la Universidad Nacional de Córdoba (SECYT-U.N.C.) and
phenotype and induced the outgrowth of 1–3 axons in 15–20% of
Ministerio de Salud, República Argentina (to S.Q.); by the US National Institutes
the neurons observed (Fig. 2e). Cotransfection with a dominant- of Health (grant R01 NS41029 to K.H.P.); by a Fogarty International Research
negative form of the same myc-tagged GTPase, however, did not Collaboration Award (1R03 TW05763 to K.H.P. and S.Q.); by the Agencia
produce any noticeable effects in the IGF-1R–suppressed neurons Nacional de Promoción Cientı́fica y Tecnológica, Argentina (A.C., S.Q. and G.P.);
(Supplementary Fig. 7 online). and by the International Research Scholar Program of the Howard Hughes
Medical Institute (HHMI 75197-553201 to A.C.).
AG1478, an inhibitor of the epidermal growth factor receptor
(EGFR) but not of IGF-1R, inhibits axon formation in cultured COMPETING INTERESTS STATEMENT
hippocampal neurons2. Differences in experimental conditions, the The authors declare that they have no competing financial interests.
involvement of additional receptor tyrosine kinases and/or cross-talk
Published online at http://www.nature.com/natureneuroscience
(such as transactivation8) between the IGF-1R and the EGFR are Reprints and permissions information is available online at http://npg.nature.com/
possible explanations for this apparent discrepancy. More recently it reprintsandpermissions/
has been suggested that activation of TrkA by nerve growth factor
(NGF) is important for axon specification in hippocampal pyramidal
1. Craig, A.M. & Banker, G. Annu. Rev. Neurosci. 17, 267–310 (1994).
cells, in particular in neurons overexpressing plasma membrane 2. Shi, S.H., Jan, L.Y. & Jan, Y.N. Cell 112, 63–75 (2003).
ganglioside-sialidase (cultured in the presence of a high insulin con- 3. Ménager, C., Arimura, N., Fukata, Y. & Kaibuchi, K. J. Neurochem. 89, 109–118
centration, sufficient to stimulate the IGF-1R)9. Challenge with NGF, 4. Nishimura, T. et al. Nat. Cell Biol. 7, 270–277 (2005).
however, did not stimulate axonal outgrowth in IGF-1R–suppressed 5. Quiroga, S., Garofalo, R. & Pfenninger, K.H. Proc. Natl. Acad. Sci. USA 92, 4309–4312
cells (Fig. 1f). This result is consistent with the fact that mice lacking (1995).
6. Mascotti, F., Cáceres, A., Pfenninger, K.H. & Quiroga, S. J. Neurosci. 17, 1447–1459
TrkA exhibit only mild alterations in CNS development10. By contrast, (1997).
IGF-1R knockout mice have serious defects in CNS development, 7. Schwamborn, J.C. & Puschel, A.W. Nat. Neurosci. 7, 923–929 (2004).
including a marked decrease in the number of axons11. IGF-1 and 8. Roudabush, F.L., Pierce, K.L., Maudsley, S., Khan, K.D. & Luttrell, L.M. J. Biol. Chem.
275, 22583–22589 (2000).
BDNF stimulate their neuronal and growth cone receptors selectively. 9. Da Silva, J.S., Hasegawa, T., Miyagi, T., Dotti, C.G. & Abad-Rodriguez, J. Nat. Neurosci.
The fact that only IGF-1 seems to be essential for axon specification is 8, 606–615 (2005).
10. Smeyne, R.J. et al. Nature 368, 246–249 (1994).
intriguing because both growth factors activate receptor tyrosine 11. Liu, J.P., Baker, J., Perkins, A.S., Robertson, E.J. & Efstratiadis, A. Cell 75, 59–72
kinases with similar secondary signaling pathways. A possible explana- (1993).
tion for this difference may be that stimulation of PI3k by BDNF is 12. Zheng, W.H. & Quirion, R. J. Neurochem. 89, 844–852 (2004).
13. Pfenninger, K.H. et al. J. Cell Sci. 116, 1209–1217 (2003).
weak and transient whereas activation of PI3k by IGF-1 is robust and 14. Laurino, L. et al. J. Cell Sci. 118, 3653–3663 (2005).
long lasting in hippocampal neurons12. Yet, other explanations are 15. Klein, R. et al. Cell 75, 113–122 (1993).


mRNA (Fig. 1a and ref. 6). We also observed expression of MAGE-D1
UNC5A promotes neuronal in a pattern similar to that of UNC5A (Fig. 1b). The spinal cord
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

apoptosis during spinal cord contains motoneurons that can be identified using Islet-1 and Islet-2
antibodies, nuclear markers that label motoneurons ventrally and a
development independent of subpopulation of interneurons dorsally (Fig. 1c)7. We double-labeled
sections using these markers and determined that both UNC5A-positive
netrin-1 and MAGE-D1–positive cells coexpressed the Islet markers (Fig. 1d,e).

Megan E Williams1,6, Xiaowei Lu2,5,6, William L McKenna1,6,
Raesha Washington1, Adam Boyette1, Phyllis Strickland1, a b c
Allison Dillon3, Zaven Kaprielian3, Marc Tessier-Lavigne4 &
Lindsay Hinck1

In addition to their role as chemorepellent netrin-1 receptors,
UNC5 proteins may mediate cell death because they induce
UNC5A 200 µm MAGE-D1 Islet-1/2
apoptosis in cultured cells. To test this in vivo, we generated
Unc5a (formerly Unc5h1) knockout mice and found that this d e f
deletion decreased apoptosis and increased the number of
neurons in the spinal cord. In contrast, loss of netrin-1 (Ntn1)
did not affect the amount of apoptosis, suggesting that NTN1
is not required for neuronal apoptosis in vivo.

UNC5A induces apoptosis in cell culture through a mechanism 10 µm
UNC5A /Islet-1/2 MAGE-D1/Islet1/2 Ctrl ab/no Tdt
involving the MAGE-D1 protein (formerly NRAGE)1. MAGE-D1
interacts specifically with the juxtamembrane region of UNC5A, but g h i
not UNC5B or UNC5C, and induces apoptosis through activation of
the proapoptotic c-jun N-terminal kinase and caspase signaling cas-
cades2, and through inhibition of prosurvival signaling events mediated
by X-linked inhibitor of apoptosis (XIAP)3. Despite evidence that
UNC5 family members induce apoptosis in cultured cells1,4, there is 10 µm
no evidence that UNC5 receptors regulate apoptosis in vivo.
Using an antibody specific to UNC5A (ref. 5), we showed UNC5A j Sac I
9.5 kb
Sac I

expression in the ventral cervical spinal cord at embryonic day (E) 12 in IG TSP TSP TM

an expression pattern similar to that previously reported for Unc5a
3 4 5 6 7 8 9 10 11 12 13 14 15 16
Sac I Probe Sac I
6 kb

Figure 1 Endogenous expression and targeted disruption of the Unc5a gene IRES-TauLacZ Neo
locus. (a–e) Immunohistochemistry on E12 spinal cords showed the loxP loxP

expression of (a) UNC5A, (b) MAGE-D1, (c) Islet-1/2, (d) UNC5A (brown)
and Islet-1/2 (black), and (e) MAGE-D1 (brown) and Islet-1/2 (black). k l m
(f–i) Costaining for TUNEL (black) and (g) UNC5A (brown) or (h) MAGE-D1
(brown) on E12 spinal cords. Control antibodies and no terminal transferase 9.5 kb
showed no signal (f), whereas DNase pretreatment showed strong TUNEL (i).
(j) Unc5a targeting strategy. IRES, internal ribosomal entry site; IG,
+/+ +/– –/–
immunoglobulin domain; TSP, thrombospondin repeat; TM, transmembrane 6.0 kb
domain. Black bars, exons. Gray double arrows, genotyping PCR products. 200 µm
(k) Southern blot analysis. (l) RT-PCR using primers to the Unc5a transcript.
(m) UNC5A immunohistochemistry showing Unc5a–/– spinal cord. Compare +/+ +/– Unc5a
with wild type in a. Unc5a

1Department of Molecular, Cell and Developmental Biology, University of California Santa Cruz, Santa Cruz, California 95064, USA. 2Department of Biological Sciences,

Howard Hughes Medical Institute, Stanford University, Stanford, California 94305, USA. 3Departments of Neuroscience and Pathology, Albert Einstein College of Medicine,
Bronx, New York 10461, USA. 4Genentech Inc., 1 DNA Way, South San Francisco, California 94080, USA. 5Present address: Department of Cell Biology, University of
Virginia, Charlottesville, Virginia 22908, USA. 6These authors contributed equally to this work. Correspondence should be addressed to L.H. (hinck@biology.ucsc.edu).
Received 8 May; accepted 14 June; published online 9 July 2006; doi:10.1038/nn1736


a Unc5a +/+ Unc5a –/– b c
14 14 Islet-1/2 Islet-1/2
χ cleaved caspase-3


χ TUNEL-positive
12 12

nuclei per section
nuclei per section

10 10
8 8 INB
6 6
4 ** 4 ** * Vl
2 2 200 µm
0 0 +/+
C1–C3 brainstem C1–C3 brainstem C1–C3 C1–C3 C1–C3 E14 Unc5a E14 Unc5a –/–
E12 E16 E12 E14 E18
d e f
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

50 BEN/UNC5A BEN/Islet1/2
cells per hemisection

χ Islet-1/2–positive

20 **
200 µm 20 µm 20 µm 200 µm
Dm Vl IN B
+/+ +/+ +/+ –/–
E12 Unc5a E12 Unc5a E14 Unc5a E14 Unc5a

Figure 2 Loss of Unc5a results in decreased apoptotic cell death, g 1.2 h
supernumerary neurons and abnormal spinal architecture. 1.1

Relative cervical
spinal cord size
200 µm
(a) Quantification of cells expressing cleaved caspase-3, in Unc5a–/– 1.0
mice versus wild-type littermates (C1–C3: 9.9 ± 1.0 versus 4.1 ± 0.5, 0.9 **
**P o 0.0001; brainstem: 3.3 ± 0.4 versus 3.6 ± 0.4; n Z 24 0.7
sections, mean ± s.e.m). No differences were found at E16 (C1–C3: mtl
mtl 0.6
1.4 ± 0.3 versus 1.4 ± 0.2; brainstem: 1.6 ± 0.2 versus 1.5 ± 0.2; mg mg 0.5
Total area mg width
n Z 33 sections). (b) Quantification of TUNEL-positive nuclei in the +/+ –/– E1
E12 E14 E14
E14 Unc5a E14 Unc5a
C1–C3 region (at E12: 11.0 ± 1.3 versus 3.0 ± 0.6, **P o 0.0001; at
E14: 6.6 ± 1.3 versus 3.6 ± 0.7, *P ¼ 0.042). No difference was found at E18 (1.8 ± 0.3 versus 1.9 ± 0.3; n Z 30 sections). (c) E14 cervical sections
from Unc5a+/+ and Unc5a–/– littermates labeled with Islet-1/2 antibodies. Boxes indicate pools of Islet-positive neurons. Dm, dorsal medial; Vl, ventral lateral;
INB, in between. (d) Quantification of Islet-1/2–positive cells at E14, in Unc5a–/– mice versus wild-type littermates (INB: 2.9 ± 1.5 versus 17 ± 2.9,
**P ¼ 0.0003; Vl: 29 ± 4.9 versus 40 ± 5.8, *P ¼ 0.023). No difference was found in Dm (45 ± 4.3 versus 46 ± 4.4). n Z 24 hemisections. (e) Fluorescent
coimmunostaining of UNC5A (red) or Islet-1/2 (red) with BEN (green) on E12 cervical spinal cords. Arrows, UNC5A/BEN colocalization. Arrowhead, dorsally
projecting SACMN axons. Asterisk, lateral exit point of longitudinally projecting spinal accessory nerve (SAN). (f) E14 cervical spinal cords immunostained
with BEN antibodies. Arrows, regions of immunoreactivity. (g) E14 cervical spinal cords stained with hematoxylin. Outer black line, total spinal cord area.
Red arrows, marginal zone width. Black arrows, points of narrowing in the marginal zone. mtl, mantle layer; mg, marginal layer. (h) Spinal cord morphometric
quantification. Data normalized such that relative cervical spinal cord size in Unc5a+/+ mice equals 1. **P o 0.0001. n Z 20 area measurements and
n Z 40 marginal zone measurements.

There is a period of motoneuron apoptosis during spinal cord Because UNC5A is expressed in this region6, we quantified the number
development that peaks between E11 and E15 (ref. 8). We used terminal of cells expressing activated caspase-3 in the brainstem at E12 and E16.
transferase dUTP-mediated nick end labeling (TUNEL) to identify In contrast to our findings in the spinal cord, we observed no difference
apoptotic cells at E12 (Fig. 1f–i), and immunostaining to examine in the number of apoptotic cells between Unc5a–/– and littermate
whether apoptotic cells also express UNC5A and MAGE-D1 control mice (Fig. 2a).
(Fig. 1g,h). We observed that most apoptotic cells, including those The reduced level of cell death in Unc5a–/– mice at E12 and E14
that were ventrally located and were likely to be motoneurons, suggested that supernumerary neurons may exist in the cervical spinal
expressed UNC5A and MAGE-D1 (Fig. 1g,h). To determine if region of Unc5a–/– mice by E14. To investigate this, we counted the
UNC5A is required for this apoptosis in vivo, we generated Unc5a number of Islet-1/2–expressing cells in sections from Unc5a–/– and
knockout mice (Fig. 1j; details in Supplementary Methods online). wild-type E14 littermates. Most Islet-1/2–expressing cells lay within
Southern analysis confirmed successful gene targeting (Fig. 1k), reverse three pools distinguished by physical location: ventral lateral (Vl),
transcription polymerase chain reaction (RT-PCR) showed that Unc5a dorsal medial (Dm) and a small pool in between (INB) (Fig. 2c).
mRNA was not present in these mice (Fig. 1l), and immunohisto- Notably, this INB subpopulation had five times more Islet-1/2–positive
chemistry demonstrated a loss of UNC5A immunoreactivity (compare neurons in Unc5a–/– mice compared to wild-type littermates
Fig. 1m with Fig. 1a). Together, these results indicated that the Unc5a (P ¼ 0.0003; Fig. 2c,d). The Vl region in Unc5a–/– mice also had
gene was successfully targeted to generate Unc5a–/– mice. significantly more Islet-1/2–expressing neurons compared to littermate
We analyzed the effects of eliminating Unc5a on apoptosis, using controls (P ¼ 0.023; Fig. 2c,d). Only the Dm pool, present in a region
immunohistochemistry with an antibody to cleaved caspase-3. In the where there was little UNC5A expression at E12 (Fig. 1a), appeared
cervical spine (cervical vertebrae, C, 1–3) at E12, we found a 59% unaffected by the loss of Unc5a (Fig. 2c,d).
decrease in the number of apoptotic cells in Unc5a–/– mice compared to These analyses defined a specific region (the INB) of the develop-
wild-type littermate mice (Fig. 2a). We confirmed this decrease using ing cervical spinal cord that is markedly affected by the loss of Unc5a
TUNEL and observed a 73% decrease at E12 and a 45% decrease at E14 (Fig. 2a–d). This region houses the spinal accessory motoneurons
(Fig. 2b). In contrast, at E16 and E18, there were no differences in the (SACMNs), a population of motoneurons that selectively innervate the
number of cells undergoing apoptosis between Unc5a–/– and wild-type sternocleidomastoid and trapezius muscles in the neck and back9.
littermates (Fig. 2a,b). A previous study reported elevated levels of Using antibodies to the SACMN marker BEN (DM-GRASP, MuSC,
TUNEL-positive cells in a brainstem section from a Ntn1–/– mouse4. SC1 and ALCAM), we identified these dorsally directed motoneurons


Figure 3 Loss of Ntn1 does not affect apoptosis in vivo. (a) Section from
a b 10 Ntn1 +/+ Ntn1 –/– Ntn1–/– mice labeled for Ntn1 expression (blue), UNC5A (brown) and TUNEL

χ cleaved caspase-3
nuclei per section
8 (black). Inset, magnified example of an individual cell. (b) There was no
20 µm 6
difference in the number of cleaved caspase-3–positive cells per section in
Ntn1+/+ mice compared to Ntn1–/– littermates, at E12 (C1–C3: 2.4 ± 0.3
versus 1.7 ± 0.3; brainstem: 2.0 ± 0.2 versus 2.1 ± 0.3). n Z 34 sections.
2 (c) No difference in the number of TUNEL-positive nuclei per section
C1–C3 brainstem
between Ntn1+/+ and Ntn1–/– littermates. Hip, hippocampus; Str, striatum;
200 µm
E12 Cer, cerebellum. n Z 29 sections. For all figures, three different Unc5a+/+
E12 and Unc5a–/– littermates were examined for each age and region. Error bars
c 14
represent s.e.m. P values were calculated using unpaired, two-tailed Mann-
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

χ TUNEL–positive
nuclei per section

10 Ntn1 +/+ Ntn1 –/–
Whitney tests. All experiments were done in accordance with the policies of
8 the Animal Care Committee at the University of California, Santa Cruz.
in vivo evidence that UNC5A is required for naturally occurring apopto-
C1–C3 Hip Cortex Str Cer Hip Cortex Str Cer sis in the embryonic cervical spinal cord, through the generation and
E12 E16 E18
analysis of Unc5a–/– mice. A decrease in the number of apoptotic cells in
Unc5a–/– mice was accompanied by a substantial increase in the number
in E12 cervical sections and found that they expressed UNC5A and of Islet-1/2–positive cells, an increase in BEN-labeled neurons, and an
Islet-1/2 (Fig. 2e). At E14, we detected excess BEN-positive neurons in altered morphology of the cervical spinal cord. We were surprised to find
Unc5a–/– mice compared to wild-type mice, although the marker was that Ntn1–/– mice did not show changes in apoptosis in the cervical
no longer specific for SACMNs at this age (Fig. 2f, arrows). Together spinal cord, brainstem or other regions of the developing nervous
with the increased number of Islet-1/2–positive neurons (Fig. 2c,d), system, arguing against a necessary role for NTN1 as a ligand for UNC5A
these results were consistent with the conclusion that UNC5A is a in this process. Thus, our data do not provide support for the proposal
proapoptotic receptor in vivo. that NTN1 functions as a dependence ligand in the developing CNS
We also noted a morphological change in Unc5a–/– mice by E14. (refs. 4,12). The data, however, do not exclude the possibility that
The marginal layer (mg) of the cervical spinal cord was consistently UNC5A-mediated apoptosis is regulated by a different ligand, by a
smaller and abnormally shaped compared to that in wild-type ligand-independent process or by NTN3, which is expressed at low levels
mice (Fig. 2g). We quantified this observation and found that the in the spinal cord10, and this issue remains to be elucidated. These
width of the marginal layer in Unc5a–/– mice was significantly reduced results, together with a recent study on ephrin/Eph signaling14, show that
(P o 0.0001) compared to that in the wild type (Fig. 2h), although the guidance receptors also serve as physiological triggers of apoptosis and
total area of the spinal cord at E12 and E14 did not differ (Fig. 2h). function to regulate both the size and shape of the nervous system.
These data, together with our quantification of Islet-positive neurons
Note: Supplementary information is available on the Nature Neuroscience website.
(Fig. 2d), suggested that reduced apoptosis in Unc5a–/– mice increased
the size of the cell soma–dense mantle layer at the expense of the axon- ACKNOWLEDGMENTS
rich marginal layer. In sum, our results indicated that UNC5A is an Antibodies to spinal accessory neurons (802C11) were a gift from S.C. Fujita
important regulator of developmental cell death in the spinal cord, as (Mitsubishi Kagaku Institute of Life Sciences, Tokyo, Japan). Supported by grants
from the US National Institutes of Health (NS39572- to L.H., MH12813 to
genetic deletion of Unc5a decreased developmental apoptosis, increased
M.W.), the Damon-Runyon Cancer Research Fund (X.L.) and the Howard
neuron number and disrupted the morphology of the spinal cord. Hughes Medical Institute (M.T.-L.).
To investigate the mechanism of UNC5A-dependent apoptosis in the
spinal cord, we investigated the possible role of netrin-1 is this process. COMPETING INTERESTS STATEMENT
Netrins are the only identified ligands for UNC5s (refs. 6,10), and the The authors declare that they have no competing financial interests.
floorplate of the developing spinal cord is a region of high NTN1 Published online at http://www.nature.com/natureneuroscience
expression11. Moreover, it has been postulated that NTN1 functions as Reprints and permissions information is available online at http://npg.nature.com/
a survival factor for cells expressing UNC5s (ref. 4). We labeled cervical reprintsandpermissions/
sections from E12 Ntn1–/– and wild-type littermates for UNC5A
expression and apoptosis (Fig. 3a). On the same sections, we also
1. Williams, M.E., Strickland, P., Watanabe, K. & Hinck, L. J. Biol. Chem. 278, 17483–
detected Ntn1 expression in Ntn1–/– mice with b-galactosidase staining 17490 (2003).
(Fig. 3a). This triple labeling allowed us to correlate changes in 2. Salehi, A.H., Xanthoudakis, S. & Barker, P.A. J. Biol. Chem. 277, 48043–48050
apoptosis directly with UNC5A and/or NTN1 expression (Fig. 3a, 3. Jordan, B.W. et al. J. Biol. Chem. 276, 39985–39989 (2001).
inset). We assayed for changes in apoptosis by comparing TUNEL or 4. Llambi, F., Causeret, F., Bloch-Gallego, E. & Mehlen, P. EMBO J. 20, 2715–2722
cleaved caspase-3 immunostaining in Ntn1–/– and wild-type littermates (2001).
5. Williams, M.E., Wu, S.C., McKenna, W.L. & Hinck, L. J. Neurosci. 23, 11279–11288
in several regions of the nervous system, including the brainstem, (2003).
where NTN1 has been suggested to function as a survival factor 6. Leonardo, E.D. et al. Nature 386, 833–838 (1997).
(Fig. 3b,c). Whereas there was more apoptosis in the spinal cord 7. Tsuchida, T. et al. Cell 79, 957–970 (1994).
8. Oppenheim, R.W. Annu. Rev. Neurosci. 14, 453–501 (1991).
compared to other regions, we found no difference in the number of 9. Dillon, A.K. et al. J. Neurosci. 25, 10119–10130 (2005).
apoptotic cells in Ntn1–/– versus wild-type mice. Therefore, our results 10. Wang, H., Copeland, N.G., Gilbert, D.J., Jenkins, N.A. & Tessier-Lavigne, M. J. Neurosci.
19, 4938–4947 (1999.).
indicated that NTN1 is not a necessary survival factor for UNC5A- 11. Kennedy, T.E., Serafini, T., de la Torre, J.R. & Tessier-Lavigne, M. Cell 78, 425–435
expressing neurons in vivo (Fig. 3b,c). (1994).
A growing body of literature shows that the NTN1 receptors UNC5s 12. Mehlen, P. et al. Nature 395, 801–804 (1998).
13. Tanikawa, C., Matsuda, K., Fukuda, S., Nakamura, Y. & Arakawa, H. Nat. Cell Biol. 5,
and DCC can regulate apoptosis, based on studies in a variety of neuro- 216–223 (2003).
nal and non-neuronal cell lines1,4,12,13. Here, we provided the first direct 14. Depaepe, V. et al. Nature 435, 1244–1250 (2005).


Specification of auditory frequency of the tones was adjusted to match the individual best
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

frequency of each receiver (Supplementary Methods online). The
sensitivity by Drosophila intensity of the tones was monitored as the stimulus particle velocity
at the receiver’s position, and the resulting mechanical response of the
TRP channels receiver was measured as the Fourier amplitude of its displacement at
the frequency of stimulation. This phase-locked response linearly scaled
Martin C Göpfert, Jörg T Albert, B Nadrowski & A Kamikouchi with intensity for stimulus particle velocities less than B0.1 mm s–1 and
greater than B10 mm s–1 (Fig. 1a). For intermediate particle velocities
Ears achieve their exquisite sensitivity by means of mechanical (0.1–10 mm s–1), however, a nonlinear scaling was found: the response
feedback: motile mechanosensory cells through their active of the receivers increased with only the two-third power of the stimulus
motion boost the mechanical input from the ear. Examination intensity, leading to an B10-fold gain in sensitivity to low intensity
of the auditory mechanics in Drosophila melanogaster mutants sounds (Fig. 1a,b). This nonlinear amplification, which was frequency
shows that the transient receptor potential (TRP) channel specific (Supplementary Fig. 1 online) like that in vertebrate hear-
NompC is required to promote this feedback, whereas the ing11,12, was affected by the disruption of TRPs. The null allele nompC3
TRP vanilloid (TRPV) channels Nan and Iav serve to control the abolished nonlinear amplification (mean sensitivity gain E 1, Fig. 1b),
feedback gain. The combined function of these channels leading to linear responses of constantly low sensitivity throughout the
specifies the sensitivity of the fly auditory organ. intensity range (Fig. 1a and Supplementary Fig. 2 online). This loss of
amplification was observed in homozygous nompC3cn bw mutants and
In mechanosensory cells for hearing, several transient receptor poten- when the nompC3 mutation was uncovered by a deficiency (Df(2L)clh2),
tial (TRP) ion channels coexist. Drosophila auditory neurons, for but not in homozygous cn bw and balanced nompC3cn bw/Cy cn and
example, putatively express NompC (also known as TRPN1)1—a Df(2L)clh2/Cy cn controls (Fig. 1a,b). The opposite effect—that is,
channel that has not yet been definitely shown to localize to these facilitation of amplification—was found for null alleles in nan and
cells1–3—and demonstrably express the two interdependent vanilloids iav (Fig. 1): in homozygous nan36a and iav1 mutants, the receiver’s
(TRPVs) Nan (ref. 2) and Iav (encoded by CG4536, or iav; ref. 3), response nonlinearly increased with approximately the one-third power
which are deemed to form a heteromultimeric Nan-Iav channel3. for intermediate particle velocities, boosting the sensitivity gain to B85
Likewise, in vertebrate auditory hair cells, NompC (refs. 4,5) and/or
TRPA1 (ref. 6) seem to occur along with TRPV1 (ref. 7) and TRPV4
(ref. 8). Most of these TRPs have been implicated in mechanosensory a 104
nompC mutants
nompC controls

transduction1–8, yet their relative positions in the auditory pathway and 2/3 2/3

their respective contributions to the process of hearing are little under- 102 102 102
Phase-locked response (nm)

stood9. Mechanosensory transduction in the context of hearing is linked
100 100 100
to nonlinear amplification: the mechanosensory cells that mediate
hearing actively generate motions to specifically augment the minute, 10–2 100 102 10–2 100 102 10–2 100 102
sound-induced vibrations that they transduce10. We have assayed this nan and iav nan and iav controls nompC; nan
104 mutants 104 104 double mutants
mechanical feedback in the ears of Drosophila nompC, nan and iav 1/3
mutants to gain insights into the diverse auditory roles of TRPs. 102 102 102
To test for nonlinear amplification, we exposed the antennal sound
receivers of wild-type flies to pure tones of different intensities. The 100 100 100

10–2 100 102 10–2 100 102 10–2 100 102
Stimulus particle velocity (mm s–1)
Figure 1 Nonlinear amplification. (a) Plots, in log-log coordinates, of the dis- b Wild-type nompC nompC nan and iav mutants nan and iav nompC; nan
Sensitivity gain

placement of the fly’s antennal sound receiver as a function of the stimulus mutants controls *** controls double
100 mutants
particle velocity. Displacements and particle velocities are given as Fourier
amplitudes at the stimulus frequency, which was adjusted to the individual *** ***
best frequency of each receiver. Orange lines, linear regimes; red lines, 1

v1 b
na a


f(2 w/ w

no av 1 621 4


bw 2.2

n 36 362
cl h

n 36
n dy

cl h c

n dy M6


pC P{ /FM

nonlinear regimes; arrows, gain in sensitivity due to nonlinear amplification.


L) Cy



cn v +}


na /T

na iav





(b) Sensitivity gain (means ± 1 s.d., logarithmic scaling) in mutant and







control strains, as determined from the data in a. A gain of one indicates the



absence of nonlinear amplification. ***P o 0.05, two-tailed Mann Whitney

U-tests against Oregon R wild-type strain (N ¼ 5–8 receivers per strain).

Volkswagen Foundation Research Group, Institute of Zoology, University of Cologne, Weyertal 119, 50 923 Cologne, Germany. Correspondence and requests for materials
should be addressed to M.C.G. (m.gopfert@uni-koeln.de).
Received 12 May; accepted 13 June; published online 2 July 2006; doi:10.1038/nn1735


a b Wild-type nompC
nan and iav mutants
nan and iav
nompC; nan
Wild-type (Oregon R ) 105

Power (nm2)
Wild-type nan and
nompC mutant (nompC3 cn bw ) 103 ***
10–2 iav controls ***
Power spectral density (nm2 Hz–1)

Power spectral density (nm2 Hz–1)
nompC control (nompC 3 cn bw/Cy cn) 101




v1 b
cl h cn


L) cn w


v 3 6 M4


n a C 3 .2

n 36

n 3 6 v 3 62
n dy

a bw
2 /

n d y M6

cl h bw

D pC 3 cn b

iav mutant (iav1)



m v+

; P /F

cn cn

f (2 w / C

n 36 cn

na /T

v 1 21




n o {ia

10–2 nompC controls




f (2





nan mutant (nan 36a)
10–2 Figure 2 Self-sustained oscillations. (a) Time course (middle) and power
nompC mutants
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

spectra (left and right) of the receiver’s vibrations measured in the absence
iav control (iav1;P {iav + } 2.2)
100 iav
of external stimulation. Black traces, spectrum for one receiver per strain;
gray, spectra obtained from other receivers of that strain. (b) Power of the
10–2 nompC; nan nompC; nan double mutant
(nompC 3 cn bw;nan36a)
10–2 receiver vibration in mutant and control strains, as determined by integrating
800 nm
double mutants
100 300 1,000 100 300 1,000 the power spectra in a (means ± 1 s.d., logarithmic scaling). ***P o 0.05,
Frequency (Hz)
20 ms
Frequency (Hz) two-tailed Mann Whitney U-tests against Oregon R wild-type strain
(N ¼ 5–8, same receivers as in Fig. 1).

(Fig. 1a,b). This excessive amplification, which was also observed in much as is found for mutations in nompC. Hence, both the requirement
homozygous nandy5 and iav3621 mutants but not in balanced controls, for feedback amplification and the functional placement of NompC in
regressed toward that in the wild type (gain E 14) when a single wild- the auditory pathway (Supplementary Fig. 3) support NompC, but not
type transgene of iav (P{iav+}2.2)3 was expressed in the iav1 mutant Nan-Iav, as a candidate transducer channel for hearing in the Drosophila
background (Fig. 1b). Hence, whereas NompC is required for ampli- ear. As disrupting NompC reduces, but does not abolish, afferent nerve
fication, Nan-Iav is required to adjust the amplificatory gain. The excess responses14, this channel may well participate in transduction, but it is
amplification caused by the disruption of Nan-Iav was found to depend unlikely to be the only transducer component the fly uses to hear
on NompC function: in nompC3; nan36a double mutants, amplification (Supplementary Fig. 3). The complete loss of nerve responses in nan
was completely abolished (mean sensitivity gain E 1.0), leading to a and iav mutants, in turn, suggests that signaling by Nan-Iav is bidirec-
linear response as found for the single mutation nompC3 (Fig. 1). tional, controlling mechanical amplification in a NompC-dependent
nompC3 is thus epistatic to nan36a, placing NompC downstream of manner and propagating electrical signals to the nerve (Supplementary
Nan-Iav in the regulatory pathway that, by controlling the extent of Fig. 3). Both functions may be Ca2+ mediated: like mammalian TRPV4
nonlinear amplification, specifies the mechanical sensitivity of the ear (ref. 8), heterologously expressed Nan and Iav form Ca2+-permeable
(Supplementary Fig. 3 online). channels and promote Ca2+ spikes2,3. As Ca2+-regulated amplification
A striking signature of amplification in hearing is sound that is and TRP-dependent mechanosensation also occur in vertebrate hear-
spontaneously emitted by some ears. This ringing sound, which in rare ing4,10,15, the auditory roles of TRPs revealed by this study may not be
cases can be loud enough to be heard by passersby, is deemed to reflect restricted to flies: spontaneous sound emissions from the ears of
self-sustained oscillations caused by excessive mechanical feedback in vertebrate TRPV knockouts have not yet been reported, but listening
the ear10,11. We found that self-sustained oscillations arise from the for such sounds seems warranted now that our ears are tuned in.
disruption of TRPVs. In nan and iav mutants—but not in nompC3
Note: Supplementary information is available on the Nature Neuroscience website.
mutants, wild-type flies or controls—the receiver continuously oscil-
lated in the absence of acoustic stimulation, with displacement ampli- ACKNOWLEDGMENTS
tudes as large as 0.5 mm (Fig. 2). The oscillations were sharply tuned We thank M.J. Kernan (Stony Brook University, Stony Brook, New York) and
(quality factor Q ¼ 50 ± 27 in nan36a and iav1 mutants versus 1.3 ± 0.4 Y.-D. Chung (University of Seoul, Seoul, Korea) for providing nan and iav
mutants and controls, and R.G. Walker (Oregon Health Sciences University,
in nan and iav controls, N ¼ 6–8 per strain), giving rise to a sharp peak
Portland, Oregon) for providing nompC mutants and controls. This work was
in the power spectrum at frequencies between approximately 100 Hz supported by the Volkswagen Foundation (grant I 79 147 to M.C.G.). A.K. is
and 250 Hz (Fig. 2a) and increasing the fluctuation power, measured as supported by the Japanese Cell Science Research Foundation and the Alexander
the total power spectral density in the frequency band between 100 Hz von Humboldt Foundation.
and 1,500 Hz, by a factor of B155 (Fig. 2b). In nompC3; nan36a double
mutants, the oscillations were abolished and the fluctuation power was The authors declare that they have no competing financial interests.
nearly as low as in single nompC3 mutants (Fig. 2b). Hence, both the
Published online at http://www.nature.com/natureneuroscience
excess amplification and the self-sustained oscillations arising from the Reprints and permissions information is available online at http://npg.nature.com/
disruption of Nan-Iav require NompC function; by negatively control- reprintsandpermissions/
ling the gain of NompC-dependent amplification, Nan-Iav prevents
oscillations in the ear (Supplementary Fig. 3). 1. Walker, R.G., Willingham, A.T. & Zuker, C.S. Science 287, 2229–2234 (2000).
2. Kim, J. et al. Nature 424, 81–84 (2003).
Mutations in nan and iav are the first genetic defects shown to cause 3. Gong, Z. et al. J. Neurosci. 24, 9059–9066 (2004).
excessive mechanical feedback in hearing and ringing in the ear. This 4. Sidi, S., Friedrich, R.W. & Nicolson, T. Science 301, 96–99 (2003).
ringing, like many chronic forms of tinnitus13, is associated with hearing 5. Shin, J.-B. et al. Proc. Natl. Acad. Sci. USA 102, 12572–12577 (2005).
6. Corey, D.P. et al. Nature 432, 723–730 (2004).
loss: nan and iav mutants reportedly lack sound-induced afferent nerve 7. Zheng, J. et al. J. Neurophysiol. 90, 444–455 (2003).
responses and, accordingly, are deaf2,3. Because of this deafness and the 8. Liedtke, W. et al. Cell 103, 525–535 (2000).
9. Lin, S.-Y. & Corey, D.P. Curr. Opin. Neurobiol. 15, 350–357 (2005).
pressure activation of heterologously expressed Nan and Iav, the 10. Fettiplace, R. & Hackney, C.M. Nat. Rev. Neurosci. 7, 19–29 (2006).
prevailing view is that these TRPVs serve as the fly’s transducer channel 11. Robles, L. & Ruggero, M.A. Physiol. Rev. 81, 1305–1352 (2001).
for hearing2,3,9. Our findings do not support this view: transducers are 12. Martin, P. & Hudspeth, A.J. Proc. Natl. Acad. Sci. USA 98, 14386–14391 (2001).
13. Baguley, D.M. Br. Med. Bull. 63, 195–212 (2002).
essential components of amplificatory feedback loops; loss of transducer 14. Eberl, D.F., Hardy, R.W. & Kernan, M.J. J. Neurosci. 20, 5981–5988 (2000).
function will inevitably disrupt the feedback and abolish amplification, 15. Frolenkov, G., Mammano, F. & Kachar, B. Cell Calcium 33, 185–195 (2003).


