Diffusion Barriers Constrain Receptors at Synapses

The flux of neurotransmitter receptors in and out of synapses depends on receptor interaction with scaffolding molecules. However, the crowd of transmembrane proteins and the rich cytoskeletal environment may constitute obstacles to the diffusion of receptors within the synapse. To address this question, we studied the membrane diffusion of the γ-aminobutyric acid type A receptor (GABAAR) subunits clustered (γ2) or not (α5) at inhibitory synapses in rat hippocampal dissociated neurons. Relative to the extrasynaptic region, γ2 and α5 showed reduced diffusion and increased confinement at both inhibitory and excitatory synapses but they dwelled for a short time at excitatory synapses. In contrast, γ2 was ∼3-fold more confined and dwelled ∼3-fold longer in inhibitory synapses than α5, indicating faster synaptic escape of α5. Furthermore, using a gephyrin dominant-negative approach, we showed that the increased residency time of γ2 at inhibitory synapses was due to receptor-scaffold interactions. As shown for GABAAR, the excitatory glutamate receptor 2 subunit (GluA2) of the α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor (AMPAR) had lower mobility in both excitatory and inhibitory synapses but a higher residency time at excitatory synapses. Therefore barriers impose significant diffusion constraints onto receptors at synapses where they accumulate or not. Our data further reveal that the confinement and the dwell time but not the diffusion coefficient report on the synapse specific sorting, trapping and accumulation of receptors.


Introduction
Concentration of neurotransmitter receptors in the postsynaptic membrane critically determines the efficacy of fast neurotransmission. Lateral diffusion in and outside synapses plays a key role in the regulation of receptor number at synapses [1,2,3]. Receptors continuously exchange between synaptic and extrasynaptic membranes. They display rapid free Brownian diffusion in extrasynaptic membrane and are slowed down and confined (restricted in space) at synapses. The restricted motion at synapses result from transient interactions of receptors with scaffolding molecules directly or indirectly bound to the cytoskeleton, a phenomenon also referred as the ''diffusion-capture'' mechanism (recently reviewed by [2]). Besides, the presence of obstacles in the synapse may also reduce the mobility of receptors. Obstacles to diffusion are created by a crowd of transmembrane proteins immobilized at the synapse through binding to the cytoskeleton [4]. These proteins include the receptors themselves as well as adhesion molecules that connect pre-and post-synaptic membranes such as the neurexin-neuroligin complex [5] and synaptic cell adhesion molecules with homophilic interactions (SynCam [6], Sidekicks [7] and cadherins [8,9]). Also, cadherins by interacting with b-catenins can organize nanometers sized subdomains around the synapse [10]. The cytoskeletal fences constituted by the high density of cytoplasmic cytoskeletal elements present in the postsynaptic density are another source of obstacles to the diffusion of receptors. The membrane-associated portion of cytoskeletal proteins forms a corral from which transmembrane proteins can only escape by hop diffusion or by passing through gaps when the cytoskeleton is discontinuous (refs. in [11]). Last, receptor diffusion at synapses may be limited by their transient association with specialized lipid microdomains, the so-called ''lipid rafts'' [12,13].
Although it has been demonstrated that obstacles alter the diffusion of lipids at synapses [14], this question has not been addressed for neurotransmitter receptors. Here we used quantum dot (QD)-based single particle tracking (SPT) as well as fluorescence recovery after photobleaching (FRAP) to compare diffusion behavior of GABA A Rs and AMPARs at excitatory and inhibitory synapses. Our main results showed that GABA A R and AMPAR diffusion is significantly reduced in synapses whatever the neurotransmitter contained in the presynaptic element. This suggests the lateral diffusion of receptors at mismatched synapses is hindered by the presence of diffusion barriers such as pickets and fences. Although the diffusion coefficients were similar at matching and mismatched PSDs, the explored area and dwell time of reflected receptor trapping at matching PSDs. Moreover, interfering with the clustering of the main inhibitory scaffolding protein gephyrin revealed that the receptor-scaffold interaction was responsible for the increased confinement and dwell time of GABA A R at inhibitory synapses, although no noticeable changes were observed in the diffusion coefficients. Thus, the explored area and dwell time, but not diffusion coefficient, are correlated with the synaptic sorting, trapping and concentration of receptors.

