080869u165

Neuron, Vol. 21, 1223–1229, December, 1998, Copyright 1998 by Cell Press
Previews
memory code essentially a relational code. According
Linkage at the Top
to this view, memory networks, after they have been
formed, are defined by their cortico-cortical connectiv-
ity, are exquisitely specific with regard to their content,
Around the last turn of the century, neuroscientists were
and presumably, to a degree, are topographically idio-
heatedly debating the associative functions of the large
syncratic for each individual. In the course of behavior,
regions of the cerebral cortex that lie between primary
reasoning, or speech, memory networks are succes-
sensory and motor areas. Primary areas had recently
sively activated—ignited, to use Braitenberg’s term
been identified anatomically, and their functions were
(1978)—by recall, recognition, or the need to retain them
being unveiled by lesion and electrical methods. Little
in short-term memory. Working memory, the kind of
or nothing was known, however, about those other areas
short-term memory needed to perform a sequential task,
between them, which in the human included wide ex-
may simply be the temporary activation of a widespread
panses of the cortex of the occipital, temporal, and pari-
cortical network of long-term memory for prospective
etal lobes, as well as the prefrontal cortex, by itself
action. That prospective action is what determines the
making up nearly one-third of the neocortex. Flechsig
role of the prefrontal cortex in that state of memory, for
(1901), noting that those large areas developed late in
this cortex contains the action-related associations of
phylogeny and ontogeny, proposed that they served to
networks originating in posterior cortex and itself plays
mediate new associations of sensation with movement
a critical role in the organization of sequential actions
and also certain associative functions of the mind, such
toward a goal (Fuster, 1997b).
as memory. Despite respectable support from clinical
To better understand the interactions between the
observations and animal experiments, those ideas were
prefrontal and other cortices in behavior, it is useful to
dismissed, even ridiculed, as unfounded attempts to
view them all in the context of the perception–action
revitalize, by neurologizing it, a fading doctrine of associ-
cycle. Briefly, this cycle is the circular flow of neural
ationist psychology. For most of this century, the neuro-
information by which an organism relates to its environ-
physiologists of the cortex have ignored them. The thala-
ment, a basic principle of biological cybernetics. The
mus has been widely considered the key to cortical
figure shows very schematically the cortical stages of
physiology, and the sensory and motor cortices the key
that cycle and their interconnections (human brain in
to the physiology of the areas beyond.
the inset, arrows symbolizing aggregates of fiber con-
That began to change some 30 years ago. Since then,
nections demonstrated in the monkey). Those cortical
some of the old ideas have come back, this time bol-
stages are the upper levels of two parallel hierarchies
stered by solid anatomical and physiological evidence.
of neural structures, one sensory and the other motor,
We have discovered the previously unsuspected rich-
that extend through the entire length of the nerve axis,
ness and specificity of cortico-cortical axonal connec-
from the spinal cord to the highest cortex of association.
tions (Pandya and Yeterian, 1985). Even in primary visual
(The unlabeled stages represent intermediary cortical
cortex, which critically depends on the thalamus, Ͻ5%
of the terminal axons have been found to be of thala-
mic—geniculate—origin; the vast majority are of cortical
origin, local or remote (Peters and Payne, 1993). High-
lighting the importance of cortical connectivity for cogni-
tive functions, microelectrode recording studies in the
behaving monkey have revealed the widespread cortical
activation of neurons during the memorization of an
event or an object. We are rapidly making the transition
from modular cognition to network cognition. The new
perspective is that memory representations are com-
prised of widely distributed cortical networks that tran-
scend areas and modules by any anatomical definition
(reviewed by Fuster, 1997a). Memory networks are prob-
ably hierarchically organized, overlapping anatomically,
and profusely interconnected. Accordingly, any neuron
or group of neurons, anywhere in the cortex, can be
part of many networks and thus many memories.
The making of those networks follows certain princi-
ples of synaptic modulation that are not yet fully under-
stood. It appears almost certain, however, that learning
and the acquisition of memory are based on the synaptic
linkage of elementary cortical representations or nets
into complex networks. Those new or expanded net-
works represent new cognitive structures or gestalts.