Cross-modal regulation of Loss of vision is usually accompanied by the increased functionality of
other sensory modalities1,2. Systems-level analyses of cross-modal
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

synaptic AMPA receptors in plasticity have revealed anatomical and functional rewiring of cortical
circuits3. However, little is known about the cellular and molecular
primary sensory cortices by mechanisms underlying this type of plasticity. Here we examined
whether manipulation of visual experience can induce bidirectional
visual experience cross-modal plasticity of synaptic function in primary sensory cortices,
and investigated the molecular mechanisms underlying this form
Anubhuthi Goel1,2,5, Bin Jiang3–5, Linda W Xu1, Lihua Song1, of plasticity.
Alfredo Kirkwood3 & Hey-Kyoung Lee1,2 To study cross-modal changes in synaptic function by visual depri-
vation, we dark-reared 4-week-old Long-Evans rats for a period of
Lack of a sensory input not only alters the cortical circuitry 1 week and then measured AMPA receptor (AMPAR)-mediated
subserving the deprived sense, but also produces compensatory miniature excitatory postsynaptic currents (mEPSCs) in layer 2/3
changes in the functionality of other sensory modalities. Here pyramidal neurons in slices from primary visual, somatosensory and
we report that visual deprivation produces opposite changes in auditory cortex (Supplementary Methods online). In visual cortex,
synaptic function in primary visual and somatosensory cortices dark rearing produced an increase in mESPC amplitude that was
in rats, which are rapidly reversed by visual experience. This reversed by re-exposing the rats to lighted conditions for 2 d (nor-
type of bidirectional cross-modal plasticity is associated with mal-reared (NR): 10.7 ± 0.6 pA, n ¼ 8; dark-reared (DR): 12.4 ± 0.4
changes in synaptic AMPA receptor subunit composition. pA, n ¼ 16; re-exposure to light (L): 10.7 ± 0.4 pA, n ¼ 13; analysis of

Figure 1 Cross-modal changes in AMPAR-
mediated synaptic transmission by visual
experience. (a–d) Left, cumulative probability of a Visual cortex
AMPAR-mediated mEPSC amplitudes. Middle, 1.00 NR DR L Scaled 4
average mEPSC traces. Right, change in mEPSC

Frequency (Hz)
0.75 15 # 3
Cum. prob.

frequency with visual experience. NR, rats reared
Ampl. (pA)

normally (red solid line); DR, rats reared in the 0.50 5 2
dark for 1 week (black solid line); L, rats reared in pA
0.25 0
the dark for 1 week and then re-exposed to light N D L
0 0
for 2 d (black dotted line). Insets in cumulative 0 20 40 60 NR DR L
probability graphs, average mEPSC amplitudes Amplitude (pA)
from NR (N), DR (D) and L groups. Results are
from (a) visual, (b) somatosensory, (c) auditory b Somatosensory cortex

1.00 NR DR L Scaled 5
and (d) frontal cortex. In visual, somatosensory
Frequency (Hz)

and auditory cortex, mEPSC of the DR group was 20 4
Cum. prob.

Ampl. (pA)

15 #
significantly different from that of the NR and L 5 3
0.50 10
groups (Kolmogorov-Smirnov test: P o 0.001). In 5 pA 2
visual, somatosensory and frontal cortex, there 0.25 0 1
N D L 20
was no change in mEPSC kinetics across groups. 0 ms 0
In auditory cortex, there was a significant increase 0 20 40 60 NR DR L

in decay time constant (t) in the DR group (NR: Amplitude (pA)
4.0 ± 0.25 ms, n ¼ 17; DR: 4.9 ± 0.36 ms,
n ¼18; t-test: P o 0.04). In all four cortices, c Auditory cortex d Frontal cortex

there was no significant change in mEPSC 1.00 NR 5 1.00 NR 5
DR Scaled DR Scaled
Frequency (Hz)

frequency with visual experience. #P o 0.03 by
Frequency (Hz)

0.75 20 4 20 4
Cum. prob.

Cum. prob.
Ampl. (pA)

Ampl. (pA)

15 15
ANOVA, Fisher’s protected least significant * 3 3
0.50 10 0.50 10
difference (PLSD) post-hoc test. *P o 0.04 by 5
5 2 5
5 2
t-test. Error bars represent s.e.m. All experiments 0.25 pA 0.25 pA
0 1 0 1
were approved by the Institutional Animal Care 0 10 0 0 10 0
0 20 40 60 ms N D 0 20 40 60 N D
and Use Committees (IACUCs) of the University ms
Amplitude (pA) Amplitude (pA)
of Maryland and Johns Hopkins University.

1Department of Biology, University of Maryland, College Park, Maryland 20742, USA. 2Neuroscience and Cognitive Science (NACS) Program, University of Maryland,

College Park, Maryland 20742, USA. 3Department of Neuroscience, The Mind/Brain Institute, Johns Hopkins University, Baltimore, Maryland 21218, USA. 4Current
address: Brain Science Institute, Riken, Wako City, Saitama, Japan. 5These authors contributed equally to this work. Correspondence should be addressed to H.-K.L.
Received 2 May; accepted 26 May; published online 2 July 2006; doi:10.1038/nn1725


P 4 0.3); somatosensory cortex: NR ¼ 2.9 ±
a Visual cortex b Visual cortex
0.5 Hz, DR ¼ 1.6 ± 0.1 Hz, L ¼ 2.3 ± 0.6 Hz
GluR1 GluR2 GluR1 GluR2
NR DR NR DR DR L DR L (ANOVA: F2,40 ¼ 1.755, P 4 0.1); auditory
R1/R2 ratio R1/R2 ratio
cortex: NR ¼ 3.8 ± 0.5 Hz, DR ¼ 2.8 ± 0.3 Hz
250 * * 150 (t-test: P 4 0.1); frontal cortex: NR ¼ 3.1 ±

* 0.5 Hz, DR ¼ 2.8 ± 0.6 Hz (t-test: P 4 0.6).
of NR


of DR
100 50 The observed increase in mEPSC amplitude in
0 0 visual cortex of dark-reared rats is consistent
with homeostatic plasticity reported pre-
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

c Somatosensory cortex
d Somatosensory cortex viously4 (Supplementary Fig. 1 online).
GluR1 GluR2 GluR1 GluR2 Moreover, visual deprivation decreased
mEPSC amplitude in both primary somato-
R1/R2 ratio R1/R2 ratio sensory and auditory cortices. In addition,
150 500 *

our results suggest that visual experience can
of NR

of DR
100 300 * *
* * 200
bidirectionally modify synapses in the visual
100 cortex and can produce the opposite changes
0 0
NR DR NR DR NR DR DR L DR L DR L in somatosensory cortex.
To examine whether bidirectional changes
e Visual cortex 1.0
in mEPSC amplitude by visual experience are
Vh (mV) 0.5 due to the regulation of postsynaptic
Inward rectification

4 #
(I–60 mV/I+40 mV)

–60 –40 –20 AMPARs, we biochemically isolated post-
0 20 40 synaptic density (PSD) fractions from both
20 20 20
2 –0.5 visual and somatosensory cortices (Supple-
pA pA pA

40 40 40 mentary Methods and Supplementary Fig. 2
ms ms ms
0 online). In visual cortex, 1 week of dark rear-
NR DR L –1.5 ing increased GluR1 (NR: 100 ± 12% of
average NR, n ¼ 9; DR: 190 ± 34% of NR,
f Somatosensory cortex 1.0
n ¼ 9; t-test: P o 0.04) but not GluR2 (NR:
Vh (mV) 0.5 100 ± 21% of NR, n ¼ 9; DR: 92 ± 11% of NR,
Inward rectification

(I–60 mV/I+40 mV)

–60 –40 –20
n ¼ 9; t-test: P 4 0.7) (Fig. 2a), resulting in a
# 0 20 40
20 20 20 significant increase in the ratio of GluR1 to
2 –0.5
pA pA pA GluR2 (GluR1/GluR2 ratio; NR: 100 ± 10% of

1 40 40 40
–1.0 ms ms ms NR, n ¼ 9; DR: 180 ± 29% of NR, n ¼ 9;
NR DR L t-test: P o 0.03). Re-exposing dark-reared
rats to light for 2 d reversed the increase in
Figure 2 Cross-modal changes in synaptic AMPAR subunit composition by visual experience. (a) Dark the GluR1/GluR2 ratio (DR: 100 ± 11% of
rearing for 1 week (DR) increased the GluR1/GluR2 (R1/R2) ratio in PSDs of visual cortex, by increasing average DR, n ¼ 14; L: 69 ± 14% of DR,
GluR1 over that in the NR group. (b) Re-exposing DR rats to light for 2 d (L) decreased the GluR1/GluR2 n ¼ 14; t-test: P o 0.05; Fig. 2b). In contrast,
ratio in the PSD of visual cortex. (c) Dark rearing decreased GluR1/GluR2 ratio in somatosensory cortical in somatosensory cortex, dark rearing
synapses by decreasing GluR1 content. (d) Re-exposure to light for 2 d increased the GluR1/GluR2 ratio
as compared to that in the DR group. Both GluR1 and GluR2 content were significantly increased in the
decreased GluR1 (NR: 100 ± 11% of NR,
L group. (e,f) Left, inward rectification of current through synaptic AMPARs. Middle, AMPAR I-V curves n ¼ 12; DR: 51 ± 11% of NR, n ¼ 12; t-test:
from NR (white circles), DR (black circles) and L (gray triangles). Right, representative AMPAR-mediated P o 0.005) without changing GluR2 (NR: 100
current traces at –60 mV and +40 mV. In visual cortex (e), dark rearing increased the inward rectification ± 23% of NR, n ¼ 12; DR: 101 ± 24% of NR,
of current through AMPARs, whereas in somatosensory cortex (f), it resulted in a more linear current. n ¼ 12; t-test: P 4 0.9), resulting in a
Both were reversed by 2 d of light exposure. *P o 0.05 by t-test. #P o 0.01 by ANOVA, Fisher’s PLSD significant decrease in the GluR1/GluR2
post-hoc test. Error bars represent s.e.m.
ratio (NR: 100 ± 22% of NR, n ¼ 12; DR:
41 ± 10% of NR, n ¼ 12; t-test: P o 0.03;
Fig. 2c), which was reversed by 2 d of light
variance (ANOVA): F2,34 ¼ 5.968, P o 0.01; Fig. 1a). Notably, we exposure (DR: 100 ± 13% of DR, n ¼ 12; L: 224 ± 42% of DR, n ¼ 12;
observed the opposite changes in somatosensory cortex, where 1 week t-test: P o 0.02; Fig. 2d). These changes at the PSD were not reflected
of dark rearing decreased the amplitude of mEPSCs and 2 d of light in the total homogenate and were accompanied by changes in GluR1
exposure reversed this effect (NR: 13.8 ± 0.8 pA, n ¼ 12; DR: 11.3 ± 0.7 phosphorylation (Supplementary Fig. 3 online), suggesting that post-
pA, n ¼ 16; L: 14.1 ± 0.9 pA, n ¼ 16; ANOVA: F2,40 ¼ 3.830, P o 0.04; translational mechanisms may be involved. Notably, GluR1 serine 845
Fig. 1b). Changes in synaptic transmission by dark rearing seems to be phosphorylation correlated with an increase in mEPSC amplitude in
general for primary sensory cortices, as dark rearing also reduced both visual and somatosensory cortex (Supplementary Fig. 3).
mEPSC amplitudes in auditory cortex (NR: 13.0 ± 0.9 pA, n ¼ 17; DR: AMPARs lacking or having reduced copies of GluR2 display inward
10.7 ± 0.5 pA, n ¼ 18; t-test: P o 0.04; Fig. 1c), but not in frontal cortex rectification of current5,6. Therefore, we assessed the GluR1/GluR2
(NR: 13.0 ± 1.2 pA, n ¼ 11; DR: 11.7 ± 1.2 pA, n ¼ 9; t-test: P 4 0.3; ratio electrophysiologically by determining an inward rectification
Fig. 1d). There was no significant change in mEPSC frequency across index (IR ¼ (I at –60 mV)/(I at +40 mV); Supplementary Methods)
groups in any of the cortical areas (Fig. 1a–d): visual cortex: NR ¼ 1.4 ± of AMPAR synaptic responses evoked by layer 4 stimulation. Consistent
0.2 Hz, DR ¼ 1.5 ± 0.2 Hz, L ¼ 2.0 ± 0.3 Hz, (ANOVA: F2,34 ¼ 1.222, with our biochemical data, in visual cortex, dark rearing produced an


increase in inward rectification that was reversed by re-exposure to light common downstream mechanism for both synapse-specific and global
(NR ¼ 1.88 ± 0.15, n ¼ 10; DR ¼ 3.42 ± 0.20, n ¼ 22; L ¼ 1.68 ± 0.14, homeostatic plasticity in vivo.
n ¼ 11; ANOVA: F2,40 ¼ 25.929, P o 0.001; Fig. 2e), whereas opposite
Note: Supplementary information is available on the Nature Neuroscience website.
changes were observed in somatosensory cortex (NR ¼ 3.87 ± 0.46,
n ¼ 10; DR ¼ 1.76 ± 0.07, n ¼ 18; L ¼ 3.07 ± 0.17, n ¼ 9; ANOVA: ACKNOWLEDGMENTS
F2,34 ¼ 23.440, P o 0.001; Fig. 2f). Inward rectification was dependent The authors would like to thank E.M. Quinlan for helpful discussions and
on intracellular spermine (Supplementary Fig 4 online). Incidentally, comments on this manuscript. This work was supported by a US National
Institutes of Health grant (R01-EY014882) and a Sloan Research Fellowship
we noticed that neurons in somatosensory cortex from normal-reared
to H.-K.L., a US National Institutes of Health grant (R01-EY012124) to A.K.
rats showed larger inward rectification and mEPSC amplitudes than and a Howard Hughes Medical Institute Undergraduate Research Fellowship
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

cells from the visual cortex of these rats (Fig. 2 and Supplementary Fig. to L.W.X.
5 online), consistent with basal differences in AMPAR subunit compo-
sition between these two cortical areas (Supplementary Fig. 6 online). AUTHOR CONTRIBUTIONS
A.G. and B.J. conducted the electrophysiology experiments (mEPSC recordings
Collectively, our biochemical and electrophysiological data suggest that and rectification measurements, respectively) and assisted in writing the
cross-modal plasticity may be mediated by changes in the subunit manuscript; L.W.X. and L.S. performed the biochemistry experiments; A.K.
composition of synaptic AMPARs. oversaw the electrophysiology (rectification measurements), contributed to
Our results demonstrate that manipulation of visual experience not discussions on experimental designs and collaborated on manuscript writing;
H.-K.L. designed the studies, oversaw experiments, contributed to the
only bidirectionally regulates synaptic AMPARs in visual cortex, but
electrophysiology (mEPSC recordings) and biochemistry and wrote
also produces complementary changes in the somatosensory cortex. the manuscript.
These changes are rapid, as dark rearing for only 1 week produced
similar changes in AMPARs as dark rearing from birth (Supplemen- COMPETING INTERESTS STATEMENT
tary Fig. 5). It remains to be determined whether the cross-modal The authors declare that they have no competing financial interests.
plasticity induced in somatosensory cortex is due to the altered cortical Published online at http://www.nature.com/natureneuroscience
processing of tactile inputs (that is, top-down) or to differences in Reprints and permissions information is available online at http://npg.nature.com/
tactile experience (that is, bottom-up). The former mechanism would reprintsandpermissions/
engage cortico-cortical inputs, whereas the latter involves thalamocor-
tical inputs for triggering homeostatic plasticity. In any case, similar 1. Goldreich, D. & Kanics, I.M. J. Neurosci. 23, 3439–3445 (2003).
changes in AMPAR function in auditory cortex suggest that these 2. Lessard, N., Pare, M., Lepore, F. & Lassonde, M. Nature 395, 278–280 (1998).
changes may occur globally across sensory cortices. The bidirectional 3. Bavelier, D. & Neville, H.J. Nat. Rev. Neurosci. 3, 443–452 (2002).
4. Desai, N.S., Cudmore, R.H., Nelson, S.B. & Turrigiano, G.G. Nat. Neurosci. 5, 783–789
synaptic changes we observed are consistent with a homeostatic (2002).
plasticity mechanism, where chronic deprivation of inputs increases 5. Verdoorn, T.A., Burnashev, N., Monyer, H., Seeburg, P.H. & Sakmann, B. Science 252,
1715–1718 (1991).
AMPAR function whereas a prolonged increase in activity decreases it7.
6. Washburn, M.S., Numberger, M., Zhang, S. & Dingledine, R. J. Neurosci. 17, 9393–
In vitro studies suggest that homeostatic synaptic plasticity is associated 9406 (1997).
with changes in the synaptic content8–10 and subunit composition11,12 7. Turrigiano, G.G. & Nelson, S.B. Nat. Rev. Neurosci. 5, 97–107 (2004).
8. Lissin, D.V. et al. Proc. Natl. Acad. Sci. USA 95, 7097–7102 (1998).
of AMPARs. We not only provide evidence that these changes also 9. O’Brien, R.J. et al. Neuron 21, 1067–1078 (1998).
occur in vivo by natural experience, but suggest that these mechanisms 10. Turrigiano, G.G., Leslie, K.R., Desai, N.S., Rutherford, L.C. & Nelson, S.B. Nature 391,
can be recruited cross-modally. Taken together with a recent study 892–896 (1998).
11. Ju, W. et al. Nat. Neurosci. 7, 244–253 (2004).
demonstrating synapse-specific regulation of AMPARs by sensory 12. Thiagarajan, T.C., Lindskog, M. & Tsien, R.W. Neuron 47, 725–737 (2005).
experience13, our results suggest that AMPAR regulation may be a 13. Clem, R.L. & Barth, A. Neuron 49, 663–670 (2006).


Maternal presence serves as During early life when pups are confined to the nest (the ‘sensitive
period’), they exhibit potentiated preference learning and attenuated
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

a switch between learning fear aversion learning, characterized by odor preferences induced by con-
ditioning with an odor and a 0.5-mA shock6–8. This paradoxical
and attraction in infancy learning does not reflect the pups’ inability to feel pain or threshold
differences9, but reflects the inability of odor-shock conditioning to
Stephanie Moriceau & Regina M Sullivan engage the amygdala8,10–12. The sensitive period ends as the pups’
ability to walk emerges and life outside the nest begins (at age 10 d),
Odor-shock conditioning produces either olfactory preference with a rapid transition to independence by age 21–23 d. In this
or aversion in preweanling (12–15 days old) rats, depending on ‘postsensitive period’, preweanling rats are in a transitional period
the context. In the mother’s absence, odor-shock conditioning from dependence to independence. At this stage, the pups need both
produces amygdala activation and learned odor avoidance. continued interactions with the mother as well as the engagement of
With maternal presence, this same conditioning yields an contingency-dependent learning for survival outside the nest. The
odor preference without amygdala activation. Maternal effects of maternal presence on odor-pain conditioning may ensure
presence acts through modulation of pup corticosterone and that pups continue to only learn approach responses to her odors,
corticosterone’s regulation of amygdala activity. Over-riding whereas in her absence they learn complex contingencies required for
maternal suppression of corticosterone through intra-amygdala survival outside the nest.
corticosterone infusions permits fear conditioning and Here we present data illustrating that odor-shock learning (0.5-mA
amygdala activation. shock) in pups accommodates their changing developmental needs.
Odor-shock conditioning resulted in an odor preference at an age when
Here we show two circuits for odor-shock conditioning, with maternal pups were confined to the nest (Fig. 1a; 8 d sensitive period; analysis of
presence providing the ‘switch’ by lowering pups’ corticosterone variance (ANOVA), F3,16 ¼ 3.917, P o 0.005; post-hoc Fisher tests
levels. Because pups must learn the diet-dependent maternal odor between each group). However, pups between 12 and 15 d old (that is,
for interactions with the mother (such as nipple attachment and postsensitive period), an age that represents a transition from nest life
approach), this system ensures that pups only learn to approach to independent life, learned an odor preference while with the mother
maternal odor. The mother’s ability to modify fear learning circuitry and an odor aversion while alone (Fig. 1b; ANOVA, F3,28 ¼ 25.563,
may provide clues to abusive attachment and predisposition for P o 0.0001; post-hoc Fisher tests between each group). In pups of
mental illness and altered emotional expression later in life1–3. weaning age (21–23 d old), odor-shock conditioning produced an odor
The validity of an animal model of abusive attachment is strength- aversion with or without the mother (Fig. 1c; ANOVA, F3,15 ¼ 9.404,
ened by the wide phylogenetic representation of abusive attachment, P o 0.005; post-hoc Fisher tests between each group). This dual
which has been documented in chicks, infant dogs, rodents learning system may ensure that pups still only learn to approach the
and nonhuman primates4,5. Moreover, these data provide insight maternal odor, but also learn to avoid odors they encounter outside the
into the timing and mechanisms of functional emergence of brain nest. The work presented here explored the mechanisms responsible for
areas during development. pups’ dual learning system using a systems-level analysis.

a 5
Sensitive period b 5 Postsensitive period c 5
Number of choices

Number of choices

Number of choices

4 * 4 4
toward odor
toward odor

toward odor

3 3 3

2 2 2
* *
1 1 1
0 0 0
Paired Paired Unpaired Odor only Paired Paired Unpaired Odor only Paired Paired Unpaired Odor only
Without With Without With Without With
maternal presence maternal presence maternal presence maternal presence maternal presence maternal presence

Figure 1 Pup learning from odor-shock conditioning (0.5-mA shock) changes over development and is influenced by maternal (anesthetized) presence.
Behavior was examined using a Y-maze test. (a) 8-d-old rats learned to prefer an odor paired with a shock, with or without maternal presence. (b) When
conditioned without maternal presence, 12- to 15-d-old pups subjected to paired odor-shock learned an odor aversion. Pups that were conditioned with
maternal presence learned an odor preference. (c) Pups of weaning age (21 to 23 d old) learned odor avoidance with or without maternal presence.
*P o 0.05. Error bars represent s.e.m.

Department of Zoology, University of Oklahoma, Norman, Oklahoma 73019, USA. Correspondence should be addressed to R.M.S. (rsullivan@ou.edu).
Received 30 March; accepted 15 June; published online 9 July 2006; doi:10.1038/nn1733


Paired without maternal
Paired with maternal
a Paired with maternal and corticosterone
Unpaired with maternal
b c Saline

Number of choices toward odor
Odor only with maternal 1.5

Mean 2-DG relative uptake
* * *
Mean 2-DG relative uptake

* * * * 4

3 1

0.5 *
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

1 1

0 0 0
Olfactory bulb Lateral nucleus Basolateral nucleus Paired Paired Unpaired Odor only
Amygdala Without With
maternal presence maternal presence

Figure 2 Maternal presence activates a non–amygdala dependent odor-shock circuit and yields odor preference. (a) Olfactory bulb activity during odor-shock
acquisition was assessed by relative 14C 2-deoxyglucose (14C 2-DG) uptake. Enhanced uptake was found in pups subjected to paired odor-shocks with
maternal presence that expressed an odor preference. (b) Activity in the basolateral and lateral nuclei of the amygdala were enhanced during odor-shock
presentation only without maternal presence, as assessed by relative 14C 2-DG uptake. Additional amygdala nuclei and a representative 14C 2-DG/Nissl-stained
amygdala section are shown in Supplementary Figs. 1 and 2, respectively. (c) Reversibly silencing the amygdala with the GABA agonist muscimol disrupted
the odor aversion learning in pups subjected to odor-shock pairings without maternal presence but had no effect on the pups subjected to odor-shock pairings
with maternal presence. *P o 0.05. Error bars represent s.e.m.

Our rationale for assessing how the mother could function as a received either the odor only or unpaired presentations of odor and
‘switch’ between the two learning systems was based on previous data. shock (details in Supplementary Methods online). Pups were condi-
First, the termination of the sensitive period is coincident with the tioned in either the presence or the absence of an anesthetized mother
gradual decline of the pups’ ‘stress hyporesponsive period’ when and were tested the next day in a Y-maze (conditioned odor versus
stressors such as shock begin to produce a surge in corticosterone the familiar odor of clean bedding). Pups subjected to odor-shock
release11,13. In preweanling pups, odor-shock conditioning requires without maternal presence learned to avoid the odor. In contrast,
that corticosterone produce odor aversion learning and basolateral pups subjected to odor-shock with maternal presence developed the
amygdala plasticity11,12. Indeed, giving corticosterone to 7-d-old (that paradoxical shock-induced odor preference (Fig. 1b). The olfactory
is, sensitive period) pups permits aversion learning and engages the bulb only participated in the conditioning for the ‘paired with maternal
amygdala, whereas depleting 12-d-old (that is, postsensitive period) presence’ pups that expressed an odor preference (Fig. 2a). This
pups of corticosterone (by adrenalectomy) reinstates the sensitive enhanced responding of the olfactory bulb is typical of learning-
period11,12. Second, maternal presence suppresses shock-induced cor- associated changes in younger (sensitive period) pups and is associated
ticosterone release in preweanling pups14. with learning the maternal odor4,13 (ANOVA, F4,21 ¼ 7.798, P o0.001;
Here we used 12- to 15-d-old (postsensitive period) pups in an odor- post-hoc Fisher tests between each group). We performed auto-
shock fear conditioning protocol (0.5-mA shock) similar to one that radiographic analysis of the basolateral complex (Fig. 2b) and other
engages the amygdala in adult rats15. For the paired presentations, pups nuclei (Supplementary Figs. 1 and 2 online) of the amygdala during
were administered 11 0.5-mA, 1-s-long tail shocks during the last conditioning. The amygdala’s cortical, medial, basolateral and lateral
second of a 30-s-long presentation of a peppermint odor. Controls nuclei only participated in odor-shock conditioning in the mother’s

a b5 Saline
c 5 Cholesterol
Number of choices toward odor

Number of choices toward odor

Corticosterone * Corticosterone
Corticosterone levels (ng ml )

4 4
80 *
* 3 3
2 2
40 *
1 1 *

0 0 0
Paired Paired Paired Unpaired Odor only Paired Unpaired Odor only Paired Unpaired Odor only
Without With With maternal presence With maternal presence
maternal presence maternal presence

Figure 3 Assessment of the association between corticosterone, learning and the amygdala. (a) Radioimmunoassay (RIA) corticosterone levels were low in pups
receiving shock with maternal presence, but high in those receiving shock without maternal presence. (b) Pups subject to paired odor-shock with maternal
presence were given systemic corticosterone 30 min before conditioning. These pups showed odor aversion learning. (c) Intra-amygdala corticosterone
permitted these pups to learn an odor aversion. *P o 0.05. Error bars represent s.e.m.


absence (ANOVAs: basolateral, F4,20 ¼ 16.577, P o 0.0001; lateral, between each group; placement of infusion cannulae are shown in
F4,20 ¼ 27.940, P o 0.0001; cortical, F4,19 ¼ 10.796, P o 0.0001; Supplementary Fig. 4 online).
medial, F4,20 ¼ 7.425, P o 0.001; post-hoc Fisher tests between each In summary, our data suggest that preweanling pups have two odor-
group; details in Supplementary Methods). Next, we reversibly shock learning circuits, with maternal presence providing suppression
silenced the amygdala with muscimol (0.5 nmol; GABAA receptor of stress-induced corticosterone release and engaging the odor-shock
antagonist) and found a causal relationship between maternal presence, circuit for odor preference learning supporting infant-mother attach-
odor-shock conditioning and amygdala participation. The amygdala ment. These data provide insight into the timing and mechanisms
silencing disrupted odor aversions learned from odor-shock pairings in of functional emergence of the amygdala and suggests ways in
the ‘without maternal presence’ condition, suggesting that the which the functional maturation of brain development may be dis-
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

amygdala is important in aversion-induced odor-shock conditioning. rupted by stress.
However, this manipulation did not disrupt paired conditioning in
Note: Supplementary information is available on the Nature Neuroscience website.
the ‘with maternal presence’ condition (Fig. 2c; ANOVA, F3,24 ¼ 3.667,
P o 0.05; post-hoc Fisher tests between each group; placements of ACKNOWLEDGMENTS
infusion cannulae shown in Supplementary Fig. 3 online). Supported by grants to R.M.S. from the US National Institute of Child Health
Due to the important role of corticosterone in infant rat learning and Human Development (HD33402), the US National Science Foundation
(IOB0544406) and the Oklahoma Center for the Advancement of Science and
and in the mother’s ability to reduce shock-induced corticosterone
Technology; and by funds to S.M. from the University of Oklahoma.
release in pups, we assessed corticosterone’s effect on learning
and amygdala activity10,11. Corticosterone levels during conditioning COMPETING INTERESTS STATEMENT
were significantly higher in pups subjected to paired condition- The authors declare that they have no competing financial interests.
ing without maternal presence than in those with maternal presence,
Published online at http://www.nature.com/natureneuroscience
including shock and nonshock groups (Fig. 3a; ANOVA F3,16 ¼ 11.794, Reprints and permissions information is available online at http://npg.nature.com/
P o 0.0005; post-hoc Fisher tests between each group). Maternal reprintsandpermissions/
effects on corticosterone and fear learning were further supported
through systemic and intra-amygdala infusions of corticosterone. 1. Teicher, M.H. et al. Neurosci. Biobehav. Rev. 27, 33–44 (2003).
In pups subjected to paired odor-shock with maternal presence, 2. Heim, C. & Nemeroff, C.B. Biol. Psychiatry 49, 1023–1039 (2001).
3. Pollak, S.D. & Kistler, D.J. Proc. Natl. Acad. Sci. USA 99, 9072–9076 (2002).
the systemic administration of corticosterone (3 mg) 30 min before 4. Sullivan, R.M. Ann. NY Acad. Sci. 1008, 122–131 (2003).
conditioning enabled odor aversion learning, as well as the incorpo- 5. Maestripieri, D., Lindell, S.G., Ayala, A., Gold, P.W. & Higley, J.D. Neurosci. Biobehav.
ration of the amygdala into the learning circuit12,13 (Fig. 3b; Rev. 29, 51–57 (2005).
6. Haroutunian, V. & Campbell, B.A. Science 205, 927–929 (1979).
behavior ANOVA, F2,12 ¼ 11.400, P o 0.005; post-hoc Fisher 7. Camp, L.L. & Rudy, J.W. Dev. Psychobiol. 21, 25–42 (1988).
tests between each group; olfactory bulb and amygdala data are 8. Sullivan, R.M. et al. Nature 407, 38–39 (2000).
9. Barr, G.A. NIDA Res. Monogr. 158, 172–201 (1995).
included in Figs. 2a and 2b, respectively, along with the associated 10. Roth, T.L. & Sullivan, R.M. Biol. Psychiatry 57, 823–831 (2005).
statistics). Furthermore, in these pups (odor-shock with maternal 11. Levine, S. Eur. J. Pharmacol. 405, 149–160 (2000).
presence), direct infusion of corticosterone (50 ng) into the amygdala 12. Moriceau, S., Wilson, D.A., Levine, S. & Sullivan, R.M. J. Neurosci. 26, 6737–6748.
13. Moriceau, S. & Sullivan, R.M. Behav. Neurosci. 118, 274–281 (2004).
during conditioning resulted in the learning of an odor aversion12 14. Stanton, M.E. & Levine, S. Dev. Psychobiol. 23, 411–426 (1990).
(Fig. 3c; ANOVA, F2,20 ¼ 35.362, P o 0.0001; post-hoc Fisher tests 15. Fanselow, M.S. & LeDoux, J.E. Neuron 23, 229–232 (1999).