Results
Diffusion of the GABA A Rs c2 and a5 Subunits at Inhibitory Synapses GABA A Rs diffuse laterally on the neuronal plasma membrane and rapidly exchange between extrasynaptic and synaptic loci [15,16,17,18,19,20,21]. Using QD-based SPT [22,23], we analyzed the mobility of the GABA A R c2 subunit enriched at inhibitory postsynaptic site [24,25,26] and of the GABA A R a5 subunit found almost exclusively at extrasynaptic sites [27,28,29,30]. The surface endogenous GABA A R c2 and a5 subunits were labeled with specific antibodies directed against their extracellular N-terminus regions, subsequently labeled with specific intermediate biotinylated Fab fragments and streptavidin-coated QD (see Materials and Methods). The cell surface exploration of c2 and a5 was visualized on trajectories reconstructed from 38.4 s recording sequences (e.g. Fig. 1A). To identify inhibitory synapses in live cells, neurons were transfected with a venus-gephyrin construct as done before [31,32]. Extrasynaptic vs. synaptic trajectories were segregated by comparison with merge fluorescence images of venus-gephyrin and FM 4-64. Trajectories were at inhibitory synapses when overlapping with venus-gephyrin and FM 4-64 punctae (e.g. grey in Fig. 1A1, 1B1), or extrasynaptic for trajectories two pixels (440 nm) away (e.g. black in Fig. 1A1, 1B1) [22]. As exemplified in Fig. 1A1 for c2, the surface exploration of the same QD was restricted to a smaller area at the synaptic vs. extrasynaptic compartment. This was also visible on the mean-square displacement function (MSD) versus time relation which showed a steeper slope in the extrasynaptic compartment (Fig. 1A2, same trajectory as in Fig. 1A1). Furthermore, the same QD had significantly lower mobility in synaptic vs. extrasynaptic compartment as seen on the instantaneous diffusion coefficient Dinst (Fig. 1A3). Quantitative analysis performed on a whole population of trajectories confirmed the reduced surface exploration of c2 at inhibitory synapses (10 cultures; Kolmogorov-Smirnov KS test, p,10 23 ; Fig. 1C-D; Table 1) and the reduced lateral diffusion (KS test, p,10 23 ; Fig. 1E; Table 1) of c2 at inhibitory synapses. These results are in agreement with earlier SPT work [18,19,20,21] and are coherent with a capture of c2 by the inhibitory scaffold. We then characterized the diffusion properties of a5. Despite their preferential non synaptic localization, a5-containing GABA A R also displayed reduced mobility at synapses (Fig. 1B1-3). Both the explored area (Fig. 1C-D and  Table 1; 6 cultures; KS test, p = 0.002) and diffusion coefficient ( Fig. 1E and Table 1; KS test, p,10 23 ) were reduced at inhibitory synapses vs. the extrasynaptic region. Therefore, GABA A Rs clustered (c2) or not (a5) at inhibitory synapses showed a drop in their diffusion coefficient and an increase in their confinement at inhibitory synapses. However, the explored area of a5 was 2.1 and 3.4 fold larger than that of c2 in extrasynaptic region and at inhibitory synapses, respectively (KS test, p,10 24 and p = 10 23 , respectively; Fig. 1D and Table 1), indicating a stronger confinement for c2.
Since the time receptors spend at the synapse influences synaptic receptor content [2], we compared the synaptic dwell time of GABA A R c2 and a5 subunits at inhibitory synapses. The mean dwell time of c2 was 3-fold that of a5 (t-test, p,10 24 ; Fig. 1F), indicating a faster exchange of a5 between extrasynaptic and synaptic locations. This is consistent with the notion that c2 but not a5 is enriched at inhibitory synapses. Actually, this means that the dwell time of a5 reflected the time required to get across an inhibitory synapse without being trapped by the scaffolding apparatus. As 95% of a5 trajectories had dwell time values less than 5.9 s, we used this threshold to segregate ''trapped'' receptors (dwell time.5.9 s) from ''passing'' ones (dwell time, = 5.9 s) (e.g. Fig. 2A). In agreement with a higher accumulation of c2 at inhibitory synapses, a larger proportion of c2 than a5 was trapped at inhibitory synapses (c2, 31.766.4%; a5, 5.162.0%; t-test, p = 0.04; Fig. 2B). Surprisingly, the proportion of trapped c2 was in minority at inhibitory synapses meaning that most c2 were passing or were interacting only transiently with the scaffold.
Altogether our results suggest the reduction of the mobility of a5 at inhibitory synapses involve obstacles and fences while that of c2 further implicate receptor-scaffold interactions. Furthermore, the dwell time but not the diffusion coefficient calculated on the global population of synaptic QDs (independent of their trapped vs passing behavior) allows distinguishing GABA A Rs accumulated (c2) or not (a5) at synapses.