All memory would, therefore, be associative and the
The Cortical Anatomy of the Perception–Action Cycle
Neuron
1224
areas or subareas of adjacent labeled regions.) All con-
and thus also to become engaged in the interactions at
nections between stages are bidirectional, providing
the summit of the perception–action cycle. In general,
feedforward as well as feedback.
however, as sensory–motor associations become rou-
In the course of new or recently acquired behavior,
tine, they are presumably relegated to lower stages of
sensory information is processed along the sensory hi-
the cycle. That is probably why, with overlearning, corti-
erarchy—both serially and in parallel. In the cortex, that
cal activations disappear from tomographic screens,
information translates into action, which is processed
and the neurons described by Asaad et al. seem to
down the motor hierarchy to produce change in the
lose their interest in old or familiar associations. The
environment, which leads to sensory change, which is
experimental approach of these investigators is uniquely
processed through the sensory hierarchy and then mod-
suited to reveal these changes. Indeed, somewhat para-
ulates further action, and so on. The prefrontal and pos-
doxically, the microelectrode remains the best tool to
terior association cortices are in the cycle inasmuch and
explore neural mechanisms in distributed cortical net-
for as long as the behavior contains novelty, uncertainty,
works with thousands if not millions of neurons.
or ambiguity and has to bridge time spans with short-
term memory. As those constraints disappear and be-
Joaquı´n M. Fuster
havior becomes automatic (e.g., walking, skilled rou-
Neuropsychiatric Institute
tines), the action is integrated in lower structures (e.g.,
University of California
premotor cortex, basal ganglia) and sensory processing
Los Angeles, California 90024
shunted at lower levels of the cycle.
Asaad et al. (1998 [this issue of Neuron]) take us closer
Selected Reading
than ever before to understanding how those action-
Braitenberg, V. (1978). In Theoretical Approaches to Complex Sys-
related associations are formed in the prefrontal cortex,
tems, R. Heim and G. Palm, eds. (Berlin: Springer), pp. 171–188.
at the top of the cycle. Their experimental animal, a
Flechsig, P. (1901). Lancet 2, 1027–1029.
monkey, is trained in a delay task, where a particular
Funahashi, S., Bruce, C.J., and Goldman-Rakic, P.S. (1989). J. Neu-
visual stimulus calls for a particular movement of the
rophysiol. 61, 331–349.
eyes after a short delay. This delay makes the task a
Fuster, J.M. (1997a). Trends Neurosci. 20, 451–459.
memory task, requiring the subject to recognize and
Fuster, J.M. (1997b). The Prefrontal Cortex (Philadelphia: Lippincott-
retain a stimulus for subsequent action. Based on previ-
ous research, so-called memory cells are expectedly
Fuster, J.M., and Alexander, G.E. (1971). Science 173, 652–654.
found, which fire faster during the delay than during
Fuster, J.M., and Jervey, J. (1981). Science 212, 952–955.
intertrial baseline periods; the discharge of some of
Miller, E.K., Li, L., and Desimone, R. (1993). J. Neurosci. 13, 1460–
these cells is stimulus preferential, that is, higher in reac-
tion to a given stimulus than to another. In other cells
Niki, H. (1974). Brain Res. 70, 346–349.
nearby, the discharge is related to the movement. Most
Pandya, D.N., and Yeterian, E.H. (1985). In Cerebral Cortex, A. Peters
notable is the finding of cells that are related to both
and E.G. Jones, eds. (New York: Plenum), pp. 3–61.
the cue and the response, or a particular combination of
Peters, A., and Payne, B.R. (1993). Cereb. Cortex 3, 69–78.
the two. As the learning of a new association progresses,
Zhou, Y., and Fuster, J.M. (1997). Exp. Brain Res. 116, 551–555.
activity in prefrontal cells related to the direction of
impending movement develops progressively earlier.
Thus, the authors demonstrate in an elegant manner
that prefrontal neurons become part of cortical networks
containing and representing associations between vi-
sual stimuli and movements.
Touch Channels Sense
Because memory cells were observed first in the pre-
Blood Pressure
frontal cortex and repeatedly reencountered in it (Fuster
and Alexander, 1971; Niki, 1974; Funahashi et al., 1989),
such cells have long been considered the substrate of
its specific role in working memory. There is now ample
Although we can all cite examples of individuals that
evidence, however, that this state of memory activates
seem to operate without perfusing their brains, this is
also other broad and widely dispersed areas of the cor-
just an illusion. Nature has installed pressure sensors
tex with which the prefrontal cortex is connected. In
(baroreceptors) to ensure relatively constant blood flow
addition to prefrontal neurons, the short-term retention
through their arteries. Imbedded in the walls of the arch
of visual stimuli elicits the sustained activation of neu-
of the aorta and the carotid sinus, arterial baroreceptor
rons in inferotemporal cortex (Fuster and Jervey, 1981;
nerve termini form intricate networks that fire in re-
Miller et al., 1993) and even in somatosensory cortex if
sponse to changes in blood pressure. These nerves
the task is visuo-haptic (Zhou and Fuster, 1997). In sum,
report to the brain stem respiratory centers located in
therefore, the memory-active prefrontal cells are part of
the solitary tract nucleus. In turn, these centers regulate
extensive networks that span posterior as well as frontal
blood vessel tone and heart pumping effectiveness
cortex. There is evidence that their sustained activation
through the sympathetic nervous system. In this issue
in working memory results from the dynamic interac-
of Neuron, Drummond et al. (1998) provide evidence
tions between those cortices at or near the top of the
that the mechanotransducers for the arterial pressure
perception–action cycle (Fuster, 1997a). The cells that
sensors are members of the degenerin (DEG)/ENaC fam-
Asaad et al. describe seem to become part of those
ily of cation channels.