Anterior cingulate cortex underpinnings of social rejection and expectancy violation. Here we
demonstrate that the dACC is sensitive to expectancy violations
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

responds differentially to independent of social rejection, whereas the vACC responds specifically
to social feedback.
expectancy violation and Forty-two right-handed subjects participated in two functional
magnetic resonance imaging (fMRI) experiments in which we mea-
social rejection sured neural activity while subjects viewed a series of unfamiliar faces.
Several weeks before scanning, subjects were photographed and led to
Leah H Somerville, Todd F Heatherton & William M Kelley believe that individuals at other institutions would be forming impres-
sions of them during this interim period. In the scanner, subjects
This study investigated human anterior cingulate cortex (ACC) viewed faces and were asked to either form a first impression of these
involvement during a task that dissociated expectancy violation ‘participants’ from other universities (that is, ‘‘Do you think you would
from social rejection. Across two studies, participants like this person?’’, experiment 1, n ¼ 20) or to predict whether the
underwent functional magnetic resonance imaging while ‘participant’ would accept or reject them (that is, ‘‘Do you think this
making social judgments and receiving fictitious feedback that person would like you?’’, experiment 2, n ¼ 22; Fig. 1). The task
was either positive or negative and consistent or inconsistent judgment in experiment 2 provided a more explicit assessment of
with their expectations. The results demonstrate that the expectancy violation. All subjects provided written informed consent
dorsal ACC is sensitive to expectancy violations, whereas the for their participation.
ventral ACC is differentially responsive to social feedback. In both experiments, subjects were given false feedback for some of
the faces, indicating that the person in the photo had previously formed
The desire to avoid social rejection is a powerful motive. Humans a negative or positive first impression of the subject. This approach
are a social species, and individuals who were shunned by
their social groups were unlikely to have survived on their own. Over
the course of human evolution, therefore, it is likely that the brain TIME STIMULUS ACTION
adapted neural circuits to detect and cope with rejection by group
members. Given the fundamental importance of social pain, it is
perhaps not surprising that brain regions commonly associated with CUE Subject decides,
0–3,000 ms "Would I like this person?"
physical pain have been implicated as crucial for the experience of
social pain1. Specifically, a recent study found that a region of the dorsal
anterior cingulate cortex (dACC) was responsive during a video
game thought to elicit feelings of social rejection when the virtual Subject response
players suddenly and surprisingly stopped cooperating with the 1,000–4,000 ms
Yes displayed to the left
research participant. of the face
Although notable, the results of this previous study challenge a
prevalent theory of a dorsal-cognitive/ventral-emotional functional
dissociation within anterior cingulate cortex2. Specifically, activity in Feedback from
the dACC often signals the occurrence of cognitive conflicts during a Yes No target individual
2,000 ms
displayed on the right
variety of tasks that encourage response competition (for example, the
Stroop task), including those that involve the commission of errors3,4.
By contrast, activity in the ventral ACC (vACC) is more typically Figure 1 Representation of the subcomponents of a complete trial. Time
associated with social and emotional processes2,5. A critical issue indicates the duration of each subcomponent. During the CUE period,
that complicates the interpretation of the social pain study1 is subjects viewed a target face and responded to the question, ‘‘Would I like this
that the method used to induce social rejection probably violated person?’’ (experiment 1) with a ‘yes’ or ‘no’ button press. The DELAY period
research participants’ expectations. Put simply, the participants began immediately following the button press, during which the subject’s
expected to participate in a ball-tossing game, and, when this judgment appeared to the left of the face. Following the DELAY, subjects were
given fictitious FEEDBACK (made up by the experimenters and believed by
expectation was violated, it probably created a situation high in
the subjects) indicating whether the subject was accepted or rejected by the
cognitive conflict. Thus, the resulting activation patterns may have pictured individual. This example represents an incongruent rejection trial.
reflected either cognitive conflict or social pain. The current study Experiment 2 trials were identical except for the CUE judgment, during which
was designed to allow for an independent examination of the neural subjects answered the question, ‘‘Would this person like me?’’

Department of Psychological and Brain Sciences and Center for Cognitive Neuroscience, Dartmouth College, Hanover, New Hampshire 03755, USA. Correspondence should
be addressed to W.M.K. (william.kelley@dartmouth.edu).
Received 10 February; accepted 25 May; published online 2 July 2006; doi:10.1038/nn1728


line; whole-brain activation patterns are sum-
a b Region x expectancy violation c Region x feedback type
marized in Supplementary Table 1 online).
0.6 dACC vACC 0.8 dACC vACC
Taken together, these findings support a

Signal change
Signal change
0.4 0.6
0.2 0.4
0 0.2 * general role for dACC in the processing of
–0.2 0 cognitive conflict and demonstrate a more
–0.4 * Congruent –0.2 Rejected
Expectancy Feedback Incongruent Accepted specific role for vACC in social and emotional
violation type –0.6 –0.4
evaluation—both of which are consistent with
Figure 2 Differential ACC response to expectancy violation and social feedback. (a) A three-dimensional current theories2,3 of ACC functioning. To the
rendering of the medial surface of the brain illustrates a functional dissociation between dorsal (dACC) extent that people expect consistency in social
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

and ventral (vACC) anterior cingulate. A whole-brain voxel-by-voxel analysis of variance (ANOVA) was exchange6,7, dACC activity reported in the
used to identify voxels that showed a significant main effect (P o 0.001, uncorrected) of expectancy present study and elsewhere1 may well reflect
violation (blue) and a main effect of feedback type (yellow). (b,c) Voxels in the dACC (BA 32: –6, 28,
violations of the fundamental expectation of
32; 13 voxels) demonstrated greater sensitivity to expectancy violation (incongruent 4 congruent) (b),
whereas voxels in the vACC (BA 32/10: –6, 49, –13; 16 voxels) demonstrated greater sensitivity social inclusion.
to feedback (accepted 4 rejected) (c). Error bars denote s.e.m.
Note: Supplementary information is available on the
Nature Neuroscience website.

permitted a factorial analysis that examined neural responses specific to ACKNOWLEDGMENTS
feedback as a function of expectancy violation (that is, when feedback We thank E. Cross, J. Dobson, T. Johnstone, N. Magoon, T. Moran, J. Van Horn
and P. Whalen. Supported by the National Institute of Mental Health (MH59282,
matched versus did not match subjects’ first impressions or predic-
MH66720). L.H.S. is a National Science Foundation Graduate Research Fellow.
tions) and social feedback (that is, when feedback was negative versus
positive; details in Supplementary Methods online).
Results revealed a double dissociation between dorsal and ventral COMPETING INTERESTS STATEMENT
ACC regions (Fig. 2a). Whereas a dACC region (Brodmann’s area, The authors declare that they have no competing financial interests.
BA, 32) was sensitive to expectancy violation (incongruent 4 congru- Published online at http://www.nature.com/natureneuroscience
ent), a region of the vACC was sensitive to feedback type (accepted 4 Reprints and permissions information is available online at http://npg.nature.com/
rejected). This was evidenced by significant interactions between region reprintsandpermissions/
and expectancy violation (F1,40 ¼ 8.0, P o 0.01; Fig. 2b) and between
1. Eisenberger, N.I., Lieberman, M.D. & Williams, K. Science 302, 290–292 (2003).
region and feedback type (F1,40 ¼ 12.3, P o 0.005; Fig. 2c). Addition- 2. Bush, G., Luu, P. & Posner, M.I. Trends Cogn. Sci. 4, 215–222 (2000).
ally, these interactions were not modulated by study (region  3. Botvinick, M.M., Cohen, J.D. & Carter, C.S. Trends Cogn. Sci. 8, 539–546 (2004).
expectancy violation  study: F o 1; region  feedback  study: 4. Carter, C.S. et al. Science 280, 747–749 (1998).
5. Whalen, P.J. et al. Biol. Psychiatry 44, 1219–1228 (1998).
Fo1). That is, the functional dissociation between dACC and vACC 6. Condon, J.W. & Crano, W.D. J. Pers. Soc. Psychol. 54, 789–797 (1988).
was independently present in both studies (Supplementary Fig. 1 on- 7. Newcomb, T.M. The Acquaintance Process (Holt, Reinhart and Winston, New York, 1961).


BDNF-mediated neurotransmission relies upon a
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

myosin VI motor complex
Hiroko Yano1, Ipe Ninan2, Hong Zhang2, Teresa A Milner3, Ottavio Arancio2 & Moses V Chao1

Brain-derived neurotrophic factor (BDNF) has been implicated in higher-order cognitive functions and in psychiatric disorders
such as depression and schizophrenia. BDNF modulates synaptic transmission and plasticity primarily through the TrkB receptor,
but the molecules involved in BDNF-mediated synaptic modulation are largely unknown. Myosin VI (Myo6) is a minus end–
directed actin-based motor found in neurons that express Trk receptors. Here we report that Myo6 and a Myo6-binding protein,
GIPC1, form a complex that can engage TrkB. Myo6 and GIPC1 were necessary for BDNF-TrkB–mediated facilitation of long-term
potentiation in postnatal day 12–13 (P12–13) hippocampus. Moreover, BDNF-mediated enhancement of glutamate release
from presynaptic terminals depended not only upon TrkB but also upon Myo6 and GIPC1. Similar defects in basal synaptic
transmission as well as presynaptic properties were observed in Myo6 and GIPC1 mutant mice. Together, these results define
an important role for the Myo6-GIPC1 motor complex in presynaptic function and in BDNF-TrkB–mediated synaptic plasticity.

In addition to neuronal survival and differentiation, neurotrophins involving synapsin phosphorylation20. However, the underlying
have a central role in synaptic transmission and plasticity1–4. Expres- mechanisms for how TrkB receptors are used by BDNF to mediate
sion and secretion of neurotrophins, such as nerve growth factor the facilitation of glutamate release from presynaptic terminals remain
(NGF), BDNF and neurotrophin (NT)-3, are regulated in an activ- largely unknown.
ity-dependent manner. Much attention has focused upon the actions of Recent studies have suggested that actin has an important role in
BDNF and its receptor TrkB, which is located at pre- and postsynaptic neurotransmitter release23. Movement of synaptic vesicles along actin
sites5,6, in directly contributing to changes in synaptic strength and filaments in the nerve terminal is believed to represent an essential step
plasticity4,7. For example, the involvement of BDNF and TrkB in long- in the release of neurotransmitters and may require the action of actin-
term potentiation (LTP) has been confirmed by many studies using based motor proteins. Actin-based motors are likely to be involved in
BDNF and TrkB knockout mice8–10, function-blocking antibodies to directing membrane trafficking events. Whether these proteins are
TrkB (ref. 11), and TrkB-IgG receptor bodies that inhibit ligand- involved in neurotrophin-regulated synaptic vesicle recycling has not
induced signaling12. Alterations in the availability of BDNF through yet been investigated.
processing and activity-dependent secretion have been proposed to In this study, we report that a presynaptic actin motor complex
underlie the pathophysiology of depression, bipolar disease and containing Myo6 and a PDZ (postsynaptic density-95/Discs large/zona
Huntington disease13–15. occludens-1) domain–containing adaptor protein, GIPC1 (ref. 24), is
The effects of BDNF upon synaptic transmission have been studied responsible for BDNF-TrkB–induced transmitter release. Myo6 is an
in both the peripheral and central nervous systems1–4. Acute exposure unconventional minus end–directed actin motor25,26. The interaction
of neuromuscular synapses to BDNF results in an increase in the of TrkB with Myo6 depends upon GIPC1, which can associate with
frequency, but not the amplitude, of spontaneous synaptic currents both Myo6 and Trk simultaneously. Using biochemical, genetic and
(miniature excitatory postsynaptic current, mEPSC)16, suggesting a electrophysiological approaches, we demonstrate that both Myo6 and
presynaptic locus for BDNF modulation. Similarly, BDNF rapidly GIPC1 are essential for BDNF-TrkB–mediated neurotransmitter release
enhances spontaneous neurotransmitter release in dissociated cultures in hippocampal neurons.
of hippocampal neurons17,18. Another effect of BDNF upon
hippocampal synaptic transmission is acute presynaptic enhancement RESULTS
of evoked glutamate release19,20. These effects have been attributed Trk receptor forms a protein complex with GIPC1 and Myo6
to TrkB receptors that are localized at presynaptic sites21. Notably, The BDNF receptor TrkB is widely expressed in the CNS and is largely
BDNF has been implicated in the mobilization and/or docking responsible for activity-dependent synaptic changes2,4,7. The juxta-
of synaptic vesicles to the active zones22, events possibly membrane region of the Trk receptor tyrosine kinase binds several

1Molecular Neurobiology Program, Skirball Institute of Biomolecular Medicine, Departments of Cell Biology, Physiology and Neuroscience, New York University School of

Medicine, 540 First Avenue, New York, New York 10016, USA. 2Department of Pathology and the Taub Institute for Research on Alzheimer’s Disease and the Aging Brain,
Columbia University, 630 West 168th Street, New York, New York 10032, USA. 3Department of Neurology and Neuroscience, Weill Medical College of Cornell University,
411 East 69th Street, New York, New York 10021, USA. Correspondence should be addressed to M.V.C. (chao@saturn.med.nyu.edu).
Received 3 April; accepted 30 May; published online 2 July 2006; doi:10.1038/nn1730


unique proteins involved in receptor signaling and trafficking. Yeast TrkB, regardless of the presence of the ligand BDNF (Fig. 1a, top). Next,
two-hybrid screening and biochemical analysis have shown that the we examined the endogenous GIPC1-Myo6 interaction in cultured
PDZ domain–containing GIPC1 protein (GAIP-interacting protein, hippocampal neurons. Myo6 efficiently coimmunoprecipitated with
C terminus27; also known as GLUT1CBP (ref. 28) and SEMCAP-1 GIPC1 (Fig. 1a, bottom). These results indicated that GIPC1 interacts
(ref. 29)) interacts directly with the juxtamembrane domains of TrkB with TrkB and Myo6 endogenously in primary neurons. To assess
and the NGF receptor TrkA (ref. 24). The GIPC1 protein can also bind whether TrkB could form a protein complex with Myo6 and GIPC1,
directly to Myo6, a minus end–directed actin motor protein28,30. we carried out immunoprecipitation of TrkB receptors after transfection
Because GIPC1 has been reported to bind to both Trk and Myo6, with Flag-tagged GIPC1 in 293 cells. Endogenous Myo6 could be
we investigated the possibility that a complex is formed among the coimmunoprecipitated with TrkB and Flag-GIPC1 (Fig. 1b). Myo6
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

three proteins. was not observed in control immunoprecipitations lacking TrkB recep-
To determine if TrkB associates with GIPC1 endogenously, we tors (Fig. 1b). In addition GIPC1 can form a complex with Trk receptors
performed immunoprecipitation and western blot experiments in in Myo6-deficient primary neurons, and GST-Trk could not pull down
primary neurons. GIPC1 proteins coimmunoprecipitated with Myo6 in brain lysates lacking GIPC1 (Supplementary Fig. 1 online).

a IP Lysate b e

Myo6 (1,004–1,254)
Myo6 (1,036–1,254)

Myo6 (1,004–1,254)
Myo6 (1,036–1,254)
Flag-GIPC1 + +

TrkB – +



WB: Myo6 150 Myo6


WB: GIPC1 37
50 100
IP: TrkB WB: Flag Flag-GIPC1 35S-GIPC1
37 75
WB: Trk 150 37 50
25 37
WB: pTrk 150 WB: TrkB TrkB 25
15 20
IP Endogenous Coomassie blue
WB: Myo6 Myo6


Lysate WB: Flag Flag-GIPC1
Myo6 Motor domain
1,004–1,254 (Myo6-C)
WB: Myo6 150 1,036–1,254
WB: TrkB TrkB

WB: GIPC1 37

c 35S-GIPC1-:
81 PDZ
FL –333 +C N C PDZ
35S-GIPC products















50 100
37 50
GFP-Myo6 Flag-GIPC1 Overlay 20 25
d 15

Myo6 TrkA
GIPC1 PDZ binding binding24
1 81 125 225 333
GFP-Myo6 FL +
GFP-Myo6 GIPC1-DsRed GIPC1-DsRed 81–333 ++ +
PDZ+C ++ +
N – –
C ++ – Myo6
PDZ – +

Trk Overlay

Figure 1 Trk associates with GIPC1-Myo6 motor complex. (a) Endogenous Trk-GIPC1 and GIPC1-Myo6 interactions. Rat hippocampocortical (top) and
hippocampal (bottom) lysates were incubated with anti-Trk and antibody to GIPC1, respectively, along with control normal rabbit IgG. Immunoprecipitates were
immunoblotted with indicated antibodies. BDNF: 50 ng ml–1. For Trk-GIPC1 interaction, cells were first crosslinked with a cell-permeable cross-linker,
dithiobis(succinimidylpropionate) (DSP). (b) TrkB-GIPC1-Myo6 complex. Lysates from 293 cells transfected with Flag-GIPC1 and TrkB constructs were subjected
to immunoprecipitation with anti-Trk, and immunoprecipitates were immunoblotted with indicated antibodies. Control immunoprecipitates were performed with
lysates lacking TrkB. (c,d) Colocalization of Trk, GIPC1 and Myo6. GFP-Myo6 was coexpressed with either Flag-GIPC1 (c) or GIPC1-DsRed (d) in primary
hippocampal neurons (days in vitro, DN, 14–17). Cells were fixed and immunostained with anti-Flag or anti-Trk followed by appropriate fluorescent secondary
antibodies. Fluorescence signals were analyzed by confocal microscopy. GFP-Myo6 colocalized with GIPC1 in the cell body and along processes (yellow). Higher
magnification images reveal colocalization of all three molecules, GFP-Myo6, GIPC1-DsRed and Trk, as seen in white (d). Scale bar, 10 mm. (e) Mapping the
interaction between GIPC1 and Myo6. In vitro translated, 35S-methionine–labeled full-length (FL) GIPC1 or its various truncation mutants were incubated with
GST-Myo6 fusion proteins or GST alone. The bound 35S radioactivity was visualized by autoradiography. Direct interaction between GIPC1 and Myo6 (1,004–
1,254) (Myo6-C) was found. Shorter Myo6 peptides (1,036–1,254) bound GIPC1 with weaker affinity. Comparable levels of various 35S-GIPC1 mutants were
produced (input). Coomassie blue staining confirmed similar amounts of GST fusion proteins. Schematic of the GIPC1 truncation mutants and summary of
GST-Myo6-C binding results are shown. Summary includes previous mapping results with GST-TrkA juxtamembrane (adapted from ref. 24) for comparison.


sm Figure 2 Synaptic localization of GIPC1. (a,b) Immunohistochemical analysis
a b of GIPC1 in rat hippocampal coronal sections using anti-GIPC1 (981) (light
microscopy). hi, hilus; sg, stratum granulosum; sm, stratum moleculare; so,
sg so
stratum oriens; sp, stratum pyramidale; sr, stratum radiatum. GIPC1
CA3 immunoreactivity was present in interneurons in the dentate gyrus (a), in
CA1 (b) and CA3 (a) pyramidal neurons in hippocampus. Neurons in hilus of
dentate gyrus showed darkly labeled somata and processes (a). In the CA1
sr region, neuronal somata and apical dendrites extending into stratum
radiatum were labeled (b). (c,d) No immunoreactivity was observed in
Gipc1–/– hippocampal sections (d) compared to those of the wild type (c).
(e–i) Ultrastructural analysis of GIPC1 in adult rodent hippocampus.
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

Immuno-EM with anti-GIPC1 revealed pre- and postsynaptic localization of
c d GIPC1 (e–h). GIPC1 proteins were also expressed in a dendritic shaft (i).
Primary and secondary antibodies were visualized by immunoperoxidase
(a–f) or immunogold (g–i) methods. Scale bars: 50 mm in a,b; 100 mm in c,d;
e uT
Sp f 500 nm in e,f,h,i; 250 nm in g. Sp, spine; SSV, small synaptic vesicles;
T, labeled nerve terminal; uT, unlabeled terminal; D, dendrites.

Sp uT
mutants. The C-terminal domain of GIPC1 bound directly to Myo6,
uT T whereas the N-terminal and PDZ domains of GIPC1 were not involved
(Fig. 1e). Therefore, Myo6 and GIPC1 interact through their
C-terminal domains, supporting a previous report using a different
assay30. Along with the colocalization and binding results (Fig. 1a–d),
these data indicate that GIPC1 can form a protein complex with Myo6
g h i and Trk through discrete nonoverlapping domains (Fig. 1e).
Lack of Myo6 or GIPC1 impairs basal synaptic transmission
Sp Myo6 is localized to both pre- and postsynaptic structures in the
Sp hippocampus32. To determine where GIPC1 is localized in the hippo-
T campal formation, we performed light and electron microscopic (EM)
analyses (Fig. 2). By light microscopy, GIPC1 immunoreactivity was
found in hilar interneurons in the dentate gyrus and in CA3 and CA1
pyramidal cell neurons (Fig. 2a,b). In CA1 stratum radiatum, GIPC1
Together these results suggest that Trk receptors form a protein immunoreactivity was in dendritic processes (Fig. 2b). Analysis of
complex with Myo6 through GIPC1. GIPC1 immunoreactivity in GIPC1-deficient mice resulted in no
We then examined the localization of Trk, GIPC1 and Myo6 in immunostaining in any of these regions (Fig. 2c,d). EM analysis of
hippocampal neurons to confirm our immunoprecipitation results. We CA1 stratum radiatum revealed GIPC1 immunoreactivity in
expressed recombinant tagged proteins, GFP-Myo6 and either Flag- postsynaptic dendritic spines as well as presynaptic axon terminals
GIPC1 or GIPC1-DsRed, in primary neurons and immunostained (Fig. 2e–h). Within these terminals, GIPC1 immunoreactivity, as
endogenous Trk (Fig. 1c,d). We observed a very similar pattern identified by immunogold staining, was affiliated with small synaptic
of expression for GIPC1 and Myo6 (Fig. 1c,d). GFP-Myo6 punctae vesicles (Fig. 2g). We also observed GIPC1 immunoreactivity in
colocalized with GIPC1 in the cell soma and the processes (Fig. 1c,d). A dendritic shafts (Fig. 2i). These results indicate that GIPC1, like
small fraction of these punctae was seen to be motile in transfected Myo6, is expressed in both pre- and postsynaptic sites.
hippocampal neurons by live imaging analysis (Supplementary The Snell’s waltzer mutant mouse (Myo6sv/Myo6sv, hereafter referred
Fig. 2 online). Detailed analysis of endogenous GIPC1 and Myo6 to as Myo6sv/sv) possesses a deletion of the Myo6 gene, resulting in
immunostaining showed a similar punctate distribution in the cell deafness33. A deficiency in Myo6 resulted in fewer synapses and
body and throughout the processes of primary hippocampal neurons dendritic spines in the hippocampus32. To examine if a loss of
(data not shown). Trk receptors were widely distributed in the cell Myo6 gives rise to other synaptic deficits, we pursued further
body and processes, and a subset of these receptors colocalized with analysis by performing electrophysiological measurements of the Snell’s
GIPC1-Myo6 punctae (Fig. 1d). waltzer mice.
Myo6 has distinct protein binding domains. The N-terminal motor Myo6-deficient hippocampal slices were generated and first exam-
domain of Myo6 directly binds to actin26, whereas the C-terminal ined for basal synaptic transmission at Schaffer collateral–CA1 synapses
domain is regarded as a cargo-binding region31. To define the domain (Fig. 3a). Input-output curves were obtained by extracellular field
of Myo6 that binds to GIPC1, we carried out an in vitro binding assay recording. Basal synaptic transmission was reduced in Myo6-deficient
using 35S-methionine–labeled GIPC1 and glutathione S-transferase mice compared to wild-type littermates (Myo6sv/sv at 35 V: 63% of
(GST)-fusion proteins of Myo6 (Fig. 1e). We observed a direct wild-type littermate control; P o 0.05; Fig. 3a). This is the first
interaction between GIPC1 and the C-terminal domain of Myo6 reported evidence for a pronounced defect in synaptic transmission
(amino acids 1,004–1,254), consistent with earlier observations28,30. in the Snell’s waltzer mice.
Thus, the binding domain of GIPC1 to Myo6 is different from its To investigate the physiological significance of GIPC1 in vivo, we
binding to actin. generated a mouse strain lacking expression of GIPC1 by homologous
GIPC1 binds to the juxtamembrane region of the Trk receptor recombination (Fig. 3b,c). The targeting vector was constructed by
through its PDZ domain24. To identify how GIPC1 binds Myo6, we replacing exons 2–4 encoding the entire N-terminal, PDZ and a part of
used the GST-Myo6 C terminus and different 35S-GIPC1 deletion the C-terminal domain with a neo-resistant gene (Fig. 3b). Expression


of GIPC1 in homozygous null mice was not detected by western blot Presynaptic deficits in Myo6- and GIPC1-deficient mice
analysis of brain lysates (Fig. 3d). Gipc1 knockout mice were viable and To determine the roles for Myo6 and GIPC1 in synaptic function,
fertile with normal life span. An analysis of Gipc1–/– adult brains we conducted additional electrophysiological measurements in
revealed normal morphology in the hippocampus (data not shown). hippocampal slices from Myo6- and GIPC1-deficient mice. We first
Coimmunoprecipitation with antibody to GIPC1 (anti-GIPC1) examined two presynaptically driven forms of short-term synaptic
from adult mouse brain lysates confirmed an endogenous inter- plasticity, paired-pulse facilitation (PPF) and post-tetanic potentiation
action between Myo6 and GIPC1, which was absent in Gipc1–/– lysates (PTP) at Schaffer collateral–CA1 excitatory synapses.
(Fig. 3e). These results verified that GIPC1 and Myo6 form a PPF is thought to reflect an accumulation of residual Ca2+ due to the
protein complex. first action potential depolarizing of the terminal, leading to enhanced
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

To assess if GIPC1 is involved with upon hippocampal synaptic transmitter release at the arrival of a second stimulus. In slices from
function, we carried out field recordings on Gipc1–/– mice. Basal wild-type controls, we observed PPF at interstimulus intervals of
synaptic transmission was reduced in hippocampal slices prepared 20–300 ms (Fig. 4a,b). Notably, slices from Myo6sv/sv and Gipc1–/–
from Gipc1–/– mice (Fig. 3a; Gipc1–/– at 35 V: 54% of wild-type mice showed increased PPF at interstimulus intervals of 20–500 ms
littermate control; P o 0.05) to a similar extent as in the Myo6- compared with wild-type littermate controls (two-way analysis
deficient mice (Fig. 3a). The reduction in basal synaptic transmission of variance (ANOVA): Myo6sv/sv F1,17 ¼ 7.691, P o 0.05; Gipc1–/–
from Myo6- and GIPC1-deficient mice may stem from a number of F1,15 ¼ 5.656, P o 0.05; Fig. 4a,b).
different reasons, including defects in presynaptic function, a lack of We also examined PTP, which is thought to be caused by enhanced
response from postsynaptic sites or a reduction in synaptic proteins presynaptic transmitter release due to loading of the presynaptic
or structure. terminal with Ca2+ after high frequency stimulation34. PTP was

a 3.5
WT n = 9
WT n = 9 d
3.0 Myo6 sv/sv n = 10 3.0 Gipc1 –/– n = 7
WT Gipc1–/–
fEPSP slope (V s–1)

fEPSP slope (V s–1)

2.5 2.5
2.0 2.0
1.5 1.5 Myo6 150
1.0 1.0 100

0.5 0.5 Trk 150

0.0 0.0 100
0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35
Stimulus (V) Stimulus (V) Actin

Sp Sa 5′-P X X Ex2 Ex3 Ex4 K 3′-P Sp X
WT allele e
10.7 kb WT Gipc1–/–
12.4 kb GIPC1


IP: WTGipc1
Sp X
Neor TK 250
1.2 kb 2.4 kb WB: Myo6 150
Sp Sa 5′-P X Sp X K 3′-P Sp X
Mut allele Neor
8.2 kb
5.5 kb WB: GIPC1
c Adult mouse brain




12.4 kb (WT)

10.7 kb (WT)
8.2 kb (Mut) 450 bp (WT)
5.5 kb (Mut) 300 bp (Mut)

5′-probe 3′-probe PCR
Spe I Xba I

Figure 3 Impaired basal synaptic transmission in Myo6- and GIPC1-deficient mice. (a) Input-output curves showing decreased basal synaptic transmission in
Myo6sv/sv and Gipc1–/– mice. Input-output curves of field excitatory postsynaptic potential (fEPSP) slope (V s–1) versus stimulus (V) at Schaffer collateral–CA1
synapses in adult hippocampal slices from Myo6sv/sv (left) and Gipc1–/– (right) mice, and their control wild-type littermates. Data are presented as mean ±
s.e.m. Similar results were obtained when the fEPSP slope was plotted versus the amplitude of the fiber afferent volley (data not shown). Filled circles, wild
type (‘WT’); open circles, knockout. (b) Schematic of wild-type Gipc1 allele, targeting construct, and mutant (‘Mut’) allele. Neor, neomycin-resistant gene;
TK, thymidine kinase gene. Sp, Sa, X and K denote the restriction enzymes SpeI, SacI, XbaI and KpnI, respectively. 5¢-P and 3¢-P represent the location of
Southern probes. (c) Embryonic stem cell clones that underwent homologous recombination in Gipc1 gene locus were confirmed by Southern blot analysis
using 5¢-probe (left) and 3¢-probe (middle) depicted in b, as well as an internal neo probe (data not shown). DNA was digested with SpeI (left) or XbaI (middle)
for hybridization. PCR genotyping using mouse tail DNAs is shown (right). Primer sites used for genotyping are shown in b (arrows). (d) Absence of GIPC1
protein in the Gipc1–/– mouse was verified by immunoblot analysis of wild-type and Gipc1–/– adult brain lysates. (e) Coimmunoprecipitation of Myo6 with
anti-GIPC1 was found in the wild-type, but not in Gipc1–/–, adult brain lysates.


