Regulated RalBP1 Binding to RalA and PSD-95 Controls AMPA Receptor Endocytosis and LTD

A two step mechanism was identified that regulates receptor endocytosis during the development of long-term depression (LTD), a long-lasting decrease in synaptic transmission.


Introduction
Long-term depression (LTD), a long-lasting activity-dependent decrease in synaptic strength, has been implicated in brain development, learning and memory, drug addiction, and mental retardation [1]. NMDA receptor (NMDAR)-dependent LTD, an extensively studied form of LTD, involves a rise in postsynaptic Ca 2+ concentration, activation of a serine-threonine protein phosphatase cascade, and clathrin-dependent rapid endocytosis of AMPA receptors (AMPARs) [1][2][3][4][5][6][7][8][9][10]. However, little is known about the identity of key phosphatase substrates and, more importantly, how their dephosphorylation promotes AMPAR endocytosis during NMDAR-dependent LTD. Previous studies have identified several phosphatase substrates associated with NMDAR-dependent LTD and AMPAR endocytosis, including the GluR1 subunit of AMPARs [11][12][13], stargazin [14], and PSD-95 [15]. However, their dephosphorylation has not been clearly linked to the regulation of AMPAR endocytosis. In addition, the clathrin adaptor complex AP2, which directly binds to AMPARs and is required for NMDAR-dependent AMPAR endocytosis and LTD [5,16], should be brought close to AMPARs in a tightly regulated manner, but little is known about the underlying mechanism.
AMPA receptors have two distinct auxiliary subunits, TARP/ stargazin and cornichons, that regulate AMPAR trafficking and gating [17,18]. Of these, TARP/stargazin is known to anchor AMPARs to synapses by directly interacting with PSD-95, an abundant postsynaptic scaffolding protein implicated in the regulation of excitatory synaptic structure, function, and plasticity [19][20][21]. Accordingly, PSD-95 is a key determinant of synaptic levels of AMPARs [21,22]. PSD-95, as a protein that is directly coupled to the AMPAR/TARP complex, is in an ideal position to determine or regulate the fate of AMPARs during their activitydependent trafficking. Indeed, mice null for PSD-95 and those carrying truncated PSD-95 show markedly enhanced LTP [23,24]. In addition, acute knockdown of PSD-95 in brain slices impairs LTD [25,26]. These results suggest that PSD-95 is important for the regulation of synaptic plasticity, but the underlying molecular mechanisms are only just beginning to be understood [15,[26][27][28].
The RalA small GTPase is an important regulator of bidirectional membrane traffic (endocytosis and exocytosis) and regulates diverse biological processes, including cell migration, apoptosis, transcription, proliferation, differentiation, and oncogenesis [29,30]. Whether RalA promotes endocytosis or exocytosis depends on RalA interaction with downstream effectors. Endocytosis occurs when active RalA associates with RalBP1/RLIP76 [31,32], an endocytic adaptor that directly interacts with the m2 subunit of the endocytic AP2 complex [33] and EH domain proteins POB1 and Reps1 [34,35]. RalA promotes exocytosis when it associates with Sec5 and Exo84 subunits of the exocyst complex [29,30]. However, little is known about how RalA binding to RalBP1 and the exocyst complex is coupled to the endocytosis and exocytosis of specific target membrane proteins, respectively.
We found that NMDAR activation induces RalA activation and that activated RalA binds and translocates RalBP1 to dendritic spines. In addition, NMDAR activation leads to the dephosphorylation of RalBP1, which promotes RalBP1 binding to PSD-95. Our data suggest that these two regulated interactions are required and sufficient for the induction of AMPAR endocytosis during NMDAR-dependent LTD.

