ARTICLES

GABAergic synapses are formed without the

© 2008 Nature Publishing Group http://www.nature.com/natureneuroscience

involvement of dendritic protrusions

Corette J Wierenga, Nadine Becker & Tobias Bonhoeffer

Synaptogenesis and the role of dendritic protrusions in this process are well studied in glutamatergic synapses. Much less is
known about the formation of GABAergic synapses, which are located predominantly on the dendritic shaft. We used genetically
labeled interneurons in mature hippocampal slice cultures and two-photon laser-scanning microscopy to examine contact
formation between GABAergic axons and the dendrites of CA1 pyramidal cells. Dendritic protrusions distinguished and selected
between glutamatergic and GABAergic boutons. In contrast with contacts with glutamatergic boutons, which can be long lasting,
the contacts of dendritic protrusions with GABAergic boutons were always short lived. Similarly, the contacts made by GABAergic
axonal protrusions were always transient. New putative GABAergic synapses were formed exclusively by new boutons appearing at
pre-existing axon-dendrite crossings without the involvement of any dendritic or axonal protrusions. These findings imply that
fundamentally different mechanisms underlie the generation of GABAergic and glutamatergic synapses.

Chemical synapses are the key connective elements in neuronal net-
works. They are not only crucial for information processing, but their
plasticity also endows the brain with its outstanding capacity for
adaptation to the environment. Understanding synapses and how
they are formed is therefore a fundamentally important task in
neuroscience. So far, the majority of studies have focused on the
formation of glutamatergic synapses, whereas the formation of the
other major type of synapse in the brain, which uses the inhibitory
transmitter GABA, remains less examined, in spite of the fact that
GABAergic synapses form 10–20% of all synapses in the brain and are
indispensable for the proper and stable functioning of the brain1.
Glutamatergic synapses on excitatory neurons in the hippocampus
occur almost exclusively on dendritic spines2,3. It is thought that these
spines and their precursors, dendritic filopodia, are actively involved in
the formation of new glutamatergic synapses. Dendrites grow small
protrusions that make contact with presynaptic axons and boutons4–8.
Only a subset of these initial contacts is selected through filopodial
signaling and forms mature, functional synapses9–11. During further
maturation of a new synaptic contact, the dendritic protrusion
shrinks as its head expands, turning into a mature spine4,12,13. In this
scenario, dendritic filopodia precede fully mature spines and it has
been shown that long, thin dendritic protrusions that are reminiscent
of filopodia can have (sometimes even multiple) synaptic contacts5,6,14.
Although the above is probably not the only way to form new synapses
on spines15, the studies cited above provide ample evidence of a role for
outgrowing protrusions in the formation of glutamatergic synapses.
In contrast to glutamatergic synapses, GABAergic synapses are
usually not located on spines but are found directly on the dendritic
shaft1,3,16. It has been suggested that dendritic protrusions can also
serve to guide presynaptic axons to the dendritic shaft, thereby
mediating the formation of shaft synapses5,6. However, it is unresolved
to date how often this occurs and whether alternative scenarios,
possibly involving a more active role of the GABAergic axon, are also
important in the formation of GABAergic contacts.
We used high-resolution two-photon imaging of genetically labeled
GABAergic neurons and their postsynaptic counterparts to examine
how new GABAergic shaft synapses are formed in the CA1 area of
organotypic hippocampal cultures. We found that new GABAergic
boutons, probably indicating new GABAergic synapses, were formed
exclusively at pre-existing axon-dendrite crossings, without the in-
volvement of either dendritic or axonal protrusions. This indicates that
there are fundamentally different mechanisms for the formation of
glutamatergic and GABAergic synapses.
RESULTS
To visualize GABAergic axons, we used hippocampal slice cultures
of GAD65-GFP mice. In these mice, 30–50% of GABAergic inter-
neurons in the hippocampus express GFP at embryonic ages and
well into adulthood17. Dendrites of 2–5 CA1 pyramidal neurons
were filled with Alexa Fluor 594 through a patch pipette before
imaging. This resulted in a sparse overlap between GFP-labeled
GABAergic axons and Alexa 594–labeled dendrites, allowing the
contacts between GABAergic axons and boutons with labeled dendritic
structures to be resolved in detail with two-photon laser-scanning
microscopy (Fig. 1a,b). We took high-resolution image stacks every
30 min for a total period of 3–6 h and examined the formation
of contacts between GABAergic axons and apical dendrites of CA1
pyramidal neurons. In total, we analyzed the time-lapse series

Max Planck Institute of Neurobiology, Cellular and Systems Neurobiology, Am Klopferspitz 18, 82152 Martinsreid, Germany. Correspondence should be addressed to C.J.W.
(wierenga@neuro.mpg.de).
Received 19 March; accepted 30 June; published online 24 August 2008; doi:10.1038/nn.2180