Figure 4 Presynaptic deficits in Myo6- and
a 250 PPF b 250
PPF GIPC1-deficient mice. (a–d) Extracellular field
recording at Schaffer collateral–CA1 synapses
Percentage of facilitation

Percentage of facilitation
200 WT n = 9 200 WT n = 10 in hippocampal slices from Myo6sv/sv (a,c) and
Myo6 sv/sv n = 10 Gipc1–/– n = 7 Gipc1–/– (b,d) mice was carried out to examine
paired-pulse facilitation (PPF) (a,b) and post-
150 150
tetanic potentiation (PTP) (c,d). PPF was
measured as the ratio of fEPSP slopes in response
100 100 to two successive stimuli delivered to the Schaffer
collateral inputs. PPF was augmented in slices
from both Myo6sv/sv and Gipc1–/– mice compared
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

50 50
1 10 100 1,000 1 10 100 1,000 with wild-type littermates. After three ten-burst
Stimulus interval (ms) Stimulus interval (ms) trains in the presence of 50 mM AP5 to block
c 300 PTP d 300 PTP LTP-inductive mechanisms, PTP was decreased
in both Myo6sv/sv and Gipc1–/– slices compared

fEPSP slope (% of baseline)
fEPSP slope (% of baseline)

WT n = 6 WT n = 6
250 Myo6 sv/sv n = 6 250 Gipc1–/– n = 7 with slices from wild-type littermates. Data are
presented as mean ± s.e.m.
200 200

150 150

F1,6 ¼ 0.4177, P 4 0.05; Gipc1–/–: F1,6 ¼
0.04055, P 4 0.05; data not shown). These
50 50
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
findings suggest that Myo6 and GIPC1
Time (s) Time (s) are dispensable for the induction of LTP in
adult hippocampus.
Exogenous BDNF can promote the induc-
induced in the presence of the NMDA receptor antagonist D(-)-2- tion of LTP by tetanic stimulation in hippocampal slices from P12–13,
amino-5-phosphonovaleric acid (AP5, 50 mM) to block LTP-inductive an age at which LTP is observed only after BDNF application12.
mechanisms (Fig. 4c,d). PTP was reduced in both Myo6sv/sv and Therefore, we examined the effects of BDNF (50 ng ml–1, 2.5 h) on
Gipc1–/– mice (66% and 77%, respectively, of PTP in control wild- LTP of Schaffer collateral–CA1 synapses at P12–13. Deletion of Myo6
type littermates at the first time point after the tetanus; P o 0.05 for or GIPC1 blocked the BDNF-mediated facilitation of LTP (two-way
both; Fig. 4c,d). These results suggest that GIPC1- and Myo6-deficient ANOVA; Myo6sv/sv: F1,11 ¼ 0.09138, P 4 0.05; Gipc1–/–: F1,11 ¼
mice manifest presynaptic deficits. 0.01427, P 4 0.05; wild-type: F1,15 ¼ 11.61, P o 0.005; Fig. 5a).
To further analyze potential presynaptic effects of GIPC1 and Taken together, these findings indicate that GIPC1 and Myo6 act
Myo6 deletion, we examined synaptic fatigue, another form of downstream of BDNF in P12–13 hippocampus and support the
short-term plasticity that occurs during high frequency stimulation. hypothesis that association of TrkB with GIPC1 and Myo6 mediates
Synaptic fatigue is characterized by a decrease in synaptic strength, the effects of BDNF on synaptic plasticity.
due in part to a depletion of the release-ready pool of vesicles34.
As opposed to PPF and PTP, synaptic fatigue is not necessarily Increase in EPSC amplitude by BDNF requires Myo6 and GIPC1
dependent upon Ca2+ buildup within the terminal35. We applied Exposure of cultured hippocampal neurons to exogenous
two different high frequency protocols, a 30-s train at 10 Hz and a BDNF increases EPSC amplitude17,18. To establish if Myo6 and
0.4-s train at 100 Hz, in the presence of AP5 (50 mM) onto slices from GIPC1 proteins are involved in BDNF-TrkB regulation of trans-
Myo6- and GIPC1-deficient mice. The data revealed no difference in mission rather than synaptic transmission per se, we applied BDNF
synaptic fatigue between wild-type and mutant mice (two-way (100 ng ml–1) to wild-type, Myo6sv/sv and Gipc1–/– primary hippocam-
ANOVA; Myo6sv/sv: F1,6 ¼ 0.0387 for 30-s train, F1,6 ¼ 0.6114 for pal cultures. We measured the amplitude of EPSCs at synapses
0.4-s train, P 4 0.05 for both; Gipc1–/–: F1,8 ¼ 0.05102 for 30-s between pairs of hippocampal neurons. We found that BDNF increased
train, F1,8 ¼ 0.07499 for 0.4-s train, P 4 0.05 for both; Supplementary EPSC amplitude in wild-type neurons but not in neurons from mutant
Fig. 3 online). Taken together, these electrophysiological assays mice (two-way ANOVA, F3,15 ¼ 2.269, P o 0.05; Fig. 5b). EPSC
suggest that both GIPC1 and Myo6 are required for a subset of amplitude in mutant neurons treated with BDNF was not markedly
presynaptic functions. different from that of mutant neurons treated with vehicle (data
not shown). These results indicate that the effects of BDNF on
BDNF-induced LTP in P12–13 mice requires Myo6 and GIPC1 synaptic responses rely upon GIPC1 and Myo6 proteins downstream
To test whether GIPC1 and Myo6 have effects upon long-lasting of its TrkB receptor.
increases in synaptic strength, we assessed LTP at Schaffer collateral–
CA1 synapses in hippocampal slices from 3- to 5-month-old TrkB signaling is grossly normal without Myo6 and GIPC1
Myo6- and GIPC1-deficient mice. Measurements of LTP using a tetanic The impairment of BDNF-induced synaptic transmission in Myo6-
stimulation protocol (four pulses at 100 Hz with the burst repeated ten and GIPC1-deficient mice could be due to decreased TrkB expression at
times at 5 Hz) did not show significant differences between Myo6sv/sv or the cell surface in mutant neurons or to impaired signaling. We
Gipc1–/– mice compared to wild-type littermates (two-way ANOVA; therefore conducted a cell-surface biotinylation assay of TrkB receptors
Myo6sv/sv: F1,11 ¼ 0.1216, P 4 0.05; Gipc1–/–: F1,13 ¼ 0.8520, P 4 0.05; on Myo6-deficient primary neurons. Surface TrkB receptors from wild-
Supplementary Fig. 3). We obtained similar results when we elicited type and Myo6sv/sv neurons were first biotinylated and then isolated
LTP with a stronger tetanus consisting of four pulses at 100 Hz, with streptavidin beads. The amount of cell-surface or total TrkB
with bursts repeated at 5 Hz and each tetanus including in Myo6-deficient neurons was equivalent to that in wild-type
three ten-burst trains separated by 15 s (two-way ANOVA; Myo6sv/sv: neurons (Fig. 6a). Thus, the impairment of BDNF-induced synaptic


a Young (P12-13) hippocampal slices
WT Myo6 sv/sv Gipc1–/–
200 Vehicle, n = 9 200 Vehicle, n = 7 200 Vehicle, n = 6
fEPSP slope (% of baseline)

BDNF, n = 8 BDNF, n = 6 BDNF, n = 7

150 150 150

100 100 100
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

50 50 50
0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50
Time (min) Time (min) Time (min)

b 240
Figure 5 GIPC1 and Myo6 are necessary for BDNF-induced synaptic transmission. (a) BDNF failed to WT + BDNF

EPSC amplitude (% of baseline)
promote induction of LTP in slices from both Myo6sv/sv and Gipc1–/– mice in contrast to slices from wild- Myo6 sv/sv + BDNF
200 Gipc1–/– + BDNF
type littermates. Exogenous BDNF (50 ng ml–1) was applied to P12–13 hippocampal slices for 2.5 h WT + vehicle
before the y-burst stimulation (arrows). Each point of the graphs represents the average of three
successive events. These experiments were interleaved with each other. These mice had normal basal 160

synaptic transmission, as indicated by the input-output curves we obtained (data not shown). (b) Pairs 140
of monosynaptically connected hippocampal neurons in culture were recorded using the whole-cell 120
technique. Application of BDNF (100 ng ml–1) enhanced EPSCs in wild-type neurons but not in neurons 100
from Myo6sv/sv and Gipc1–/– mice. Each point represents the average of five successive trials, normalized 80
to the average value during the 10 min before BDNF application in each experiment. The average EPSC
amplitudes were 26.06 ± 2.01 pA (wild type, n ¼ 9), 20.12 ± 3.07 pA (Myo6sv/sv, n ¼ 5), and 21.15 ± BDNF
4.81 pA (Gipc1–/–, n ¼ 5) and were not markedly different. Data are presented as mean ± s.e.m. –10 –5 0 5 10 15
Time (min)

transmission in Myo6-deficient neurons was not caused by reduced boutons, an indicator of the number of neurotransmitter release sites.
surface expression of TrkB in these cells. Primary hippocampal neuronal cultures were prepared from Myo6sv/sv
Upon BDNF binding, cell-surface TrkB receptors are activated, and Gipc1–/– neonatal (P1) mice and stained with FM 1-43, a
leading to the stimulation of downstream kinases, such as the fluorescent styryl dye used for labeling presynaptic sites36. Synapses
extracellular signal–regulated kinase (Erk) and Akt families. Erk1/2 that released FM dye upon a second depolarization stimulus were
have been implicated in the BDNF enhancement of evoked glutamate counted, quantified and expressed as the number of active synaptic
release by virtue of the fact that Erks are involved in the phosphoylation boutons per unit length (Fig. 7a–c). Cultures from wild-type litter-
of presynaptic proteins such as synapsin20, a molecule implicated in mates were used as control. The numbers of active boutons per unit
synaptic vesicle cycling. We therefore determined if Myo6 or GIPC1 length in Myo6sv/sv and Gipc1–/– cultures were significantly reduced
affect BDNF-TrkB–mediated signaling pathways. However, we compared to wild-type littermate cultures (wild-type: 14 ± 0.5, n ¼ 42
observed comparable levels of Erk1/2 and Akt activation between neurites; Myo6sv/sv: 9.6 ± 0.47, n ¼ 42; *P o 0.01, Fig. 7b) (wild-type:
wild-type and Myo6-deficient primary neurons after BDNF treatment 13.74 ± 0.45, n ¼ 36; Gipc1–/–: 9.95 ± 0.53, n ¼ 39; *P o 0.001, Fig. 7c).
(Fig. 6b). Similar results were observed with GIPC1-deficient cultured This reduction in the number of active synaptic boutons may be
neurons (Fig. 6c). These results indicate that GIPC1 and Myo6 do not associated with the changes in PPF and PTP observed in Myo6- and
interfere with basic BDNF-induced TrkB signaling. GIPC1-deficient mice.

Fewer active boutons in Myo6sv/sv and Gipc1–/– neurons Rapid synaptic modulation by BDNF requires GIPC1 and Myo6
To determine the basis for the physiological role of Myo6 and GIPC1 in In addition to the effect of BDNF on LTP induction in P12–13
normal presynaptic function, we assessed the number of active synaptic hippocampus (ref. 12 and Fig. 5a), BDNF can also rapidly enhance

Figure 6 TrkB cell surface expression and
a b c

Myo6 sv/sv WT Gipc1 –/– Gipc1 +/–

signaling. (a) The levels of surface-expressed TrkB
W 6

BDNF (min) 0 10 20 40 0 10 20 40 BDNF (min)

0 5 15 30 60 120 0 5 15 30 60 120
in wild-type and Myo6sv/sv neurons were similar.

Avidin 150 Surface pY490-TrkB Level of cell-surface TrkB in Myo6-deficient
pull down TrkB TrkB
primary hippocampal and cortical neurons was
TrkB Trk evaluated by standard surface biotinylation
TrkB experiments followed by SDS–polyacrylamide gel
pErk1/2 pErk1/2 electrophoresis (SDS-PAGE) and western blotting.
Myo6 150 (b,c) Levels of BDNF-induced phosphorylation of
Lysate pS473-Akt
Erk1/2 TrkB, Erk1/2 and Akt in Myo6sv/sv (b) and Gipc1–/–
Actin Akt (c) neurons were similar to those in control
pS473-Akt littermates. Primary hippocampal and cortical
α-tubulin 50 Actin neurons prepared from Myo6sv/sv, Gipc1–/–, and
GIPC1 control wild-type or heterozygous littermate
Myo6 embryos were treated with 25 ng ml–1 BDNF for
the indicated time. Lysates were analyzed by
immunoblotting with the indicated antibodies.


Figure 7 Reduction in the number of neurotransmitter release sites in
a FM 1-43 perfusion 45 s
15 s K+ /45 s normal
Myo6sv/sv and Gipc1–/– mice. (a) The protocol for FM 1-43 staining and
K perfusion 45 s destaining of synaptic vesicles. ADVASEP-7: an anionic cyclodextrin
ADVASEP-7 complexing agent used for enhanced removal of the dye from the external
Washing 5 min medium. Two images were taken after washing (I) and destaining (II). The
I II difference in the number of FM 1-43–stained boutons between the images
Image capture indicates the number of active boutons (neurotransmitter release sites).
(b,c) Both Myo6- and GIPC1-deficient mice had reduced numbers of active
b 16

Number of boutons per 75 µm
synaptic boutons. Activity-dependent FM 1-43 staining in cultured
WT Myo6 sv/sv 14
hippocampal neurons from Myo6sv/sv (b) and Gipc1–/– mice (c). Scale bar,
* 5 mm. The graphs represent the average of all the experiments performed on
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

Myo6sv/sv and Gipc1–/– neurons compared to wild-type littermate neurons. FM
8 1-43–loaded synapses that released the dye upon subsequent depolarization
6 with high concentration of K+ were counted and quantified, and are indicated
4 as the number of active boutons per unit length (75 mm). Data represent
2 means + s.e.m.
WT Myo6 sv/sv

Number of boutons per 75 µm

WT Gipc1 –/– 14 experiments, Fig. 8b). The total releasable neurotransmitter induced by
* high concentration of K+ was similar among all conditions (data not
shown). These results confirm that TrkB kinase activation is required
for BDNF-induced neurotransmitter release.
To examine whether GIPC1 and Myo6 are involved in BDNF-TrkB–
2 induced glutamate release, similar experiments were conducted using
0 hippocampal neurons derived from Myo6- and GIPC1-deficient mice
WT Gipc1 –/– (Fig. 8c). Whereas an increase in neurotransmitter release was observed
in wild-type cultured neurons (Fig. 8c), BDNF-induced glutamate
release was completely blocked in hippocampal neurons lacking GIPC1
evoked synaptic transmission and spontaneous glutamate release in (n ¼ 123 boutons from five experiments, Fig. 8c) or Myo6 (n ¼ 132
excitatory central synapses in cultures2,17,18,21,37. This acute action boutons from five experiments, Fig. 8c). The total releasable neuro-
of BDNF is initiated when it binds to its receptor, TrkB, at transmitter induced by high concentration of K+ was similar between
presynaptic terminals21. wild-type and knockout neurons (data not shown). Similar results were
The finding that both Myo6- and GIPC1-deficient mice showed obtained in the presence of tetrodotoxin (1 mM), ruling out the
impairments in basal synaptic transmission and presynaptic properties possibility that BDNF has a direct effect on membrane depolarization
(Fig. 3a and Fig. 4) raises the possibility that glutamate release from the that might enhance neurotransmitter release by itself. Tetrodotoxin did
nerve terminals may depend upon Myo6 and GIPC1 proteins. Also, not affect BDNF-induced neurotransmitter release (releasable fluores-
given that all three proteins, GIPC1, Myo6 and TrkB receptor, reside in cence was 18.33 ± 2.71, n ¼ 71 for vehicle-treated wild-type; 28.6 ± 3.1,
presynaptic nerve terminals (Fig. 2f,g and refs. 5,32) and can form a n ¼ 59 for BDNF-treated wild-type; 19.00 ± 2.95, n ¼ 54 for vehicle-
protein complex (Fig. 1a,b), we investigated the possibility that GIPC1 treated Gipc1–/–; 17.43 ± 4.02, n ¼ 66 for BDNF-treated Gipc1–/–; 13.91
and Myo6 are involved in BDNF-TrkB–induced glutamate release. ± 3.05, n ¼ 55 for vehicle-treated Myo6sv/sv; and 10.53 ± 2.1, n ¼ 72 for
Localization of Trk receptors at active presynaptic boutons (neuro- BDNF-treated Myo6sv/sv groups; two experiments per condition, data
transmitter release sites) in primary hippocampal neurons was first not shown). These results are consistent with the reported effect of
examined using AM 1-43, a fixable analog of FM 1-43, as a fluorescent BDNF on mEPSCs in the presence of tetrodotoxin in cultured
marker for active synaptic boutons. The nerve terminals loaded with neurons18. These findings indicate that both GIPC1 and Myo6 are
AM 1-43 in the presence of high concentration of K+ were fixed, involved in the regulation of glutamate release by BDNF.
coimmunostained with antibody to Trk (anti-Trk) followed by a Cy5- Notably, a deficiency of Myo6 (n ¼ 133 boutons from five experi-
conjugated secondary antibody, and analyzed by confocal microscopy ments), but not GIPC1 (n ¼ 129 boutons from five experiments),
(Fig. 8a). We observed the colocalization of AM 1-43 and Trk in active resulted in lower spontaneous glutamate release (Fig. 8c). These results
boutons. Together with the previous EM localization of TrkB in nerve suggest that this motor protein is required for a basal level of synaptic
terminals5, these results confirm that TrkB is expressed in active vesicle recycling. The role of GIPC1 may therefore be to mediate
presynaptic terminals. BDNF-induced glutamate release by linking TrkB to basal synaptic
To examine BDNF-induced glutamate release, we quantified relea- machinery governed by Myo6.
sable FM 1-43 fluorescence intensity at individual active synaptic We also measured the kinetics of FM dye destaining in neurons from
terminals. FM dye–loaded synapses were treated with or without GIPC1- and Myo6-deficient mice and its modulation by BDNF.
BDNF for 10 min. The difference in FM dye fluorescence intensity Application of BDNF for 10 min enhanced release kinetics in wild-
before and after treatment with BDNF or vehicle was defined as type cultured neurons (30 s into the destaining stimulus, which
releasable FM fluorescence (Fig. 8b). The releasable fluorescence with- consisted of action potentials at 20 Hz, about 47% and 60% of the
out BDNF treatment corresponded to a basal low level of spontaneous endocytosed vesicles were released in the vehicle-treated wild-type
glutamate release (n ¼ 150 boutons from four experiments, Fig. 8b). (n ¼ 102 boutons from five experiments) and in the BDNF-treated
Exposure to BDNF increased glutamate release (n ¼ 118 boutons from wild-type (n ¼ 103 boutons from five experiments) groups, respec-
five experiments, Fig. 8b). In the presence of the Trk tyrosine kinase tively; P o 0.001, Fig. 8d). However, cultured hippocampal neurons
inhibitor K252a (100 nM), the releasable FM fluorescence with BDNF from both Myo6sv/sv and Gipc1–/– mice demonstrated slower
treatment was completely inhibited (n ¼ 113 boutons from three release kinetics than wild-type cultures after 30 s, 30% and 37% of


a b FM 1-43 perfusion 45 s
15 s K+ /45 s normal
c FM 1-43 perfusion 45 s
15 s K+ /45 s normal
K+ perfusion 45 s K+ perfusion 45 s
Washing 5 min Washing 5 min
BDNF 10 min BDNF 10 min
K252a 20 min
I II III Image capture
Image capture

Releasable FM 1-43 fluorescence
35 35 35

Releasable FM 1-43 fluorescence
30 * 30 * 30 *

(arbitrary units)
25 25 25

(arbitrary units)
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

20 # 20 20
15 15 15
10 10 10

AM 1-43 Trk Overlay 5 5 5
0 0
1 2 3 4 1 2 3 4 1 2 3 4






























Gipc1 –/– Myo6 sv/sv


Figure 8 Requirement of GIPC1 and Myo6 for BDNF-induced glutamate release. (a) Localization of Trk in WT + vehicle
AM 1-43–loaded active synaptic boutons by confocal microscopy. AM 1-43–loaded neurons were fixed and d WT + BDNF
Myo6 + vehicle
immunostained with anti-Trk followed by Cy5-conjugated secondary antibody. Scale bar, 5 mm. (b) TrkB activation Myo6
is required for BDNF-induced neurotransmitter release. Cultured mouse hippocampal neurons (DIV 10) were 120 Gipc1 + vehicle
Gipc1 + BDNF
preincubated with or without the Trk kinase inhibitor K252a (100 nM) and treated with BDNF as depicted. Three 20 Hz

Percent FM1-43 fluorescence
images were taken after washing (I), BDNF treatment (II), and destaining (III). Releasable FM 1-43 fluorescence 100
intensity (difference between images I and II) was quantitatively measured. *P o 0.01, significant difference
from vehicle control; #P o 0.01, significant difference from BDNF treatment alone. (c) Recycling synaptic 80
vesicles in cultured neurons (DIV 10) from Gipc1–/– or Myo6sv/sv P1 hippocampus were labeled with FM 1-43.
Releasable FM 1-43 fluorescence intensity with or without BDNF stimulus was quantitatively measured as in b.
*P o 0.05, significant difference from vehicle control; #P o 0.05, significant difference from wild-type vehicle 40
control. (d) Release kinetics were measured by the decrease in FM 1-43 fluorescence intensity during application
of a destaining stimulus in wild-type, Myo6sv/sv and Gipc1–/– cultured hippocampal neurons after exposure with or 20
without BDNF. Application of BDNF for 10 min enhanced release kinetics in wild-type neurons. Both Myo6sv/sv
and Gipc1–/– neurons demonstrated slower release kinetics than the wild type. Release kinetics in Myo6sv/sv 0
neurons were significantly slower than in Gipc1–/– neurons. BDNF was no longer capable of increasing the –10 0 10 20 30 40 50 60
Time (s)
unloading rate in Gipc1–/– and Myo6sv/sv neurons. Data in b–d represent means ± s.e.m.

endocytosed vesicles had been released in the Myo6sv/sv (n ¼ 112 can use GIPC1 and the Myo6 actin motor to modulate neurotrans-
boutons from four experiments) and the Gipc1–/– (n ¼ 109 boutons mitter release at presynaptic locations.
from four experiments) groups, respectively (Fig. 8d). Notably, the Neurotransmission of dopamine, glutamate, GABA and acetylcho-
release kinetics in Myo6-deficient neurons were slower than that in line is modulated by neurotrophins and Trk receptors through quantal
GIPC1-deficient neurons (Fig. 8d). Most importantly, BDNF was no release mechanisms4,19,37. In this study, we found that GIPC1 and
longer capable of increasing the FM dye unloading rate in Gipc1–/– (n ¼ Myo6 proteins also exert an effect upon BDNF-mediated LTP in
102 boutons from four experiments) and Myo6sv/sv (n ¼ 111 boutons postnatal (P12–13) CA3-CA1 synapses, but not in adult hippocampus.
from four experiments) neurons (Fig. 8d). These findings show Therefore, the effects of Myo6 and GIPC1 upon LTP are restricted to a
compromised vesicle recycling in neurons lacking GIPC1 and Myo6, specific time window: P12–13. The unique age-dependent effects of
which do not respond to BDNF stimulation. BDNF on hippocampal plasticity were described previously12 but have
not been fully molecularly delineated. Our data suggest that BDNF-
DISCUSSION mediated neurotransmitter release mechanisms for LTP differ during
The actions of BDNF upon synaptic plasticity occur through increased different developmental stages and that in adults, proteins other than
Bdnf gene expression, processing and release, leading to binding and Myo6 and GIPC1 are involved in modulating the effects of TrkB
activation of TrkB receptors in both pre- and postsynaptic sites3,4. Here receptors during synaptic transmission.
we demonstrate that BDNF-induced neurotransmitter release relies These findings also confirm the important role of actin-based
upon Myo6 and the GIPC1 adaptor protein. GIPC1 binds directly to motors in neurotransmitter release. Presynaptic nerve terminals are
the Trk receptor24 as well as to Myo6 (refs. 28,30). A notable comprised of actin-rich cytoskeletal networks. Actin-dependent vesicle
colocalization of GIPC1 and Myo6 exists in hippocampal neurons, transport is mediated by several different myosin proteins38,39. To date,
which overlapped with the expression of TrkB receptors. Loss of either at least 20 structurally and functionally distinct classes of the myosin
GIPC1 or Myo6 protein led to similar defects in presynaptic properties, superfamily have been identified and are involved in various cellular
as measured by PPF and PTP. In addition to hippocampal slice functions26. Nearly all direct traffic toward the plus end of actin
measurements, electrophysiological and FM 1-43 staining experiments filaments, but Myo6 moves toward the minus end of actin26. Recent
using primary hippocampal cultures verified that Myo6 and GIPC1 studies have suggested important roles of actin filaments in neuro-
were required for vesicle recycling in response to BDNF. Hence, BDNF transmitter release23. Indeed, we found that Myo6 is also necessary for


BDNF-induced spontaneous and evoked glutamate release in hippo- Myo6-deficient Snell’s waltzer mice33 were obtained from Jackson laboratory
campal neurons. Myo6 may serve as an anchoring protein between and were bred with C57BL/6.
synaptic vesicles and the actin cytoskeleton, or it may be involved in the
Electrophysiological analysis of hippocampal slices. Experiments were
transport of vesicles in a minus end direction. Alternatively, given that
carried out on male hippocampi, which were cut into 400-mm transverse slices
Myo6 has been implicated in endocytotic events31, it may be involved with a tissue chopper (EMS) and maintained in an interface chamber
in the retrieval of synaptic vesicles after exocytosis at the terminal or in at 29 1C for 90 min before recording, as previously reported46. Briefly,
Trk internalization. The plus end–directed myosin V motor can CA1 fEPSPs were recorded by placing both the stimulating and recording
associate with synaptic vesicles and may be responsible for the move- electrodes in CA1 stratum radiatum. Basal synaptic transmission was
ment of synaptic vesicles in axon terminals38–42. Because of their assayed either by plotting the stimulus voltages (V) against slopes of fEPSP
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

opposite directionality, myosin V and Myo6 might cooperatively or by plotting the peak amplitude of the fiber volley against the slope of the
regulate synaptic vesicle trafficking. fEPSP to generate input-output relations. PPF was tested using nine inter-
Several other synaptic functions are affected by the loss of GIPC1 and stimulus intervals and defined as the second fEPSP slope expressed as a
percentage of the first. For PTP measurements, three ten-burst trains were
Myo6. For example, a reduction in basal synaptic transmission was
applied in the presence of 50 mM AP5. For LTP (P12–13), tetanic
observed both in Myo6sv/sv and in Gipc1–/– mice. Also, we found a
stimulation consisted of four pulses at 100 Hz, with bursts repeated ten times
reduction in the numbers of active synaptic boutons in Myo6sv/sv and at 5 Hz following 2.5 h of exposure to 50 ng ml–1 BDNF. The protocols for
Gipc1–/– hippocampal neurons. These common defects, in addition to synaptic fatigue and LTP experiments in adult mice are described in the
the similarity of the deficits in presynaptic properties in Myo6sv/sv and Supplementary Methods.
Gipc1–/– mice, verify functional as well as physical interactions between
these proteins. Electrophysiological analysis of hippocampal cultures. The pre- and post-
Given that the C-terminal domain of Myo6, a putative cargo- synaptic hippocampal neurons (DIV 10–15) were maintained under ruptured
binding domain, also interacts with other adaptor proteins, including patch whole-cell voltage clamp (–60 mV) throughout the experiments47, and
the input resistances of both cells were periodically checked for constancy.
SAP97 and Dab2 (ref. 31), Myo6 may be involved in functions
EPSC amplitude was measured between the peak and the mean of the
independent of GIPC1 through different binding partners. SAP-97,
baseline just before the start of the EPSC using Clampfit 9 software (Axon
which can bind to the GluR1 AMPA receptor subunit, has been Instruments). The protocols for these recordings are described in the
proposed to link postsynaptic AMPA receptors to the Myo6 motor43. Supplementary Methods.
Indeed, a recent report has implicated Myo6 in AMPA receptor
endocytosis at postsynaptic sites32. These findings, together with FM 1-43 staining. Loading and unloading of FM 1-43 (Biotium) was carried
our data showing that presynaptic TrkB receptors are directly and out as previously described48 on primary neurons from P1 hippocampi. In
functionally linked to Myo6, indicate that Myo6 is critically involved Figure 7, the difference between the number of boutons before and after FM
in synaptic function. These different Myo6 motor complex pools dye destaining yielded the number of active boutons36. Neurite lengths of
75 mm located 20 mm from the soma were randomly selected, and individual
provide a plausible explanation for the marked reduction in the
punctae (diameter 0.5–2 mm) were measured for fluorescence intensity. In
number of synapses observed in Myo6sv/sv mice, as demonstrated by Figure 8b,c, FM 1-43–loaded presynaptic boutons were exposed to BDNF
a recent study32. (100 ng ml–1, 10 min). BDNF-induced releasable fluorescence of each active
How does TrkB signaling control glutamate release? One possibility bouton was analyzed using MATLAB7.1 software.
is that GIPC1 and Myo6 are regulated by TrkB tyrosine kinase activity. To study the effect of BDNF on release kinetics, presynaptic boutons were
However, there is currently no evidence for increased phosphorylation loaded with FM 1-43 by passing 1-ms-long current pulses (50 action potentials
of these proteins by TrkB. An earlier study proposed that synapsins may at 20 Hz), yielding fields of B10 V cm–1, through the chamber via a platinum
be mediators of BDNF-enhanced evoked neurotransmitter release20. electrode (Cell MicroControls). The cultures were then washed for 5 min.
Phosphorylation of synapsins by Erk1/2 was implicated in the A destaining stimulus of 1,200 action potentials at 20 Hz was delivered
presynaptic facilitation of glutamate release by BDNF (ref. 20). Synap- after exposure of the culture to BDNF for 10 min. Images were acquired every
5 s, and fluorescence intensity at individual boutons were measured. The
sin proteins can associate with synaptic vesicles as well as actin
percentage of FM 1-43 fluorescence was calculated based on the fluorescence
filaments, and have been implicated in the availability and mobilization intensity of presynaptic boutons immediately before the application of the
of synaptic vesicles44,45. Up to now, there has been little attention paid destaining stimulus.
to how synaptic vesicles are mobilized and docked by BDNF in Image acquisition and data analysis as well as loading of AM 1-43 (Biotium)
presynaptic active zones. The present study points to a critical role are described in the Supplementary Methods.
for a specific actin-based motor in the release of neurotransmitter by
BDNF. An important direction for the future will be to link BDNF- Statistical analysis. Statistical analysis was performed with t-test, ANOVA and
TrkB signal transduction pathways to the machinery that controls post-hoc comparisons. Where applicable, data were normalized to the basal
vesicle docking and neurotransmitter release. values. Results are expressed as mean ± s.e.m.

Note: Supplementary information is available on the Nature Neuroscience website.
Cell and molecular biology. Detailed methods for immunoprecipitation, ACKNOWLEDGMENTS
immunoblotting, cell surface biotinylation, immunohistochemistry including We are grateful to N. Jenkins (National Cancer Institute, Frederick, Maryland)
EM, GST pull-down assays, cell culture, DNA transfection, and a list of reagents for porcine Myo6 cDNAs; T. Hasson (University of California, San Diego) for
can be found in the Supplementary Methods online. anti-Myo6 antibody; I. Taniuchi and D. Littman (New York University) for
advice on the generation of Gipc1 mutant mice; A. Auerbach and staff members
of the Transgenic Animal Facility (New York University) for assistance;
Mutant mice. To generate a targeting construct for Gipc1 knockout mice, a
R. Rajagopal, J.C. Arevalo, M. Beyna, D.B. Pereira and A.H. Kim for reagents
genomic library (Stratagene) from 129SV/J mice was screened using mouse and advice; D.C. Powell for writing the program for Matlab analysis; and
Gipc1 cDNA. Exons 2–4, which encode the entire N terminus, PDZ domain, I. Orozco and P.K. Hsu (Columbia University, New York) for assistance
and a part of the C-terminal domain of GIPC1 protein, were replaced by during electrophysiological experiments. This work was funded by US National
a neomycin-resistance gene (schematic in Fig. 3b). Details on the generation Institutes of Health grants to M.V.C. (HD23315 and NS21072), O.A. (NS40045)
and genotyping analysis of the mice are in the Supplementary Methods. and T.M. (HL-18974 and DA08259).


AUTHOR CONTRIBUTIONS 22. Pozzo-Miller, L.D. et al. Impairments in high-frequency transmission, synaptic vesicle
H.Y. performed the cell and molecular biology experiments and the mouse docking, and synaptic protein distribution in the hippocampus of BDNF knockout mice.
genetics. T.A.M. conducted the electron microscopy experiments. O.A. supervised J. Neurosci. 19, 4972–4983 (1999).
and I.N and H.Z. performed FM dye and electrophysiological studies. 23. Dillon, C. & Goda, Y. The actin cytoskeleton: integrating form and function at the
synapse. Annu. Rev. Neurosci. 28, 25–55 (2005).
24. Lou, X., Yano, H., Lee, F., Chao, M.V. & Farquhar, M.G. GIPC and GAIP form a complex
COMPETING INTERESTS STATEMENT with TrkA: a putative link between G protein and receptor tyrosine kinase pathways. Mol.
The authors declare that they have no competing financial interests. Biol. Cell 12, 615–627 (2001).
25. Roberts, R. et al. Myosin VI: cellular functions and motor properties. Phil. Trans. R. Soc.
Published online at http://www.nature.com/natureneuroscience Lond. B 359, 1931–1944 (2004).
Reprints and permissions information is available online at http://npg.nature.com/ 26. Krendel, M. & Mooseker, M.S. Myosins: tails (and heads) of functional diversity.
reprintsandpermissions/ Physiology (Bethesda) 20, 239–251 (2005).
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

27. De Vries, L., Lou, X., Zhao, G., Zheng, B. & Farquhar, M.G. GIPC, a PDZ domain
containing protein, interacts specifically with the C terminus of RGS-GAIP. Proc. Natl.
1. Thoenen, H. Neurotrophins and neuronal plasticity. Science 270, 593–598 (1995). Acad. Sci. USA 95, 12340–12345 (1998).
2. McAllister, A.K., Katz, L.C. & Lo, D.C. Neurotrophins and synaptic plasticity. Annu. Rev. 28. Bunn, R.C., Jensen, M.A. & Reed, B.C. Protein interactions with the glucose transporter
Neurosci. 22, 295–318 (1999). binding protein GLUT1CBP that provide a link between GLUT1 and the cytoskeleton.
3. Poo, M.M. Neurotrophins as synaptic modulators. Nat. Rev. Neurosci. 2, 24–32 (2001). Mol. Biol. Cell 10, 819–832 (1999).
4. Lu, B. BDNF and activity-dependent synaptic modulation. Learn. Mem. 10, 86–98 29. Wang, L.H., Kalb, R.G. & Strittmatter, S.M.A. PDZ protein regulates the distribution of
(2003). the transmembrane semaphorin, M-SemF. J. Biol. Chem. 274, 14137–14146 (1999).
5. Drake, C.T., Milner, T.A. & Patterson, S.L. Ultrastructural localization of full-length trkB 30. Reed, B.C. et al. GLUT1CBP(TIP2/GIPC1) interactions with GLUT1 and Myo6: evidence
immunoreactivity in rat hippocampus suggests multiple roles in modulating activity- supporting an adapter function for GLUT1CBP. Mol. Biol. Cell 16, 4183–4201 (2005).
dependent synaptic plasticity. J. Neurosci. 19, 8009–8026 (1999). 31. Hasson, T. Myosin VI: two distinct roles in endocytosis. J. Cell Sci. 116, 3453–3461
6. Aoki, C. et al. Localization of brain-derived neurotrophic factor and TrkB receptors to (2003).
postsynaptic densities of adult rat cerebral cortex. J. Neurosci. Res. 59, 454–463 32. Osterweil, E., Wells, D.G. & Mooseker, M.S. A role for Myo6 in postsynaptic structure and
(2000). glutamate receptor endocytosis. J. Cell Biol. 168, 329–338 (2005).
7. Purcell, A.L. & Carew, T.J. Tyrosine kinases, synaptic plasticity and memory: insights 33. Avraham, K.B. et al. The mouse Snell’s waltzer deafness gene encodes an unconven-
from vertebrates and invertebrates. Trends Neurosci. 26, 625–630 (2003). tional myosin required for structural integrity of inner ear hair cells. Nat. Genet. 11,
8. Korte, M. et al. Hippocampal long-term potentiation is impaired in mice lacking brain- 369–375 (1995).
derived neurotrophic factor. Proc. Natl. Acad. Sci. USA 92, 8856–8860 (1995). 34. Zucker, R.S. & Regehr, W.G. Short-term synaptic plasticity. Annu. Rev. Physiol. 64,
9. Patterson, S.L. et al. Recombinant BDNF rescues deficits in basal synaptic transmission 355–405 (2002).
and hippocampal LTP in BDNF knockout mice. Neuron 16, 1137–1145 (1996). 35. Janz, R. et al. Essential roles in synaptic plasticity for synaptogyrin I and synaptophysin
10. Minichiello, L. et al. Essential role for TrkB receptors in hippocampus-mediated I. Neuron 24, 687–700 (1999).
learning. Neuron 24, 401–414 (1999). 36. Ryan, T.A. et al. The kinetics of synaptic vesicle recycling measured at single presynaptic
11. Kang, H., Welcher, A.A., Shelton, D. & Schuman, E.M. Neurotrophins and time: different boutons. Neuron 11, 713–724 (1993).
roles for TrkB signaling in hippocampal long-term potentiation. Neuron 19, 653–664 37. Tyler, W.J., Perrett, S.P. & Pozzo-Miller, L.D. The role of neurotrophins in neurotrans-
(1997). mitter release. Neuroscientist 8, 524–531 (2002).
12. Figurov, A., Pozzo-Miller, L.D., Olafsson, P., Wang, T. & Lu, B. Regulation of synaptic 38. Bridgman, P.C. Myosin-dependent transport in neurons. J. Neurobiol. 58, 164–174
responses to high-frequency stimulation and LTP by neurotrophins in the hippocampus. (2004).
Nature 381, 706–709 (1996). 39. DePina, A.S. & Langford, G.M. Vesicle transport: the role of actin filaments and myosin
13. Egan, M.F. et al. The BDNF val66met polymorphism affects activity-dependent secre- motors. Microsc. Res. Tech. 47, 93–106 (1999).
tion of BDNF and human memory and hippocampal function. Cell 112, 257–269 40. Prekeris, R. & Terrian, D.M. Brain myosin V is a synaptic vesicle-associated motor
(2003). protein: evidence for a Ca2+-dependent interaction with the synaptobrevin-synaptophy-
14. Nestler, E.J. et al. Neurobiology of depression. Neuron 34, 13–25 (2002). sin complex. J. Cell Biol. 137, 1589–1601 (1997).
15. Gauthier, L.R. et al. Huntingtin controls neurotrophic support and survival of neurons 41. Evans, L.L., Lee, A.J., Bridgman, P.C. & Mooseker, M.S. Vesicle-associated brain
by enhancing BDNF vesicular transport along microtubules. Cell 118, 127–138 myosin-V can be activated to catalyze actin-based transport. J. Cell Sci. 111, 2055–
(2004). 2066 (1998).
16. Lohof, A.M., Ip, N.Y. & Poo, M.M. Potentiation of developing neuromuscular synapses by 42. Watanabe, M. et al. Myosin-Va regulates exocytosis through the submicromolar Ca2+-
the neurotrophins NT-3 and BDNF. Nature 363, 350–353 (1993). dependent binding of syntaxin-1A. Mol. Biol. Cell 16, 4519–4530 (2005).
17. Lessmann, V., Gottmann, K. & Heumann, R. BDNF and NT-4/5 enhance glutamatergic 43. Wu, H., Nash, J.E., Zamorano, P. & Garner, C.C. Interaction of SAP97 with minus-end-
synaptic transmission in cultured hippocampal neurones. Neuroreport 6, 21–25 directed actin motor Myo6. Implications for AMPA receptor trafficking. J. Biol. Chem.
(1994). 277, 30928–30934 (2002).
18. Li, Y.X., Zhang, Y., Lester, H.A., Schuman, E.M. & Davidson, N. Enhancement of 44. Greengard, P., Valtorta, F., Czernik, A.J. & Benfenati, F. Synaptic vesicle phosphopro-
neurotransmitter release induced by brain-derived neurotrophic factor in cultured teins and regulation of synaptic function. Science 259, 780–785 (1993).
hippocampal neurons. J. Neurosci. 18, 10231–10240 (1998). 45. Sudhof, T.C. The synaptic vesicle cycle. Annu. Rev. Neurosci. 27, 509–547
19. Schinder, A.F. & Poo, M. The neurotrophin hypothesis for synaptic plasticity. Trends (2004).
Neurosci. 23, 639–645 (2000). 46. Petrone, A. et al. Receptor protein tyrosine phosphatase á is essential for hippocampal
20. Jovanovic, J.N., Czernik, A.J., Fienberg, A.A., Greengard, P. & Sihra, T.S. Synapsins as neuronal migration and long-term potentiation. EMBO J. 22, 4121–4131 (2003).
mediators of BDNF-enhanced neurotransmitter release. Nat. Neurosci. 3, 323–329 47. Arancio, O., Kandel, E.R. & Hawkins, R.D. Activity-dependent long-term enhancement
(2000). of transmitter release by presynaptic 3¢,5¢-cyclic GMP in cultured hippocampal neurons.
21. Li, Y.X. et al. Expression of a dominant negative TrkB receptor, T1, reveals a requirement Nature 376, 74–80 (1995).
for presynaptic signaling in BDNF-induced synaptic potentiation in cultured hippocam- 48. Ninan, I. & Arancio, O. Presynaptic CaMKII is necessary for synaptic plasticity in
pal neurons. Proc. Natl. Acad. Sci. USA 95, 10884–10889 (1998). cultured hippocampal neurons. Neuron 42, 129–141 (2004).