Diffusion of the GABA A Rs c2 and a5 Subunits at Excitatory Synapses
We then asked whether the membrane diffusion of GABA A R c2 and a5 subunits was reduced within excitatory postsynaptic differentiations. The localization of QD trajectories at excitatory synapses was determined by comparison with merge fluorescence images of venus-gephyrin and FM 4-64. QD trajectories were respectively at inhibitory or at excitatory synapses when overlapping with FM 4-64 loaded presynaptic boutons colocalized or not with recombinant venus-gephyrin clusters. Trajectories localized on areas devoid of FM 4-64 and venus-gephyrin labeling were considered extrasynaptic. Relative to extrasynaptic regions, the lateral diffusion of c2 and a5 was reduced in excitatory synapses (10 cultures; KS test, p,10 24 ; a5, 6 cultures; KS test, p,10 23 ; Fig. 3 A-B; Table 1). Furthermore, the diffusion coefficient of c2 and a5 did not significantly differ in excitatory and inhibitory synapses (KS test, p = 0.2 and 0.8 respectively; Fig. 3 A-B; Table 1). However, the reduction of the explored area of c2 was more pronounced in inhibitory vs. excitatory synapses (KS test, p = 0.0002; Fig. 3C; Table 1). Furthermore, c2 resided ,2.9-fold longer in inhibitory synapses (t-test, p,10 23 ; Fig. 3D; Table 1), indicating a faster escape from excitatory synapses. This is coherent with the notion that c2 is anchored to the inhibitory but not to the excitatory postsynaptic scaffold. In contrast, the dwell time of a5 was not significantly different at excitatory and at inhibitory synapses (t-test, p = 0.3; Fig. 3D; Table 1). Interestingly, the average synaptic dwell time of a5 at inhibitory synapses was close to the one found for c2 at excitatory synapses (t-test, p = 0.8;  Table 1). We therefore concluded that the diffusion of c2 and a 5 was also hindered by the presence of obstacles in excitatory synapses. Moreover, the dwell time but not the diffusion coefficient calculated on the entire population of synaptic QDs unraveled the sorting of GABA A Rs at inhibitory synapses.
To verify that the constrained diffusion of GABA A R c2 at excitatory synapses was not due to the presence of QDs, we performed Fluorescence Recovery After Photobleaching (FRAP) experiments of the sensitive superecliptic pHluorin GABA A R c2 subunit (SEP-c2) in neurons double-transfected with Red fluorescent protein from Discosoma species (DsRed)-tagged homer1c. SEP-c2 and DsRed-homer1c fluorescent punctae (Fig. 4A) identified inhibitory and excitatory synapses, respectively. SEP-c2 fluorescence was ,3-fold higher in inhibitory vs. excitatory synapses or in extrasynaptic membrane (Fig. S1), indicating a specific concen-tration at inhibitory synapses. The SEP fluorescence is pHsensitive: it exhibits bright fluorescence when exposed at the cell surface and lower fluorescence in internal acidified trafficking organelles [33]. As shown before [16], the SEP-c2 fluorescence specifically reported surface c2 since neuronal exposure to pH4 buffer caused an instantaneous loss of nearly all fluorescence without main changes in monomeric Red Fluorescent Protein (mRFP)-tagged gephyrin fluorescence in co-transfected neurons (Fig. S1). Moreover, the eclipsed fluorescence rapidly returned in pH 7.4 buffer (Fig. S1). After photobleaching, the fluorescence of SEP-c2 clusters recovered to ,35% (34.361.3, n = 65) of the initial value within 2 min (Fig. 4 B), meaning that most SEP-c2 at inhibitory synapses was part of a stable pool that did not exchange during the course of the experiment. This contrasted with the fluorescence recovery in the extrasynaptic membrane which  Table 1. Diffusion properties of GABAAR c2 and a5 subunits, and of AMPAR GluA2 subunit in relation with excitatory and inhibitory synapses.  reached 8361% of the initial fluorescence value within 1 min (Fig. 4B). In other words, the stable pool of SEP-c2 represented most (60%) of the postsynaptic receptors whereas it corresponds to a minority (15%) of the whole extrasynaptic population. Interestingly, the stable pool of SEP-c2 at excitatory synapses was twice as much the stable pool of receptors in the extrasynaptic membrane. This difference was highly significant (t-test, p,10 23 ; Fig. 4C). In line with the SPT data, these results indicate higher diffusion constraints on c2 at excitatory synapses as compared with the extrasynaptic membrane. Therefore, molecular constraints rather than QD detection method account for the increased confinement of c2 at excitatory synapses. FRAP curves were then fitted with a double exponential function with two different kinetic constants (fast and slow), as previously done [34,35]. We found that the fast pool recovered faster at extrasynaptic location as compared with excitatory and inhibitory synapses (extrasynaptic: 5.4460.81 s; inhibitory synap-ses: 11.4662.40 s; t test p = 0.04, excitatory synapses: 9.8761.36 s, t test p = 0.008; Fig. 4D), suggesting increased diffusion constraints for SEP-c2 at both excitatory and inhibitory synapses. The significant reduction in the fast pool recovery at excitatory synapses vs the extrasynaptic compartment suggests the fast pool reports on freely diffusing receptors and receptors encountering weak molecular constraints. Interestingly, the time constant of the fast pool was not significantly different at excitatory vs inhibitory synapses (p = 0.43; Fig. 4D). In contrast, for the slow pool the slowest recovery was obtained at inhibitory synapses (extrasynaptic: 467690 s; inhibitory synapses: 30596396 s p = 0.0003; excitatory synapses: 6956141 s, p = 0.17; Fig. 4E). The time constant of the slow pool did not significantly differ at excitatory synapses and at extrasynaptic site (Fig. 4E), meaning that, contrary to inhibitory synapses, c2 did not encounter strong constraints at excitatory ones. Therefore, FRAP results confirmed our SPT data demonstrating that c2 encountered weak constraints probabilities of diffusion coefficients of c2 and a5 in inhibitory (green, inh) and excitatory synapses (red, exc) vs. the extrasynaptic membrane (black, ex). Note the reduced diffusion of c2 and a5 in inhibitory and excitatory synapses as compared to the extrasynaptic membrane (KS test, ***p,10 23 ). (C-D) Decreased explored surface area (C, median EA 625%-75% IQR; KS test, *p,5610 22 ; ***p,10 23 ) and dwell time (D, mean 6 SEM; t-test, ns: not significant, ***p,10 23 ) for c2 in inhibitory (green) vs. excitatory (red) synapses. In contrast, the dwell time of a5 did not significantly differ in inhibitory (green) and in excitatory (red) synapses. doi:10.1371/journal.pone.0043032.g003 at both excitatory and inhibitory synapses whereas the strong constraints appeared only at inhibitory synapses.