networks as they are formed or expanded by learning
Although the baroreceptor reflex is well understood,
Previews
1225
little is known about the basic mechanosensory process
Mammalian homologs, BNaC1, BNaC2, and DRASIC,
that senses distension of the arterial wall. Ion channels
have also been cloned from nervous tissue (reviewed
whose gating is responsive to changes in plasma mem-
by Tvernarakis and Driscoll, 1997; Snyder et al., 1998,
brane tension are primary candidates for these mecha-
and references therein). Mutations near the second trans-
notransducers. In the cardiovascular system, mechano-
membrane domain result in the DEG/ENaC channels
sensitive channels have been recorded from endothelial
being constitutively open, allowing the unobstructed en-
cells lining the lumen of arteries and from cardiac myo-
try of cations into the cell. The flood of cations results
cytes (reviewed by Sachs and Morris, 1998). The energy
in degeneration of the mechanosensory neurons of C.
needed to gate mechanosensitive channels may be col-
elegans.
lected by the membrane-associated cytoskeleton. But,
Can the DEG/ENaC channels bridge the mechanosen-
to date, the only cloned channel that is an unequivocal
sory gap between arterial blood pressure and barore-
mechanosensor is the bacterial MscL protein (reviewed
ceptor discharge? Until recently, the evidence linking
by Sukharev et al., 1997). The bacterial channel is unique
baroreceptor mechanotransduction with the DEG/ENaC
in that it is a hexameric protein complex that can be
channels was purely circumstantial. Mechanosensitive
gated by membrane tension independent of cytoskeletal
gating of the DEG/ENaC channels has not been unequiv-
elements.
ocally shown in their native tissues, in part due to the
Our first glimpse at the molecular structure of a mech-
relative inaccessibility of the baroreceptive nerve termi-
anosensitive channel in eukaryotes was obtained from
nals buried within the arterial wall. Nonetheless, mecha-
genetic studies conducted in the worm Caenorhabditis
nosensory responses have been observed from dissoci-
elegans. These worms move away in response to light
ated baroreceptor neurons isolated from the nodose
touch of the nose or body. Using genetic approaches,
ganglion, which innervates the aortic arch. These re-
ف400 mutants were isolated that were defective in the
sponses included macroscopic Ca2ϩ entry in cells in
touch response but still capable of locomotion (reviewed
response to membrane distortion by a puff of solution
by Tvernarakis and Driscoll, 1997). From these mutants,
(Sullivan et al., 1997) and single channel cation currents
16 genes were identified that when mutated gave rise
activated by suction applied through a recording elec-
to the aberrant mechanosensory phenotype, Mec. The
trode (Kraske et al., 1998). Although these responses
Mec mutations involve proteins localized in a network
were blocked by the trivalent gadolinium previously shown
of six neurons and associated cytoskeletal and extracel-
to block mechanosensitive channels in other prepara-
lular components. These proteins are distributed across
tions, their sensitivity to amiloride was not demonstrated
the long axis of the worm and comprise what are now
(Hamill and McBride, 1996). Also, since the site of mech-
known as touch receptors. Interestingly, mutations within
anotransduction is at the nerve terminals imbedded in
a subset of these genes also result in neuronal cell death
the arterial wall, the significance of mechanosensitive
and are hence also broadly referred to as degenerins
responses measured on the soma is questionable.