1044 VOLUME 11 [ NUMBER 9 [ SEPTEMBER 2008 NATURE NEUROSCIENCE


a b
© 2008 Nature Publishing Group http://www.nature.com/natureneuroscience
30 µm
c d GFP
Alexa 594
GFP
GFP
Biocytin
1 µm
Alexa 594 Gephyrin
Live imaging Post hoc immunostaining Live imaging
of 21 image stacks, in which we observed 4400 pre-existing, stable
contacts (Table 1).
From the high-resolution images, we could clearly identify
the locations where a GABAergic bouton was in contact with a labeled
dendrite (see Methods for our criteria). Although synapses require
the presence of an axonal bouton, axonal varicosities or swellings
do not necessarily imply the presence of synapses18,19. To examine
whether the physical contacts between GABAergic boutons and
dendrites corresponded to synapses, we carried out post hoc immu-
nostaining on contacts between boutons and dendrites that were
previously identified in two-photon image stacks. We found that
80% of the contacts (56 out of 70, 5 slices) that were classified as
making physical contact on the basis of the two-photon images
did indeed show postsynaptic gephyrin staining, indicating that the
majority of physical contacts represent GABAergic synapses (Fig. 1c,d).
For 10 of the 14 gephyrin-negative contacts, the bouton showed
gephyrin staining that did not overlap with the labeled dendrite,
suggesting that the GABAergic bouton made a synapse to a neighbor-
ing unlabeled structure. Using the mirror image of the gephyrin
channel reduced the number of gephyrin-positive contacts from 80%
to 47%, indicating that gephyrin staining was specific (P o 0.001,
w2 test). We also examined locations where a bouton was close to a
labeled dendrite, but the overlap between the red and green channel
in the two-photon images was too small to be classified as a contact by
our criteria. However, 19% of these ‘noncontacts’ showed gephyrin
staining (10 out of 54, 5 slices). This indicates that our criteria for
identifying GABAergic synapses from our two-photon images might
lead to an error in assessing synaptic contacts of no more than B20%
in both directions, which should be unproblematic for the conclusions
presented below.
From the immunostaining experiments, we therefore inferred that
at least 80% of boutons contacting a dendrite, identified on the
basis of the two-photon images, represent actual synaptic contacts.
The locations where we observed a GABAergic axon crossing
or touching a dendrite, but which lacked a bouton, are probably
nonsynaptic contact points18,19 and will be referred to as axon-
dendrite crossings.
NATURE NEUROSCIENCE VOLUME 11 [ NUMBER 9 [ SEPTEMBER 2008
ARTICLES
Figure 1 Imaging GABAergic synapses in
GAD65-GFP slice cultures. (a) Maximal intensity
projection image of the CA1 region, showing
GFP-labeled GABAergic interneurons (green) and
Alexa 594–labeled pyramidal neurons (red).
(b) A single section from an image stack (raw
data) from the boxed area indicated in a showing
clear labeling of GABAergic axons and boutons
and their overlap with labeled dendrites. (c,d) Two
examples of synaptic contacts (white arrows)
between GABAergic boutons (green) and labeled
dendrites (red). Single sections of two-photon
image stacks are shown to the left with the
corresponding confocal sections after
immunofluorescence staining for gephyrin (blue)
10 µm
to the right. Areas with all three labels appear
white. In these images (but nowhere else), two-
GFP photon images were filtered to allow for better
Biocytin comparison with the confocal images.
Gephyrin
Transient contacts are formed by
protrusions
1 µm
In principle, new GABAergic synapses could
Post hoc immunostaining be formed with or without the involvement of
dendritic or axonal protrusions. During the
3–6-h imaging period, we observed many
dendritic protrusions on labeled dendrites forming, retracting or
changing size and/or shape. In 19 cases, a dendritic protrusion could
be seen making contact with a GFP-labeled GABAergic bouton (two
examples are shown in Fig. 2a,b). In the first example, a dendritic
protrusion (Fig. 2a) grew out from the shaft and touched the
GABAergic bouton of a neighboring axon (Fig. 2a). At the next time
point, 30 min later, the dendritic protrusion had almost completely
retracted to the shaft. In the second example (Fig. 2b), the dendritic
protrusion grew out from a spine that was present during the entire
imaging period. As in the first example (Fig. 2a), the contact between
the dendritic protrusion and the GABAergic bouton lasted for only a
single time point (complete time series for these two examples are
shown in Supplementary Fig. 1 online).
The majority of contacts between a new dendritic protrusion and a
GABAergic bouton (12 out of 19) were present at only one time point
(0.5 h) and none lasted longer than 1.5 h. There was no difference in the
lifetime of the contact between protrusions that grew directly from the
shaft (13 out of 19) and protrusions that grew from stable spines (6 out
of 19). These observations indicate that dendritic protrusions do make
contact with GABAergic boutons in a similar fashion to glutamatergic
boutons. However, these contacts were always transient, and therefore,
Table 1 Number of observations
Pre-existing stable boutons 423 (78%)
Transient contacts 44
Dendritic protrusion contacts GABAergic bouton 19 (3.5%)
Axonal protrusion contacts labeled dendrite 25 (4.6%)
New boutons 42
Transient bouton appears at crossing 30 (5.5%)
Stable bouton appears at crossing 12 (2.2%)
Lost boutons 12
Bouton disappears at crossing 10 (1.8%)
Dendrite retraction 2 (0.4%)
Contact rearrangements 24
New contact by movement of bouton or dendrite 10 (1.8%)
Lost contact by movement of bouton or dendrite 14 (2.6%)
Data from 21 experiments with imaging durations of 3–6 h.
1045
 ARTICLES

Figure 2 Transient contacts by dendritic and axonal protrusions. Three-

a 0h 0h b dimensional representations (left) and single sections (right) of two-photon
image stacks at the indicated time points. GABAergic axons are shown in
green (GFP) and dendrites in red (Alexa 594). Overlap (yellow) of the red
and green channels indicates contact between the axonal and dendritic
0.5 h 0.5 h structures. Some nonrelevant structures that are visible in the single sections
were removed from the three-dimensional reconstructions for clarity. Three-
1h 1h
dimensional reconstructions were made for visualization purposes only;
DP
analysis was always done on individual image sections. (a) A dendritic
DP
protrusion (DP) grew out from the dendritic shaft and made transient contact
with a GABAergic bouton. (b) A DP emerged from a stable spine and made
© 2008 Nature Publishing Group http://www.nature.com/natureneuroscience

1h 1h
transient contact with a nearby GABAergic bouton. (c) An axonal protrusion
(AP) grew out from a GABAergic bouton that was contacting the labeled
1.5 h 1.5 h
dendrite. The protrusion made contact with the dendrite twice (see
Supplementary Fig. 1), but the contact was not present at the end of the
imaging period. (d) An AP grew out and formed a transient contact with
the labeled dendrite. Both boutons did not form contacts with the labeled
1.5 h 1.5 h
dendrite. Scale bars represent 1 mm.

Red dendrites of CA1 pyramidal cells dendritic or axonal protrusion makes contact
Green GABAergic axons/boutons protrusion without contact protrusion (and its contact) was not present the next day and the axon
c d 0h 0h and dendrite appeared to be unchanged. This shows that, even in our
longest-lasting contact between axonal protrusion and dendritic shaft,
the contact did not seem to be transformed into a stable synapse. These
0h 0h
observations suggest that axonal protrusions grow out and make
transient contacts with the dendritic shaft, but they are not transformed
3h 3h
into long-lasting, stable contacts.
AP

New stable boutons are formed on dendritic shafts

1.5 h 1.5 h
4.5 h 4.5 h
AP
AP
3h 3h
5.5 h 5.5 h
5.5 h 5.5 h
in contrast with glutamatergic synapses, the formation of stable
GABAergic synapses is probably not mediated by dendritic protrusions.
We also looked at the outgrowth of axonal protrusions from labeled
GABAergic axons during the 3–6-h imaging period. Consistent with
previous reports20–22, these were usually smaller and shorter lived than
dendritic protrusions and occurred exclusively at boutons. In 25 cases,
we observed an axonal protrusion touching a labeled dendrite (two
examples are shown in Fig. 2c,d; complete time series for these two
examples are shown in Supplementary Fig. 1). In the first example
(Fig. 2c), a GABAergic bouton contacted the labeled dendrite during
the entire imaging period. After 1.5 h, an axonal protrusion (Fig. 2c,d)
grew out from the bouton along the dendritic shaft of the labeled
neuron but retracted before the end of the imaging period. The axonal
protrusion in the second example (Fig. 2d) grew from a bouton that
was not in contact with the labeled dendrite. It appeared to form a new
bouton on contact with the dendrite (t ¼ 4.5 h). As with the first
example, however, the contact did not last. In all cases where an axonal
protrusion made contact with a labeled dendrite, the contact was only
transient (o2.5 h for all but one). There was no difference in lifetime of
the contact between small protrusions (such as in Fig. 2c; 17 out of 26)
or longer filopodia-like protrusions (Fig. 2d; 9 out of 26). Only once
did we observe a new contact by an axonal protrusion that lasted until
the end of the imaging period (4.5 h). In this case, we could image the
same contact the following day (see below). We found that the axonal
In addition to transient contacts by protrusions, we also observed long-
lasting changes in GABAergic contacts during the 3–6-h imaging
period (Fig. 3). At locations where a labeled GABAergic axon crossed
a labeled dendritic shaft, a new bouton occasionally appeared that then
stayed present for the rest of the imaging period. We observed this in 12
cases (Figs. 3a,b and 4a). The appearance of a new bouton did not seem
to be influenced by the presence (4/12; Fig. 3a) or absence (8/12;
Fig. 3b) of a neighboring bouton that was already contacting the
dendritic shaft (complete time series of the two examples in Fig. 3 are
shown in Supplementary Fig. 2 online).
Not all new GABAergic boutons were stable from their first appear-
ance. Some boutons were present for only a short period (Figs. 3c and
Fig. 4b), whereas others disappeared and reappeared during the
imaging period (Figs. 3d and Fig. 4b). We observed 30 of these
transient boutons. In some cases, the new bouton appeared to make
small movements to and from the dendrite before and after making
contact (Fig. 3c and Supplementary Fig. 2). In other cases, the new
bouton appeared or disappeared at the location of the contact without
noticeable movement (Fig. 3d).
The detailed time courses of the appearance of individual boutons
on axon-dendrite crossings clearly shows the difference between
boutons that were stable from their first appearance (Fig. 4a) and
transient boutons, which disappeared and sometimes reappeared
during the imaging period (Fig. 4b). These observations indicate that
stable GABAergic boutons can be formed over a period of several hours
on a similar time scale as has been reported for the formation of
glutamatergic synapses23–25. However, in contrast with glutamatergic
contacts, dendritic or axonal protrusions do not seem to be involved in
the establishment of long-lasting contacts. Instead, new stable GABAer-
gic boutons appear at pre-existing axon-dendrite crossings.
Although the majority of inhibitory synapses in the brain are located
directly on the dendritic shaft, some of them occur on dendritic
spines3,26,27. In hippocampal slice cultures of GAD65-GFP mice, in
which only a subset of GABAergic interneurons are labeled, 21% of the
labeled GABAergic contacts that were present before and during the