Impact of Overexpression of Gephyrin Dominantnegative Constructs on the Lateral Diffusion of the GABA A R c2 Subunit
We then studied the role of the main inhibitory scaffolding molecule gephyrin in the lateral diffusion of c2 at inhibitory synapses. The gephyrin sequence contains amino and carboxy terminal G and E domains, connected by a central linker C [36]. Expression of the isolated G and E domains of gephyrin interfere with the oligomerization and synaptic clustering of the full length scaffolding protein in spinal cord neurons [37]. We first analyzed the impact of venus-tagged G(2) and E domains on gephyrin clustering in hippocampal neurons dissociated from mRFPgephyrin knock-in mouse allowing direct imaging of mRFPgephyrin clusters [38]. Neurons were transfected with venus-G(2), venus-E chimera or with a plasmid encoding Green Fluorescent Protein (GFP) alone for a control. In control conditions, mRFPgephyrin formed numerous clusters along dendrites of GFP transfected neurons (Fig. 5A), and many of them were colocalized with GABA A R c2 clusters (overlay, Fig. 5A). As reported previously [37], the isolated G(2) and E domains were diffusely distributed in the cytoplasm of transfected neurons (Fig. 5A). Dendritic portion of a neuron co-transfected with SEP-GABA A Rc2 (green) and DsRed-homer1c (red) constructs. Scale, 1 mm. Note that SEP-GABA A R c2 (green) form clusters that are not colocalized with DsRed-homer1c (red) fluorescent clusters. (B) Normalized SEP-GABA A Rc2 FRAP fluorescence recovery curves at inhibitory synapses (green, inh), excitatory synapses (red, exc), and in extrasynaptic compartment (black, ex). (C) Percentage (mean 6 SEM) of SEP-c2 fluorescence recovery at inhibitory synapses (inh, green), excitatory synapses (exc, red) and extrasynaptically (ex, black). Note the increase in the size of the pool of slowly mobile c2 at excitatory synapses as compared with the extrasynaptic compartment (t-test, ***p,10 23 ). (D-E) Time constants (mean 6 SEM) for the fast (D) and slow (E) pool, obtained from the double exponential fit (see Materials and Methods) applied to the data in B for inhibitory (green, inh), excitatory (red, exc) synapses or extrasynaptic area (black, ex) (t test, ns: not significant, *: p,0.05, **: p,0.01; ***: p,0.001). doi:10.1371/journal.pone.0043032.g004 Overexpression of the G(2) and E domains noticeably reduced the number of mRFP-gephyrin and GABA A R c2 clusters as well as the fluorescence intensity of the remaining clusters (Fig. 5A). As seen on the overlays, gephyrin clusters were rarely colocalized with GABA A R c2 staining in neurons transfected with the dominantnegative constructs (Fig. 5A). Isolated G(2) domains decreased the intensity of the remaining gephyrin clusters by ,2 fold, (73-78 cells, 2 cultures; t test, p,0.0001; Fig. 5B-C). Overexpression of the E domain also efficiently interfered with gephyrin aggregation, decreasing the number per 10 mm dendritic length of gephyrin clusters and their intensity (73-82 cells, 2 cultures; t test, p = 0.0002 and p,0.0001; Fig. 5B-C). However, the influence of the G(2) and E domains were more pronounced respectively on the fluorescence intensity of gephyrin clusters and the density of gephyrin clusters. These data demonstrate that the isolated G(2) and E domains exert a dominant-negative effect on the clustering of mRFP-gephyrin in hippocampal neurons. This is reminiscent of what was found in spinal cord neurons [37]. In parallel, the isolated G(2) and E domains efficiently reduced the density of c2 clusters (t test, p = 0.001 and p,0.0001, respectively; Fig. 5D) and the fluorescence intensity of c2 clusters (t test, p,0,0001 and p = 0.0027, respectively; Fig. 5E), consistent with the notion that c2 containing GABA A Rs require gephyrin for their clustering [16,39,40,41,42].
We next investigated the effects of the gephyrin dominantnegative constructs on the lateral diffusion of GABA A R c2. Trajectories were considered as synaptic when overlapping with mRFP-gephyrin clusters (e.g. green trajectory in Fig. 6A). Overexpression of the isolated G(2) and E domains did not change the diffusion coefficient of the entire population (trapped + passing) c2at inhibitory synapses 2 cultures; KS test p = 0.11 and p = 0.34; Fig. 6B and Table 2) or in extrasynaptic membranes 2 cultures; KS test p = 0.06 and p = 0.06; Fig. 6B and Table 2). In contrast, the chimeras increased the c2 explored area at inhibitory synapses ( Fig. 6C and Table 2), showing reduced confinement. This effect was more pronounced with the E domain (t test p,0.0001 and p = 0.2 for the E and G(2) domains, respectively). Moreover, the expression of both chimera reduced by ,2 fold the dwell time (t test, p,0.05; Fig. 6D and Table 2) and the proportion of trapped c2 (t test, p,0.05; Fig. 6E and Table 2) at inhibitory synapses. Thus, the explored area and dwell time but not diffusion coefficient of the whole population of QDs are correlated with gephyrin-mediated trapping and accumulation of c2 at inhibitory synapses.