Drummond et al. (1998) use reverse transcriptase
A subset of the DEG proteins (MEC-4, MEC-10) share
polymerase chain reaction (RT–PCR) to show that ␤ and
homology with the amiloride-sensitive sodium channel
␥ subunits of the epithelial amiloride sodium channel
subunits previously described in the epithelial layers of
(ENaC) are present in isolated baroreceptor cells of the
the kidneys, lungs, and intestines of vertebrates (Palmer,
nodose ganglion. Since nodose ganglia contain non-
1992). The epithelial amiloride-sensitive sodium channel
baroreceptor cells, this result was corroborated by im-
(ENaC) is a multimeric protein complex composed of
munostaining for ␥ENaC in baroreceptor neurons spe-
three subunits (␣, ␤, and ␥), each of which is thought
cifically labeled with the fluorescent lipophilic dye Di-I.
to be represented three times in the channel complex
Di-I applied to the aortic arch retrogradely labeled a
(Snyder et al., 1998). This finding inspired the notion that
majority (80%) of nodose cells that had also stained
MEC-4 and MEC-10 comprise subunits of a mechani-
positively for ␥ENaC. Anterogradely labeled nodose
cally gated ion channel related to the amiloride-sensitive
ganglia stained small nerve terminals in the aortic arch
epithelial sodium channel, and, indeed, amiloride is known
with both Di-I and anti-␥ENaC, and the labeled nerve
to block certain classes of mechanosensitive channels
terminals had complicated morphologic features pre-
(Hamill and McBride, 1996). As a family, these proteins
viously associated with baroreceptor nerve terminals.
have been termed the DEG/ENaC cation channels.
Surprisingly, ␣ENaC subunit could not be demonstrated
Structurally, each DEG/ENaC channel subunit con-
in nodose ganglia, raising the possibility that ␥ and
tains two hydrophobic transmembrane segments, a large
␤ENaC subunits might be associating with an unidenti-
extracellular loop containing three cysteine-rich regions,
fied third channel subunit. This result may underlie the
a domain with homology to venom neurotoxins, and
differences in mechanosensitive channel conductance
cytoplasmic N and C termini through which the channel
and selectivity previously observed in a variety of tissue
is thought to associate with the cytoskeleton. Interest-
types (Sachs and Morris, 1998). Finally, mechanosen-
ingly, the bacterial MscL channel also contains two
sory responses, such as puff-induced Ca2ϩ entry in
membrane-spanning domains and cytoplasmic N and
retrogradely labeled nodose cells and baroreflex nerve
C termini. Other MEC proteins include tubulin-based
cytoskeletal proteins (MEC-2, MEC-7, MEC-12) and com-
discharge in response to artery distention, could be re-
ponents of the extracellular matrix (MEC-5 and MEC-9).
versibly inhibited by amiloride and its analog. Although
DEG/ENaC homologs also exist in C. elegans but are
not demonstrating mechanosensitive gating of DEG/
not confined to the touch receptor complexes. UNC-8
ENaC channels directly, these results do strengthen the
and DEL-1 are DEG/ENaC homologs expressed in motor
evidence that these channels are the basic mechano-
neurons, while UNC-105 is expressed in muscle cells.
transducers in the baroreceptor nerve terminals.
Neuron
1226
What is the role of the membrane-associated cyto-
brain are likely to be complex, and the tools we posses
skeleton in gating the DEG/ENaC channels? The Unc-
are relatively coarse. In this light, the fact that scientists
105 mutant in C. elegans is characterized by hypercon-
generally are clever enough to think of mechanistic sce-
tracted muscle resulting from unabated cation entry.
narios that cannot be disproved by existing empirical
Unc-105 interacts with Let-2, which encodes collagen
tools complicates the search. Furthermore, the imbal-
IVa2. The Unc-105 mutation can be counteracted by
anced impact of positive results over negative results,
mutations in Let-2, further reinforcing the notion that
or the natural bias of scientists to champion their own
the cytoskeleton is important in the gating of mechano-
point of view, can prolong the discourse. Whatever the
sensitive channels. In humans, X-linked Becker’s and
source, the field of LTP has been mired with LTC to the
Duchenne’s muscular dystrophies are associated with
point that most consider it a long-term tar pit (LTTP).
a faulty myoplasmic Ca2ϩ handling somehow resulting
How does one escape eternal fossilization? It can only
from the disruption of the cell cytoskeleton (Anderson
be hoped that over time different groups, using different
and Kunkel, 1992). By analogy, recordings of mechano-
techniques and asking questions related to different as-
sensitive channels from skeletal muscle from a mouse
pects of synaptic transmission modulation, will provide
model of human X-linked muscular dystrophy (mdx) ex-
the cleansing solvent.
hibit constitutively active channels at rest (Franco and
Toward this end, a number of groups have been
Lansman, 1990) and elevated Ca2ϩ entry (Turner et al.,
scouring the biophysical underpinnings of some scenar-
1991). It will be interesting to see if mutations of the DEG/
ios proposed to explain LTP in CA1 hippocampus. This
ENaC channels cause human disorders not previously
month, Gomperts et al. (1998 [this issue of Neuron])
understood on the molecular level.