Vesicular proteins exocytosed and subsequently
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

retrieved by compensatory endocytosis are nonidentical
Martin Wienisch & Jurgen Klingauf

Upon exocytosis, synaptic vesicle proteins are released into the plasma membrane and have to be retrieved by compensatory
endocytosis. When green fluorescent protein–labeled versions of the vesicle proteins synaptobrevin-2 and synaptotagmin-1 are
overexpressed in rat hippocampal neurons, up to 30% are found on axonal membranes under resting conditions. To test whether
and to what extent these plasma membrane–stranded proteins participate in exo-endocytic cycling, a new proteolytic approach
was used to visualize the fate of newly exocytosed proteins separately from that of the plasma membrane–stranded ones. We
found that both pools were mixed and that endocytosed vesicles were largely composed of previously stranded molecules. The
degree of nonidentity of vesicular proteins exo- and endocytosed depended on stimulus duration. By using an antibody to the
external domain of synaptotagmin-1, we estimated that under physiological conditions a few percent of vesicular proteins
were located near the active zone, from where they were preferentially recycled upon stimulation.

Following neurotransmitter release, clathrin-mediated endocytosis is from electron micrographs of frog neuromuscular junctions quick-
thought to be the major pathway for synaptic vesicle recycling1–4. frozen at different time points after stimulation19.
Alternatively, a fast ‘kiss and run’ mechanism, whereby vesicles connect What distinguishes both scenarios fundamentally is the molecular
only briefly to the plasma membrane without full collapse5–8, identity of vesicles exo- and endocytosed by the same stimulus.
might retrieve vesicular components more efficiently. In addition, a Maintaining molecular identity is the hallmark of a kiss and run
slow pathway via large infoldings and endosomes9 may contribute type of mechanism, whereas nonidentity would be the essential feature
to recycling. and kinetic advantage of a preassembled pool at the periphery of the
Most of our knowledge of the kinetics of stimulated endocytosis in active zone.
synaptic boutons comes from experiments using the fluorescent vesicle Indeed, up to 30% of the synaptic vesicle protein spH is present on
marker FM1-43 (refs. 7,10–13) and synapto-pHluorin (spH), a fusion the external bouton membrane of transfected hippocampal neurons20.
construct of the vesicle protein synaptobrevin/VAMP-2 with a pH- Immuno-electronmicroscopy has revealed that endogenous synapto-
sensitive green fluorescent protein (GFP) variant8,14,15. However, brevin is also present in the synaptic membrane, at a density that is only
neither technique allows us to distinguish between different molecular three times lower than that in synaptic vesicles under resting condi-
mechanisms of membrane retrieval. tions21. Exocytosed spH and synaptobrevin-eGFP disperse into the
When visualizing the redistribution of enhanced GFP (eGFP)-tagged axonal plasma membrane during synaptic stimulation20,22, prompting
clathrin in hippocampal boutons16, it was found that stimulation leads the question of whether this stranded plasma membrane pool of spH
to a transient increase of fluorescence in boutons, with decay kinetics participates in synaptic vesicle retrieval during exo- and endocytosis.
notably similar to previous estimates of the endocytic time course. Here we tested whether the same molecules that are exocytosed are
However, the increase in fluorescence is preceded by a long delay, retrieved during compensatory endocytosis. We introduced a tobacco
suggesting that endocytosis during the first few seconds of continuing etch virus (TEV) protease cleavage site between the synaptobrevin and
stimulation is not mediated by the slow (B10 s) formation of clathrin- GFP moieties. This site is only accessible to external enzyme if this
coated pits17,18. The rate of endocytosis, however, is fastest for short protein is in the plasma membrane. This TEV motif–containing spH
stimuli15. One likely interpretation is that the early phase of endocytosis (spH-TEV) behaved identically to the original spH. We found that
does not depend on clathrin, but is of the ‘kiss and run’ type. mostly digested molecules were recycled during compensatory endo-
An alternative explanation, however, could be a pool of preas- cytosis—that is, synaptobrevin molecules deposited on the plasma
sembled ‘ready to go’ coats at the periphery of the active zone, membrane before digestion. Identical results were obtained with a
supporting a first wave of endocytosis. Such a first wave of endocytosis, synaptotagmin-1-pHluorin construct. We conclude that most recycling
via coated vesicles and lasting for only 10 s, has been observed in the synaptic vesicles lose their protein complement during exocytosis. By
first reconstruction of the time course of synaptic endocytosis obtained using an antibody to the external domain of synaptotagmin-1, we

Department of Membrane Biophysics, Max-Planck-Institute for Biophysical Chemistry, Am Fassberg 11, D-37077 Goettingen, Germany. Correspondence should be
addressed to J.K. (J.Klingauf@mpi-bpc.mpg.de).
Received 4 April; accepted 16 June; published online 16 July 2006; doi:10.1038/nn1739


Figure 1 Synaptobrevin-TEV-pHluorin (spH-TEV)
a spHTEV FM5-95
is targeted to functional boutons and the plasma
membrane–stranded fraction can be effectively
cleaved. (a) Axonal branches of hippocampal
neurons in primary culture overexpressing
spH-TEV. Difference image (left) obtained by
subtraction of the picture recorded before from
that recorded during a stimulation with 200 action
potentials. Transient fluorescence increases at
sites that can be stained and destained with the
activity marker FM5-95 (right) identify them as
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

0 functional presynaptic boutons. Scale bar, 5 mm.
pH 5.5 NH4Cl (b) Left, individual fluorescence transients of
b 20 Hz
1.6 20 Hz 2.5 exemplar boutons marked by arrowheads in a;
Fluorescence intensity (a.u.)

1.5 middle, average fluorescence signal (red line
140 2.0
1.4 represents single exponential fit with t ¼ 26.9 s);
120 right, average fluorescence signal from boutons

1.3 1.5
sequentially superfused with acidic solution
1.0 (pH 5.5) and with ammonium chloride solution
80 (50 mM). Comparing the size of the fluorescence
60 0.5 change during the acidic pulse with the
fluorescence change during the ammonium
0 20 40 60 0 20 40 60 0 10 20 30 40
chloride pulse reveals that 33.2 ± 0.9% of
Time (s) Time (s) Time (s)
spH-TEV are stranded in the plasma membrane
c pH 5.5 0 min 1 min 3 min 5 min (n ¼ 3 experiments, 450 boutons each, error
bars represent s.e.m.). (c) When neurons
overexpressing spH-TEV were exposed to 0.12 U
ml–1 TEV protease at room temperature, the
pHluorin moiety was effectively cleaved, within
7 min 9 min 11 min 13 min 15 min 15 min, from molecules residing in the plasma
membrane. Vesicular spH-TEV is inaccessible to
the protease (bottom, left). This decreased the
plasma membrane fluorescence down to a level
comparable to that obtained with acidic solution
TEV protease
(bottom, right). Scale bar, 10 mm.
Surface fluorescence (a.u.)


400 moieties that is only accessible to external
300 enzyme if spH-TEV is in the plasma mem-
brane. Like spH, spH-TEV was specifically
targeted to synaptic boutons. It colocalized
100 pH 5.5
with the synaptic activity marker FM5-95
0 2 4 6 8 10 12 14 (Fig. 1a), and fluorescence transients upon
Time (min)
stimulation with 200 action potentials were
indistinguishable from those obtained with
estimated that under physiological conditions a small percentage of spH (Fig. 1b). As found for spH, up to 30% of spH-TEV was found
vesicular proteins are located near the active zone, constituting a pool on the plasma membrane of boutons at rest, as revealed by fluorescence
of preferentially recycled molecules. A stimulus of 30–40 action quenching by short superfusion with an acidic solution of pH 5.5
potentials led to the compensatory endocytosis of about 5–10 vesicles, (Fig. 1b right, and first two images in Fig. 1c). For investigating the fate
which is remarkably close to the number of vesicles in the readily of vesicular SNAREs after fusion, the plasma membrane pool of spH-
releasable or docked pool of vesicles that are rapidly released by a TEV, which normally masks the spatiotemporal dynamics of vesicular
similar number of action potentials (ref. 23). SNARES during stimulation, was almost completely fluorescently
‘silenced’ by exposure to TEV protease for 15 min (Fig. 1c).
Tracking the fate of vesicular proteins after fusion Vesicular proteins are lost after fusion
Overall endocytic activity can be measured with spH, a fusion construct When axons expressing spH-TEV were stimulated, postdigest boutons
of the vesicle protein synaptobrevin and a pH-sensitive form lit up as vesicular spH-TEV was dequenched upon vesicle fusion
(pHluorin) of GFP (ref. 15). The pHluorin in the acidic vesicle (Fig. 2a, left). For analysis of the spatiotemporal dynamics of vesicular
lumen is dequenched upon exocytosis, and reacidification of endo- spH-TEV, we created a maximum projection image in which each pixel
cytosed vesicles is accomplished within about 5 s (ref. 24). Up to 30% of contained the maximum value over all images in the stack at that
spH is present on the external axonal membrane of transfected particular pixel; we drew a line of interest (LOI) along an axonal stretch
hippocampal neurons (ref. 20 and Fig. 1a,b), regardless of the absolute and for each image plotted the fluorescence along this LOI for a width
level of spH expression. of 3 pixels (615 nm) in a kymograph image (Fig. 2a, right). Superfusion
To test whether synaptic vesicles maintain their identity with respect with acidic (pH 5.5) solution after digestion revealed that most spH-
to their protein composition during exo-endocytic cycling, we intro- TEV in the plasma membrane had been digested (some spH-TEV
duced a TEV protease cleavage site between the VAMP and GFP became stranded again during the protease washout period).


a Maximum projection 0 b t = 10 s
t = 16 s
c d = 0.5 µm
d = 3.2 µm

pH 5.5
t = 38 s pH 5.5 20 Hz d = 5.0 µm
t = 60 s d = 6.6 µm
400 d = 8.5 µm
20 t = 96 s d = 10.8 µm

20 Hz
300 d = 12.6 µm

F(t) (a.u.)

Time (s)

F(d ) (a.u.)

© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience


0 0
0 2 4 6 8 10 12 0 2 4 6 8 10 12 0 20 40 60 80
Distance (µm) Distance (µm) Time (s)

Figure 2 Vesicular proteins are lost after fusion. (a) Maximum projection image (left) of an axonal segment with three functional spH-TEV–expressing boutons
stimulated with 400 action potentials. Scale bar, 5 mm. For further analysis, an LOI (red solid line) was placed along the axon to create a waterfall plot (right).
Fluorescence intensities at each distance of the LOI (origin in upper left corner of maximum projection image) are integrated over a width (normal to the LOI)
of three pixels. Vertical bars on the right indicate times of acidic pulse and stimulation. (b) Exemplar spatial profiles along the LOI in a, at different times
indicated by arrowheads in a. (c) Exemplar time courses at different positions along the LOI in a, colorcoded according to the colors of arrows in a.

Upon stimulation, vesicular spH-TEV dequenched and, we were (Supplementary Fig. 1 online), and indeed led to incomplete recovery
surprised to find, spread quickly in the axonal plasma membrane; this (40.1% at 58 s after stimulus end) of fluorescence transients (Fig. 3a,b).
can be easily seen by comparing spatial (horizontal) and temporal Because vesicular proteins, once released, tend to diffuse away from the
(vertical) profiles of the kymograph at different times and distances release sites (Fig. 2), a likely explanation is that during compensatory
(Fig. 2b,c). Because the fluorescence increase in axonal segments was endocytosis, a portion of the plasma membrane–stranded spH-TEV
not substantially delayed from the bouton signals (Fig. 2c), it is likely has been retrieved, instead of the spH-TEV released during stimulation.
that vesicular SNARES diffuse rather freely in the plasma membrane. Before digestion, plasma membrane–stranded spH-TEV will contri-
After stimulation, vesicular SNAREs were retrieved and quenched by bute equally to the dequenching signal during reacidification of
endocytosis in the boutons only, with maybe some accumulation of endocytosed vesicles; however, after digestion, these stranded
exocytosed spH-TEV remaining at the periphery of the boutons at spH-TEV were fluorescently silent (Fig. 3b).
t ¼ 95 s (Fig. 2b, compare blue and green profile). To test this hypothesis, we first restored the relative stoichiometry of
molecules in the synaptic vesicles and the plasma membranes that had
Vesicular proteins exo- and endocytosed are not identical been subjected to proteolysis, by turning over the recycling pool of
Notably, recovery of the spH-TEV transients in the three boutons synaptic vesicles with a prolonged train of action potentials (800 action
depicted appeared to be incomplete after digestion (Fig. 2c), as potentials at 5 Hz) in the same set of boutons and then stimulating
opposed to that of the transients before digestion (Fig. 1b). Thus, we again with 100 action potentials. Indeed, the transient nature of the
next analyzed the recovery of fluorescence transients in a given set of spH signals was fully recovered (Fig. 3c). Thus, it seems that the
boutons before and after digestion, upon stimulation with 100 action majority of synaptic vesicles lose their molecular identity during exo-
potentials (Fig. 3). Digestion with TEV protease increased the rela- and endocytic cycling. Alternatively, our results can be explained if we
tive amplitude (DF/F0; that is, the change in fluorescence over assume that synaptic vesicles do indeed maintain their molecular
initial fluorescence), as expected, but not the absolute amplitudes identity in raft-like structures that diffuse slowly to the periphery of

a Predigest
20 Hz
b Postdigest 4.0 20 Hz
c Postmixing 1.7 20 Hz 1.0
20 Hz Predigest

0.8 Postmixing
3.5 1.6
1.3 0.6
3.0 1.5 0.4
1.2 1.4



2.5 0.0
1.3 0 20 40 60 80
2.0 1.2 Tims (s)
1.5 1.1
1.0 1.0 1.0
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
Time (s) Time (s) Time (s)

Figure 3 Vesicular proteins exo- and endocytosed during stimulation are not identical. (a) Left, image of axonal arborizations expressing spH-TEV before TEV
protease digestion. Right, upon stimulation with 100 action potentials at 20 Hz, control fluorescence transients at boutons (shown in left image) recovered
completely within 80 s after stimulation. (b) Left, image of axonal arborizations depicted in a after digestion with TEV protease; very little plasma membrane
fluorescence was seen. Right, upon application of 100 action potentials at 20 Hz, the fluorescence transients of boutons (shown in left image) had increased
relative amplitudes, as expected, but recovery was incomplete. (c) Left, image of axonal arborizations depicted in a after vesicle and plasma membrane pools
of spH-TEV were mixed by using a prolonged train of action potentials (800 action potentials at 5 Hz); substantial fluorescence of plasma membrane–stranded
spH-TEV was seen. Right, upon application of 100 action potentials at 20 Hz, fluorescence transients of boutons (image at left) again displayed complete
recovery, demonstrating that endocytosis was not blocked by TEV protease digestion. Inset, normalized responses for 100 action potentials at 20 Hz,
predigestion (a), postdigestion (b) and postmixing (n ¼ 3 experiments, 450 boutons each, error bars represent s.e.m.). Scale bar in a–c, 10 mm.


a 180
pH 5.5 NH4Cl
b 180
pH 5.5 NH4Cl c Prebleach 20 Hz
160 160
Fluorescence (a.u.)

Fluorescence (a.u.)
140 140
120 120 1.4

100 100 1.3
80 80 1.2
60 60 1.1
40 40 1.0
20 20
0 5 10 15 20 25 0 5 10 15 20 25 0 20 40 60 80
Frame number Frame number Time (s)
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

20 Hz
d Postbleach
20 Hz e Postmixing 1.5

2.5 1.4




1.5 1.1

0 20 40 60 80 0 20 40 60 80
Time (s) Time (s)

Figure 4 Nonidentity is also observed after photobleaching. (a,b) The photobleaching of plasma membrane–stranded spH-TEV occurred ten times as fast
as that of quenched vesicular spH-TEV. Applying sequential pulses of an acidic solution and an ammonium chloride solution, before (a) and after (b) 90 s
of repeated laser scanning at high intensity, showed that 88.9% of the unquenched plasma membrane spH-TEV pool, but only 8.5% of the quenched
vesicular spH-TEV pool, was photobleached, as seen by the reduction of the fluorescence change during the acid or the ammonium pulse, respectively.
Each frame is 2.9 s long. Error bars represent s.e.m. n ¼ 79 and n ¼ 80 boutons, respectively. (c) Left, image of axonal arborizations expressing spH-TEV.
Right, fluorescence transients before photobleaching (200 action potentials at 20 Hz). (d) Left, image of axonal arborizations depicted in c after selective
photobleaching of the plasma membrane–stranded spH-TEV pool by repetitive laser scanning at 488 nm for 90 s. Right, fluorescence transients upon
stimulation (200 action potentials at 20 Hz). (e) Left, image of axonal arborizations depicted in c after mixing vesicle and plasma membrane pools of
spH-TEV, by using a prolonged train of action potentials (800 action potentials at 5 Hz). Right, fluorescence transients upon stimulation (200 action
potentials at 20 Hz), error bars represent s.e.m. (all conditions). Scale bar in c–e, 10 mm.

the active zone, but are internalized during another round of stimula- Nonidentity is also observed for synaptotagmin-1
tion. Digestion would render such membrane patches fluorescently Is the observed nonidentity, with respect to synaptobrevin, of exo- and
silent, thereby leading to incomplete fluorescence recovery if these endocytosed vesicles specific to this protein, or are other vesicle
patches at endocytic sites are retrieved preferentially over those freshly proteins also lost after fusion? To answer this question, we constructed
inserted at the release sites. a chimera of a signal sequence–tagged pHluorin fused to the lumenal
N terminus of synaptotagmin-1 bridged by a TEV protease cleavage
Degree of nonidentity depends on stimulus duration site (sytpH-TEV). The sytpH-TEV chimera was properly targeted to
If the degree of recovery reflects the amount of vesicular spH-TEV presynaptic vesicles (Fig. 5) and colocalized with other presynaptic
released relative to the amount of stranded spH-TEV, it should depend vesicle proteins (data not shown).
on the stimulus duration. Indeed, a stimulus of 200 action potentials Electrical stimulation gave rise to stimulation-dependent fluores-
yielded substantially more recovery (50% at 58 s after stimulus end) of cence transients at synaptic boutons, indistinguishable from spH-TEV
the transients after digestion (Fig. 3c, inset). However, different degrees responses (Fig. 5d). Similar to that found for spH-TEV, acid quenching
of recovery for differing ratios of vesicular and stranded spH-TEV may and alkalinization of the vesicular lumen with ammonium revealed that
also be a consequence of a competition between cleaved and non- 23.9 ± 5.6% of sytpH-TEV was stranded (n ¼ 3 experiments, 450
cleaved forms of spH-TEV for binding at sites of endocytosis. In fact, boutons each) (Fig. 5b). When most of the plasma membrane–
the cleaved form lacking the 27-kDa pHluorin moiety may be pre- stranded sytpH-TEV was silenced by protease treatment (Fig. 5c),
ferentially retrieved for steric reasons. stimulation produced, as expected, fluorescence transients that only
To test this, we sought to silence the fluorescence of the stranded pool partially recovered (Fig. 5d). After mixing stranded and vesicular pools
by selective photobleaching8, using repeated laser scanning at high by stimulation with 800 action potentials at 5 Hz, transients once again
intensity to minimize bleaching time and thereby any turnover by spon- recovered completely in the same set of boutons (Fig. 5d). For
taneous release. By using sequential pulses of acidic and high ammo- both spH-TEV and sytpH-TEV, the sizes of the stranded pools were
nium solution, we indeed found that bleaching of unquenched stranded similar and the degree of recovery after digestion was the same (Fig. 5d)
spH-TEV was about tenfold as effective as that of quenched vesicular for a given stimulus protocol (here 200 action potentials at 20 Hz).
spH-TEV (Fig. 4a,b). We next stimulated spH-TEV–expressing boutons Thus, the loss of vesicle proteins is a more general property of fusing
(n ¼ 4 experiments, 450 boutons each) with 200 action potentials at synaptic vesicles.
20 Hz, before and after bleaching as well as following mixing after
bleaching (with 800 action potentials at 5 Hz). Qualitatively and quanti- Stranded pool size is under modulatory control
tatively, the same results were obtained as for digestion (Fig. 4c–e, Interactions between pairs of synaptic vesicle proteins can play a role in
Fig. 3c inset, and Supplementary Fig. 2 online). This demonstrates that synaptic vesicle biogenesis25. Synaptobrevin/VAMP-2 forms a complex
neither the protease treatment nor the presence of the large pHluorin on the synaptic vesicle membrane with synaptophysin-1 (ref. 26).
moiety affects synaptobrevin recycling or rates of endocytosis. Therefore the change in the stoichiometry of vesicular proteins by


Figure 5 Nonidentity is also observed for
a Stranded syt Quenchable syt b 2.0 pH 5.5 NH4Cl
synaptotagmin-1-TEV-pHluorin. (a) Images of
1.8 axonal arborizations expressing sytpH-TEV,
1.6 showing the fluorescence distribution of plasma

1.4 membrane–stranded and vesicular sytpH-TEV.
1.2 Image at left was computed by subtracting
pictures before and during a pulse with acidic
solution, and that at right was computed by
subtracting picures during and after a pulse with
0 10 20 30
Time (s)
ammonium chloride solution. (b) Fluorescence
profiles of boutons during sequential pulses
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

1.2 20 Hz
c d of acidic solution and ammonium chloride

(F – F0)/Fpeak
0 min TEV protease 15 min TEV protease 20 Hz 0.6
solution are similar to those of spH-TEV (error
1.0 0.2 bars represent s.e.m.). (c) Images of axonal
arborizations of a after 0 and 15 min of TEV

(F – F0)/Fpeak
0 20 40 60
Time (s) protease treatment at room temperature.
(d) Normalized sytpH-TEV fluorescence transients
of boutons (shown in a) upon application of
0.2 Postdigest 200 action potentials at 20 Hz, predigest,
0.0 postdigest and postmixing. The degree of
0 20 40 60 80 incomplete recovery after digestion is the
Time (s) same as that seen with spH-TEV under the same
conditions (inset). Scale bars in a and c, 10 mm.

synaptobrevin (or spH) overexpression might artificially introduce a sizes of vesicular and putative plasma membrane–stranded pools in the
membrane pool of ‘missorted’ synaptobrevins/spH (ref. 27). Thus, to absence of any overexpression. To this end, we used a polyclonal
control for a potential overexpression artefact, we cotransfected spH antibody to the ecto- or luminal domain of synaptotagmin-1
along with synaptophysin-mRFP (Fig. 6a). Synaptophysin-eGFP has (ref. 30). Incubation of living neurons (nonpermeabilized) with this
successfully been used as a synaptic vesicle marker28, so regions with antibody at 4 1C in saline with 1 mM tetrodotoxin (TTX) and without
overlapping green (spH-TEV) and red (synaptophysin-mRFP) fluor- calcium indeed yielded specific staining of synaptic boutons (Fig. 6d,
escence indicate the locations of presynaptic boutons. This did not top), albeit with weak fluorescence intensities when compared to the
abolish targeting of a substantial fraction of spH-TEV to the plasma signals obtained with the same primary and secondary antibodies after
membrane; however, it resulted in an about threefold change of the fixation and permeabilization (Fig. 6d, bottom). Although we cannot
vesicular membrane–to–plasma membrane distribution ratio (percent fully rule out antibody uptake by spontaneous activity during the
stranded spH-TEV ± s.e.m., control 33.2 ± 0.9, + Syp 10.1 ± 0.8; n ¼ 3 staining procedure for nonpermeabilized cells, the fluorescence inten-
and 5 experiments, respectively) (Fig. 6b). Accordingly, the relative sity profiles along axons with specific staining at interconnecting axon
amplitudes of fluorescence transients upon stimulation (F/F0) were segments (note that the feeding glial cell layer is not stained) suggest
increased (also about threefold) when synaptophysin-mRFP was coex- that a measurable fraction of vesicular proteins are stranded at the
pressed (Fig. 6c, left). But both the absolute fluorescence intensity of plasma membrane under native conditions as well. The measured levels
synapses during ammonium application, a measure for the total of about 10 and 80 fluorescence arbitrary units for nonpermeabilized
expression of spH-TEV in boutons, and the size of the absolute and permeabilized conditions (10.6 ± 0.9 arbitrary units for unper-
amplitudes after stimulation with 200 action potentials were meabilized cells, 6 images with 450 boutons each, and 83.9 ± 5.9
unchanged when both synaptophysin-mRFP and spH-TEV were over- arbitrary units for permeabilized cells, 8 images with 450 boutons
expressed (total expression levels in arbitrary units ± s.e.m. were 1,006 each) (Fig. 6e), however, may be misleading because epitope accessi-
± 177 in control and 905 ± 156 in + Syp; absolute amplitudes in bility may be very different and because of some possible spontaneous
arbitrary units ± s.e.m. were 127 ± 32 in control and 88 ± 23 in + Syp; activity even at 4 1C.
n ¼ 3 experiments for control, 5 for synaptophysin coexpression,
450 boutons each). This suggests that the number of spH-TEV copies A ‘readily retrievable pool’ of vesicles?
per vesicle is similar for both conditions. Peak-normalized signals, Our data show that vesicular proteins are lost after fusion and diffuse
however, revealed that the degree of recovery was slightly larger in the away from sites of release. This raises the question of whether vesicular
presence of synaptophysin-mRFP (Fig. 6c, right). These data suggest proteins after fusion freely mix with the fraction of vesicular proteins
that the stoichiometry of proteins inside vesicles is almost independent stranded at the plasma membrane, or whether the stranded proteins are
of the relative expression levels—that is, only a certain number of actually preferentially retrieved during compensatory endocytosis. In
‘slots’ can be occupied by each protein species per vesicle. But the the latter case, stranded vesicle proteins might constitute a ‘readily
relative steady-state concentrations of vesicular proteins stranded in the retrievable pool’ of partially preassembled structures that are preferen-
bouton membrane are modified by the retrieval efficiency for a given tially endocytosed upon exocytosis.
protein species29. This also identifies vesicle formation at the bouton To test this, we stimulated boutons after digestion with different
plasma membrane as a major sorting step in synaptic vesicle recycling numbers of action potentials and assessed the retrieval efficiencies of
and biogenesis. freshly exocytosed proteins by measuring relative recovery (that is,
Although overexpression of spH-TEVor sytpH-TEV alone obviously relative to recovery after mixing) at a time when most endocytic activity
leads to artificially increased pools of plasma membrane–stranded has ceased but errors due to focus or bouton drift or bleaching should
proteins, the existence of a pool of a small percentage of all vesicular still be minimal (Fig. 7). If stranded proteins were indeed preferen-
components at the bouton membrane may be of high functional tially retrieved, a plot of relative recovery versus the number of action
significance. To test this more rigorously, we sought to assess the potentials should intercept the abscissa at a positive number of action


Figure 6 Stranded pool size is under modu-
a SynaptopHluorin Synaptophysin-mRFP Merge latory control and is about 10% of total pool
size under native conditions. (a) Images of
axonal arborizations expressing spH-TEV (left)
and synaptophysin-mRFP (middle). Colocalization
was seen at presynaptic boutons (right).
(b) Relative fluorescence intensities of plasma
membrane–stranded spH-TEV at boutons that
b c 3.0 20 Hz 20 Hz did not (‘Control’) and did (‘+ Syp’) overexpress
Percentage stranded spH

1.0 synaptophysin-mRFP (along with spH-TEV),
25 2.5

(F – F0)/Fpeak
Control 0.8 Control determined by sequential pulses of acidic
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

20 + Syp + Syp
F/F0 2.0 0.6 solution or ammonium chloride solution
0.4 (error bars represent s.e.m.; P o 0.0001).
10 1.5 (c) Relative (left) and peak-normalized (right)
5 0.2
spH-TEV fluorescence time courses at boutons
0 1.0 0.0
Control + Syp 0 20 40 60 80 0 20 40 60 80
that did not and did overexpress synaptophysin-
Time (s) Time (s) mRFP (as well as spH-TEV) upon stimulation with
200 action potentials at 20 Hz (n ¼ 3 and 5
d Unpermeabilized
255 40
e 120 experiments, respectively; 450 boutons each,
Fluorescence (a.u.)

30 error bars represent s.e.m.). (d) Images of axonal
arborizations stained with antibodies to the

Average bouton fluorescence (a.u.)
20 100
ectodomain of synaptotagmin-1 without and with
10 permeabilization (details in Methods). Detailed
images for both conditions show punctate bouton
0 5 10 15 staining. For nonpermeabilized cells, specific
Distance (µm) 60 staining was also observed at the interconnecting
255 axonal segments, better seen in the fluorescence
Permeabilized 140
Fluorescence (a.u.)

120 40 intensity profiles (right) along the LOIs (shown
100 as red solid lines in the detailed images).
80 20 (e) Quantification of bouton fluorescence
60 intensities using areas of interest of a larger
set of boutons. Error bars represent s.e.m.
Scale bars, 10 mm in a, 5 mm in d.