Diffusion of the GluA2 Subunit of the AMPA Receptor at Excitatory and Inhibitory Synapses
We reasoned that excitatory receptors should also be restricted in their diffusion at mismatched synapses. QD-based SPT experiments of the GluA2 subunit of the a-amino-3-hydroxy-5methyl-4-isoxazolepropionic acid receptor (AMPAR) were performed in neurons transfected with GFP-coupled homer1c and mRFP-gephyrin to label excitatory (GFP-homer1c fluorescent clusters) and inhibitory (mRFP-gephyrin fluorescent punctae) synapses, respectively. As visualized on individual trajectories (e.g. Fig. 7A), the surface exploration of QDs decreased at excitatory and inhibitory synapses as compared with the extrasynaptic membrane. As shown for these examples, the diffusion dropped while the confinement increased when QDs went across excitatory and inhibitory synapses (Fig. S2). Quantitative analysis performed on a large number of QDs confirmed GluA2 was significantly slowed down when passing through synapses whatever the neurotransmitter contained in the presynaptic element (3 cultures; KS test, excitatory synapses vs. extrasynaptic, p,10 23 ; inhibitory synapses vs. extrasynaptic, p,5610 22 ; Fig. 7B; Table 1). Furthermore, GluA2 diffusion was not significantly different at excitatory vs. inhibitory synapses when considering the whole QDs population (KS test, p = 0.5). In contrast, the dwell time of GluA2 was increased by ,1.2-fold at excitatory synapses (t-test, p,10 22 ; Fig. 7C; Table 1). Using the criteria of distinction of trapped vs. passing QDs, we found the exploratory behavior of trapped GluA2 was restricted to a smaller area of the excitatory synapse (Fig. 7 D,  G-H), indicating increased confinement. The proportion of GluA2 trapped at excitatory synapses (1563.3%, n = 52) was 3.3-fold that found at inhibitory synapses (4.661.7%, n = 37; Fig. 7E), informing on local anchoring. It is worth noting that most GluA2 containing AMPARs encountered weak molecular constraints at excitatory synapses since only 15% of the QDs were trapped (Fig. 7E). Relative to passing GluA2, the diffusion of trapped GluA2 was reduced at excitatory synapses (trapped, median D = 0. Altogether, our data show GABA A R and AMPAR are significantly hindered in their diffusion at mismatched synapses, and the confinement and dwell time but not the diffusion coefficient report on local anchoring at matched synapses.