address the biophysical basis of “silent” synapses, a
sticky issue currently at the fulcrum of the debate over
Alfredo Franco-Obrego´n and David E. Clapham
whether LTP is due to a pre- or postsynaptic modifica-
Children’s Hospital
tion. “Silent” synapses refer to excitatory transmission
Harvard Medical School
mediated purely by NMDA receptors (NMDARs): due
Department of Basic Cardiovascular Research
to the voltage-dependent properties of NMDARs, such
Boston, Massachusetts 02115
transmission will produce no postsynaptic response at
resting potentials; hence, it is termed silent. Addition of
Selected Reading
AMPARs (which are functional at resting potentials) to
synapses with only NMDARs was proposed as a possi-
Anderson, M.S., and Kunkel, L.M. (1992). Trends Biochem. Sci. 17,
ble postsynaptic mechanism to explain the (consistently
289–292.
observed) decrease in synaptic failures during LTP, evi-
Drummond, H.A., Price, M.P., Welsh, M.J., and Abboud, F.M. (1998).
Neuron 21, this issue, 1435–1441.
dence that is traditionally interpreted as a presynaptic
Franco, A., and Lansman, J.B. (1990). Nature 344, 670–673.
change (Liao et al., 1992). Support for such a process,
relying on the difference in variability between AMPAR-
Hamill, O.P., and McBride, D.W. (1996). Pharmacol. Rev. 48,
231–252.
and NMDAR-mediated responses, was initially detected
Kraske, S., Cunningham, J.T., Hajduczok, G., Chapleau, M.W., Ab-
by Kullmann (1994). This view was strengthened by di-
boud, F.M., and Wachtel, R.E. (1998). Am. J. Physiol. 275, H1497–
rect observations of pure NMDAR-mediated synaptic
responses and a conversion of silent synapses to func-
Palmer, L.G. (1992). Annu. Rev. Physiol. 54, 51–66.
tional synapses during LTP (Isaac et al., 1995; Liao et
Travernarakis, N., and Driscoll, M. (1997). Annu. Rev. Physiol. 59,
al., 1995). Thus, a simple postsynaptic model emerged
659–689.
that could largely explain the existing data on LTP, even
Turner, P.R., Fong, P.Y., Denetclaw, W.F., and Steinhardt, R.A.
those data classically interpreted as a change in presyn-
(1991). J. Cell Biol. 115, 1701–1712.
aptic function. If nothing else, this model is attractive
Sachs, F., and Morris, C.E. (1998). Rev. Physiol. Biochem. Pharma-
because it requires only established intracellular sig-
cology 132, 1–77.
naling mechanisms. It has been well accepted that post-
Snyder, P.M., Cheng, C., Prince, L.S., Rogers, J.C., and Welsh, M.J.
synaptic processes initiate LTP; now well-established
(1998). J. Biol. Chem. 273, 681–684.
intracellular second messenger mechanisms (such as
Sukharev, S.I., Blount, P., Martinac, B., and Kung, C. (1997). Annu.
protein phosphorylation or membrane trafficking) can
Rev. Physiol. 59, 633–657.
explain the longer-lasting modification.
Sullivan, M.J., Sharma, R.V., Wachtel, R.E., Chapleau, M.W., Waite,
L.J., Bhalla, R.C., and Abboud, F.M. (1997). Circ. Res. 80, 861–867.
However, this model requires the existence of syn-
apses with only NMDARs. While few doubt that pure
NMDAR responses exist, an alternative mechanism to
the silent synapse hypothesis has been proposed based
on a series of experimental findings (reviewed by Kull-
mann and Asztely, 1998). In this scenario, all excitatory
Silencing the Controversy in LTP?
synapses have both AMPA and NMDA receptors. Pure
NMDA responses onto cell A are due to the “spillover”
of transmitter from a synapse directly contacting cell B.
Why has there been such long-term controversy (LTC)
The concentration of transmitter, once it reaches cell
over the mechanisms underlying long-term potentiation
A, is sufficient to activate NMDARs but not AMPARs
(LTP)? The inability to resolve this debate may have
because of their lower affinity for transmitter. Gomperts
many sources, including intrinsically empirical as well
et al. test this model by examining excitatory transmis-
as sociological factors. Certainly, the regulatory mecha-
sion in a preparation where an individual neuron is cul-
nisms underlying modification of transmission in the
tured in isolation and makes synapses only on itself. In
Previews
1227
Selected Reading
this case, every presynapse in the preparation forms a
direct contact on the recorded neuron. If every synapse
Diamond, J.S., Bergles, D.E., and Jahr, C.E. (1998). Neuron 21,
has both AMPA and NMDA receptors, even a spillover
425–433.
response will always be accompanied by a direct re-
Gomperts, S.N., Rao, A., Craig, A.M., Malenka, R.C., and Nicoll, R.A.
sponse, i.e., a response with an AMPA component.