PSD-95 Interacts with RalBP1
Yeast two-hybrid screens with PSD-95 identified RalBP1 as a novel PSD-95-interacting protein, which is coupled to other endocytic proteins including POB1, RalA, AP2, epsin, and Eps15 ( Figure 1A) [34,36]. The PDZ-binding motif at the RalBP1 Cterminus interacted with the PDZ domains of PSD-95 ( Figure 1B). RalBP1 binding to PSD-95 was further confirmed by in vitro and in vivo pull down and coimmunoprecipitation assays ( Figure 1C-1H). In addition to interacting with PSD-95, RalBP1 formed an in vivo complex with PSD-93/chapsyn-110 (a PSD-95 relative), but no associations of other PSD-95 family proteins (SAP97 and SAP102) with RalBP1 were detected. RalBP1 also formed a complex with POB1 and a-adaptin (a subunit of the AP2 complex) in the brain ( Figure 1I and 1J). Notably, RalBP1 tightly associated with POB1, but relatively weakly with a-adaptin and PSD-95.
Expression Patterns of RalBP1, RalA, and POB1 in the Brain In situ hybridization revealed that mRNAs of RalBP1, RalA, and POB1 are widely expressed in various brain regions ( Figure  S1). The three different mRNAs showed overlapping as well as distinct distribution patterns.
In the brain, RalBP1 and POB1 antibodies recognized a single and double band, respectively ( Figure S2A and S2B). RalBP1, RalA, and POB1 proteins were most abundant in the brain, compared to other tissues ( Figure S2C). RalBP1, RalA, and POB1 proteins were detected in various brain regions ( Figure S2D), consistent with the in situ results. Expression levels of RalBP1 and RalA proteins remained largely unchanged during postnatal brain development, whereas POB1 and a-adaptin showed age-dependent increases ( Figure S2E).
In subcellular brain fractions, RalA was mainly detected in crude synaptosomal (P2) and synaptic membrane (LP1) fractions (P21 and 6 wk), similar to PSD-95 ( Figure S2F). This is consistent with the previous reports that RalA is present in postsynaptic protein complexes [37,38]. In contrast, RalBP1 and POB1 were largely found in cytosolic (S3) and microsomal (P3) fractions. RalA was detected in the PSD I fraction and weakly in PSD II and III fractions ( Figure S2G), suggesting that RalA is not tightly associated with the PSD, although it is mainly detected in synaptic fractions. RalBP1 was minimally detected in PSD fractions.

Phosphorylation of RalBP1 at C-terminal Thr 645 Inhibits PSD-95 Binding
The limited subcellular overlap between RalBP1 and PSD-95 ( Figure S2F) suggests that their interaction might be regulated. Because PDZ interactions can be regulated by phosphorylation [39], we tested if Thr 645 (22 position) at the RalBP1 C-terminus can be phosphorylated (Figure 2A). To this end, we generated a RalBP1 antibody that selectively recognizes RalBP1 proteins with phosphorylated Thr 645. This antibody could detect wild-type (WT) RalBP1 but not a RalBP1 T645A mutant ( Figure 2B and 2C). l-Phosphatase digestion of RalBP1 proteins expressed in vitro and in vivo substantially weakened protein detection by phosphospecific antibodies ( Figure 2D). An upstream mutation in RalBP1 (R642A; 25 position) abolished the phosphorylation ( Figure 2E), indicating that R642 is important for RalBP1 phosphorylation. Of note, the RKET sequence in the RalBP1 C-terminus (25 to 22 positions) matches the consensus sequence for protein kinase A (PKA) phosphorylation. Consistent with this possibility, PKA could directly phosphorylate RalBP1 in vitro ( Figure 2F).