Discussion
The main conclusions of this work are that i) diffusing neurotransmitter receptors do cross PSDs where they do not usually accumulate, ii) obstacles and fences significantly reduce the lateral diffusion of neurotransmitter receptors at synapses, and iii) the explored area and dwell time but not the diffusion coefficient inform on synapse specific sorting, trapping and accumulation of receptors.

Relevant Diffusion Parameters for Synaptic Sorting, Trapping and Accumulation of Receptors
We found that the measurement of diffusion coefficients on the whole population of QDs (trapped + passing) did not allow distinguishing receptors slowdown by diffusion barriers and obstacles from receptors interacting with the scaffold. This was shown for GABA A R c2 and AMPAR GluA2 subunits by comparing the cumulative distributions of diffusion coefficients at excitatory vs inhibitory synapses. Furthermore, we showed that preventing GABA A R c2 and gephyrin interaction with a dominant negative approach did not change the diffusion coefficients of c2. This observation was unexpected since receptor-scaffold interactions do lead to a significant reduction in the diffusion coefficient of GABA A R (Fig. 2) and AMPARs (Fig. 7) trapped at the synapse as compared to passing ones. This is in agreement with previous data showing a shift in the diffusion coefficient toward lower values for receptors diffusing in the scaffolding molecule enriched zone (e.g. GlyR-gephyrin: [43]; GABA A R-gephyrin: [44]; mGluR-homer: [45]; GluA1 AMPAR-PSD95: [46]). However, we compared the distribution of diffusion coefficients on the whole population of QDs detected at synapses, independently of their trapped vs passing behavior. We showed here that most receptors detected at matched synapses are passing receptors i.e. receptors not captured by the scaffold or receptors interacting only transiently with the scaffold. Indeed, we found that only ,30% of GABA A R c2 and ,15% of GluA2 were trapped at inhibitory synapses and excitatory synapses respectively. The predominance of passing receptors at synapses thus condition the average behavior of QDs. Moreover, the changes in mobility resulting from the transitions between states such as bound and unbound to a scaffolding element can be difficult to identify with the classical MSD analysis (refs in [47]). Therefore, Figure 5. Gephyrin dominant-negative chimera decreased gephyrin and GABA A R c2 clustering. (A) Hippocampal cultured neurons from mRFP-gephyrin knock-in mice transfected with GFP, venus-G(2), or venus-E constructs, and stained for c2. Scale, 2 mm. Note that mRFP-gephyrin and GABA A R c2 formed numerous clusters along dendrites of GFP transfected neurons. Many mRFP-gephyrin clusters were colocalized with GABA A R c2 clusters (arrows). Over-expression of the venus-G(2) and venus-E chimera resulted in a marked reduction of the number and intensity of mRFPgephyrin and GABA A R c2 clusters. The remaining mRFP-gephyrin clusters were rarely colocalized with c2 clusters (arrows). when looking at global population of QDs, the diffusion coefficient measurement does not inform on receptor anchoring. Calculation of diffusion coefficient over a sliding window (Dinst, Figs. 1C and F) could be more appropriate, but the error in the calculation of the diffusion coefficient is very important for short trajectories [48]. Therefore, Dinst does not provide a more realistic value because of its high uncertainty.
Interestingly, GluA2 was more mobile, explored a larger area and resided shorter time in excitatory synapses relative to GABA A Rs at inhibitory synapses. AMPARs are more mobile within excitatory synapses than NMDA receptors [49,50]. GABA A Rs also display lower diffusion constraints than glycine receptors (GlyR) in inhibitory synapses formed between spinal cord neurons [18]. These differences are likely attributable to differences in binding affinity to the postsynaptic scaffold. Actually there was a higher proportion of trapped GABA A Rs (,30%) than AMPARs (15%), suggesting that the affinity of AMPARs for their scaffolding molecules is lower than that of GABA A Rs.
Furthermore, we found that the median explored area values were very variable in extrasynaptic membranes for GABA A R c2, GABA A R a5 and GluA2 (Table 1). These differences were accompanied by variations in the median diffusion coefficient values. These differences in mobility indicate that not all receptors are equivalent with respect to diffusion in extrasynaptic membrane i.e. for their capacity to interact with obstacles and diffusion barriers. This can be due to various incorporations into multimolecular complexes that impact the size of the tracked molecules, or various interactions with extrasynaptic scaffolding molecules.