1046 VOLUME 11 [ NUMBER 9 [ SEPTEMBER 2008 NATURE NEUROSCIENCE

 ARTICLES

Figure 3 New GABAergic boutons appear at axon-dendrite crossings. Three-

a 0h 0h b 0h 0h dimensional representations (left) and single sections (right) of two-photon
microscopy image stacks at the indicated time points (see Fig. 2 for details).
(a) A stable bouton (SB) appeared at an axon-dendrite crossing and stayed in
contact with the dendrite for the rest of the imaging period. The lower bouton
formed a stable contact with the dendrite for the entire imaging period.
1h 1h 2h 2h (b) An SB appeared at an axon-dendrite crossing at t ¼ 2.5 h and stayed in
SB contact with the dendrite for the rest of the imaging period. (c) A transient
appearance of a bouton (TB) at an axon-dendrite crossing. (d) A TB at an
axon-dendrite crossing appeared, disappeared and reappeared several times.
Scale bars represent 1 mm.
© 2008 Nature Publishing Group http://www.nature.com/natureneuroscience

2.5 h 2.5 h
4h 4h
SB SB quantified the duration of each episode of contact or bouton appear-
ance, the number of repeated contacts or bouton appearances during
the imaging period, and the total percentage of the imaging period in
5h 5h 4h 4h which the contact or bouton was present. The durations for all types of
SB transient events were fairly similar and transient events were much
SB shorter compared to the time a new stable bouton was present (Fig. 4c).
Notably, repeated appearance and disappearance of a transient bouton
Red dendrites of CA1 pyramidal cells stable or transient bouton with contact
at an axon-dendrite crossing was fairly common, whereas contacts
Green GABAergic axons/boutons bouton without contact made by protrusions, either dendritic or axonal, were usually made
c d 0h only once during the imaging period (Fig. 4d). When a transient
bouton appeared at an axon-dendrite crossing, it was present, on
TB
average, for almost 40% of the total imaging period, whereas a transient
0h 0h 0h
contact made by a dendritic protrusion was present for o20% of the
1h time (Fig. 4e). These data support the notion that new GABAergic
synapses are formed by the occurrence of a new GABAergic bouton at
an existing axon-dendrite crossing, whereas contacts made by protru-
1h
sions, either axonal or dendritic, have a very low probability of being
0.5 h 0.5 h
2h transformed into stable synapses.
TB
Long-lasting contacts
TB We next examined whether GABAergic boutons that were formed
2h 2h 2h
3.5 h
during the imaging period persisted into the next day. After a night in
the incubator, the Alexa 594 dye was often cleared by the CA1
pyramidal neurons and had gathered mostly in the soma, leaving
only weakly staining in the dendrites. In five experiments, the dendrites
4h 4h 3.5 h
4h
were bright enough to image the same region with sufficient detail the
next day (examples are shown in Supplementary Fig. 3 online). The
TB GABAergic axons and boutons were as bright as the day before and had
largely unchanged morphology.
4h
Of the 86 contacts between GABAergic boutons and labeled dendrites
that were present before and during the entire imaging period on the
entire imaging period were located on spines (90 out of 423 pre-existing first day, 81 were also observed the next day (94%), corroborating their
contacts). We asked whether the appearance of a new bouton prefer- classification as pre-existing stable contacts. Two stable boutons had
entially occurred at axons crossing the dendritic shaft or whether they formed during the experiment at an axon-dendrite crossing that had
also occurred at spines. Two of the 12 new stable boutons and 5 of the 30 stayed present for the rest of the imaging period on the first day. Both
transient boutons appeared at dendritic spines. The fraction of new boutons were still present and in contact with the red dendrite the
boutons appearing on spines versus shafts was not significantly different following day, indicating that the newly formed boutons were indeed
from the fraction of pre-existing GABAergic contacts made on spines long lasting. Of the 15 boutons that qualified as transient on the first
(P 4 0.5, w2 test). This suggests that the formation of new GABAergic day, seven were present the next day at the location where the bouton
synapses occurs through the appearance of a new bouton at pre-existing had first occurred. In the other eight cases, the axon-dendrite crossing
axon-dendrite crossings, irrespective of whether the new bouton is was still present, but a bouton could not be discerned. This suggests that
made on a spine or on the dendritic shaft of the postsynaptic neuron. at least a subset of transient boutons is transformed over time into stable
boutons, possibly forming synapses. Of the transient contacts made by
Lifetime of transient contacts and new boutons 11 axonal protrusions and 3 dendritic protrusions that were observed on
As described above, we observed three types of transient events. the first day, none were present the next day and the axons and dendrites
Transient contacts were initiated by the dendrite through the out- appeared to be unchanged (see Table 2). Although it is possible that
growth of a dendritic protrusion that touched a GABAergic bouton but small dendritic protrusions were missed as a result of the weak dendrite
could also be initiated by the outgrowth of an axonal protrusion labeling on the second day, these observations are consistent with the
contacting the dendrite. Furthermore, boutons transiently appeared idea that transient contacts formed by axonal or dendritic protrusions
at an existing axon-dendrite crossing. For each of these events, we are generally not transformed into GABAergic synapses.