(1998). Neuron 21, this issue, 1443–1451.
Gomperts et al. detect pure NMDA responses in this
Isaac, J.T., Nicoll, R.A., and Malenka, R.C. (1995). Neuron 15,
preparation in several ways. First, they note that evoked
427–434.
responses have a larger NMDAR component than spon-
Kullmann, D.M. (1994). Neuron 12, 1111–1120.
taneous miniature responses that are selected based
Kullmann, D.M., and Asztely, F. (1998). Trends Neurosci. 21, 8–14.
on having an AMPAR component. This suggests that
Liao, D., Jones, A., and Malinow, R. (1992). Neuron 9, 1089–1097.
a significant number of spontaneous events were not
Liao, D., Hessler, N.A., and Malinow, R. (1995). Nature 375, 400–404.
selected that have an NMDAR component and no
Liao, D., Zhang, X., O’Brien, R., Ehlers, M., and Huganir, R.L. (1999).
AMPAR component. Furthermore, they are able to pick
Nat. Neurosci., in press.
out spontaneous events that, when averaged, have a
Lu¨scher, C., Malenka, R.C., and Nicoll, R.A. (1998). Neuron 21,
435–441.
slow time course similar to that of a pure NMDAR re-
Mainen, Z.F., Jia, Z., Roder, J., and Malinow, R. (1998). Nat. Neurosci.
sponse. These results strongly suggest that pure NMDA
7, 579–586.
responses can be detected in this preparation and thus
Nusser, Z., Lujan, R., Laube, G., Roberts, J.D., Molnar, E., Somogyi,
argue that there must be some mechanism, other than
P. (1998). Neuron 21, 545–559.
spillover, to account for pure NMDA responses.
Petralia, R.S., Esteban, J.A., Wang, Y.-X., Partridge, J.G., Zhao,
Gomperts et al. examine this further: using immuno-
H.-M., Wenthold, R.J., and Malinow, R. (1999). Nat. Neurosci., in
labeling techniques, they make two important observa-
tions. First, all presynaptic boutons have a cluster of
adjacent postsynaptic receptors. Thus, indeed, any
spillover response would also produce a direct re-
sponse. Secondly, they show that a significant fraction
of synaptic connections have NMDA and lack AMPA
Eph Receptors, Ephrins, and PDZs
receptor immunolabeling and can thus account for the
pure NMDAR transmission.
Gather in Neuronal Synapses
Thus, for this preparation, the authors argue that trans-
mitter spillover cannot account for the pure NMDAR
responses, and they provide anatomical evidence for
Efficient intercellular communication depends on the
synapses with only NMDARs. This, along with another
localization of specific signaling proteins to particular
recent study (Liao et al., 1999), indicates that cultured
sites on the cell surface. The synaptic junction, which
neuronal preparations have silent synapses that can be
mediates rapid communication between neurons, pro-
accounted for by synapses with only NMDARs. Such
vides a striking example in which specific proteins ac-
synapses have also been identified in the experimentally
cumulate at membrane specializations on both sides of
more hostile terrain of the intact brain with immunogold
the synapse. For instance, ionotropic glutamate re-
electron microscopy (Nusser et al., 1998; Petralia et al.,
ceptors are highly concentrated in the postsynaptic
1999). Pure NMDAR synapses were found to be more
membrane of excitatory synapses. What is the molecu-
prevalent in CA1 hippocampus early in postnatal devel-
lar mechanism underlying such localized clustering of
opment, supporting the view that initial synapses my
membrane proteins? Recent studies have highlighted
the role played by proteins that contain PDZ domains
be silent and become AMPAfied during development
(Sheng, 1997; Ziff, 1997). PDZ domains are modular pro-
through an activity-dependent process (Nusser et al.,
tein interaction domains that typically recognize short
peptide sequences of four or more amino acids at the
Finding that silent transmission can be due to action at
very C terminus of its ligands, and different PDZ domains
synapses with only NMDARs enhances our knowledge
recognize different C-terminal sequences. For example,
about basic excitatory transmission in the brain. Further-
PDZ domains in the PSD-95/SAP90 family of postsynap-
more, this provides an important element to a postsyn-
tic density proteins bind to the C-terminal -ESDV peptide
aptic model for expression of LTP. These results come at
sequence of NR2 subunits of the NMDA receptor. On
the heel of several studies arguing against presynaptic
the other hand, GluR2/3 subunits of AMPA receptors
changes during LTP. Three independent groups, using
bind via their C termini (-SVKI) to GRIP, a protein con-
synaptic (Mainen et al., 1998) or peri-synaptic (Diamond
taining seven PDZs (Dong et al., 1997). Studies of PDZ-
et al., 1998; Lu¨scher et al., 1998) detectors of synaptic
based interactions in synapses have naturally focused
transmitter release, found no increase after LTP. While
on neurotransmitter receptors and ion channels, which
an optimistic observer may thus conclude that the tar
are known to be concentrated in synaptic junctions. By
is thinning, and that the LTC of LTP is getting resolved,
contrast, little is known about receptor tyrosine kinases
there may (always) be more clever scenarios to consider.