Author Summary
Neurons adapt over time in order to dampen their response to prolonged or particularly strong stimuli. This process, termed long-term depression (LTD), results in a long-lasting decrease in the efficiency of synaptic transmission. One extensively studied form of LTD requires the activation of a specific class of receptors known as NMDA glutamate receptors (NMDARs). A key molecular event initiated by NMDA receptor activation is the stimulation of protein phosphatases. Another key event is internalization via endocytosis of synaptic AMPA glutamate receptors (AMPARs). However, the mechanism by which protein dephosphorylation is coupled to AMPAR endocytosis has remained unclear. Here, we help to define this mechanism. We show that endocytic proteins, including RalBP1, are widely distributed in neurons under normal conditions. Upon NMDAR activation, the small GTPase RalA becomes activated and binds to RalBP1, resulting in the translocation of RalBP1 and RalBP1-associated endocytic proteins to synapses. At the same time, RalBP1 becomes dephosphorylated, which promotes its binding to the postsynaptic scaffold protein PSD-95, a protein that itself associates with AMPARs. This concerted recruitment of endocytic proteins to the vicinity of AMPARs results in AMPAR endocytosis. On the basis of our data, we propose a model in which dual binding of RalBP1 to both RalA and PSD-95 following RalBP1 dephosphorylation is essential for NMDAR-dependent AMPAR endocytosis during LTD. human) interacts with PDZ domains of PSD-95 family proteins (PSD-95, PSD-93/chapsyn-110, SAP97, and SAP102) in yeast two-hybrid assays. Point mutation at the last residue (I655A) of RalBP1 eliminates its PDZ interaction with PSD-95 family proteins. PDZ domains from Shank1, GRIP1, and ZO-1 were used as controls. b-Galactosidase (b-gal) activity: +++, ,45 min; ++, 45-90 min; +, 90-240 min; 2, no significant b-gal activity. HIS3 activity: +++, .60%; ++, 30%-60%; +, 10%-30%; 2, no significant growth. (C) GST fusion proteins of RalBP1 (aa 410-655) pull down PSD-95 family proteins expressed in HEK293T cells. GRIP2, a control PDZ protein; Myc, EGFP, and Flag, epitope tags. (D, E) RalBP1 forms a complex with PSD-95 (D) and SAP97 (E) in HEK293T cells. RalBP1 DC, a mutant RalBP1 that lacks the last four residues and, thus, PSD-95 interaction; Trans, Transfection; WT, wild type; IP, immunoprecipitation. (F) GST-RalBP1 pulls down PSD-95, PSD-93, and POB1, but not GRIP1 or synaptophysin (SynPhy; negative controls) from brain extracts. Deoxycholate (1%) extracts of the crude synaptosomal (P2) fraction, or the S2 fraction (supernatant after P2 precipitation), of adult (6 wk) rat brain were pulled down by GST-RalBP1 and immunoblotted with the indicated antibodies. (G, H) RalBP1 coprecipitates with PSD-95 and PSD-93 but does not form a detectable complex with SAP97, SAP102, GRIP1, and S-SCAM in the brain. Deoxycholate (1%) extracts of the crude synaptosomal fraction of adult rat brain were immunoprecipitated and immunoblotted. Boiled, boiled PSD-95 or RalBP1 antibodies; NR2B, NMDA glutamate receptor subunit 2 (a positive control). (I, J) RalBP1 forms a tight complex with POB1 in the brain. The S2 fraction of adult rat brain was immunoprecipitated and immunoblotted. a-adaptin, a subunit of the AP2 adaptor complex known to associate with POB1 (a positive control). doi:10.1371/journal.pbio.1000187.g001 A phosphomimetic RalBP1 mutant (T645E) failed to interact with PSD-95 in yeast two-hybrid and in vitro coprecipitation assays ( Figure 2G and 2H). RalBP1 proteins phosphorylated at T645 showed reduced biochemical association with PSD-95 in HEK293T cells, compared to total (phosphorylated+non-phosphorylated) RalBP1 ( Figure 2I). These results suggest that RalBP1 is phosphorylated at Thr 645 in vivo, and this inhibits PSD-95 binding.