NATURE NEUROSCIENCE VOLUME 11 [ NUMBER 9 [ SEPTEMBER 2008 1047

 ARTICLES
a Stable boutons b Transient boutons
1.0
Normalized overlap
Normalized overlap
0.8
0.6
0.4
0.2
0.0
–3 –2 –1 0 1 2 3 4 5 h 1d
Time
© 2008 Nature Publishing Group http://www.nature.com/natureneuroscience
Repeated contacts
c Episode duration d or appearances
4.5 ***
***
4.0
2
3.5
3.0
Time (h)
2.5
2.0
1.5
1.0
0.5
19 25 30 12 19 25
0.0 1 0%
DP AP TB SB DP AP TB
Lifetime of protrusions
We noticed that dendritic protrusions often retracted soon after
contacting a GABAergic bouton, whereas axonal protrusions appeared
to live longer. We examined whether this was the result of a general
difference in lifetime between dendritic and axonal protrusions or
whether the transient contact with the GABAergic bouton or dendritic
shaft, respectively, affected the lifetimes of the protrusions. We there-
fore measured the lifetime of an arbitrarily chosen population of newly
emerging (that is, not present at the first time point) axonal and
dendritic protrusions that were not contacting any labeled structure.
Axonal protrusions that had contacted a labeled dendrite had a range
of lifetimes that was indistinguishable from the range of lifetimes of the
population of arbitrarily chosen GABAergic axonal protrusions (Fig. 5a).
For dendritic protrusions, the picture was very different. Although the
arbitrarily chosen dendritic protrusions included both short-lived and
long-lived protrusions, all dendritic protrusions that contacted a
GABAergic bouton had short lifetimes (Fig. 5b). This suggests that
only short-lived dendritic protrusions contact GABAergic boutons or,
somewhat more likely, that contacting a GABAergic bouton shortens the
lifetime of a dendritic protrusion. Alternatively, some of the arbitrarily
chosen protrusions contacted something that prolonged their lifetimes.
The arbitrarily chosen population of dendritic protrusions probably
contacts a rather inhomogeneous selection of structures, including
GABAergic and glutamatergic axons and boutons, glial processes, and
other dendrites28. We therefore tested more specifically whether the
lifetime of dendritic protrusions depends on whether they contact a
glutamatergic or GABAergic bouton. To this end, we labeled glutama-
tergic axons from CA3 neurons with a bolus-loading technique (N.B.,
C.J.W., Fonseca, R., T.B. and Nägerl, U.V., unpublished data). In these
experiments, we observed 20 dendritic protrusions growing from
labeled CA1 pyramidal neurons that contacted labeled glutamatergic
boutons. In contrast with the protrusions contacting GABAergic
boutons, a substantial fraction of these protrusions had lifetimes that
were greater than 2 h (Fig. 5b). In fact, in 5 of the 20 cases, the
protrusion was still present and contacting the glutamatergic bouton at
the last imaging time point (Fig. 5c). In many cases, the dendritic
protrusions stayed in contact for much longer than was observed for
dendritic protrusions contacting a GABAergic bouton. Nevertheless,
not all contacts between dendritic protrusions and glutamatergic
boutons were long lasting. Retraction of dendritic protrusions shortly
1048
Figure 4 Lifetime of transient contacts and boutons. (a,b) Time course of new
stable (a, n ¼ 12) and transient boutons (b, n ¼ 30). Individual boutons are
1.0
represented in different colors, and the average time course is given by the
0.8 thick black curve. The number of overlapping pixels between bouton and
0.6 shaft (normalized to maximum overlap) is plotted for all time points, aligned
0.4 to the time when the bouton (first) appeared. New stable boutons always
stayed in contact with the shaft, whereas transient boutons disappeared and
0.2
sometimes reappeared at later time points. In some experiments, we
0.0
–3 –2 –1 0 1 2 3 4 5 h 1d examined the boutons on the day following the imaging period (1 d, see also
Time Table 2). (c) Average duration of single episodes of contact by dendritic and
axonal protrusions and transient and stable bouton occurrences. (d) Number
e Total contact time of repeated contacts or bouton appearances. (e) Percentage of the imaging
Percentage of total imaging period
80% *** period that the transient contact or bouton was present. The numbers at
70% the base of the bars reflect the number of observations. * P o 0.05,
60% *** P o 0.001, ANOVA with post hoc Tukey HSD test. Error bars reflect s.e.
***
50% *
40% after contacting glutamatergic boutons also occurred, similar to pro-
30% trusions contacting GABAergic boutons. Taken together, these data
20%
suggest that some, but not all, of the contacts with glutamatergic
10%
30 12 19 25 30 12
boutons were transformed into stable glutamatergic spine contacts,
SB DP AP TB SB consistent with the proposed role of dendritic protrusions in the
formation of glutamatergic spine synapses4–6,9. Our finding that
dendritic protrusions distinguish between GABAergic and glutamater-
gic boutons is further supported by the individual time courses of
contact formation by dendritic protrusions (Fig. 5d,e). These data
strongly suggest that, although dendritic protrusions are important in
the formation of glutamatergic spine contacts, contacts between
dendritic protrusions and GABAergic boutons are short lived and do
not normally result in new GABAergic synapses.
New boutons can form synapses in several hours
The results described above suggest that new GABAergic boutons
appearing at pre-existing axon-dendrite crossings may be important
for establishing new GABAergic synapses. However, our two-photon
data show anatomical apposition as well as a morphological change
into a bouton, but they do not conclusively demonstrate that these new
boutons actually form functional synapses, nor do they indicate at what
time scale these new synapses form. For glutamatergic synapses, it has
been shown that the presynaptic active zone can be functional in
30–60 min after the initial contact between axon and dendrite and that
postsynaptic scaffold proteins and receptors are recruited soon there-
after23,25,29,30 (but also see refs. 7,8). At present, detailed information
on the time course of the recruitment of synaptic molecules to new
GABAergic synapses is almost completely lacking31–33.
To start addressing this issue, we carried out time-lapse imaging of
GFP-labeled GABAergic axons, followed by post hoc immunostaining for
GABAergic pre- and postsynaptic markers, vesicular GABA transporter
(VGAT) and gephyrin (Fig. 6). For comparison, we immunostained slice
cultures without previous time-lapse imaging and found that the vast
majority of GABAergic boutons contained VGAT and were associated
with gephyrin (Fig. 6c,d). In fact, only 2.4% of GFP-positive boutons
Table 2 Contacts that were present the following day
Pre-existing stable boutons 81 of 86 (94%)
Transient contacts
Dendritic protrusions 0 of 3* (0%)*
Axonal protrusions 0 of 11 (0%)
New boutons
Transient boutons 7 of 15 (47%)
Stable boutons 2 of 2 (100%)
Data from 5 experiments. *It is unlikely but cannot be completely excluded, that small
dendritic protrusions might have been missed as a result of the weak dendrite labeling on
the second day.
VOLUME 11 [ NUMBER 9 [ SEPTEMBER 2008 NATURE NEUROSCIENCE
 ARTICLES

a Axonal protrusions b Dendritic protrusions c Glutamatergic bouton Figure 5 Lifetime of dendritic and axonal
protrusions. (a) The mean (upper) and cumulative
* 0 hr 0 hr
1.6 1.6 distributions (lower) of the lifetimes of axonal
1.4 1.4 protrusions that contacted a labeled dendrite were
1.2 1.2 not different from those of arbitrarily selected
Lifetime (h)

Lifetime (h)
1.0 1.0
0.8 0.8 axonal protrusions. (b) The mean lifetime (upper)
0.6 0.6 1.5 hr 1.5 hr of dendritic protrusions that contacted a
0.4 0.4 DP GABAergic bouton was significantly shorter than
0.2 25 85 0.2 19 110
0.0 0.0 that of the arbitrarily chosen dendritic protrusions
Contacting Arbitrarily Contacting Arbitrarily (P ¼ 0.03, Mann-Whitney U test). The cumulative
labeled dendrite chosen GABAergic bouton chosen
distributions also clearly show the shorter
© 2008 Nature Publishing Group http://www.nature.com/natureneuroscience

2 hr 2 hr
DP lifetimes of GABA-contacting protrusions (lower,
1.0 1.0 gray solid lines; P o 0.01, w2 test). Dendritic
Cumulative probability

Cumulative probability

0.8 0.8 protrusions that contacted a glutamatergic bouton

showed a much broader range of lifetimes, with
0.6 0.6 5 hr 5 hr
Contacting GABAergic
some short- and some long-lived protrusions
DP
0.4
Contacting labeled
0.4 bouton (lower, gray dotted line; significantly different from
Arbitrarily chosen
0.2 dendrite 0.2 gray solid line, P o 0.001, w2 test). (c) Example
Contacting
Arbitrarily chosen glutamatergic bouton of a long-lasting contact made by a dendritic
0.0 0.0
0 1 2 3 4 5 6 0 1 2 3 4 5 6 Dendritic protusion makes contact protrusion and a glutamatergic bouton. The
Lifetime (h) Lifetime (h) Protrusion without contact dendritic protrusion and the contact were still
present at the end of the imaging period. Scale
d Protrusions contacting GABAergic boutons e Protrusions contacting glutamatergic boutons bar represents 1 mm. (d,e) Time course of contact
1.0 1.0
formation for dendritic protrusions with GABAergic
(d, n ¼ 19) and glutamatergic (e, n ¼ 20)
Normalized overlap

Normalized overlap

0.8 0.8 boutons. Individual protrusions are represented in

0.6 0.6 different colors and the average time course is
given by the thick black curve. The numbers of
0.4 0.4
overlapping pixels between protrusion and bouton
0.2 0.2 (normalized to maximum overlap) are plotted for
all time points, aligned to the time of first contact.
0.0 0.0
Contacts with GABAergic boutons were always
–3 –2 –1 0 1 2 3 4 5 –3 –2 –1 0 1 2 3 4 5
Time (h) Time (h)
short lived, whereas longer-lived and repeated
contacts occurred with glutamatergic boutons.