Barriers to Diffusion and Receptor Accumulation at Synapses
Reduced diffusion of lipids [35] and neurotransmitter receptors (the present study) at both excitatory and inhibitory synapses confirm the presence of diffusion barriers at both types of synapses. Although constraining the diffusion of receptors, the diffusion barriers at synapses are permeable. Our report of constrained diffusion of GABA A R c2 and a5 at excitatory synapses and AMPAR at inhibitory synapses is reminiscent of what was found for the lipid rafts markers, cholera toxin (ChTx) and glycophosphatidylinositol-anchored green fluorescent protein (GPI-GFP) and the non raft marker, Phosphatidylethanolamine (DOPE) [14,35].
Lipids are associated with the outer part of the membrane leaflet while receptors have intracellular tails that protrude below the membrane. Based on the analysis of the mobility of lipids at synapses, it was proposed that the diffusion is more constrained at inhibitory synapses than at excitatory ones. Indeed, the diffusion of lipids dropped, and their confinement and dwell time increased at inhibitory synapses vs. excitatory ones [35]. On the contrary, we found the GABA A R a5 subunit, not enriched at synapses, had similar diffusion coefficient and dwell time at both inhibitory and excitatory synapses. This means the permeability of the barrier varies in function of the nature of the diffusing molecule (membrane anchored vs. membrane spanning). A high density in different molecular components of the barrier may define its permeability and its function. For instance, an increased density of specific isoforms of ankyrin G and spectrin are required for the segregation of the axonal and the dendritic plasma membrane at the axon initial segment [51]. Furthermore, the scaffolding molecule septin7, enriched at the base of dendritic spines may act as a diffusion barrier to physically separate and compartmentalize dendritic shaft and spines [52].
The reduction in diffusion does not lead to the protein accumulation at synapses if the net flux of molecules equals zero (i.e. at steady state, see [2]). However, obstacles may occasionally permit the local accumulation of proteins by increasing their confinement and dwell time. Receptors have been occasionally found clustered at mismatched synapses. In the cerebellum, the GABA A R a6, c2, and b2/3 subunits are concentrated at the inhibitory Golgi synapses as well as at some excitatory glutamatergic mossy synapses together with functional AMPA-type glutamate receptors [53,54]. It was proposed that GABA A R clustering at excitatory postsynaptic locus mediates inhibition via GABA spillover from nearby Golgi terminals [53,55]. Therefore obstacles may have a functional implication in synaptic transmission. This will be a way for some synapses to use both GABAergic and glutamatergic signaling without neurotransmitter co-release from the same terminal.

Ethics Statement
All animal procedures in this study were performed in accordance with the guidelines issued by the French Ministry of Agriculture and approved by the Direction Départamentale des services Vétérinaires de Paris (Ecole Normale Supérieure, Animalerie des Rongeurs, license B 75-05-20). All efforts were made to minimize animal suffering and to reduce the number of animals used.

Fluorescence Image Acquisition and Analysis
Images were acquired with a cooled Micromax CCD camera (Princeton Instruments, Trenton, NJ) mounted onto a Leica (Nussloch, Germany) DRM upright epifluorescence microscope equipped with a 636 objective (NA 1.32) using MetaView software (MetaImaging, Downingtown, PA). Exposure time was determined on highly fluorescent cells to avoid pixel saturation. Quantitations were performed using MetaMorph software (Meta Imaging). A user-defined intensity threshold was applied to select clusters and avoid their coalescence. For quantification of the number and fluorescence intensity of gephyrin and GABA A R c2 clusters, clusters comprising at least 3 pixels were considered.

QD Imaging
Cells were imaged at 37uC in an open chamber mounted onto an Olympus IX71 inverted microscope equipped with a 606 objective (NA 1.45; Olympus, Tokyo, Japan). The fluorescent probes were detected using a Xe lamp and appropriate filter sets (ref in [23]). GFP, FM 4-64 images and QD real time recordings were acquired with a EMCCD camera (Cascade 512B; Roper Scientific, Evry, France) and MetaView software (Meta Imaging). Real time fluorescence images were obtained with an integration time of 75 ms with 360-600 consecutive frames. Cells were imaged within 30 min after FM 4-64 staining. Acid wash assays indicated that GABA A Rc2 and a5 subunits are mostly localized at the cell surface during this recording period (data not shown).