0 5 10 15 20


Distance (µm)

potentials. The corresponding number of vesicles would be the size of compensatory endocytosis lose their protein complement during
the preferentially retrieved or readily retrievable pool. Our measured exocytosis (Figs. 1–3 and 5), making it essential that a sorting step
relative recoveries slightly favor the existence of a preferentially occurs during and maybe also after endocytosis (that is, by an
retrieved pool of stranded proteins. For 40 action potentials, we endosome). Whereas cotransfection with synaptophysin-mRFP
found virtually no recovery after digestion (Fig. 7, inset), although affected spH-TEV targeting to the plasma membrane, there appeared
there may have been some recovery when synaptophysin-mRFP was to be a small pool of stranded synaptic vesicle components under native
coexpressed (open circle in Fig. 7, Supplementary Fig. 3 online). The conditions also (Fig. 6), from which new vesicles were retrieved
slight slow downward trend of the signal may reflect, however, sources preferentially (Fig. 7). Because the large pHluorin moiety did not
other than specific uptake by compensatory endocytosis, such as focus affect vSNARE recycling or rates of endocytosis, this preferential
drift, bleaching, diffusional spread and so on. The finding that relative endocytosis of stranded vesicle proteins cannot be attributed to sterical
recoveries plateaued for high numbers of action potentials is in hindrance (Fig. 4).
agreement with the finding that synaptic vesicle turnover reaches a The existence of a surface population of vesicle proteins is in
maximum at about 200 action potentials (ref. 31). In addition, because agreement with previous studies that showed the expression of synap-
of endocytosis during stimulation32, the peak fluorescence at the end of tobrevin in hippocampal neurons, which was interpreted as a result of
the stimulus is increasingly underestimated for longer stimuli, leading either small inefficiencies during retrieval16,22 or ectopic recycling
to an underestimation of relative recovery. This may explain why we during synaptogenesis33,34 . The delivery of synaptic vesicle proteins
observed an even smaller recovery (albeit only slightly so) for 400 like synaptotagmin to the plasma membrane may be important during
action potentials than for 200 action potentials. development30, during synaptic vesicle biogenesis and for functional
maintenance of mature synapses.
DISCUSSION Here we show that this surface-stranded pool actively participates in
We tested whether synaptic vesicles lose their identity with respect to exo- and endocytic synaptic vesicle cycling. The pool was depleted by
their protein composition during stimulus-evoked exo- and endocy- stimuli of more than 40 action potentials (Fig. 7). Notably, it has been
tosis, using a fusion construct of the vSNARE synaptobrevin with the shown that a stimulus of 40 action potentials at 20 Hz also depletes the
pH-sensitve GFP variant pHluorin (synapto-pHluorin). We intro- readily releasable pool of docked vesicles, comprising about 5–10
duced a TEV protease cleavage site between both moieties that was vesicles23,35. Thus, the readily releasable pool of vesicles may be
accessible only if spH-TEV was in the plasma membrane. We found counterbalanced by a readily retrievable pool of preassembled endo-
that mostly digested molecules (that is, spH deposited on the plasma cytic structures (Fig. 7). Results from experiments in which both
membrane before digestion) were recycled during compensatory synaptophysin and spH-TEV were overexpressed, however, argue that
endocytosis. This suggests that synaptic vesicles that are recycled by the size of the stranded pool of vSNARES is highly overestimated in our


1.2 20 Hz potentials, kiss and run is the dominant mechanism of vesicle

(F – F0)/Fpeak
1.0 1.0
retrieval7,38. Because molecular identity is the major hallmark of kiss
Relative recovery (postdigest/postmixing)
0.8 Predigest
0.6 Postdigest
and run recycling, our data argue for full collapse of all fusing synaptic
0.8 0.2
0.0 vesicles during stimulation with 40 action potentials. If, however,
0 20 40 60 80 reacidification was much faster than has been recently reported
Time (s)
(B5 s; ref. 24), kiss and run events might add only to the noise in
the rising phase of our averaged fluorescence signals (Fig. 7, inset) and
0.4 would thus go mostly undetected. But even then, one would expect to
spHTEV, digest
observe a sharp decrease in the averaged fluorescence after the last
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

0.2 spHTEV, bleaching action potential of the stimulus, reflecting the kiss and run events
sytTEV, digest
spHTEV + syp triggered by the last few action potentials. We detected no such
0.0 fluorescence drop (Fig. 7, inset).
0 100 200 300 400
Number of APs
Our observations are best explained by a model in which synaptic
vesicles fully collapse after fusion and vesicle proteins disperse in the
Figure 7 Stranded synaptic vesicle proteins are preferentially recycled. bouton membrane and are rather slowly re-sorted at specialized sites of
Postdigest or postbleach recovery (relative to postmixing recovery at t ¼ 58 s endocytosis, which might be located at the periphery of the bouton.
after stimulus end) of bouton fluorescence transients for spH-TEV, spH-TEV + Alternatively, vesicles after collapse may form raft-like patches25,39,40
synaptophysin-RFP and sytpH-TEV, as a function of the number of action
that slowly diffuse in toto to sites of retrieval. In fact, it has been
potentials applied. Inset, normalized responses for 40 action potentials at
20 Hz predigest, postdigest and postmixing (n Z 3 experiments for all reported8 that after photobleaching of the membrane-stranded spH
conditions, error bars represent s.e.m.). pool, single exo-endocytic events are characterized by fluorescence
increases and decreases of the same size, implying that all exocytosed
spH molecules of a single vesicle stay together, irrespective of whether
measurements relying on the overexpression of spH-TEV or sytpH- they are retrieved fast (o1 s, kiss and run) or slowly.
TEV. On the other hand, the ‘stranded’ pool size measured with acidic This is contrary to our findings of fast diffusional dispersal and
and ammonium solution pulses (Figs. 1 and 5) may not be identical to preferential uptake of bleached/cleaved spH for stimuli of less than
the ‘functional’ retrievable pool size measured by determining the 40 action potentials.
stimulus yielding zero recovery (Fig. 7). In fact, with about 200 vesicles We think, however, that our finding of fast diffusion after fusion in
per bouton23, a stranded pool of 30% should correspond to 60 vesicles, all directions, also observed previously16,22, favors dispersion and free
whereas a functional retrievable pool of 5–10 vesicles would correspond mixing of vesicular and bouton membrane components. During one
to 2–5% fluorescence on the bouton membrane only—that is, less than exo-endocytic cycle, initially only the presorted membrane patches at
our synaptotagmin immunocytochemistry data indicate to be present these specialized sites of endocytosis would be retrieved by compensa-
under native conditions (Fig. 6d,e). Thus, overexpression of one vesicle tory endocytosis, leading to the observed nonidentity of exocytic and
protein species may merely result in an oversupply of stranded or endocytic vesicles as long as the presorted pool was not depleted—that
orphan proteins without changing the functional retrievable pool size. is, for stimuli of less than 40 action potentials. For prolonged stimula-
This is supported by our finding that coexpression of synaptophysin tion, a mixture of fast retrieval of presorted structures and slow retrieval
markedly decreased the steady-state stranded pool of spH-TEV of newly inserted vesicle components was observed, resulting in pro-
(Fig. 6b), but did not affect the number of spH-TEV molecules per gressively slower endocytic kinetics with stronger or longer stimuli16.
vesicle, nor the number of vesicles released during stimulation In non-neuronal cells, numerous stationary clathrin structures are
(Fig. 6c), and only slightly altered retrieval kinetics and recovery often observed19,41,42. In presynaptic boutons, however, such clathrin
efficiency (Fig. 7). Likewise, in presynaptic boutons of the tempera- structures have thus far not been detected. But spatial segregation of
ture-sensitive Drosophila dynamin mutant shibire, which was kept at a exocytic and endocytic sites in presynaptic boutons has been observed
nonpermissive temperature to block synaptic vesicle endocytosis, (see ref. 43 for review). In Drosophila neuromuscular junction36,44,45,
dispersion of excess synaptic vesicle molecules in the plasma membrane for example, components of the endocytic machinery, like dynamin
by lateral diffusion did not alter the spatial organization of the and a-adaptin, are highly enriched at ‘hot spots’, indicating that the
endocytic machinery36,37. Thus the endocytic machinery at sites of molecular players of the endocytic machinery are anchored at specia-
retrieval may define the size of the functional stranded pool as opposed lized sites and thereby act as spatial organizers of endocytosis. The
to that of the total stranded pool, which may also comprise excess or future challenge will be to dissect the general principles of the exo- and
orphan cargo molecules. endocytic spatial coordination and their tight temporal coupling. If it is
We found that a stimulus of 40 action potentials resulted in almost not the exocytosed vesicle proteins themselves that trigger their own
complete diffusional loss of vesicle proteins after fusion; this makes for retrieval, how might this tight coupling occur? Maybe the arrival of new
an attractive hypothesis—namely, that a small pool of readily retrie- proteins in the periactive zone forces compensatory endocytosis of the
vable vesicle patches exists near the active zones of small CNS synapses readily retrievable pool. Alternatively, a physical link between mem-
under native conditions. It has been shown that the rate of endocytosis brane tension and endocytosis or the trigger for exocytosis—that is, a
is fastest for this stimulus16. A readily retrievable pool could nicely rise of intracellular calcium concentration—may couple compensatory
explain a first, fast and depletable endocytic component at synapses endocytosis to exocytosis.
without having to postulate a clathrin-independent mechanism like
kiss and run. Such a pool of presorted or preassembled endocytic METHODS
structures could also explain the fact that the first peak of endocytic pits Cell culture. Hippocampal neurons of the CA3/CA1 region from 1- to 3-d-old
occurs at 10 s in the frog neuromuscular junction, a finding from the Wistar rats were prepared in a rather sparse culture (B2,000–5,000 cells per
electron microscopic reconstruction of the endocytic time course at coverslip) and transfected at 4–6 days in vitro (DIV) by a modified calcium
this site10. It has been suggested that for the stimulus of 40 action phosphate transfection procedure46. Microscopy was performed at 14–16 DIV.


Plasmid constructs. Superecliptic pHluorin-synaptobrevin-2 (synapto- subjected to an à trous wavelet transformation with the level k ¼ 4 and
pHluorin; ref. 15) was provided by G. Miesenböck (Yale University, New detection level ld ¼ 1.0 (ref. 49), resulting in a segmented mask image. Spots on
Haven). A DNA fragment encoding the cleavage site for recombinant TEV mask images, each representing putative functional boutons, were identified,
protease (rTEV) flanked by spacer arms (amino acid sequence SGGSGGDY and only masks with areas between 4 and 20 pixels were accepted for calculating
DIPTTENLYFQGELKTVDAD), according to the manufacturer’s recommenda- bouton fluorescence transients. All identified masks and calculated time courses
tions (Invitrogen), was introduced into the linker region between the synapto- were visually inspected for correspondence to individual functional boutons.
brevin-2 and the pHluorin moiety by polymerase chain reaction (PCR).
The synaptotagmin-1-TEV-pHluorin (sytpH) expression construct was Note: Supplementary information is available on the Nature Neuroscience website.
made by first fusing the rTEV target site flanked by spacer arms (amino acid ACKNOWLEDGMENTS
sequence DYDIPTTENLYFQGELKTVDAD) to rat synaptotagmin-1. In a We are grateful to O. Kochubey for his assistance with data analysis, R. Nehring
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

second step, the cDNAs encoding the signal sequence of rat preprotachykinin for advice in molecular biology, E. Neher for support, A.K. Boegle and
and the pHluorin were fused to the amino-terminal domain of the construct29. T. Groemer for critically reading the manuscript, all lab members for fruitful
Monomeric red fluorescent protein47 was used to replace eGFP in a plasmid discussions and M. Pilot for expert technical assistance. This work was
encoding for a GFP-tagged p38 (ref. 28; gift of A. Iliev, European Neuroscience supported by grants from the Deutsche Forschungsgemeinschaft (SFB 523,
Institute, Göttingen, Germany) to generate the synaptophysin-mRFP construct. J.K.), the Human Frontier Science Project (J.K.) and the Boehringer Ingelheim
All constructs were cloned into a modified version of the pcDNA3 expres- Fonds (M.W.).
sion plasmid (Invitrogen) carrying a neuron-specific human synapsin-1 gene
promoter48 (obtained from S. Kügler, University of Göttingen, Germany) and The authors declare that they have no competing financial interests.
were verified by dideoxynucleotide sequencing.
Published online at http://www.nature.com/natureneuroscience
Enzymatic tag removal. Proteolytic cleavage was performed at room tempera- Reprints and permissions information is available online at http://npg.nature.com/
ture (23 1C) by adding 60 U AcTEV protease and 1 mM dithiothreitol reprintsandpermissions/
(Invitrogen) directly to the living neurons for 15 min. The progress of cleavage
was assayed by imaging the loss of fluorescence. After digestion, cells were 1. Heuser, J.E. & Reese, T.S. Evidence for recycling of synaptic vesicle membrane during
transmitter release at the frog neuromuscular junction. J. Cell Biol. 57, 315–344
washed for 2 min. (1973).
2. Cremona, O. & De Camilli, P. Synaptic vesicle endocytosis. Curr. Opin. Neurobiol. 7,
Immunocytochemistry. To assess the fraction of noninternalized surface- 323–330 (1997).
stranded synaptotagmin-1 (syt1), live neurons were labeled with monoclonal 3. Brodin, L., Low, P. & Shupliakov, O. Sequential steps in clathrin-mediated synaptic
vesicle endocytosis. Curr. Opin. Neurobiol. 10, 312–320 (2000).
syt1 lumenal domain antibody (mouse monoclonal, ascites, 604.4, 1:100; 4. Brodsky, F.M., Chen, C.Y., Knuehl, C., Towler, M.C. & Wakeham, D.E. Biological basket
Synaptic Systems) at 4 1C in nominally Ca2+-free solution in the presence of weaving: formation and function of clathrin-coated vesicles. Annu. Rev. Cell Dev. Biol.
TTX and 4% goat serum for 1 h to allow for specific binding. Cells were 17, 517–568 (2001).
washed, fixed with 4% formaldehyde (freshly prepared from paraformalde- 5. Ceccarelli, B., Hurlbut, W.P. & Mauro, A. Turnover of transmitter and synaptic vesicles at
the frog neuromuscular junction. J. Cell Biol. 57, 499–524 (1973).
hyde) at 4 1C for 30 min, and stained with Alexa 546 secondary antibodies 6. Koenig, J.H., Yamaoka, K. & Ikeda, K. Omega images at the active zone may be
(1:1,000; Invitrogen). endocytotic rather than exocytotic: implications for the vesicle hypothesis of transmitter
For labeling the total pool of syt1, cells were fixed, permeabilized with release. Proc. Natl. Acad. Sci. USA 95, 12677–12682 (1998).
0.4% saponin (Sigma) and labeled with the primary and secondary anti- 7. Aravanis, A.M., Pyle, J.L. & Tsien, R.W. Single synaptic vesicles fusing transiently and
successively without loss of identity. Nature 423, 643–647 (2003).
bodies as above. Images were acquired using equal light exposure times for 8. Gandhi, S.P. & Stevens, C.F. Three modes of synaptic vesicular recycling revealed by
both conditions. single-vesicle imaging. Nature 423, 607–613 (2003).
9. Takei, K., Mundigl, O., Daniell, L. & De Camilli, P. The synaptic vesicle cycle: a single
Epifluorescence and confocal microscopy of living neurons. A modified vesicle budding step involving clathrin and dynamin. J. Cell Biol. 133, 1237–1250
Tyrode solution (150 mM NaCl, 4 mM KCl, 1 mM MgCl2, 2 mM CaCl2, 10. Ryan, T.A. et al. The kinetics of synaptic vesicle recycling measured at single presynaptic
10 mM glucose, 10 mM HEPES buffer, pH 7.4) was used for all experiments boutons. Neuron 11, 713–724 (1993).
unless otherwise indicated. Synaptic boutons were stimulated by electric field 11. Ryan, T.A., Smith, S.J. & Reuter, H. The timing of synaptic vesicle endocytosis. Proc.
stimulation (platinum electrodes, 10-mm spacing, 1-ms pulses of 50 mA and Natl. Acad. Sci. USA 93, 5567–5571 (1996).
12. Wu, L.G. & Betz, W.J. Nerve activity but not intracellular calcium determines the time
alternating polarity); 10 mM 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX) course of endocytosis at the frog neuromuscular junction. Neuron 17, 769–779 (1996).
and 50 mM D,L-2-amino-5-phosphonovaleric acid (AP5) were added to prevent 13. Klingauf, J., Kavalali, E.T. & Tsien, R.W. Kinetics and regulation of fast endocytosis at
recurrent activity. Fast solution exchanges were achieved by a piezo-controlled hippocampal synapses. Nature 394, 581–585 (1998).
stepper device (SF77B, Warner Instruments) using a three-barrel glass tubing. 14. Miesenbock, G., De Angelis, D.A. & Rothman, J.E. Visualizing secretion and synaptic
transmission with pH-sensitive green fluorescent proteins. Nature 394, 192–195
Ammonium chloride solution (pH 7.4) was prepared by substituting 50 mM (1998).
NaCl in normal saline with NH4Cl, while all other components remained 15. Sankaranarayanan, S. & Ryan, T.A. Real-time measurements of vesicle-SNARE recy-
unchanged. Acidic solution with a final pH of 5.5 was prepared by replacing cling in synapses of the central nervous system. Nat. Cell Biol. 2, 197–204 (2000).
HEPES with 2-[N-morpholino]ethane sulphonic acid (pKa ¼ 6.1). 16. Mueller, V.J., Wienisch, M., Nehring, R.B. & Klingauf, J. Monitoring clathrin-mediated
endocytosis during synaptic activity. J. Neurosci. 24, 2004–2012 (2004).
In some experiments, recycling synaptic vesicles were additionally labeled 17. Merrifield, C.J., Feldman, M.E., Wan, L. & Almers, W. Imaging actin and dynamin
with FM5-95 (Invitrogen). Cells were exposed to 10 mM FM5-95 during and recruitment during invagination of single clathrin-coated pits. Nat. Cell Biol. 4,
for 1 min after stimulation and washed for 10 min before imaging. 691–698 (2002).
Images were taken by a cooled slow-scan CCD camera (SensiCam-QE, PCO) 18. Loerke, D., Wienisch, M., Kochubey, O. & Klingauf, J. Differential control of clathrin
subunit dynamics measured with EW-FRAP microscopy. Traffic 6, 918–929 (2005).
mounted on an inverted microscope (Axiovert S100TV, Zeiss) with a 63, 1.2 19. Miller, T.M. & Heuser, J.E. Endocytosis of synaptic vesicle membrane at the frog
NA water-immersion objective and an eGFP filter set (DCLP 505, BP 525/50). neuromuscular junction. J. Cell Biol. 98, 685–698 (1984).
PHluorins were excited at 470 nm with a computer-controlled monochromator 20. Sankaranarayanan, S., De Angelis, D., Rothman, J.E. & Ryan, T.A. The use of pHluorins
(Polychrom II, Till Photonics) every 2 s for 500 ms. FM5-95 was excited at for optical measurements of presynaptic activity. Biophys. J. 79, 2199–2208 (2000).
21. Taubenblatt, P., Dedieu, J.C., Gulik-Krzywicki, T. & Morel, N. VAMP (synaptobrevin) is
515 nm and imaged using 565 nm dichroic and 590 nm long-pass emission present in the plasma membrane of nerve terminals. J. Cell Sci. 112, 3559–3567
filters. Photobleaching was performed by repetitive laser scanning at 488 nm for (1999).
90 s (SP2 confocal microscope, Leica). 22. Li, Z. & Murthy, V.N. Visualizing postendocytic traffic of synaptic vesicles at hippocam-
Quantitative analysis was performed with self-written macros in Igor Pro pal synapses. Neuron 31, 593–605 (2001).
23. Schikorski, T. & Stevens, C.F. Quantitative ultrastructural analysis of hippocampal
(Wavemetrics). To avoid the bias introduced by manual selection of functional excitatory synapses. J. Neurosci. 17, 5858–5867 (1997).
boutons, an automated detection algorithm was used. The image from the time 24. Atluri, P.P. & Ryan, T.A. The kinetics of synaptic vesicle reacidification at hippocampal
series showing the maximal pHluorin response during stimulation was nerve terminals. J. Neurosci. 26, 2313–2320 (2006).


25. Thiele, C., Hannah, M.J., Fahrenholz, F. & Huttner, W.B. Cholesterol binds to synapto- 37. Roos, J. & Kelly, R.B. The endocytic machinery in nerve terminals surrounds sites of
physin and is required for biogenesis of synaptic vesicles. Nat. Cell Biol. 2, 42–49 exocytosis. Curr. Biol. 9, 1411–1414 (1999).
(2000). 38. Pyle, J.L., Kavalali, E.T., Piedras-Renteria, E.S. & Tsien, R.W. Rapid reuse of
26. Calakos, N. & Scheller, R.H. Vesicle-associated membrane protein and synaptophysin readily releasable pool vesicles at hippocampal synapses. Neuron 28, 221–231
are associated on the synaptic vesicle. J. Biol. Chem. 269, 24534–24537 (1994). (2000).
27. Pennuto, M., Bonanomi, D., Benfenati, F. & Valtorta, F. Synaptophysin I controls 39. Martin, T.F. Racing lipid rafts for synaptic-vesicle formation. Nat. Cell Biol. 2, E9–11
the targeting of VAMP2/synaptobrevin II to synaptic vesicles. Mol. Biol. Cell 14, (2000).
4909–4919 (2003). 40. Willig, K.I., Rizzoli, S.O., Westphal, V., Jahn, R. & Hell, S.W. STED microscopy reveals
28. Kaether, C., Skehel, P. & Dotti, C.G. Axonal membrane proteins are transported in that synaptotagmin remains clustered after synaptic vesicle exocytosis. Nature 440,
distinct carriers: a two-color video microscopy study in cultured hippocampal neurons. 935–939 (2006).
Mol. Biol. Cell 11, 1213–1224 (2000). 41. Gaidarov, I., Santini, F., Warren, R.A. & Keen, J.H. Spatial control of coated-pit dynamics
29. Diril, M.K., Wienisch, M., Jung, N., Klingauf, J. & Haucke, V. Stonin 2 is an AP-2- in living cells. Nat. Cell Biol. 1, 1–7 (1999).
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

dependent endocytic sorting adaptor for synaptotagmin internalization and recycling. 42. Kirchhausen, T. Clathrin. Annu. Rev. Biochem. 69, 699–727 (2000).
Dev. Cell 10, 233–244 (2006). 43. Gundelfinger, E.D., Kessels, M.M. & Qualmann, B. Temporal and spatial coordination of
30. Matteoli, M., Takei, K., Perin, M.S., Sudhof, T.C. & De Camilli, P. Exo-endocytotic exocytosis and endocytosis. Nat. Rev. Mol. Cell Biol. 4, 127–139 (2003).
recycling of synaptic vesicles in developing processes of cultured hippocampal neurons. 44. Gonzalez-Gaitan, M. & Jackle, H. Role of Drosophila alpha-adaptin in presynaptic
J. Cell Biol. 117, 849–861 (1992). vesicle recycling. Cell 88, 767–776 (1997).
31. Vanden Berghe, P. & Klingauf, J. Synaptic vesicles in hippocampal boutons recycle to 45. Roos, J. & Kelly, R.B. Dap160, a neural-specific Eps15 homology and multiple SH3
different pools in a use-dependent fashion. J. Physiol. 572, 707–720 (2006). domain-containing protein that interacts with Drosophila dynamin. J. Biol. Chem. 273,
32. Fernandez-Alfonso, T. & Ryan, T.A. The kinetics of synaptic vesicle pool depletion at 19108–19119 (1998).
CNS synaptic terminals. Neuron 41, 943–953 (2004). 46. Threadgill, R., Bobb, K. & Ghosh, A. Regulation of dendritic growth and remodeling by
33. Ahmari, S.E., Buchanan, J. & Smith, S.J. Assembly of presynaptic active zones from Rho, Rac, and Cdc42. Neuron 19, 625–634 (1997).
cytoplasmic transport packets. Nat. Neurosci. 3, 445–451 (2000). 47. Campbell, R.E. et al. A monomeric red fluorescent protein. Proc. Natl. Acad. Sci. USA
34. Sampo, B., Kaech, S., Kunz, S. & Banker, G. Two distinct mechanisms target membrane 99, 7877–7882 (2002).
proteins to the axonal surface. Neuron 37, 611–624 (2003). 48. Kugler, S. et al. Neuron-specific expression of therapeutic proteins: evaluation of
35. Murthy, V.N. & Stevens, C.F. Reversal of synaptic vesicle docking at central synapses. different cellular promoters in recombinant adenoviral vectors. Mol. Cell. Neurosci.
Nat. Neurosci. 2, 503–507 (1999). 17, 78–96 (2001).
36. Estes, P.S. et al. Traffic of dynamin within individual Drosophila synaptic boutons 49. Olivo-Marin, J.-C. Extraction of spots in biological images using multiscale products.
relative to compartment-specific markers. J. Neurosci. 16, 5443–5456 (1996). Pattern Recognit. 35, 1989–1996 (2002).


Generalization of amygdala LTP and conditioned fear
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

in the absence of presynaptic inhibition
Hamdy Shaban1, Yann Humeau1,2, Cyril Herry1, Guillaume Cassasus1, Ryuichi Shigemoto3, Stephane Ciocchi1,
Samuel Barbieri4, Herman van der Putten5, Klemens Kaupmann5, Bernhard Bettler4 & Andreas Lüthi1

Pavlovian fear conditioning, a simple form of associative learning, is thought to involve the induction of associative, NMDA receptor–
dependent long-term potentiation (LTP) in the lateral amygdala. Using a combined genetic and electrophysiological approach, we
show here that lack of a specific GABAB receptor subtype, GABAB(1a,2), unmasks a nonassociative, NMDA receptor–independent form
of presynaptic LTP at cortico-amygdala afferents. Moreover, the level of presynaptic GABAB(1a,2) receptor activation, and hence the
balance between associative and nonassociative forms of LTP, can be dynamically modulated by local inhibitory activity. At the
behavioral level, genetic loss of GABAB(1a) results in a generalization of conditioned fear to nonconditioned stimuli. Our findings
indicate that presynaptic inhibition through GABAB(1a,2) receptors serves as an activity-dependent constraint on the induction of
homosynaptic plasticity, which may be important to prevent the generalization of conditioned fear.

Induction of NMDA receptor–dependent long-term potentiation Functional GABAB receptors are generally thought to be hetero-
(LTP) in the lateral amygdala is thought to underlie the acquisition dimers containing a GABAB(1) and a GABAB(2) subunit20. The
of classical fear conditioning in rodents1–3. Projection neurons in the GABAB(1) subunit exists in two predominant differentially expressed
lateral amygdala receive converging thalamic and cortical sensory isoforms—GABAB(1a) and GABAB(1b). These differ by virtue of the fact
afferents that are simultaneously active during sensory experience1. that the GABAB(1a) isoform contains two ‘sushi repeats’ in the N
Whereas LTP at thalamic afferents is thought to be induced and terminal domain20. In the lateral amygdala, GABAB receptors are
expressed postsynaptically3–6 (but see ref. 7), associative coactivation expressed at moderate to high levels21,22 and can be activated by
of thalamic and cortical afferents induces presynaptic LTP at cortical afferent stimulation in vitro and in vivo9–11. These earlier studies
afferents involving the heterosynaptic activation of presynaptic NMDA focused on somato-dendritic GABAB receptors that inhibit the post-
receptors8. Thus, by means of these receptors, cortical afferents are able synaptic neuron by activating inwardly rectifying Kir3-type K+ chan-
to detect and integrate the coincident activity of neighboring excitatory nels, which give rise to a slow inhibitory postsynaptic potential23.
inputs. Given that neuronal activity and plasticity in the lateral Here we took a combined genetic and electrophysiological approach
amygdala is tightly constrained by local GABAergic inhibition5,9–11, to determine the role of presynaptic GABAB receptors during
this raises the question of whether presynaptic integration is also the induction of presynaptic LTP in the lateral amygdala.
subject to inhibitory control. Moreover, we analyzed whether the observed effects on LTP in the
Presynaptic GABAergic inhibition of excitatory synaptic transmis- lateral amygdala are accompanied by behavioral alterations in Pavlo-
sion can be mediated by ionotropic GABAA (ref. 12) and metabotropic vian fear conditioning.
GABAB receptors13. In the lateral amygdala, GABAA receptor–mediated
inhibition has a major role in postsynaptic integration5,9–11. Presynap- RESULTS
tic LTP at cortical afferents, however, is insensitive to GABAA receptor– GABAB blockade unmasks nonassociative homosynaptic LTP
mediated inhibition8, suggesting that presynaptic GABAA receptors do Whole-cell current-clamp recordings from projection neurons showing
not play a major role at cortico-amygdala afferents. Indeed, there is spike frequency adaptation upon depolarizing current injection were
accumulating evidence suggesting a role for GABAB receptors in obtained in the dorsal subdivision of the lateral amygdala4,5. Stimula-
regulating amygdala-dependent fear and anxiety14. However, whereas tion of afferent fibers from the internal capsule, containing thalamic
a role for GABAB receptors in presynaptic inhibition and integration at afferents4, and of those from the external capsule, containing cortical
glutamatergic synapses is well documented in many brain areas13,15–17 afferents24 (Fig. 1a), elicited monosynaptic excitatory postsynaptic
including the basolateral amygdala18,19, nothing is known about their potentials (EPSPs) of similar amplitudes and slopes. Simultaneous
role in the lateral amygdala. stimulation of cortical and thalamic afferents with a single Poisson train

1Friedrich Miescher Institute for Biomedical Research, CH-4058 Basel, Switzerland. 2Centre National de la Recherche Scientifique, Unité Mixte de Recherche 7168,

F-67084 Strasbourg, France. 3Division of Cerebral Structure, National Institute for Physiological Sciences, Myodaiji, Okazaki 444-8585, CREST, Japan Science and
Technology Corporation, Kawaguchi, Japan. 4Pharmazentrum, Department of Clinical-Biological Sciences, University of Basel, CH-4056 Basel, Switzerland.
5Novartis Institutes for Biomedical Research, Novartis Pharma AG, CH-4002 Basel, Switzerland. Correspondence should be addressed to A.L. (andreas.luthi@fmi.ch).

Received 17 April; accepted 6 June; published online 2 July 2006; doi:10.1038/nn1732


Figure 1 GABAB receptor blockade enables the induction of homosynaptic,
a Stimulation
b Cortical + thalamic
Thalamic Cortical Post nonassociative LTP at cortical afferents to the lateral amygdala.
Cortical afferents Pre
(a) Placement of stimulation and recording electrodes. (b) Time course of

EPSP slope (%)
200 synaptic changes after simultaneous tetanic stimulation of cortical and
Thalamic afferents
thalamic afferent fibers. Heterosynaptic, associative LTP was induced at
100 cortical, but not thalamic, afferents (n ¼ 11). (c) A single tetanus delivered
Cort + thal Cortical Thalamic only to the cortical afferents did not induce LTP (n ¼ 10). In the presence of
0 10 20 30 the GABAB receptor antagonist CGP55845A (10 mM), homosynaptic,
Time (min) nonassociative LTP was induced by the same protocol (n ¼ 6). (d) Induction
c Cortical only d Post Scaled Pre
of homosynaptic LTP (filled circles) was associated with a decreased paired-
pulse ratio (PPR; open circles; n ¼ 10). (e) Previous induction of
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

Control CGP Post

Cortical EPSP/PPR (%)
homosynaptic LTP occluded the subsequent induction of heterosynaptic LTP
EPSP slope (%)

200 (n ¼ 8). (f) The converse experiment revealed a partial occlusion of LTP by

previous induction of heterosynaptic LTP (n ¼ 5). Numbers above traces refer
100 100 to experimental time points as indicated. Scale bars, 4 mV and 20 ms.
Cort Control CGP55845A Cort EPSP slope PPR
0 0
0 10 20 30 0 10 20 30
Time (min) Time (min)
of the activated inputs. Thus, these experiments show that GABAB
e 1 2 3
f 1 2 3
receptor–mediated inhibition serves as a constraint for the
induction of homosynaptic, nonassociative LTP at cortical afferents,
Cortical EPSP slope (%)

CGP55845A thereby making concomitant thalamic input a necessary requirement
EPSP slope (%)

2 3 3 for LTP induction.