(total 1,317 boutons, 12 slices) lacked a synaptic marker. This indicates Axon-dendrite crossings
that, in general, GABAergic boutons reliably reflect GABAergic synapses. Our data show that new GABAergic synapses are formed at pre-existing
During 3 h of time-lapse imaging, most GABAergic boutons were axon-dendrite crossings. Taking this result into consideration, a flexible,
present for the entire imaging period and only 5–10% of boutons were modifiable GABAergic system would require a considerable fraction of
plastic, that is, appeared or disappeared. Post hoc immunostaining a axon-dendrite crossings that do not (yet) have synaptic contacts so that
bouton that appeared at t ¼ 1.5 h during the imaging period (Fig. 6a,b) new synapses can still be formed when needed. We therefore estimated
showed that, at the time of fixation (2 h after first appearance), the new the percentage of GABAergic axon-dendrite crossings with and without
bouton (yellow circle) had accumulated VGAT, while the postsynaptic boutons. We found that only 42 ± 8% (mean ± s.d.; 21 slice cultures) of
scaffolding protein gephyrin was not (yet) present (Fig. 6b). Stable axon-dendrite crossings contained one or more boutons. This indicates
boutons on the same and nearby axons (blue circles) were immuno- that there are still a substantial number of available axon-dendrite
positive for VGAT and gephyrin (Fig. 6b), presumably reflecting stable crossings where new GABAergic contacts can be formed.
GABAergic synapses. Notably, the percentage of axon-dendrite crossings with boutons was
When we examined multiple new boutons that appeared at various similar in all slices (range of 30–56%, with one exception of 21%; 21
intervals before fixation, we found that the majority of new boutons slices) and did not depend on the total number of crossings. This
acquired VGAT in the first hour after appearance (all time points are suggests that the fraction of axon-dendrite crossings that have synapses
different from background, P o 0.01, w2 test), whereas gephyrin may be tightly controlled by the GABAergic system so that the
accumulated at a slower rate (Fig. 6c). We found that B70% of new generation of new synaptic contacts is always permitted.
boutons were immunopositive for both synaptic markers 3 h after their
first appearance, presumably reflecting new GABAergic synapses DISCUSSION
(Fig. 6d). At locations where boutons had disappeared during the The overwhelming majority of studies on synapse formation in the
two-photon imaging period (at various intervals before fixation), the CNS have been carried out on glutamatergic synapses. However,
presynaptic protein VGAT was not present above background levels, besides the greater part of synapses, which are glutamatergic and
whereas postsynaptic gephyrin could occasionally be found up to 1 h often located on spines, a substantial fraction of synapses in the
after the bouton disappeared (data not shown). This might indicate that brain are GABAergic. These synapses are located mostly (although
presynaptic changes precede postsynaptic alterations during GABAergic not exclusively) on dendritic shafts. Our experiments were conducted
synapse formation and disassembly, but more detailed studies will be in vitro on a specific subset of GABAergic neurons. Therefore, care needs
needed to fully resolve these processes. Notably, our findings indicate to be taken before generalizing these results to all GABAergic axons and/
that a large fraction of newly appearing GABAergic boutons forms or the in vivo situation. Nevertheless, we demonstrate that, at least in the
new GABAergic synapses with a time constant on the order of CA1 region of the hippocampus and under our experimental circum-
a few hours. stances, GABAergic synapses are formed via a fundamentally different

NATURE NEUROSCIENCE VOLUME 11 [ NUMBER 9 [ SEPTEMBER 2008 1049

 ARTICLES
a Figure 6 New GABAergic boutons form synapses.
0h 0.5 h 1h 1.5 h
b Merge GFP
© 2008 Nature Publishing Group http://www.nature.com/natureneuroscience
c VGAT
Gephyrin d One marker
100% 100%
80%
60%
40%
20%
27 21 15 1317 575
0%
0.5–1 h 1.5–2 h 2.5–3 h No Randomized
Time present time lapse
before fixation (h)
process than glutamatergic spine synapses. Although, in the case of
glutamatergic synapses, dendrites grow small protrusions, such as spines
and filopodia, to form new synaptic contacts with nearby axons, such
protrusions do not seem to mediate the formation of GABAergic
synapses. In contrast, new GABAergic synapses appear to be formed
exclusively at locations where GABAergic axons and postsynaptic
dendrite are already in close proximity. Although the above caveat
about preparation and molecular/genetic cellular identity holds, our
data make a strong case that GABAergic and glutamatergic synapse
formation occur according to very different schemes.
Our immunofluorescence labeling (Figs. 1 and 6) showed that the
great majority of stable GABAergic boutons probably represent actual
synapses. However, from the two-photon imagery, it is impossible to
definitively tell whether a GABAergic bouton synapses onto the
labeled dendrite or onto another nonlabeled nearby dendrite. Indeed,
B15–20% of the light microscopy–identified contacts were close to
gephyrin-labeled structures that were not on the labeled dendrite
but were instead elsewhere adjacent to the bouton. This drawback,
inherent to conventional light-microscopy imaging, does not conflict
with or devaluate our main observation that new GABAergic
boutons are formed at pre-existing axon-dendrite crossings without
mediating protrusions.
Our post hoc immunostaining of newly formed GABAergic boutons
showed that boutons can acquire pre- and postsynaptic markers in a
period of several hours after their first appearance, suggesting that new
functional GABAergic synapses can be formed on a similar time scale as
has been previously shown for glutamatergic synapses23,25,29,30. How-
ever, for both types of synapses, the formation of morphologically fully
mature synapses (with membrane specializations as observed at the
electron-microscopy level) may take longer, perhaps even days7,8,23,34.
1050
(a) Time lapse two-photon images (projection of
three z sections) showing the appearance of a new
bouton (yellow circle) on a GFP-expressing
GABAergic axon. (b) Post hoc immunostaining
(projection of three z sections) of the same region
as in a. The dashed box is enlarged on the right
side (merge) and the separate channels are
2h 2.5 h 3h
shown. The new bouton (yellow circle) was
VGAT Geph immunopositive for VGAT (blue), but not for
gephyrin (red). Stable boutons (blue circles) were
immunopositive for both markers (appearing white
in merged image). Scale bars represent 1 mm.
(c) Left, percentage of new boutons that were
immunopositive for VGAT (black bars) and
gephyrin (gray bars), grouped according to their
lifetime before fixation. Right, percentage of
boutons that were immunopositive for VGAT and
No marker
gephyrin in slices in which no previous time lapse
Two markers imaging was performed. Background staining
levels of boutons were determined after horizontal
or vertical reflection of the VGAT and gephyrin
80% channels (‘randomized’). (d) Same data as in c,
*** but the bars indicate the number of synaptic
60% markers that were present at each bouton.
*** * P o 0.05, *** P o 0.001, w2 test, difference
40% from background. The numbers at the base of the
*
bars reflect the number of boutons examined.
20%
27 21 15 1317 575
0% Dendritic protrusions that contacted
0.5–1 h 1.5–2 h 2.5–3 h No Randomized
Time present time lapse GABAergic boutons had substantially shorter
before fixation (h) lifetimes than protrusions that contacted glu-
tamatergic boutons. This suggests that only
short-lived protrusions make contact with
GABAergic boutons, that contact with GABAergic boutons promotes
the retraction of dendritic protrusions or, perhaps most likely, that
contact with glutamatergic boutons can prolong the lifetime of den-
dritic protrusions. It has been shown that dendritic protrusions can
detect and grow toward glutamate35,36 and that dendritic protrusions
often make contacts with pre-existing glutamatergic boutons that are
already part of a synapse7,8. Stabilization of dendritic protrusions has
been shown to be correlated with (appearance or) growth of the
postsynaptic density, and this process depends on the activation of
glutamatergic receptors9. This suggests that contact with glutamatergic
boutons indeed stabilizes dendritic protrusions. The observation that
some spines have GABAergic contacts seems to argue against GABA as a
retraction signal, although previous studies have shown that spines with
GABAergic synapses usually also have a glutamatergic synapse3,26,37,38,
potentially stabilizing the spine and ‘protecting’ it against negative
GABAergic effects. In any case, our findings clearly indicate that
outgrowing dendritic protrusions can distinguish between potential
presynaptic partners, even before synapses are formed10,11. It should
be kept in mind that the mechanism for the formation of new
GABAergic synapses could change with changing circumstances.
For instance, during early development, when GABA is depolarizing
and can induce calcium influx, GABA might stabilize dendritic protru-
sions, possibly leading to the formation of GABAergic synapses on
dendritic protrusions.
Our main finding is that new GABAergic synapses appear to be
exclusively formed by the appearance of new boutons at pre-existing
axon-dendrite crossings. Although this is in marked contrast with the
formation of glutamatergic synapses, in which the outgrowth of
dendritic protrusions often has a major role, it is not unique, as
glutamatergic axons can also form new boutons at pre-existing spines39.
VOLUME 11 [ NUMBER 9 [ SEPTEMBER 2008 NATURE NEUROSCIENCE