Single Particle Tracking and Analysis
Single QDs were identified by their blinking property, i.e. their random alternation between emitting and non emitting state [57]. Single QD tracking and reconstruction of trajectories over the recording were performed with homemade software (Matlab; The Mathworks, Natick, MA) as described in [58]. Subtrajectories of single QDs with $30 points without blinks were retained. Spots were classified as synaptic when they overlapped or were within 2 pixels (380 nm) from FM 4-64 spots or within 1 pixel (190 nm) from venus-gephyrin, mRFP-gephyrin or GFP-homer1c clusters. Values of the mean square displacement (MSD) plot versus time were calculated for each trajectory by applying the re- , where t is the acquisition time, N is the total number of frames, n and i are positive integers with n determining the time increment. Diffusion coefficients were calculated by fitting the first four points without origin of the MSD versus time curves with the equation: MSD(nt)~4Dntzbwhere b is a constant reflecting the spot localization accuracy. Trajectories with D,10 24 mm 2 s 21 were considered immobile and were excluded from the calculation of median D. In our experiments, 3-15% of spots were immobile. For QDs exchanging between synaptic and extrasynaptic compartments, the dwell time inside the synapse, was measured as previously described [31]. Dwell times #5 frames (375 ms) were not retained. The explored area of each trajectory was defined as the MSD value of the trajectory for time intervals between 0.3 and 0.375 s [35].

Fluorescence Recovery After Photobleaching
Experiments were conducted using a FRAP system (FRAP L5D, Roper Scientific, Evry, France) run by MetaMorph software (Meta Imaging Software, Roper Scientific, Evry, France) using routines developed by the Curie Institute Imaging Center (Paris, France). It consist of an inverted microscope (Eclipse TE2000-E; Nikon) equipped with an autofocus system (Nikon), a DG-4 illumination system (Sutter Instruments) and appropriate filter sets (Semrock; Optoprim, Paris, France). Coverslips were mounted on a custommade chamber and observed with a 1006objective (Nikon; Roper Scientific, Evry, France). Chamber and objective were heated at 36uC. For FRAP, 2-5 circular regions (radius = 0.6 mm) on top of synaptic spots or on extrasynaptic membranes on different neurites of each cell were bleached by high-intensity 488 nm laser (ERROL, Paris, France) for 5 ms at 65 mW, reducing fluorescence by ,80%. Recovery was monitored by time lapse acquisitions with a CCD camera (QuantEM 512SC, Roper Scientific, Evry, France) at 1 Hz for the first 5s, then at 0.2 Hz for 60s and finally at 0.05 Hz for the subsequent 200 s. Images were analyzed with built-in functions of Metamorph software. Data were normalized and corrected for ongoing photobleaching according to the following equation [34]: Fcorrt = (Ft/F0)/(Fnbt/ Fnb0) where Ft is the fluorescence at time t, F0 is the fluorescence before bleaching, Fnbt is the average fluorescence intensity of three non-bleached spots at time t, and Fnb0 is the average fluorescence intensity of the same non-bleached spots before bleaching. Best fits of FRAP recovery curves were made according to the following equation [34]: Ft = Pf [1-(1-Fbl) exp (-t/tf) ] + (1-Pf) [1-(1 -Fbl) exp (-t/ts) where Pf is the relative size of the fast pool (expressed as a fraction of 1), Fbl is the normalized fluorescence immediately after the photobleaching procedure, and tf and ts are the recovery time constants for the fast and slow pools, respectively.

Statistical Analyses and Image Preparation
Data were compiled and analyzed using Microsoft Excel (Microsoft, Les Ulis, France). Data are presented as median or mean 6 SEM. Means were compared using the nonparametric Student's t-test. Cumulative distributions were compared using the Kolmogorov-Smirnov (KS) test. Differences were considered significant for P values above 5%. Tests were performed using StatView (SAS, Grégy-sur-Yerres, France). Images were prepared for printing using Photoshop (Adobe Systems, San Jose, CA). Figure S1 SEP-GABA A R c2 clusters are localized at the neuronal cell surface. (A) Live cell imaging of recombinant SEP-GABA A R c2 (green) and mRFP-Gephyrin (red) in hippocampal neurons transfected at DIV10. Scale bar, 1 mm. When live cell imaging was done in imaging medium at pH 7.4 (left column), SEP-GABA A R c2 formed numerous fluorescent clusters along the neurite that colocalized with mRFP-Gephyrin fluorescent clusters. After a brief wash at pH 4 (middle), most fluorescence associated with SEP-GABA A R c2 but not with mRFP-Gephyrin was eclipsed. The eclipsed SEP-GABA A R c2 fluorescence rapidly returned in pH 7.4 buffers (right column). (JPG) Figure S2 Diffusion properties of GluA2 in excitatory and inhibitory synapses. Instantaneous diffusion coefficients (A, C) and MSDs as a function of time (B, D) for GluA2 trajectories exemplified in Fig. 5 A at excitatory synapses (A-B) and at inhibitory synapses (C-D). Color code: red, green and black, QD trajectories at excitatory synapses, inhibitory synapses and at extrasynaptic site, respectively. (JPG)