200 200 2
1 1
100 100 Mechanisms of presynaptic LTP induction
Cort Cort + thal
Cort + thal Cort Because efficient glutamate uptake seems to prevent NMDA (auto-)
0 10 20 30 40 50 0 10 20 30 40 50 receptor activation by tetanic stimulation of cortical afferents alone,
Time (min) Time (min)
induction of heterosynaptic, associative LTP depends on the activation
of presynaptic NMDA receptors by glutamate released from thalamic
(45 stimuli at an average frequency of 30 Hz) resulted in the pathway- afferents8. In contrast, application of the competitive NMDA receptor
specific induction of associative LTP at cortical afferents (cortical: 157 ± antagonist 3-(R)-2-carboxypiperazin-4-propyl-1-phosphonic acid
14% of baseline, n ¼ 11, P o 0.05; thalamic: 101 ± 10%, n ¼ 11, P 4 (CPP; 20 mM) did not prevent the induction of nonassociative,
0.05; Fig. 1b and ref. 8). Stimulation of cortical afferents alone did not homosynaptic LTP (159 ± 18%, n ¼ 7, P o 0.05; Fig. 2a). Moreover,
result in long-lasting changes in synaptic efficacy (105 ± 11%, n ¼ 10, ionotropic glutamatergic transmission did not seem to be necessary for
P 4 0.05; Fig. 1c). In contrast, in the presence of the specific GABAB LTP induction, as LTP could be induced in the presence of the
antagonist CGP55845A (10 mM), homosynaptic, nonassociative LTP nonspecific glutamate receptor antagonist kynurenate (3 mM), which
could be induced by a single train delivered to cortical afferents (153 ± completely abolished excitatory postsynaptic potentials (5 ± 1% of
15%, n ¼ 6, P o 0.05; Fig. 1c). Like associative LTP induced by baseline, n ¼ 4; LTP: 191 ± 27%, n ¼ 4, P o 0.05; Fig. 2b).
costimulation of both inputs8, homosynaptic LTP was associated with a To test if postsynaptic Ca2+ signaling is required for the induction of
decrease in the ratio of the postsynaptic responses to double stimulation homosynaptic LTP, recorded neurons were loaded with the Ca2+
of cortical afferents (paired-pulse ratio, PPR; 73 ± 6% of baseline, chelator BAPTA (30 mM) for at least 15 min before LTP induction.
n ¼ 10, P o 0.05; Fig. 1d), suggesting a presynaptic expression This BAPTA concentration completely abolishes postsynaptic LTP
mechanism. To directly test whether homosynaptic and hetero- induction at thalamic inputs6. Homosynaptic LTP at cortical inputs,
synaptic LTP are mediated by the same expression mechanism, we however, was resistant to postsynaptic BAPTA perfusion (157 ± 19%,
performed occlusion experiments. Previous induction of homosynaptic n ¼ 6, P o 0.05; Fig. 2c,d). To determine if homosynaptic cortical
LTP completely occluded the subsequent induction of heterosynaptic
LTP (n ¼ 8; Fig. 1e). In the reverse experiment, previous induction
of heterosynaptic LTP partially occluded the induction of homo-
synaptic LTP (n ¼ 5; Fig. 1f), suggesting that associative LTP induction
a CPP Baseline LTP Post b Kynurenic acid
300 Pre
EPSP slope (%)

EPSP slope (%)

– Tetanic stim.
requiring NMDA receptor–dependent interactions between thalamic 300 + Tetanic stim.
and cortical afferents is more specific and is only induced at a subset 200 kyn
100 100
Cort +CGP55845A +CGP55845A
0 0
0 10 20 30 0 10 20 30
Figure 2 Induction of homosynaptic LTP at cortical afferents is independent Time (min) Time (min)
of NMDA receptor activation and postsynaptic Ca2+. (a) Homosynaptic
cortico-amygdala LTP was independent of NMDA receptor activation (n ¼ 7). c BAPTA
Post 200
(b) Complete blockade of glutamatergic transmission by kynurenate (3 mM) * *
* *
EPSP slope (%)

300 Pre
did not interfere with homosynaptic LTP induction (n ¼ 4; filled symbols). In
EPSP slope (%)

the absence of tetanic stimulation, synaptic responses recovered to baseline 200 100
levels after washout of kynurenate (n ¼ 3; open symbols). (c) LTP induction
was not affected by postsynaptic perfusion with BAPTA (30 mM; n ¼ 6). 100
Application of the membrane-permeant BAPTA-AM (50 mM) completely Cort +CGP55845A 10 6 7 6 6 6
0 0
blocked LTP induction (n ¼ 6). Scale bars, 3 mV and 20 ms. 0 10 20







(d) Quantification of pharmacological experiments. *P o 0.05. Number



Time (min)



of experiments are indicated on bars.


Figure 3 Synaptic activation of presynaptic GABAB receptors. (a) Synaptic
a 30 Hz, 1.5 s b activation of GABAB heteroreceptors on cortical terminals. Traces illustrate
150 r 2 = 0.596 suppression of EPSC amplitude and increase in PPR at cortical afferent

EPSC change (%)
Cort. synapses after tetanic stimulation of thalamic afferents. Scale bars, 100 pA
and 50 ms. (b) Consistent with a presynaptic mechanism, there was a clear
Control 50 correlation between the reduction in ESPC amplitude and an increase in
PPR (n ¼ 26). (c) CGP55845A blocked heterosynaptic depression.
0 Cumulative probability plot illustrates GABAB receptor–mediated
CGP 0 100 200
100 pA
50 ms PPF change (%) heterosynaptic depression 1 s after tetanic stimulation of thalamic afferents.
(d) Quantification of cortical EPSC ratios (EPSC amplitude after tetanic
c 1s d 1s stimulation/EPSC amplitude before tetanic stimulation) measured 1 s
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

after tetanic stimulation of thalamic afferents. CGP55845A abolished
1.0 150 ***
Cumulative prob.

heterosynaptic depression (control: n ¼ 54; CGP: n ¼ 23), whereas CPP
EPSC ratio had no effect (n ¼ 22). (e) CPP blocked heterosynaptic facilitation.
0.5 Ctrl Cumulative probability plot illustrates NMDA receptor–mediated
CPP 50 heterosynaptic facilitation 15 s after tetanic stimulation of thalamic
afferents. (f) Quantification of cortical EPSC ratios measured 15 s after
0.0 0
0 100 200 tetanic stimulation of thalamic afferents. CPP abolished heterosynaptic



facilitation (control: n ¼ 65; CGP: n ¼ 36), whereas CGP had no effect

EPSC (% of baseline)
(n ¼ 27). *P o 0.05, ***P o 0.001.
15 s 15 s
e f
1.0 125 *
Cumulative prob.

Cortical afferents show heterosynaptic facilitation mediated by
EPSC ratio

0.5 Ctrl
activation of presynaptic NMDA receptors8. Consistent with earlier
CPP 75 observations, we found that heterosynaptic facilitation did not develop
CGP immediately after stimulation of thalamic afferents. Whereas cortical
0.0 50
50 100 150 synapses were depressed after 1 s, 15 s later this depression was





EPSC (% of baseline) converted into a small but significant facilitation that was abolished
by the NMDA receptor antagonist CPP (control: 108 ± 2% of baseline,
n ¼ 65; CPP: 99 ± 2%, n ¼ 36, P o 0.05; Fig. 3e,f). CPP did not affect
LTP was nevertheless Ca2+ dependent, we bath-applied BAPTA-AM the EPSC ratio at the earlier time point (control: 63 ± 4% of baseline,
(50 mM), a membrane-permeant form of BAPTA. Stimulation strength n ¼ 54; CPP: 61 ± 6%, n ¼ 22; Fig. 3c,d). In contrast, GABAB receptor–
was set to compensate for the BAPTA-AM–induced reduction in mediated heterosynaptic depression was already terminated 15 s after
baseline EPSP amplitude. Under these conditions, BAPTA-AM com- stimulation of the thalamic afferents (control: 108 ± 2% of baseline,
pletely abolished LTP induction (96 ± 10%, n ¼ 6, P o 0.05; Fig. 2c,d). n ¼ 65; CGP: 110 ± 2%, n ¼ 27; Fig. 3e,f).
Consistent with the idea that postsynaptic depolarization is not Because presynaptic GABAB receptors on cortical afferents can be
involved in LTP induction, we found that, unlike thalamic LTP activated by synaptically released GABA, we reasoned that the level of
(ref. 5), homosynaptic cortical LTP could be induced in the absence GABAB receptor activation, and hence the threshold for homosynaptic
of the GABAA receptor antagonist picrotoxin (148 ± 17%, n ¼ 6, P o LTP induction, should be sensitive to the modulation of GABA release.
0.05; Fig. 2d). In conclusion, these results support the notion that To address this question we applied the m-opioid receptor agonist
induction of homosynaptic LTP at cortical afferents is independent of [D-Ala2, NMe-Phe4, Gly5-ol]-enkephalin (DAMGO), which is known
postsynaptic activity and requires presynaptic Ca2+ signaling. to selectively decrease GABA release in the lateral amygdala26. DAMGO
(1 mM) strongly decreased disynaptic inhibitory postsynaptic currents
Synaptic activation of presynaptic GABAB receptors (IPSCs) elicited by cortical afferent stimulation (15 ± 2% of baseline,
We next addressed the question of whether presynaptic GABAB n ¼ 5, P o 0.01) without significantly affecting the EPSC slope (85 ±
receptors at cortico-amygdala afferents can be activated by the intense 6%, n ¼ 5, P 4 0.05; Fig. 4a,b). Moreover, whereas tetanic stimulation
synaptic stimulation used for LTP induction. Given that cortical and of cortical afferents induced a small long-term depression under
thalamic afferents are thought to converge, at least in part, on the same control conditions (voltage clamp at –50 mV; no GABAB or GABAA
feedforward population of interneurons25, we chose to stimulate antagonists), homosynaptic LTP was induced during DAMGO-
thalamic instead of cortical afferents to avoid homosynaptic short- mediated suppression of disynaptic inhibition (control: 80 ± 14% of
term plasticity at cortical afferents. We compared the postsynaptic baseline, n ¼ 5; DAMGO: 132 ± 23%, n ¼ 8; P o 0.05, DAMGO versus
response to stimulation of cortical afferents 1 s before and 1 s after the control; Fig. 4c). As the induction of homosynaptic cortical LTP is not
tetanic stimulation (45 stimuli; 30 Hz) of thalamic afferents (Fig. 3a). sensitive to manipulations of postsynaptic activity (Fig. 2), this suggests
Under control conditions, tetanic stimulation of thalamic afferents that DAMGO indirectly facilitates homosynaptic LTP induction by
resulted in a transient decrease in the amplitude of cortical excitatory diminishing presynaptic inhibition.
postsynaptic currents (EPSCs; Fig. 3b–d). In line with a presynaptic
mechanism, we found an inverse correlation (r2 ¼ 0.596) between the GABAB(1a) receptors mediate presynaptic inhibition
change in EPSC amplitude and the change in PPR (PPR change: 17 ± Next, we sought to examine the role of presynaptic GABAB hetero-
7%, n ¼ 26, P o 0.05; Fig. 3b). Thalamic afferent stimulation had no receptors in vivo. Recent findings in cerebellum, hippocampus and
effect on cortical EPSCs in the presence of CGP55845A (control: 63 ± cortex indicate that the GABAB(1) subunit isoforms GABAB(1a) and
4% of baseline, n ¼ 54; CGP: 104 ± 5%, n ¼ 23, P o 0.001; Fig. 3c,d). GABAB(1b) are selectively associated with pre- versus postsynaptic
This indicates that repetitive firing of local GABAergic interneurons GABAB receptors27–30. To date, it is not possible to pharmacologically
transiently depresses release probability at cortical afferents via the discriminate between pre- and postsynaptic GABAB receptors; there-
activation of presynaptic GABAB receptors. fore, we took a genetic approach using mice in which the individual


Figure 4 Neuromodulator-mediated suppression of GABA release gates
a Control b DAMGO homosynaptic LTP induction. (a) The m-opioid receptor agonist DAMGO
1′ 100 (1 mM) selectively decreased disynaptic inhibition triggered by cortical

3′ afferent stimulation. Traces illustrate the progressive decrease of the
5′ 50 disynaptic IPSC (recorded at –50 mV) during DAMGO application. Scale
–45 mV
bars, 20 pA and 20 ms. (b) Summary plot illustrating the suppression of
0 disynaptic inhibition by DAMGO (n ¼ 5). (c) Application of DAMGO enabled
0 1 2 3 4 5
Time (min)
the induction of homosynaptic LTP in the absence of a GABAB receptor
antagonist (n ¼ 8). Traces (averages of four sweeps) illustrate EPSP-IPSP
c Control DAMGO sequences recorded during the baseline condition (‘a’), 2 min and 5 min into
DAMGO application, and after LTP induction (‘b’) as indicated. Inset,
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

a 2′ 5′ b
–50 mV increase in EPSP slope after LTP induction (‘b’) as compared to baseline
(‘a’). Scale bars, 3 mV and 200 ms.
EPSP slope (%)

a a marked reduction in presynaptic inhibition at thalamo-amygdala
100 afferents, the second major sensory input to the lateral amygdala1
Cort Control DAMGO
(wild-type: 87 ± 2% inhibition, n ¼ 21; GABAB(1a)–/–: 42 ± 4%, n ¼
0 10 20 30 18; GABAB(1b)–/–: 77 ± 2%, n ¼ 22; P o 0.01; Fig. 5b). In all genotypes,
Time (min) presynaptic inhibition induced by adenosine receptors, the presynaptic
receptors coupling to the same G proteins as the GABAB receptors, was
isoforms GABAB(1a) and GABAB(1b) were selectively ablated29,30. To test not affected (Fig. 5a). Furthermore, because GABAB autoreceptor–
the contribution of GABAB(1a)- versus GABAB(1b)-containing receptors mediated presynaptic inhibition at GABAergic terminals gates
to presynaptic inhibition, we assessed the effect of the GABAB receptor hippocampal LTP (ref. 31), we analyzed presynaptic inhibition at
agonist baclofen (50 mM) on EPSCs elicited by cortical afferent GABAergic terminals. Inhibition of GABAergic synaptic transmission
stimulation. In wild-type mice, baclofen reduced EPSC amplitude by in GABAB(1a)–/– and in GABAB(1b)–/– mice was not different from
86 ± 2% (n ¼ 20; Fig. 5a,b). This reduction was entirely mediated by that in wild-type mice (Supplementary Fig. 1 online). This demon-
presynaptic GABAB receptors, as baclofen equally reduced EPSC strates that GABAB(1a)–/– mice largely lack functional GABAB
amplitude (86 ± 3% inhibition, n ¼ 5, P 4 0.05 versus control) in receptors on glutamatergic cortical and thalamic afferents, but not at
the presence of 200 mM extracellular Ba2+, which was added to block GABAergic synapses.
G protein–coupled inwardly rectifying potassium (GIRK) channels, the To examine whether not only baclofen-induced presynaptic inhibi-
postsynaptic effectors of GABAB receptor–mediated inhibition23. tion but also synaptic activation of presynaptic GABAB receptors was
Whereas presynaptic inhibition was only slightly decreased in reduced in GABAB(1a)–/– mice, we measured heterosynaptic depression
GABAB(1b)-deficient (GABAB(1b)–/–) mice (75 ± 3% inhibition, induced by tetanic stimulation of thalamic afferents (Fig. 3). We found
n ¼ 16, P o 0.01; Fig. 5a,b), it was strongly reduced in GABAB(1a)–/– heterosynaptic depression to be significantly reduced in GABAB(1a)–/–
mice (27 ± 4% inhibition, n ¼ 18, P o 0.001 versus wild-type; mice when measured 1 s after tetanic stimulation (wild-type: 61 ± 6%
Fig. 5a,b). Likewise, GABAB(1a)–/–, but not GABAB(1b)–/–, mice showed of baseline, n ¼ 21; GABAB(1a)–/–: 86 ± 8%, n ¼ 13, P o 0.05; Supple-
mentary Fig. 2 online). In contrast, 15 s after
tetanic stimulation, when GABAB-mediated
heterosynaptic depression was terminated,
a –300 +/+ Bac CGP Adeno b 100 EPSC cortical ** c +/+
the EPSC ratio in GABAB(1a)–/– mice was not
Inhibition (%)

Bac CGP Adeno

–200 100 different from that in the wild-type mice (wild-
–100 ***
20 18 16
0 type: 117 ± 5% of baseline, n ¼ 13;
0 0
+/+ 1a–/– 1b–/– 0 10 20 30 GABAB(1a)–/–: 119 ± 7%, n ¼ 14; Supplemen-
EPSC amplitude (pA)

GABAB(1a)–/– tary Fig. 2). Thus, GABA released upon
Holding current (pA)

EPSC thalamic
Inhibition (%)

Bac CGP Adeno 100
–150 ** Bac CGP Adeno

–100 100 repetitive firing of local interneurons de-
50 *** creases release probability at cortical afferents
–50 0
21 18 22
0 0
+/+ 1a –/–
1b –/– 0 10 20 30 mainly via the activation of presynaptic
Post GABAB(1b) –/– GABAB(1a,2) receptors.
Bac CGP Adeno
Current (pA)

–150 75 Bac CGP Adeno In contrast to the predominant role of
–100 50 * * 100
GABAB(1a) subunits in the presynaptic inhibi-
–50 25
0 0
21 20 16 tion of excitatory synaptic transmission, post-
0 5 10 15 20 25 30 +/+ 1a–/– 1b–/– 0 10 20 30 synaptic inhibition mediated by the activation
Time (min) Time (min)
of GIRK-type K+ channels was equally reduced
Figure 5 GABAB(1a) receptors are the predominant presynaptic heteroreceptors at cortical afferents. in GABAB(1a)–/– and GABAB(1b)–/– mice (wild-
(a) Baclofen-induced presynaptic inhibition at cortical afferents (baclofen: 50 mM; CGP55845A: type: 61 ± 8 pA, n ¼ 21; GABAB(1)–/–: 2 ± 6 pA,
2 mM) was strongly reduced in GABAB(1a)-deficient mice (middle) as compared to wild-type (top) or n ¼ 5, P o 0.01; GABAB(1a)–/–: 37 ± 5 pA, n ¼
GABAB(1b)-deficient (bottom) mice. The effect of adenosine (100 mM) was similar in all genotypes. 20, P o 0.05; GABAB(1b)–/–: 34 ± 6 pA, n ¼ 16,
(b) Summary graphs illustrating the predominant contribution of GABAB(1a)-containing receptors to P o 0.05; Fig. 5b,c).
baclofen-induced presynaptic inhibition at cortical afferents (top) and thalamo-amygdala afferents
(middle). Postsynaptic K+ currents induced by application of baclofen (50 mM) were equally reduced
In addition to the functional analysis of pre-
in GABAB(1a)- and GABAB(1b)-deficient mice (bottom). (c) Individual experiments illustrating equal and postsynaptic GABAB receptor–mediated
reduction of postsynaptic inhibition in GABAB(1a)- and GABAB(1b)-deficient mice. *P o 0.05, inhibition, we examined GABAB(1) immunos-
**P o 0.01, ***P o 0.001. taining in GABAB(1a)–/– and GABAB(1b)–/– mice


GABAB(1a)–/–: 130 ± 8%, n ¼ 18, P o 0.01; GABAB(1b)–/–: 90 ± 11%,
a n ¼ 13, P 4 0.05; Fig. 7a,b). LTP in GABAB(1a)–/– mice was not
significantly different from that in GABAB(1)–/– mice (Fig. 7b) and was
not further enhanced by CGP55845A (139 ± 15% of baseline, n ¼ 4,
P 4 0.05). In line with the results using pharmacological GABAB
receptor blockade, cortico-amygdala LTP was associated with a
decrease in PPR (wild-type/CGP55845A: –23 ± 4% PPR change,
n ¼ 10; GABAB(1)–/–: –25 ± 6%, n ¼ 5, P 4 0.05 versus wild-type;
GABAB(1a)–/–: –13 ± 3%, n ¼ 18, P o 0.01 versus wild-type;
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

GABAB(1b)–/–: 2 ± 6%, n ¼ 13, P 4 0.05 versus wild-type; Fig. 7c).
This demonstrates that blockade of GABAB(1a)-containing receptors
b underlies the pharmacological facilitation of homosynaptic LTP.
Given the reduction in presynaptic inhibition at thalamic afferents in
GABAB(1a)–/– mice, we examined whether LTP at this input was also
facilitated. Pharmacological GABAB receptor blockade facilitated LTP
induction in C57BL/6J mice (control: 104 ± 12% of baseline, n ¼ 8;
CGP55845A: 162 ± 19%, n ¼ 11, P o 0.05; Supplementary Fig. 3
online) and in BALB/c wild-type mice (control: 101 ± 5%, n ¼ 8;
CGP55845A: 154 ± 21%, n ¼ 11, P o 0.05). Consistent with previous
reports5,6, induction of thalamic LTP required blockade of postsynaptic
GABAA receptors, NMDA receptor activation and postsynaptic Ca2+
Figure 6 Ultrastructual localization of GABAB(1a) and GABAB(1b) in the lateral influx, and was not associated with a PPR change (Supplementary
amygdala. (a,b) Immunoperoxidase reaction products for GABAB(1b) were often Fig. 3). Consistent with the finding that GABAB(1b) is the predominant
found in dendritic spines (arrows) in GABAB(1a)-deficient mice (a), whereas GABAB receptor on postsynaptic spines, the impact of GABAB receptor
those for GABAB(1a) were mainly found in axons (double arrows) and axon
blockade on thalamic LTP induction seemed to be entirely mediated
terminals (arrowheads) in GABAB(1b)-deficient mice (b). Scale bar, 500 nm.
by GABAB(1b)-containing receptors. Whereas GABAB(1)–/– and
GABAB(1b)–/– mice showed clear LTP (with no change in PPR) in the
using electron microscopy32. Pre-embedding peroxidase-labeling absence of CGP55845A (GABAB(1)–/–: 141 ± 5%, n ¼ 6, P o 0.05;
experiments in the lateral amygdala of GABAB(1b)–/– mice confirmed GABAB(1b)–/–: 148 ± 10%, n ¼ 7, P o 0.01), no LTP was induced in
that GABAB(1a) is predominantly localized to glutamatergic terminals GABAB(1a)–/– mice (104 ± 2%, n ¼ 5, P 4 0.05; Fig. 7d–f). This was
(Fig. 6). In contrast, analysis of GABAB(1a)–/– mice revealed that not due to a saturation effect, because LTP could still be induced by a
GABAB(1b) is mainly localized to postsynaptic spines (Fig. 6), suggest- different (pairing) protocol (175 ± 26%, n ¼ 5, P o 0.05).
ing differential postsynaptic compartmentalization of GABAB iso- In conclusion, by using GABAB(1a)–/– and GABAB(1b)–/– mice, we
forms, which cannot be detected by the bath application of baclofen. were able to specifically dissociate pre- and postsynaptic GABAB
Taken together, these results show that GABAB(1a) is the predominant
isoform mediating presynaptic inhibition at glutamatergic afferents to
the lateral amygdala, and that GABAB(1a)–/– mice are useful as a tool to a Cortical d Thalamic
–/– –/– –/– –/–
test whether the pharmacological facilitation of homosynaptic LTP is GABAB(1a) GABAB(1b) GABAB(1a) GABAB(1b)
Post Post
due to the blockade of presynaptic heteroreceptors on cortical afferents.
EPSP slope (%)
EPSP slope (%)

200 200
Facilitation of homosynaptic LTP in GABAB(1a)–/– mice
Because GABAB(1a)–/– and GABAB(1b)–/– mice are maintained on a 100 100
–/– –/– –/– –/–
BALB/c background, we first verified our pharmacological findings in GABAB(1a) GABAB(1b) GABAB(1a) GABAB(1b)
0 0
wild-type BALB/c mice. As in C57BL/6J mice, CGP55845A facilitated 0 10 20 30 0 10 20 30
the induction of homosynaptic LTP in BALB/c mice (control: 105 ± 4% Time (min) Time (min)

of baseline, n ¼ 18, P 4 0.05; CGP55845A: 150 ± 15%, n ¼ 10, P o b 200 +/+ –/– e 200 +/+ –/–
0.05). In the absence of CGP55845A, induction of homosynaptic LTP *
EPSP slope (%)

EPSP slope (%)

* * **
was possible in GABAB(1)–/– and GABAB(1a)–/– mice, but not in ** *
GABAB(1b)–/– mice (GABAB(1)–/–: 153 ± 17%, n ¼ 5, P o 0.05; 100 100

18 10 5 18 13 8 11 6 5 7
Figure 7 GABAB(1a)-deficient mice exhibit facilitated homosynaptic LTP 0 0
A )

A )
A )

A )

AB 1a


AB 1a


AB B(1

AB (1


induction at cortical afferents. (a) In the absence of CGP, LTP induction at


G B(


G B(





cortical afferents was possible in GABAB(1a)-deficient mice (n ¼ 18), but not

in GABAB(1b)-deficient mice (n ¼ 13). (b) Quantification of LTP at cortico- c +/+ –/– f +/+ –/–
amygdala afferents. In the presence of CGP55845A, LTP was facilitated in 15 15
PPR (% change)

PPR (% change)

BALB/c wild-type mice. In its absence, LTP could be induced in GABAB(1)–/–
0 * ** ** 0
and GABAB(1a)–/– mice, but not in GABAB(1b)–/– mice. Number of experiments
are indicated on bars. (c) Corresponding changes in PPR. (d–f) Same as a–c, –15 –15
for thalamo-amygdala afferents. In the absence of CGP55845A, LTP could be –30 –30
induced in GABAB(1)–/– and GABAB(1b)–/–, but not GABAB(1a)–/–, mice.
A 1)

AB B(1)
A )

A )

AB (1a


AB 1a






G B(


Thalamic LTP was not associated with any PPR changes. Scale bars, 4 mV





and 20 ms. *P o 0.05, **P o 0.01.




a Figure 8 Generalization of conditioned fear in GABAB(1a)–/– mice. (a) The
US 0.9 mA n.s.
80 freezing behavior of GABAB(1a)–/– mice in response to CS+ presentations was
Time spent freezing (%)
not different from that of wild-type mice (24 h after conditioning with a
60 –/–
GABAB(1a) 0.9-mA footshock), but a complete absence of stimulus discrimination as
evidenced by the fact that the time spent freezing during CS– presentations
(n ¼ 6 for each genotype) was of the same duration as that during CS+
20 presentations. (b) At a lower US intensity (0.6 mA), both wild-type and
GABAB(1a)–/– mice were able to discriminate between the CS– and the CS+
0 (n ¼ 6 for each genotype). (c) GABAB(1a)–/– mice had no difference in their
No tone CS– CS+ thresholds for footshock-induced movements or vocalization (n ¼ 10 for each
genotype). *P o 0.05.
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

US 0.6 mA
Time spent freezing (%)

60 WT * when US intensity was reduced to 0.6 mA (n ¼ 5, P o 0.05; Fig. 8b).
GABAB(1a) * Wild-type mice also demonstrated generalization if US intensity was
increased by 50%, to 1.35 mA (time spent freezing during CS–: 78 ±
20 11%; during CS+: 81 ± 14%; n ¼ 5; P 4 0.05), indicating that in
GABAB(1a)–/– mice, there is a shift in the threshold for fear general-
No tone CS– CS+ ization. Although these experiments rule out the explanation that fear
generalization in GABAB(1a)–/– mice is caused by a general hearing
c Flinching Vocalization
deficit or by a more specific deficit in frequency discrimination, the
1.0 1.0 possibility remains that GABAB(1a)–/– mice have an increased US
Cummulative probability

sensitivity. We therefore compared the threshold for US-induced
movements and vocalizations in wild-type and GABAB(1a)–/– mice.
0.5 0.5 GABAB(1a)–/– mice showed no difference in the threshold for the US to
induce rapid movements or vocalization (n ¼ 4, P 4 0.05; Fig. 8c),
GABAB(1a)–/– indicating that fear generalization in GABAB(1a)–/– mice is unlikely to
0.0 0.0 be caused by alterations in CS or US perception. Thus, in parallel to the
0.0 0.4 0.8 0.0 0.4 0.8
Current (pA) Current (pA) loss of associativity of cortical afferent LTP, GABAB(1a)–/– mice show a
US intensity–dependent deficit in associative stimulus discrimination
at the behavioral level.
receptor function at cortical and thalamic afferents to the lateral
amygdala. Moreover, our experiments demonstrate that presynaptic DISCUSSION
GABAB(1a)-containing heteroreceptors are critical in determining the In this study, we show that presynaptic GABAB receptor–mediated
associative properties of cortical afferent LTP. inhibition sets the balance between two forms of LTP at cortical
afferents to the lateral amygdala: associative, NMDA receptor–depen-
Generalization of conditioned fear in GABAB(1a)–/– mice dent LTP, and nonassociative, NMDA receptor–independent LTP. In
To assess the behavioral impact of impaired presynaptic inhibition, we the presence of presynaptic inhibition, concomitant thalamic afferent
subjected GABAB(1a)–/– mice to an auditory fear conditioning protocol. activity and activation of presynaptic NMDA receptors on cortical
Given that previous studies have implicated the cortico-amygdala afferents are needed for LTP induction8. This raises the question of why
pathway in stimulus discrimination and in the generalization of tetanic stimulation of the cortical afferents alone does not induce LTP.
conditioned fear33 (but see ref. 34), mice were trained using a differ- One factor seems to be glutamate uptake, as when glutamate uptake is
ential fear conditioning protocol: only one of two conditioned stimuli pharmacologically inhibited, cortical afferent stimulation alone is
(CS+; 7.5 kHz, 30 s, 80 dB) was paired to the unconditioned stimulus sufficient for LTP induction8. An alternative way to overcome gluta-
(US; 0.9 mA, 1.5 s); the second one (CS–; 3 kHz, 30 s, 80 dB), however, mate uptake, and thus induce homosynaptic LTP, might be to increase
was not. Training consisted of seven CS+-US pairings with seven stimulation frequency during tetanic stimulation. However, doubling
interleaved CS– presentations. When tested 24 h later, wild-type mice stimulation frequency from 30 Hz to 60 Hz did not result in homo-
discriminated between the CS+ and the CS– (n ¼ 6, P o 0.05; Fig. 8a), synaptic LTP induction. Only at a frequency of 90 Hz did we observe a
whereas GABAB(1a)–/– mice demonstrated equal freezing behavior small, but nonsignificant, enhancement of synaptic transmission (P ¼
during both CS+ and CS– presentation (n ¼ 6, P 4 0.05; Fig. 8a). It 0.08; Supplementary Fig. 5 online). In contrast, in the presence of a
is unlikely that the generalization of conditioned fear in GABAB(1a)–/– GABAB receptor antagonist, 30-Hz stimulation was sufficient for the
mice was caused by increased stress or anxiety, because they did not induction of homosynaptic LTP. Like heterosynaptic LTP, homosynap-
freeze more than wild-type mice when exposed for the first time to a tic LTP is independent of postsynaptic activity and Ca2+ influx, and is
new context (n ¼ 6, P 4 0.05; Fig. 8a). Given the differential facilitation associated with a decrease in PPR. Moreover, the two forms of LTP
of cortical and thalamic LTP in GABAB(1a)–/– and GABAB(1b)–/– mice, we occlude each other, suggesting that their presynaptic expression
also examined fear behavior in GABAB(1b)–/– mice. Unfortunately, fear mechanisms are inter-related. The fact that heterosynaptic LTP can
conditioning was completely impaired in GABAB(1b)–/– mice, thereby be induced in the presence of functional presynaptic inhibition suggests
precluding an analysis of generalization (Supplementary Fig. 4 online). that presynaptic GABAB receptors act in parallel to presynaptic NMDA
Nonassociative sensitization and generalization of conditioned fear receptors, for instance by a direct inhibition of presynaptic Ca2+
depends on US intensity35,36. We therefore tested whether GABAB(1a)–/– channels or by the activation of K+ channels20, and that NMDA
mice were able to discriminate between CS+ and CS– if the CS+ was receptor–mediated presynaptic Ca2+ signaling is sufficient to activate
conditioned to a US of lower intensity. Notably, we found that possible downstream targets such as Ca2+/calmodulin-dependent
GABAB(1a)–/– mice were able to discriminate between CS+ and CS– adenylyl cyclases37.