A recent study showed that axons appear to have preferred locations for
forming new boutons, even in the absence of postsynaptic structure40.
At present, it is not clear how the locations for new boutons are deter-
mined. One possibility is that these locations are determined by the
presence of cell adhesion or other signaling molecules. Indeed, recent
studies have shown that the formation of GABAergic and glutamatergic
synapses is regulated by different synaptic adhesion molecules41–43.
Differential expression of such molecules on dendritic shafts and
protrusions could provide an explanation for the different mechanisms
© 2008 Nature Publishing Group http://www.nature.com/natureneuroscience
underlying the formation of glutamatergic and GABAergic synapses.
Our findings imply that GABAergic axons in a mature network can
make new synapses only with postsynaptic partners that are in their
immediate neighborhood, which is in marked contrast to the way
glutamatergic synapses are made. Our findings do not address whether
GABAergic and glutamatergic synapses differ in their capacity to
undergo plasticity, but they suggest that plasticity in GABAergic
connections is more restricted than that in glutamatergic connections.
The degree of this restriction depends on the number of available axon-
dendrite crossings. An adaptable GABAergic system requires that
GABAergic axons cross and touch the dendrites of a sufficiently large
number of potential postsynaptic partners. We found that, in our slice
cultures, B60% of axon-dendrite crossings do not have a GABAergic
bouton and therefore leave ample opportunity for generating new
GABAergic connections. Obviously, plasticity can also be achieved by
changing the strength of pre-existing synapses, but so far as synaptic
strength is controlled by synaptogenesis, the different rules for plasticity
in the mature network may require different developmental strategies
for GABAergic and glutamatergic axons44,45. In this context, it should
be noted that anatomical studies have shown that many GABAergic
axons are highly complex with often tortuous paths, whereas glutama-
tergic axons tend to follow more linear trajectories45–47. A theoretical
analysis of the trajectories of axons revealed that GABAergic axons have
substantially larger overlap with the dendritic trees of their potential
target neurons than expected from chance, whereas this is not the case
for glutamatergic axons44. These findings suggest that glutamatergic
axons grow relatively straight, as nearby postsynaptic partners can later
grow spines to form synaptic contacts. Conversely, GABAergic axons
may cross and touch dendrites of many postsynaptic neurons to
enhance their future potential for synapse formation.
Taken together, our findings imply that fundamentally different
mechanisms underlie the generation of GABAergic and glutamatergic
synapses, at least as far as the involvement of dendritic and axonal
protrusions are concerned. Glutamatergic synapses use short connecting
processes such as filopodia and spines to enable new synaptic connec-
tions, whereas GABAergic synapse rely on pre-existing axon-dendrite
crossing. This puts substantial structural constraints on the generation
and plasticity of GABAergic and glutamatergic synapses, which will be
important to keep in mind when trying to understand development and
plasticity of the intricate neural networks of the brain.
METHODS
Cultures. Hippocampal slices (300 mm thick) were prepared from postnatal day
2–5 GAD65-GFP mice17 and maintained using the roller tube technique48. In
these mice, 30–50% of hippocampal GABAergic interneurons express GFP,
with a steady expression level from early embryonic age to adulthood17. Slices
were kept in culture for at least 1 week before the experiments (range, 7–20 d
in vitro; mean ± s.d., 12.2 ± 3.8 d in vitro). At this developmental stage, the
hippocampal network is mostly mature, but synapse formation still takes place.
For the experiments, cultures were transferred into a recording chamber,
where they were continuously perfused with carbogenated (95% O2, 5% CO2)
artificial cerebrospinal fluid containing 126 mM NaCl, 3 mM KCl, 2.5 mM
CaCl2, 1.3 mM MgCl2, 1.25 mM NaH2PO4, 26 mM NaHCO3, 20 mM glucose,
NATURE NEUROSCIENCE VOLUME 11 [ NUMBER 9 [ SEPTEMBER 2008
ARTICLES
1 mM pyruvate and 1 mM Trolox. The temperature was maintained at 35 1C.
CA1 pyramidal neurons were filled through a patch pipette and 250 mM Alexa
Fluor 594 was added to the pipette solution.
For experiments with labeled glutamatergic axons, slice cultures of wild-type
C57 BL/6 mice were used. Glutamatergic axons were labeled via extracellular
bolus loading from a pipette in the CA3 pyramidal layer, as will be described
elsewhere (N.B., C.J.W., Fonseca, R., T.B. and Nägerl, U.V., unpublished data)29.
Imaging. Time-lapse images were acquired using a custom-made two-photon
laser-scanning microscope based on an Olympus IX70 microscope with a 40
,
1.2-NA water-immersion objective (Olympus). GFP and Alexa Fluor 594 were
simultaneously excited using a laser beam tuned to 855 nm (5W Mira-Verdi
laser system, Coherent). Laser power at the objective was 4–5 mW. Fluorescence
was detected by external photomultipliers (Hamamatsu).
The imaged regions varied between 100 and 140 mm (1,024
1,024 pixels)
and stacks consisted of 60–99 z layers (step size Dz ¼ 0.5 mm), depending on
the overlap between GFP axons and labeled dendrites. Images were taken every
30 min, for a total of 3–6 h. Small misalignments of images between time
points as a result of drift were compensated for during the analysis.
Data analysis. Image stacks were visually inspected in ImageJ (US National
Institutes of Health) to determine all of the locations at which a GFP-labeled
axon or bouton was in close proximity to labeled dendritic structures. These
locations were subsequently examined in detail at all time points and z sections
to detect possible contacts using software written in Matlab (Mathworks). To
reduce noise, images were filtered with a 2
2-pixel Wiener filter. Analysis was
always carried out on individual image sections. Volume renderings were made
in Imaris 5.0.1 (Bitplane) and were used for illustrational purposes only.
The overlap between the red and the green channel was determined after
thresholding the channels independently. Thresholds were empirically defined
by a semi-automatic two-step procedure. In the first step, the background was
distinguished from neuronal structures such as dendrites and axons (threshold
¼ mean + 3 s.d. of intensity values of a local 200