(RTKs) in neuronal synapses. Some RTKs (MuSK and
erbB receptors) are concentrated in the vertebrate neu-
romuscular junction, but the mechanisms underlying
Roberto Malinow
this localization are unclear. No interactions between
Cold Spring Harbor Laboratory
RTKs and PDZ domains have been reported in verte-
Cold Spring Harbor, New York 11724
brates. Enter Torres et al. (1998 [this issue of Neuron])
Neuron
1228
with two significant advances. First, they report that
specific protein complex around their membrane protein
RTKs of the Eph family and their transmembrane ligands
ligands. In Drosophila photoreceptors, a physiologically
(ephrins) bind to specific PDZ domain proteins; second,
coupled “transducisome” of phototransduction signal-
certain Eph receptors and ligands are concentrated in
ing proteins is built around InaD, a protein with five
neuronal synapses, probably in association with their
PDZs (Tsunoda et al., 1997). In synapses, PSD-95 can
PDZ binding partners.
assemble a specific cytoskeletal-signaling complex that
Torres et al. show that the Eph RTK EphB2 (C-terminal
is physically linked to the NMDA receptor (Craven and
sequence -SVEV) has specific affinity for PDZ domains
Bredt, 1998). Perhaps the interaction of Eph receptors
in two different proteins, GRIP and PICK1, while EphA7
and ligands with PDZ proteins couples them to intracel-
(-GIQV) can bind to GRIP, PICK1, and a third PDZ-con-
lular signaling networks or modulatory enzymes. This
taining protein, syntenin. Ligands belonging to the
may be particularly significant for the ephrin-B ligands,
ephrin-B subfamily (-YYKV) also bind to GRIP, PICK1,
which participate in reciprocal signaling with their Eph
and syntenin. The interaction between PDZ proteins and
receptors despite lacking a catalytic domain. PICK1 has
Eph receptors/ligands is not so surprising; after all, PDZ
only one PDZ domain but was previously identified as
domains recognize just the last few amino acids of their
a protein kinase C (PKC)-binding protein (Staudinger et
ligands, and this C-terminal “zipcode” can be appended
al., 1997); thus, PICK1 could mediate the association of
onto any class of protein. Indeed, the precedent for an
PKC with specific Eph receptors and ligands. PICK1
interaction between an RTK (LET23) and a PDZ protein
also appears to be a direct substrate for the Eph RTK
(LIN-7) has been established in C. elegans epithelial cells
(Torres et al., 1998). GRIP has seven PDZs and the po-
(Kaech et al., 1998). More unexpected is the ensuing
tential to scaffold an elaborate protein architecture
finding that EphB2 and its ligand, Ephrin-B, are concen-
around Eph receptors and their ligands. Since GRIP
trated at synapses in cultured neurons, where their PDZ
was originally identified as an AMPA receptor–binding
partners GRIP and PICK1 are also localized.
protein, it will be interesting to determine whether Eph
To date, Eph receptors and their ligands have been
receptors or ligands are physically and functionally cou-
studied primarily in a developmental context. In the ner-
pled to AMPA receptors in synapses. To date, there has
vous system, these molecules are implicated in axon
been little evidence for regulation of AMPA receptors
guidance, particularly in repulsion and in establishment
by tyrosine phosphorylation.