We found that the extent of GABAB receptor–mediated presynaptic (that is, GABAB(1a)-deficient mice)—LTP at cortical afferents can be
inhibition, and hence the need for concomitant thalamic afferent induced independent of thalamic afferent activity, thereby ‘general-
activity for LTP induction, is critically determined by the activity of izing’ to inputs that would normally not be potentiated. Consistent
local interneurons. Reducing GABA release by activation of m-opioid with our conditioning protocol in which mice were exposed to both
receptors on GABAergic terminals enabled the induction of homo- CS+ and CS–, the unmasking of nonassociative cortico-amygdala LTP
synaptic LTP in the absence of a GABAB antagonist. Thus, cortical would still depend on US-mediated suppression of GABA release. In
afferent boutons are endowed with presynaptic ionotropic (NMDA) GABAB(1a)–/– mice, the threshold for the induction of nonassociative
and metabotropic (GABAB) receptor complements, allowing the detec- LTP was lowered. Lack of presynaptic inhibition in GABAB(1a)–/– mice
tion and integration of concomitant excitatory and inhibitory activity could thus contribute to a shift in the threshold for the generalization of
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

in a heterosynaptic manner. Similar properties have been reported for conditioned fear toward lower US intensities.
other types of synapses that show presynaptic long-term plasticity, such In conclusion, the modulation of associative properties of cortico-
as hippocampal mossy fiber synapses38–40 or parallel fiber–Purkinje cell amygdala LTP by presynaptic GABAB receptor–mediated inhibition
synapses in the cerebellum16, suggesting that heterosynaptic modula- suggests an attractive mechanism by which short- or long-term changes
tion of plasticity thresholds, a form of metaplasticity, may be a general in inhibitory transmission44–47, or distinct patterns of network activ-
feature of the presynaptic plasticity underlying many forms of inte- ity48, may modulate the degree of fear generalization appropriate for
grative computation. specific behavioral demands under certain conditions.
The differential contribution of GABAB(1a)- and GABAB(1b)-contain-
ing receptors to pre- and postsynaptic inhibition, and the finding that METHODS
GABAB autoreceptor–mediated inhibition is not affected in Slice preparation. Coronal slices (350 mm) were prepared from 3- to 4-week-
GABAB(1a)–/– or GABAB(1b)–/– mice, allowed us to specifically address old male C57BL/6J or BALB/c mice following a protocol approved by the
the role of presynaptic heteroreceptors in vivo. We found that in the Veterinary Department of the Canton of Basel-Stadt. Slices were maintained for
absence of a GABAB antagonist, homosynaptic LTP could be induced in 45 min at 35 1C in an interface chamber containing artificial cerebrospinal fluid
GABAB(1a)–/– mice, but not in GABAB(1b)–/– mice. Cortical LTP in equilibrated with 95% O2 and 5% CO2, and containing 124 mM NaCl, 2.7 mM
GABAB(1a)–/– mice could not be further enhanced by a GABAB KCl, 2 mM CaCl2, 1.3 mM MgCl2, 26 mM NaHCO3, 0.4 mM NaH2PO4,
18 mM glucose and 4 mM ascorbate. They were then maintained for at least
antagonist, suggesting that GABAB(1a)-containing receptors are the
45 min at room temperature (21–23 1C) before being transferred to a super-
key heteroreceptors controlling presynaptic LTP induction at cortico- fusing recording chamber.
amygdala afferents.
In contrast to cortical LTP, LTP at thalamic afferents invariably Electrophysiology. Whole-cell recordings were performed at 30–32 1C.
depended on postsynaptic NMDA receptor activation and Ca2+ influx, Neurons were visually identified with infrared videomicroscopy using an
and is sensitive to postsynaptic GABAA receptor–mediated inhibition. upright microscope equipped with a 40 objective (Olympus). Patch electro-
Whereas our functional analysis revealed an equal contribution of des (3–5 MO) were pulled from borosilicate glass tubing and filled with a
GABAB(1a)- and GABAB(1b)-containing receptors to postsynaptic inhi- solution containing 120 mM potassium gluconate, 20 mM KCl, 10 mM HEPES
bition during bath application of a GABAB receptor agonist, electron buffer, 10 mM phosphocreatine, 4 mM Mg-ATP and 0.3 mM Na-GTP (pH ¼
microscopy indicates that GABAB(1b) is the predominant receptor on 7.25; 295 mOsm). For voltage-clamp experiments, potassium gluconate was
replaced by equimolar cesium gluconate. All experiments were performed in
postsynaptic spines, the locus of LTP induction at thalamic afferents6.
the presence of picrotoxin (100 mM) unless indicated otherwise. In current-
Accordingly, we found that thalamic LTP is selectively facilitated in
clamp recordings, membrane potential was kept manually at –70 mV (not
GABAB(1b)–/– mice but not in GABAB(1a)–/– mice, suggesting that in the corrected for junction potentials). Data were recorded with an Axopatch200B
absence of GABAB(1a), local postsynaptic GABAB(1b)-mediated inhibi- (Axon Instruments), filtered at 2 kHz and digitized at 10 kHz. Monosynaptic
tion on spines is sufficient to control the induction of thalamic LTP. EPSPs or EPSCs were elicited by stimulation of afferent fibers with a bipolar
Thus, at least with the LTP induction protocol we used, GABAB(1a)–/– twisted platinum/10% iridium wire (25 mm). LTP was quantified for statistical
mice showed a pathway-specific facilitation of nonassociative LTP at comparisons by averaging and normalizing EPSP slopes during the last 5 min
cortical afferents to the lateral amygdala. of experiments relative to baseline. All values are expressed as means ± s.e.m.
At the behavioral level, GABAB(1a)–/– mice showed a generalization of Statistical comparisons were done with paired or unpaired Student’s t-test as
conditioned fear to nonconditioned stimuli. The fact that GABAB(1a)- appropriate (two-tailed P o 0.05 was considered significant).
deficient mice were able to discriminate between the CS+ and the CS–
Pre-embedding immunoelectron microscopy. Immunohistochemical labeling
after training with a lower-intensity US, and that they were equally
for electron microscopy was performed as described previously32. Mice were
sensitive to the US, makes it unlikely that the observed behavioral anaesthetized with pentobarbital (100 mg per kg body weight, intraperitone-
phenotype was caused by a sensory deficit. Previous behavioral studies ally) and perfused transcardially with 4% paraformaldehyde, 15% saturated
have implicated auditory cortex in stimulus discrimination and the picric acid and 0.05% glutaraldehyde made up in 0.1 M phosphate buffer.
generalization of conditioned fear33,41 (but see ref. 34). It is tempting to Coronal sections (50 mm) were cut with a microslicer (Leica) and were freeze-
speculate that the loss of associativity of cortico-amygdala LTP thawed in liquid nitrogen. After washing in 50 mM Tris-buffered saline (TBS,
might contribute to behavioral generalization. However, our data do pH 7.4) and blocking in TBS containing 20% normal goat serum (NGS, Vector
not allow us to causally link the two phenomena, and GABAB(1a)- Laboratories), sections were incubated with 1.0 mg ml–1 primary antibody
containing receptors may be involved in stimulus discrimination and (B17; ref. 32) to GABAB(1) in TBS containing 2% NGS. After washing in
discriminative learning in brain areas providing input to the amygdala, TBS, sections were incubated with biotinylated goat anti-rabbit IgG antibody
(1:100, Vector Laboratories), reacted with avidin-biotin peroxidase complex
such as the auditory cortex or the auditory thalamus42. Alternatively,
(ABC kit, 1:100, Vector Laboratories), and incubated with 0.025% 3,3¢-
our data are consistent with a model in which associativity of cortico- diaminobenzidine tetrahydrochloride (Sigma) and 0.003% hydrogen
amygdala LTP would be important to preserve stimulus discrimination peroxide. After treatment with OsO4, sections were stained with uranyl acetate,
during discriminative fear conditioning43. Accordingly, if GABAB dehydrated and flat-embedded in epoxy resin (Durcupan, ACM, Fluka).
receptor–mediated presynaptic control is decreased—either by the Ultrathin sections were prepared (Ultracut S, Leica) and examined with a
suppression of GABA release or by the selective ablation of GABAB(1a) 208S electron microscope (Phillips).


Behavior. Adult (2–3 months) age-matched male wild-type and mutant BALB/ 17. Porter, J.T. & Nieves, D. Presynaptic GABAB receptors modulate thalamic excitation
c mice were individually housed in plastic cages, with ad lib access to food and of inhibitory and excitatory neurons in the mouse barrel cortex. J. Neurophysiol. 92,
2762–2770 (2004).
water and a 12-h light/12-h dark cycle. All studies took place during the light 18. Asprodini, E.K., Rainnie, D.G. & Shinnick-Gallagher, P. Epileptogenesis reduces
portion of the cycle. Mice were handled gently for 2–3 min per day for 5 d. Fear the sensitivity of presynaptic gamma-aminobutyric acidB receptors on glutamatergic
conditioning and testing took place in two different contexts. The conditioning afferents in the amygdala. J. Pharmacol. Exp. Ther. 262, 1011–1021 (1992).
and testing boxes and the floor were cleaned with 70% ethanol or 1% acetic 19. Yamada, J., Saitow, F., Satake, S., Kiyohara, T. & Konishi, S. GABA(B) receptor-
mediated presynaptic inhibition of glutamatergic and GABA-ergic transmission in the
acid, before and after each session, respectively. To score freezing behavior, we basolateral amygdala. Neuropharmacology 38, 1743–1753 (1999).
used an automatic infrared beam detection system placed on the bottom of the 20. Bettler, B., Kaupmann, K., Mosbacher, J. & Gassmann, M. Molecular structure
experimental chambers (Coulbourn). The mice were considered to be freezing and physiological functions of GABA(B) receptors. Physiol. Rev. 84, 835–867(2004).
if no movement was detected for 2 s. Mice were submitted to a discriminative 21. Fritschy, J.M. et al. GABAB-receptor splice variants GB1a and GB1b in rat brain:
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

developmental regulation, cellular distribution and extrasynaptic localization. Eur. J.
fear conditioning protocol in which a 30-s tone CS+ (7.5 kHz, 80 dB) was Neurosci. 11, 761–768 (1999).
systematically paired with a 1-s US (0.6 mA or 0.9 mA footshock; seven CS+/US 22. McDonald, A.J., Mascagni, F. & Muller, J.F. Immunocytochemical localization of
pairings; inter-trial interval 20–180 s). The onset of the US coincided with the GABABR1 receptor subunits in the basolateral amygdala. Brain Res. 1018, 147–158
offset of the CS+. CS+/US pairings were interleaved with the presentation of a (2004).
23. Lüscher, C., Jan, L.Y., Stoffel, M., Malenka, R.C. & Nicoll, R.A. G protein-coupled
nonconditioned CS– tone (30 s, 3 kHz, 80 dB). Mice were tested 24 h later by inwardly rectifying K+ channels (GIRKs) mediate postsynaptic but not presynaptic
presentations of four CS– and four CS+. To examine pain sensitivity, mice were transmitter actions in hippocampal neurons. Neuron 19, 687–695 (1997).
submitted to a series of 1-s-long footshocks of increasing strength (0.1– 24. Huang, Y.Y. & Kandel, E.R. Postsynaptic induction and PKA-dependent expression of
0.8 mA). Threshold was determined by scoring the first noticeable flinching LTP in the lateral amygdala. Neuron 21, 169–178 (1998).
25. Szinyei, C., Heinbockel, T., Montagne, J. & Pape, H.C. Putative cortical and thalamic
behavior and stress-induced vocalization.
inputs elicit convergent excitation in a population of GABAergic interneurons of the
lateral amygdala. J. Neurosci. 20, 8909–8915 (2000).
Reagents. BAPTA, BAPTA-AM, CPP and NBQX were from Tocris-Cookson. 26. Sugita, S. & North, R.A. Opioid actions on neurons of rat lateral amygdala in vitro. Brain
CGP55845A was a gift from Novartis Pharma AG. All other drugs were from Res. 612, 151–155 (1993).
Fluka/Sigma. 27. Billinton, A., Upton, N. & Bowery, N.G. GABA(B) receptor isoforms GBR1a and GBR1b,
appear to be associated with pre- and postsynaptic elements respectively in rat and
Note: Supplementary information is available on the Nature Neuroscience website. human cerebellum. Br. J. Pharmacol. 126, 1387–1392 (1999).
28. Benke, D., Honer, M., Michel, C., Bettler, B. & Mohler, H. Gamma-aminobutyric acid
type B receptor splice variant proteins GBR1a and GBR1b are both associated with
GBR2 in situ and display differential regional and subcellular distribution. J. Biol.
We thank J. Cryan, B. Gähwiler, K. Vogt and all members of the Lüthi lab for Chem. 274, 27323–27330 (1999).
helpful discussions and comments on the manuscript. This work was supported 29. Vigot, R. et al. Different compartmentalization and distinct functions of GABAB receptor
by the Swiss Science Foundation, the Centre National de la Recherche variants. Neuron 50, 589–601 (2006).
Scientifique (CNRS), and the Novartis Research Foundation. 30. Perez-Garci, E., Gassmann, M., Bettler, B. & Larkum, M.E. The GABA(B1b) isoform
mediates long-lasting inhibition of dendritic Ca2+ spikes in layer 5 somatosensory
COMPETING INTERESTS STATEMENT pyramidal neurons. Neuron 50, 603–616 (2006).
The authors declare that they have no competing financial interests. 31. Davies, C.H., Starkey, S.J., Pozza, M.F. & Collingridge, G.L. GABA autoreceptors
regulate the induction of LTP. Nature 349, 609–611 (1991).
32. Kulik, A. et al. Distinct localization of GABAB receptors relative to synaptic sites
Published online at http://www.nature.com/natureneuroscience
in the rat cerebellum and ventrobasal thalamus. Eur. J. Neurosci. 15, 291–307 (2002).
Reprints and permissions information is available online at http://npg.nature.com/ 33. Jarrell, T.W., Gentile, C.G., Romanski, L.M., McCabe, P.M. & Schneiderman, N.
reprintsandpermissions/ Involvement of cortical and thalamic auditory regions in retention of differential
bradycardia conditioning to acoustic conditioned stimuli in rabbits. Brain Res. 412,
285–294 (1987).
1. LeDoux, J.E. Emotion circuits in the brain. Annu. Rev. Neurosci. 23, 155–184 (2000). 34. Armony, J.L., Servan-Schreiber, D., Romanski, L.M., Cohen, J.D. & LeDoux, J.E.
2. Maren, S. & Quirk, G.J. Neuronal signalling of fear memory. Nat. Rev. Neurosci. 5, Stimulus generalization of fear responses: effects of auditory cortex lesions in a
844–852 (2004). computational model and in rats. Cereb. Cortex 7, 157–165 (1997).
3. Rumpel, S., LeDoux, J.E., Zador, A. & Malinow, R. Postsynaptic receptor trafficking 35. Laxmi, T.R., Stork, O. & Pape, H.C. Generalisation of conditioned fear and its
underlying a form of associative learning. Science 308, 83–88 (2005). behavioural expression in mice. Behav. Brain Res. 145, 89–98 (2003).
4. Weisskopf, M.G., Bauer, E.P. & LeDoux, J.E. L-type voltage-gated calcium channels 36. Baldi, E., Lorenzini, C.A. & Bucherelli, C. Footshock intensity and generalization in
mediate NMDA-independent associative long-term potentiation at thalamic input contextual and auditory-cued fear conditioning in the rat. Neurobiol. Learn. Mem. 81,
synapses to the amygdala. J. Neurosci. 19, 10512–10519 (1999). 162–166 (2004).
5. Bissière, S., Humeau, Y. & Lüthi, A. Dopamine gates LTP induction in lateral amygdala 37. Xia, Z. & Storm, D.R. The role of calmodulin as a signal integrator for synaptic plasticity.
by suppressing feedforward inhibition. Nat. Neurosci. 6, 587–592 (2003). Nat. Rev. Neurosci. 6, 267–276 (2005).
6. Humeau, Y. et al. Dendritic spine heterogeneity determines afferent-specific Hebbian 38. Nicoll, R.A. & Malenka, R.C. Contrasting properties of two forms of long-term potentia-
plasticity in the amygdala. Neuron 45, 119–131 (2005). tion in the hippocampus. Nature 377, 115–118 (1995).
7. McKernan, M.G. & Shinnick-Gallagher, P. Fear conditioning induces a lasting potentia- 39. Lauri, S.E. et al. A critical role of a facilitatory presynaptic kainate receptor in mossy fiber
tion of synaptic currents in vitro. Nature 390, 607–611 (1997). LTP. Neuron 32, 697–709 (2001).
8. Humeau, Y., Shaban, H., Bissière, S. & Lüthi, A. Presynaptic induction of hetero- 40. Schmitz, D., Mellor, J., Breustedt, J. & Nicoll, R.A. Presynaptic kainate receptors impart
synaptic associative plasticity in the mammalian brain. Nature 426, 841–845 (2003). an associative property to hippocampal mossy fiber long-term potentiation. Nat.
9. Sugita, S., Tanaka, E. & North, R.A. Membrane properties and synaptic potentials of Neurosci. 6, 1058–1063 (2003).
three types of neurones in rat lateral amygdala. J. Physiol. (Lond.) 460, 705–718 41. Thompson, R.F. The role of the cerebral cortex in stimulus generalization. J. Comp.
(1993). Physiol. Psychol. 55, 279–287 (1962).
10. Li, X.F., Armony, J.L. & LeDoux, J.E. GABAA and GABAB receptors differentially regulate 42. Weinberger, N.M. Specific long-term memory traces in primary auditory cortex. Nat. Rev.
synaptic transmission in the auditory thalamo-amygdala pathway: an in vivo micro- Neurosci. 5, 279–290 (2004).
iontophoretic study and a model. Synapse 24, 115–124 (1996). 43. Collins, D.R. & Paré, D. Differential fear conditioning induces reciprocal changes in the
11. Lang, E.J. & Paré, D. Synaptic and synaptically activated intrinsic conductances sensory responses of lateral amygdala neurons to the CS(+) and CS(-). Learn. Mem. 7,
underlie inhibitory potentials in cat lateral amygdaloid projection neurons in vivo. 97–103 (2000).
J. Neurophysiol. 77, 353–363 (1997). 44. Mahanty, N.K. & Sah, P. Calcium-permeable AMPA receptors mediate long-term
12. Kullmann, D.M. et al. Presynaptic, extrasynaptic and axonal GABAA receptors in the potentiation in interneurons in the amygdala. Nature 394, 683–687 (1998).
CNS: where and why? Prog. Biophys. Mol. Biol. 87, 33–46 (2005). 45. Marsicano, G. et al. The endogenous cannabinoid system controls extinction of aversive
13. Thompson, S.M., Capogna, M. & Scanziani, M. Presynaptic inhibition in the hippocam- memories. Nature 418, 530–534 (2002).
pus. Trends Neurosci. 16, 222–227 (1993). 46. Bauer, E.P. & LeDoux, J.E. Heterosynaptic long-term potentiation of inhibitory inter-
14. Cryan, J.F. & Kaupmann, K. Don’t worry ‘B’ happy!: a role for GABA(B) receptors in neurons in the lateral amygdala. J. Neurosci. 24, 9507–9512 (2004).
anxiety and depression. Trends Pharmacol. Sci. 26, 36–43 (2005). 47. Zhu, P.J. & Lovinger, D.M. Retrograde endocannabinoid signaling in a postsynaptic
15. Isaacson, J.S., Solis, J.M. & Nicoll, R.A. Local and diffuse synaptic actions of GABA in neuron/synaptic bouton preparation from basolateral amygdala. J. Neurosci. 25, 6199–
the hippocampus. Neuron 10, 165–175 (1993). 6207 (2005).
16. Dittman, J.S. & Regehr, W.G. Mechanism and kinetics of heterosynaptic depression at a 48. Pelletier, J.G. & Paré, D. Role of amygdala oscillations in the consolidation of emotional
cerebellar synapse. J. Neurosci. 17, 9048–9059 (1997). memories. Biol. Psychiatry 55, 559–562 (2004).


G(o) signaling is required for Drosophila associative
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

Jacob Ferris1, Hong Ge1, Lingzhi Liu1 & Gregg Roman1,2

Heterotrimeric G(o) is one of the most abundant proteins in the brain, yet relatively little is known of its neural functions in vivo.
Here we demonstrate that G(o) signaling is required for the formation of associative memory. In Drosophila melanogaster,
pertussis toxin (PTX) is a selective inhibitor of G(o) signaling. The postdevelopmental expression of PTX within mushroom body
neurons robustly and reversibly inhibits associative learning. The effect of G(o) inhibition is distributed in both c- and a/b-lobe
mushroom body neurons. However, the expression of PTX in neurons adjacent to the mushroom bodies does not affect memory.
PTX expression also does not interact genetically with a rutabaga adenylyl cyclase loss-of-function mutation. Thus, G(o) defines
a new signaling pathway required in mushroom body neurons for the formation of associative memory.

An associative memory is one that links external stimuli to particular PTX recognition site, whereas G(o)a does; PTX will ADP-ribosylate
events, such that the stimuli come to predict the events. In the a single protein in Drosophila, as seen in western blots and after
negatively reinforced olfactory associative learning assay of Drosophila isoelectric focusing; and PTX comigrates with G(o)a and is immuno-
melanogaster, flies are presented with an odor (conditional stimulus precipitated by independent G(o)a-specific antibodies15,16,18,19.
paired, CS+) paired with an electric shock (unconditioned stimulus,
US). The flies are then presented with a second odor (conditioned RESULTS
stimulus unpaired, CS–). The associative memory is measured as the PTX inhibits the physiology of memory formation
conditioned avoidance of the CS+ in a T-maze1. The disruption of the To determine if G(o) is a mediator of associative memory, we expressed
cyclic AMP (cAMP) signaling pathway within Drosophila leads to a PtxA transgene within the mushroom body neurons. We selected the
reduced learning scores2–10. The effect of cAMP disruption has been P{UASPTX}16 transgenic line because the basal expression of PTX is
mapped back to the mushroom body neurons through the targeted low in this line and because PTX can be induced by Gal4, albeit in small
expression of a constitutively active G(s)a and by rescuing the rutabaga amounts (Fig. 1a). G(o)a47A loss-of-function mutant embryos die
type I adenylyl cyclase (rut) phenotype with targeted expression of a rut during embryogenesis owing to defects in nervous system and meso-
cDNA7–10. It is thought that the cAMP pathway controls the association derm development14. In keeping with this result, we found that the
between the CS+ and the US within the mushroom body neurons2. induction of PTX within the developing mesoderm or nervous system
The G(o) heterotrimeric protein is thought to be the most abundant also resulted in embryonic lethality, indicating that this toxin is
membrane protein in the vertebrate brain and is activated both by functional when expressed early in development (Supplementary
numerous G protein–coupled receptors (GPCRs) and by amyloid Table 1 online).
precursor protein11. Although G(o) can participate in diverse signaling We examined the role of G(o) in associative memory by inducing
pathways, only a few specific in vivo functions have been ascribed to this PTX expression within the adult mushroom bodies with the
molecule12–14. In Drosophila, G(o)a47A is the only gene encoding the P{MBSwitch}12 Gene-Switch driver10. The resulting induction abol-
alpha subunit of G(o), and it is expressed throughout the adult ished the immediate associative memory 3 min after training, which is
brain14,15. The G(o) protein is much more abundant in the heads of frequently taken as a measure of learning (Fig. 1b; P o 0.0001,
rutabaga (rut) and dunce learning mutants than in the heads of wild- Bonferroni-Dunn). Although the PTX-uninduced P{MBSwitch}12/
type flies, suggesting a possible role for G(o) in memory formation16. P{UASPTX}16 flies also showed reduced learning, their scores were
The S1 subunit of PTX from Bordetella pertussis catalyzes the transfer not significantly lower than the PTX-uninduced P{UASPTX}16/+ con-
of an ADP-ribose onto the Ga subunit of the vertebrate G(i/o/t) trol group (P ¼ 0.015; significance with the Bonferroni-Dunn post-hoc
heterotrimeric G proteins, preventing these proteins from binding to test requires P o 0.0033). The induction of PTX within the mushroom
activated GPCRs (ref. 17). In Drosophila melanogaster, PTX is a selective body did not alter naı̈ve sensitivities to either odorants or electric shock,
enzymatic inhibitor of G(o) signaling: Drosophila does not have a indicating that PTX expressed in the mushroom bodies does not affect
transducin homolog, and the G(i)a65A protein does not contain the the perception of the stimuli (Supplementary Table 2 online).

1Department of Molecular and Cellular Biology, Baylor College of Medicine, One Baylor Plaza, Houston, Texas 77030, USA. 2Present address: Department of Biology and

Biochemistry, University of Houston, Houston, Texas 77204, USA Correspondence should be addressed to G.R. (gwroman@UH.edu).
Received 8 May; accepted 14 June; published online 16 July 2006; doi:10.1038/nn1738


a 10 ng
hsG4 PTX
hsG4 PTX
b 0.8 + RU486
c Induced
– RU486 0.8 Uninduced
* 0.7
0.6 *


0.3 0.3
0.2 0.2
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

0.1 0.1

0 0
Gal80ts20/+; Gal80ts20/+; Gal80ts20/+;
p{MBSwitch} PTX/+ p{MBSwitch}
PTX/P247, PTX/Gal80ts2 P247,
30 min 37 °C No HS 12/PTX 12/+
Gal80ts2 Gal80ts2

Figure 1 PTX expression in adult mushroom bodies eliminates 3-min memory. (a) Flies of the three genotypes were either given a 30-min-long, 37 1C heat
shock, followed by a 3-h recovery period, or they were left at 18 1C for the entire period. Total protein was extracted from the flies, 100 mg was loaded onto
the gel and 10 ng of purified activated S1 subunit was loaded as a positive control. The S1 subunit of pertussis toxin was detected with the 1B7 monoclonal
antibody. The genotypes are as follows: (i) hsG4 ¼ hsGal4 (89-2-1)/+, (ii)PTX/+ ¼ P{UASPTX}16/+, and (iii) hsG4/PTX ¼ hsGal4/P{UASPTX}16. (b) Effect of
inducing PTX expression within mushroom bodies on negatively reinforced associative learning. The induction of PTX by 500 mM RU486 in the presence of the
P{MBswitch} driver inhibited performance in the negatively reinforced olfactory associative learning assay. The PTX-induced flies performed significantly worse
than any of the other control groups (Bonferroni-Dunn, P o 0.0001; n ¼ 16 for each group). (c) The induction of PTX expression in mushroom body neurons
using the Gal80ts system further demonstrated a requirement for G(o) signaling in olfactory learning. The expression of PTX was induced by a 12-h-long, 32 1C
heat shock, followed by a 3-h recovery period, before training. The PTX-induced flies performed significantly worse than any of the other control groups
(Bonferroni-Dunn, P o 0.0001; n ¼ 12–14 for each group). All values are mean ± s.e.m.

The severity of the PTX learning phenotype might result from the The PTX effect is distributed within mushroom bodies
death of the mushroom body neurons. We tested this hypothesis by We next examined whether the effect of PTX on learning was specific to
examining the integrity of the mushroom bodies after the induction of the mushroom bodies. We induced PTX in the R3 and R4d neurons of
PTX and by establishing whether the associative learning phenotype the ellipsoid body and separately in the dorsally paired medial (DPM)
was reversible. Because Gene-Switch has slow off-rate kinetics20, we neurons, which innervate the mushroom bodies23 (Fig. 3a,b). The
used the Gal80ts system with the P247 Gal4 driver9. P247 drives induction of PTX with the Gal80ts system in either set of neurons did
expression in B700 a/b- and g-lobe mushroom body neurons21. We not affect performance in the learning assay, suggesting that PTX is cell
used two independent Gal80ts transgenes to ensure more complete autonomous. Moreover, PTX induction in the DPM neurons had no
inhibition of Gal4 at 18 1C. After inducing PTX for 12 h at 32 1C, 3-min effect on 60-min memory. The inhibition of neurotransmission in
memory was almost entirely abolished (Fig. 1c). Using antibodies to DPM neurons by the shibirets transgene completely blocks 60-min
downstream of receptor kinase (DRK), which preferentially mark the memory but has no effect on 3-min memory23. Thus, PTX and shibirets
mushroom body22, we found that PTX induction did not alter either have different effects in the DPM neurons, indicating that PTX is not a
the gross structure of the mushroom bodies or the expression of DRK general inhibitor of neurotransmission (Fig. 3b).
(Fig. 2a). We found similar results using antibodies to cAMP- We next sought to determine if the requirement for G(o) signaling in
dependent protein kinase 1 (DCO; data not shown). We also found olfactory associative learning is dispersed throughout the different
that the effect of PTX was reversible (Fig. 2b): although a 2-h induction neurons of the mushroom body lobes, or if the requirement is limited
of PTX within mushroom body neurons produced significant inhibi- to a subset of these neurons. Several genes have been identified that are
tion of 3-min memory (P o 0.0001), this effect was completely preferentially expressed in the different mushroom body lobes, indicat-
reversed after 6 d (P ¼ 0.8274; Fig. 2b). Therefore, the effect of PTX ing that these lobes have distinct molecular repertoires22; however,
on learning is not due to the death of the mushroom body neurons. direct tests for lobe function have yet to provide unequivocal and

Figure 2 PTX expression does not kill the
mushroom body neurons. (a) PTX induction did
a 0.8 b Induced + 3 h recovery
G80ts2/+; PTX/P247, G80ts2 Wild type * Induced + 6 d recovery

not affect mushroom body morphology. The wild- 0.7 * Uninduced

type and PTX genotypes were incubated for 12 h 0.6
at 32 1C and given a 3-h recovery before 0.5
cryosectioning. The prolonged heat treatment

induced PTX expression and eliminated α/β lobes
performance in the negatively reinforced olfactory
associative learning protocol. The a-DRK antibody 0.2
stains the a/b and g lobes and the calyces of the 0.1
mushroom bodies. Scale bar, 50 mm. (b) The PTX 0
3-min-memory phenotype is reversible. In this Calyces Gal80ts20/+; Gal80ts20/+; Gal80ts20/+;
PTX/P247, PTX/Gal80ts2 P247,
experiment, PTX was driven by the P247 Gal80ts2 Gal80ts2
mushroom body Gal4 line and conditionally
inhibited by two Gal80ts transgenes. A 2-h
induction of PTX was followed by recovery for either 3 h or 6 d. All flies were tested at 7 d of age. Although there was a significant effect of PTX expression
3 h after the 2-h induction (P o 0.0001), there was no apparent effect 6 d later (Bonferroni-Dunn; n ¼ 18 for each group). All values are mean ± s.e.m.


a 0.8
Uninduced b 0.8
3 min induced
3 min uninduced
60 min induced
60 min uninduced c 0.8 *
12 h induced
2 h induced
0.7 0.7 0.7

0.6 0.6 0.6

0.5 0.5 0.5


0.4 0.4 0.4

0.3 0.3 0.3

0.2 0.2 0.2
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

0.1 0.1 0.1

0 0 0
Gal80ts20/+; Gal80ts20/+; Gal80ts20/+; Gal80ts20/+; Gal80ts20/+; Gal80ts20/+; Gal80ts20/+; Gal80ts20/+; Gal80ts20/+;
PTX/c232, PTX/Gal80ts20 c232, PTX/c316 PTX/+ c316/+ PTX/c772, PTX/Gal80ts20 c772,
Gal80ts20 Gal80ts20/+ Gal80ts20 Gal80ts20/+

d 0.8
* Induced
Uninduced e 0.8
Uninduced f 0.8

0.7 0.7 * 0.7
0.6 0.6 0.6

0.5 0.5 0.5


0.4 0.4 0.4

0.3 0.3 0.3

0.2 0.2 0.2

0.1 0.1 0.1

0 0 0
Gal80ts20/+; Gal80ts20/+; Gal80ts20/+; Gal80ts20/+; Gal80ts20/+; Gal80ts20/+; Gal80ts20/+; Gal80ts20/+; Gal80ts20/+;
PTX/1471 PTX/+ 1471/+ PTX/c739, PTX/Gal80ts20 c739, PTX/17d PTX/+ 17d/+
Gal80ts20 Gal80ts20/+

Figure 3 The PTX effect on learning maps to the a/b- and g-lobe neurons of the mushroom bodies. In all panels, the white bar indicates a 12-h 32 1C
induction, and the black bar represents the uninduced treatment. (a) c232. The expression of PTX in the ellipsoid body had no effect on 3-min memory. n ¼ 8
for each group. (b) c316. The expression of PTX within the DPM neurons had no significant effect on either 3-min memory (P ¼ 0.8789; first two bars) or 1-h
memory (P ¼ 0.3619; last two bars). White and gray bars, PTX induced; black and dark gray bars, PTX uninduced. n ¼ 13 for each group. (c) c772: a/b and
g lobes. This driver expresses Gal4 in B800 a/b and g lobe neurons35. Gray bars, results after a 2-h 32 1C induction. The 12-h, but not the 2-h, induction of
PTX within these 800 neurons completely ablated 3-min memory (Bonferroni-Dunn, P o 0.0001; n ¼ 13 for each group). (d,e) The induction of PTX within
the g lobes (1471, d) and the a/b lobes (c739, e) produced a significant but incomplete reduction in 3-min memory (Bonferroni-Dunn, P o 0.0001 for both;
n ¼ 12 for each group). (f) 17d: the presumptive a/b core lobes. The induction of PTX within the a/b-lobe neurons defined by this driver had no apparent effect
on 3-min memory. n ¼ 12 for each group. PTX driven by the 201Y g driver resulted in substantial defects in naı̈ve odor avoidance; hence we could not use this
driver to evaluate G(o)-dependent learning (Supplementary Table 2). All values are mean ± s.e.m.

differentiated roles for the constituent neurons in associative learn- the expression of PTX in the a/b-lobe neurons marked by c739
ing15–17,24,25. The c772 Gal4 line drives expression in B800 neurons of (Fig. 3e). Thus, G(o) signaling is required for 3-min memory in
the a/b and g lobes21. We found that a 12-h induction with c772 was both the g and a/b neurons of the mushroom body as defined by the
sufficient to ablate the associative memory, whereas a 2-h induction 1471 and c739 drivers, respectively. In contrast, PTX driven by the
was not (Fig. 3c). There were no differences in the naı̈ve avoidance of a/b-lobe driver 17d did not have an observable effect on 3-min memory
odor or shock between the c772/Gal80ts20; PTX/Gal80ts2 PTX-induced (Fig. 3f). The mushroom body neurons defined by 17d are most likely
and PTX-uninduced experimental groups (Supplementary Table 2). the core neurons of the a/b lobe26, which may be functionally distinct
There were, however, some differences in naı̈ve odor avoidance between from the other neurons of the a/b lobe as they are insensitive to the
the c772/Gal80ts20; PTX/Gal80ts2 PTX-induced group and the control effects of PTX in associative memory and have no effect on the rescue of
genotypes (Supplementary Table 2), suggesting that PTX induction in the rut learning phenotype8. The fact that associative memory forma-
non-mushroom-body neurons by c772 may affect odor perception or tion was affected by PTX induction in the a/b- and g-lobe neurons, but
discrimination and that the Gal80ts inhibition may not be complete in not in the putative a/b core neurons, defines a new requirement for
these neurons. The differences in odor avoidance may also participate G(o) signaling in these lobes for learning and memory and should
in the severe c772/PTX phenotype, although it is unlikely to have a further help dissect the memory process in these neurons.
major effect on learning as naı̈ve avoidance scores were not significantly
different in the within-genotype control group (P ¼ 0.9759 for 0.1% G(o) is independent of rutabaga
methylcyclohexanol (MCH), P ¼ 0.4937 for 0.05% MCH, P ¼ 0.9984 We next considered whether the G(o) pathway interacts genetically
for 0.1% octanol (OCT), and P ¼ 0.7673 for 0.05% OCT; Fig. 3c and with the rut adenylyl cyclase in associative memory formation. The
Supplementary Table 2). It is likely that differences in expression levels persistent activation of vertebrate G(o) may initially lead to the short-
between c772 and P247 account for the different time courses in the term inhibition of type I adenylyl cyclase, followed by the increased
inhibition of learning by PTX between these two lines. The 12-h responsiveness of this enzyme to G(s)a stimulation, known as hetero-
induction of PTX in the g-lobe neurons marked by 1471 caused a logous sensitization or supersensitization27,28. Thus, PTX may be
substantial, but not complete, loss of 3-min memory (Fig. 3d), as did interfering with the down regulation of rut activity by G(o), resulting


associative memory formation. The severity of
a 0.8 Induced b 0.8 Induced
the learning phenotype in PTX-induced flies
Uninduced Uninduced
0.7 0.7 * coupled with the lack of genetic interaction
with rut2080 strongly suggests that the function
0.6 * 0.6
of G(o) in associative learning and memory is
0.5 0.5
* largely independent of the cAMP pathway.
* * Additional members of this new associative

0.4 0.4
* learning pathway are currently unknown. One
0.3 0.3
possibility is that, similar to the role of the
© 2006 Nature Publishing Group http://www.nature.com/natureneuroscience

0.2 0.2 G(o) in the vertebrate dorsal root ganglia, the
Drosophila G(o) may participate in learning
0.1 0.1
through the inhibition of voltage-gated Ca2+
0 0 channels (VGCCs; refs. 12,29). These Ca2+
rut/+; rut/+; PTX/P247 P247/+ rut ; rut; PTX/P247 P247/+
PTX/P247 P247/+ PTX/P247 P247/+ channels are thought to be activated by the
odor-induced depolarization of the mush-
Figure 4 G(o) function in olfactory learning and memory is independent of rutabaga. All genotypes listed room body neurons, leading to the release of
also contained the Gal