200-pixel area). In the
second step, boutons were distinguished from axonal shafts (threshold ¼ mean
+ 1.5 s.d. of axon pixel intensities). In this way, thresholds were objective and
automatically corrected for bleaching (B5% in the green and 15–20% in the
red channel over the entire imaging period). Nevertheless, we also verified that
changing these thresholds only marginally changed the numbers but did not
affect our general conclusions.
For simplicity, we refer to all axonal swellings or varicosities as boutons,
although we are aware that some of them might not contain presynaptic
specializations. Conversely, some small presynaptic structures might go unno-
ticed by this procedure. Furthermore, to avoid ambiguities, all dendritic
structures that formed during our observation are simply denoted as protru-
sions, irrespective of whether they were filopodia like or spine like. We use the
term ‘spine’ for stable dendritic structures that were present before and during
the entire imaging period.
Contacts between a GABAergic bouton and a labeled dendrite were
considered to be physical contacts only if the red and the green channel had
overlapping pixels in at least two z sections and if the overlap covered at least
10% of the bouton pixels (mean ± s.d. bouton volume ¼ 126 ± 86 pixels, n ¼
221 boutons). For contact with spines, the value of 10% of the smaller structure
(that is, bouton or spine) was used. Because of their small size and low
fluorescence levels, outgrowing dendritic or axonal protrusions were considered
to be making contact when they showed any overlap (Z1 pixel) with a labeled
bouton or dendrite, respectively. The minimal volume for new or transient
boutons was set at 15 pixels.
The lifetime of transient contacts and protrusions were determined by
counting each time point at which the contact or protrusion had been present
at 30 min. In principle, for every time point, the actual lifetime could have been
between 1 and 59 min. One should also realize that lifetimes beyond the
imaging period (3–6 h) are artificially shortened by the end of the imaging
period and therefore the distributions (Fig. 5a,b) are inevitably biased toward
shorter lifetimes.
Immunofluorescence staining. For post hoc immunofluorescence analysis, the
GFP signal in GABAergic axons was amplified by immunofluorescence staining
(chicken antibody to GFP, 1:1,000, Chemicon). GABAergic synapses were
1051
 ARTICLES
visualized by fluorescent labeling of VGAT (rabbit antibody to VGAT, 1:200,
Synaptic Systems) and the postsynaptic scaffold protein gephyrin (mouse
antibody to gephyrin, 1:400, Synaptic Systems). Slice cultures were fixed in
4% paraformaldehyde (wt/vol, pre-warmed to 35 1C) for 4 h at 4 1C, washed
extensively in 0.1 M phosphate buffer and removed from their coverslips to be
processed as free-floating slices. Permeabilization and blocking was achieved by
incubation of the sections for 24 h at 4 1C in 0.1 M phosphate buffer,
0.4% Triton X-100 (vol/vol) and 10% horse serum (vol/vol). Primary anti-
bodies were applied overnight at 4 1C in 0.1 M phosphate buffer with 0.4%
Triton and 5% horse serum. Following extensive washing, appropriate second-
© 2008 Nature Publishing Group http://www.nature.com/natureneuroscience
ary antibodies (Molecular Probes, Invitrogen) were applied overnight at 4 1C at
a concentration of 1:200. For the immunostaining of contacts after two-photon
imaging, patch pipettes contained an additional 0.1% biocytin, which was
visualized after fixation by Avidin Texas Red (Vector Labs).
Immunofluorescence labeling was analyzed from high-resolution confocal
image stacks (Dz ¼ 0.35–0.4 mm). The association of GFP-labeled boutons with
pre- and postsynaptic markers was assessed in individual z sections after
filtering and thresholding each channel independently.
Statistics. Unless otherwise stated, means are always given ± standard error.
Means were compared using the Mann-Whitney U test. Distributions were
compared using the w2 test. Multiple comparisons were made by an ANOVA
test, followed by a Tukey Honestly Significant Difference (HSD) test.
Note: Supplementary information is available on the Nature Neuroscience website.
ACKNOWLEDGMENTS
We would like to thank G. Szábo for kindly providing the GAD65-GFP mice,
U.V. Nägerl for help with the experimental setup and comments on the
manuscript, N. Stöhr and C. Huber for technical assistance, and T. Mrsic-Flögel,
C. Lohmann and V. Stein for critical reading of the manuscript. This work was
supported by the Max Planck Gesellschaft, the Alexander von Humboldt
Stiftung, a Marie Curie Intra-European fellowship (C.J.W.) and the Boehringer
Ingelheim Fonds (N.B.).
AUTHOR CONTRIBUTIONS
C.J.W. designed and conducted the experiments and analyzed the data. N.B.
carried out the experiments on glutamatergic boutons. C.J.W. and T.B. conceived
the project and wrote the manuscript.
Published online at http://www.nature.com/natureneuroscience/
Reprints and permissions information is available online at http://npg.nature.com/
reprintsandpermissions/
1. Freund, T.F. & Buzsaki, G. Interneurons of the hippocampus. Hippocampus 6, 347–470
(1996).
2. Harris, K.M. & Kater, S.B. Dendritic spines: cellular specializations imparting both
stability and flexibility to synaptic function. Annu. Rev. Neurosci. 17, 341–371 (1994).
3. Megı́as, M., Emri, Z., Freund, T.F. & Gulyás, A.I. Total number and distribution of
inhibitory and excitatory synapses on hippocampal CA1 pyramidal cells. Neuroscience
102, 527–540 (2001).
4. Ziv, N.E. & Smith, S.J. Evidence for a role of dendritic filopodia in synaptogenesis and
spine formation. Neuron 17, 91–102 (1996).
5. Fiala, J.C., Feinberg, M., Popov, V. & Harris, K.M. Synaptogenesis via dendritic filopodia
in developing hippocampal area CA1. J. Neurosci. 18, 8900–8911 (1998).
6. Petrak, L.J., Harris, K.M. & Kirov, S.A. Synaptogenesis on mature hippocampal
dendrites occurs via filopodia and immature spines during blocked synaptic transmis-
sion. J. Comp. Neurol. 484, 183–190 (2005).
7. Knott, G.W., Holtmaat, A., Wilbrecht, L., Welker, E. & Svoboda, K. Spine growth
precedes synapse formation in the adult neocortex in vivo. Nat. Neurosci. 9,
1117–1124 (2006).
8. Nägerl, U.V., Kostinger, G., Anderson, J.C., Martin, K.A. & Bonhoeffer, T. Protracted
synaptogenesis after activity-dependent spinogenesis in hippocampal neurons.
J. Neurosci. 27, 8149–8156 (2007).
9. De Roo, M., Klauser, P., Mendez, P., Poglia, L. & Muller, D. Activity-dependent PSD
formation and stabilization of newly formed spines in hippocampal slice cultures. Cereb.