of boundaries between groups of cells. What are Eph
Unlike many ligands of RTKs, Eph ligands are not
receptors and ephrins doing in synapses? It is tempting
active as soluble proteins; ephrins need to be clustered
to speculate that they might be involved in synaptogene-
on the cell surface for them to stimulate their cognate
sis (like MuSK and erbB receptors) or in synaptic plastic-
Eph receptors. It is pertinent, therefore, that surface
ity, perhaps by controlling the adhesion and/or repulsion
aggregation of transmembrane proteins is a common
of pre- and postsynaptic membranes. The synaptic lo-
outcome of interaction with PDZ proteins. The ability of
calization of Eph receptors and their ligands needs to
certain PDZ proteins to cluster their binding partners
be confirmed in the brain and extended to other mem-
may reflect the propensity of PDZ-containing proteins to
bers of these protein families. Important questions will
multimerize and/or their ability to bind these membrane
include whether the various Eph receptors and ephrins
proteins in a multivalent manner. Indeed, PICK1 can
are differentially distributed among CNS synapses, and
aggregate ephrin-B1 in heterologous cells (Torres et al.,
whether receptors and ligands are segregated to pre-
1998). Clustering by PICK1 or GRIP may optimize the
and postsynaptic sides of the junction. Detailed analysis
presentation of ephrins to their Eph receptors in vivo;
of mouse knockouts of Eph receptors and ephrins may
such a mechanism offers another potential level for reg-
shed more light on the roles of these proteins in syn-
ulation of Eph signaling. PDZ-dependent clustering of
apses and in mature brain.
Eph receptors and ligands at specific subcellular sites
If it is early to speculate about the synaptic functions
(e.g., in growth cones) may also be important for Eph/
of Eph receptors and ephrins, what about the functional
ephrin function in development. Thus, following up the
significance of their interactions with PDZ domain pro-
findings of Torres et al. promises to shed new light on
teins? A prevailing idea is that the PDZ protein is impor-
the functions and mechanisms of the Eph system in
tant for the subcellular localization of its binding part-
both developing and mature brain.
ners. In Drosophila, the PSD-95 homolog Discs-large is
localized in synapses and is essential for the synaptic
Yi-Ping Hsueh and Morgan Sheng
clustering of its PDZ interactors, Shaker and Fasciclin
Howard Hughes Medical Institute and Department
II (Thomas et al., 1997; Zito et al., 1997). Genetic studies
of Neurobiology
on InaD (in Drosophila) and LIN-2/LIN-7/LIN-10 (in C.
Massachusetts General Hospital and
elegans) additionally support the idea that PDZ-medi-
Harvard Medical School
ated interactions are important for the subcellular tar-
Boston, Masachusetts 02114
geting of the interacting proteins, both at synapses and
at other specialized membrane domains (Tsunoda et al.,
1997; Kaech et al., 1998; Rongo et al., 1998). By analogy,
Selected Reading
EphB2 and ephrin-B1 localization in neuronal synapses
may depend on their binding to synaptic PDZ proteins
Craven, S.E., and Bredt, D.S. (1998). Cell 93, 495–498.
like GRIP and PICK1.
Dong, H., O’Brien, R.J., Fung, E.T., Lanahan, A.A., Worley, P.F., and
Another (not mutually exclusive) concept is that PDZ
Huganir, R.L. (1997). Nature 386, 279–284.
proteins have a scaffolding function and can assemble a
Kaech, S.M., Whitfield, C.W., and Kim, S.K. (1998). Cell 94, 761–771.
Previews
1229
Rongo, C., Whitfield, C.W., Rodal, A., Kim, S.K., and Kaplan, J.M.
(1998). Cell 94, 751–759.
Sheng, M. (1997). Nature 386, 221–223.
Staudinger, J., and Olsen, E.N. (1997). J. Biol. Chem. 272, 32019–
32024.
Thomas, U., Kim, E., Kuhlendahl, S., Ho Koh, Y., Gundelfinger, E.D.,
Sheng, M., Garner, C.C., and Budnik, V. (1997). Neuron 19, 787–799.
Torres, R., Firesrein, B.L., Dong, H., Staudinger, J., Olsen, E.N.,
Huganir, R.L., Bredt, D.S., Gale, N.W., and Yancopoulos, G.D. (1998).
Neuron 21, this issue, 1453–1463.
Tsunoda, S., Sierralta, J., Sun, Y., Bodner, R., Suzuki, E., Becker,
A., Socolich, M., and Zuker, C.S. (1997). Nature 388, 243–249.
Ziff, E.B. (1997). Neuron 19, 1163–1174.
Zito, K., Fetter, R.D., Goodman, C.S., and Isacoff, E.Y. (1997). Neuron
19, 1007–1016.

Source: http://malinowlab.cshl.edu/Malinow%20Images/PDF%20papers/Malinow%20Neuron%201998.pdf