Cortex 18, 151–161 (2008).
10. Lohmann, C., Finski, A. & Bonhoeffer, T. Local calcium transients regulate the
spontaneous motility of dendritic filopodia. Nat. Neurosci. 8, 305–312 (2005).
11. Lohmann, C. & Bonhoeffer, T. A role for local calcium signaling in rapid synaptic partner
selection by dendritic filopodia. Neuron 59, 253–260 (2008).
12. Dailey, M.E. & Smith, S.J. The dynamics of dendritic structure in developing hippo-
campal slices. J. Neurosci. 16, 2983–2994 (1996).
13. Marrs, G.S., Green, S.H. & Dailey, M.E. Rapid formation and remodeling of postsynaptic
densities in developing dendrites. Nat. Neurosci. 4, 1006–1013 (2001).
1052
14. Saito, Y., Song, W.J. & Murakami, F. Preferential termination of corticorubral axons on
spine-like dendritic protrusions in developing cat. J. Neurosci. 17, 8792–8803 (1997).
15. Yuste, R. & Bonhoeffer, T. Genesis of dendritic spines: insights from ultrastructural and
imaging studies. Nat. Rev. Neurosci. 5, 24–34 (2004).
16. Somogyi, P., Tamás, G., Lujan, R. & Buhl, E.H. Salient features of synaptic organization
in the cerebral cortex. Brain Res. Brain Res. Rev. 26, 113–135 (1998).
17. López-Bendito, G. et al. Preferential origin and layer destination of GAD65-GFP cortical
interneurons. Cereb. Cortex 14, 1122–1133 (2004).
18. Kalisman, N., Silberberg, G. & Markram, H. The neocortical microcircuit as a tabula
rasa. Proc. Natl. Acad. Sci. USA 102, 880–885 (2005).
19. Shepherd, G.M. & Harris, K.M. Three-dimensional structure and composition of CA3-
CA1 axons in rat hippocampal slices: implications for presynaptic connectivity and
compartmentalization. J. Neurosci. 18, 8300–8310 (1998).
20. Meyer, M.P. & Smith, S.J. Evidence from in vivo imaging that synaptogenesis guides the
growth and branching of axonal arbors by two distinct mechanisms. J. Neurosci. 26,
3604–3614 (2006).
21. De Paola, V. et al. Cell type–specific structural plasticity of axonal branches and boutons
in the adult neocortex. Neuron 49, 861–875 (2006).
22. Ruthazer, E.S., Li, J. & Cline, H.T. Stabilization of axon branch dynamics by synaptic
maturation. J. Neurosci. 26, 3594–3603 (2006).
23. Ahmari, S.E., Buchanan, J. & Smith, S.J. Assembly of presynaptic active zones from
cytoplasmic transport packets. Nat. Neurosci. 3, 445–451 (2000).
24. Friedman, H.V., Bresler, T., Garner, C.C. & Ziv, N.E. Assembly of new individual
excitatory synapses: time course and temporal order of synaptic molecule recruitment.
Neuron 27, 57–69 (2000).
25. Bresler, T. et al. Postsynaptic density assembly is fundamentally different from
presynaptic active zone assembly. J. Neurosci. 24, 1507–1520 (2004).
26. Kubota, Y., Hatada, S., Kondo, S., Karube, F. & Kawaguchi, Y. Neocortical inhibitory
terminals innervate dendritic spines targeted by thalamocortical afferents. J. Neurosci.
27, 1139–1150 (2007).
27. Knott, G.W., Quairiaux, C., Genoud, C. & Welker, E. Formation of dendritic spines with
GABAergic synapses induced by whisker stimulation in adult mice. Neuron 34,
265–273 (2002).
28. Risher, W.C., Ostroff, L.E. & Harris, K.M. What dendritic filopodia induced by LTP
encounter along their path through the neuropil of PN15 rat hippocampus. Abstr. Soc.
Neurosci. 135.4 (2006).
29. Ziv, N.E. & Garner, C.C. Cellular and molecular mechanisms of presynaptic assembly.
Nat. Rev. Neurosci. 5, 385–399 (2004).
30. Gerrow, K. et al. A preformed complex of postsynaptic proteins is involved in excitatory
synapse development. Neuron 49, 547–562 (2006).
31. Maas, C. et al. Neuronal cotransport of glycine receptor and the scaffold protein
gephyrin. J. Cell Biol. 172, 441–451 (2006).
32. Chattopadhyaya, B. et al. GAD67-mediated GABA synthesis and signaling regulate
inhibitory synaptic innervation in the visual cortex. Neuron 54, 889–903 (2007).
33. Di Cristo, G. et al. Activity-dependent PSA expression regulates inhibitory maturation
and onset of critical period plasticity. Nat. Neurosci. 10, 1569–1577 (2007).
34. Ahmari, S.E. & Smith, S.J. Knowing a nascent synapse when you see it. Neuron 34,
333–336 (2002).
35. Portera-Cailliau, C., Pan, D.T. & Yuste, R. Activity-regulated dynamic behavior of early
dendritic protrusions: evidence for different types of dendritic filopodia. J. Neurosci. 23,
7129–7142 (2003).
36. Richards, D.A. et al. Glutamate induces the rapid formation of spine head protrusions in
hippocampal slice cultures. Proc. Natl. Acad. Sci. USA 102, 6166–6171 (2005).
37. de Zeeuw, C.I., Ruigrok, T.J., Holstege, J.C., Jansen, H.G. & Voogd, J. Intracellular
labeling of neurons in the medial accessory olive of the cat. II. Ultrastructure of dendritic
spines and their GABAergic innervation. J. Comp. Neurol. 300, 478–494 (1990).
38. Cope, D.W. et al. Cholecystokinin-immunopositive basket and Schaffer collateral-
associated interneurones target different domains of pyramidal cells in the CA1 area
of the rat hippocampus. Neuroscience 109, 63–80 (2002).
39. Pratt, K.G., Taft, C.E., Burbea, M. & Turrigiano, G.G. Dynamics underlying synaptic
gain between pairs of cortical pyramidal neurons. Dev. Neurobiol. 68, 143–151
(2008).
40. Sabo, S.L., Gomes, R.A. & McAllister, A.K. Formation of presynaptic terminals at
predefined sites along axons. J. Neurosci. 26, 10813–10825 (2006).
41. Graf, E.R., Zhang, X., Jin, S.X., Linhoff, M.W. & Craig, A.M. Neurexins induce
differentiation of GABA and glutamate postsynaptic specializations via neuroligins.
Cell 119, 1013–1026 (2004).
42. Chih, B., Engelman, H. & Scheiffele, P. Control of excitatory and inhibitory synapse
formation by neuroligins. Science 307, 1324–1328 (2005).
43. Chubykin, A.A. et al. Activity-dependent validation of excitatory versus inhibitory
synapses by neuroligin-1 versus neuroligin-2. Neuron 54, 919–931 (2007).
44. Stepanyants, A., Tamás, G. & Chklovskii, D.B. Class-specific features of neuronal wiring.
Neuron 43, 251–259 (2004).
45. Thomson, A.M. & Morris, O.T. Selectivity in the inter-laminar connections made by
neocortical neurones. J. Neurocytol. 31, 239–246 (2002).
46. Karube, F., Kubota, Y. & Kawaguchi, Y. Axon branching and synaptic bouton pheno-
types in GABAergic nonpyramidal cell subtypes. J. Neurosci. 24, 2853–2865
(2004).
47. Shepherd, G.M., Raastad, M. & Andersen, P. General and variable features of varicosity
spacing along unmyelinated axons in the hippocampus and cerebellum. Proc. Natl.
Acad. Sci. USA 99, 6340–6345 (2002).
48. Gähwiler, B.H. Organotypic monolayer cultures of nervous tissue. J. Neurosci. Methods
4, 329–342 (1981).
VOLUME 11 [ NUMBER 9 [ SEPTEMBER 2008 NATURE NEUROSCIENCE