NMDA Treatment Induces Activation of RalA, and Activated RalA Binds and Translocates RalBP1 to Dendritic Spines Because RalA activation is regulated by Ras, Rap, and calcium/ calmodulin [29,30], which act downstream of NMDAR activation [1,42], we tested if NMDAR activation leads to the activation of RalA. NMDA treatment of cultured neurons induced RalA activation, measured by pull down assays ( Figure 4A). The RalA activation consisted of two phases; a rapid (,1 min) and small increase followed by a slow and bigger increase.
In cultured neurons, RalBP1 expressed alone showed a widespread distribution pattern in dendrites ( Figure 4B). Interestingly, coexpression of a constitutively active form of RalA (G23V) with RalBP1 induced a marked translocation of RalBP1 to dendritic spines, whereas WT and dominant negative RalA (S28N; constitutively in the GDP-bound state) had no effect ( Figure 4B). RalBP1-enriched spines were positive for PSD-95 ( Figure S3), indicating that RalBP1 was translocated to synaptic sites. RalBP1 DCC, which lacks the CC domain that is involved in RalA binding, showed no significant RalA-dependent spine translocation, indicating that the direct binding of RalBP1 to RalA is important. A RalA G23V mutant with weakened RalBP1 binding (G23VD49N; termed GVDN) induced a spine translocation of RalBP1 that is smaller than that of RalBP1 coexpressed with RalA G23V, further suggesting that RalA directly recruits RalBP1 to spines. Notably, RalBP1 DC, which lacks PSD-95 binding, showed a RalA-dependent spine translocation similar to that of WT RalBP1, indicating that activated RalA alone is sufficient to induce spine translocation of RalBP1. These results suggest that NMDAR activation induces RalA activation and that activated RalA binds and translocates RalBP1 to dendritic spines.

RalBP1 Dephosphorylation Combined with RalA Activation Further Promotes Spine Translocation of RalBP1
We next tested whether RalBP1 dephosphorylation induced by NMDAR activation affects RalA-induced spine translocation of RalBP1. NMDA treatment of neurons coexpressing RalA G23V and RalBP1 further increased RalA-dependent spine translocation of RalBP1 ( Figure 4C). In contrast, such increases were not observed in control neurons expressing RalA G23V and RalBP1 DC, indicating that NMDA-induced RalBP1 binding to PSD-95 is important. RalBP1 that was transfected alone was not translocated to spines upon NMDA treatment, indicating that dephosphorylation of RalBP1 alone is not sufficient to induce RalBP1 translocation to spines. Spine morphology, as measured by spine head area, was not changed by overexpression of RalA and RalBP1 constructs (WT and mutants), or by NMDA treatment of the transfected neurons ( Figure S4). These results suggest that binding of dephosphorylated RalBP1 to PSD-95, combined with RalBP1 binding to activated RalA, further promotes synaptic localization of RalBP1.
Biochemically, NMDA treatment of cultured neurons significantly increased coimmunoprecipitation of RalBP1 and PSD-95, but not of RalBP1 and POB1 ( Figure 4D). Whether the association between RalBP1 and RalA is affected by NMDA treatment could not be determined because RalBP1 did not coprecipitate with RalA under our experimental conditions, possibly owing to the transient nature of RalA binding to RalBP1. In support of this interpretation, RalBP1 did not form a complex with RalA WT, in contrast to the strong association of RalBP1 with RalA G23V ( Figure S5A).
The results described thus far indicate that a ternary complex containing RalA, RalBP1, and PSD-95 might be formed in a regulated manner. Because the formation of a RalA-RalBP1 complex could not be demonstrated in vivo, we tested this possibility in heterologous cells using RalA G23V. In HEK293T cells, RalA G23V formed a ternary complex with RalBP1 and PSD-95 ( Figure S5B). In addition, RalA, which is mainly associated with the plasma membrane by geranyl-geranylation, induced translocation of PSD-95 to the plasma membrane in HEK293T cells coexpressing RalBP1 WT, but not in those coexpressing RalBP1 DC that lacks PSD-95 binding ability ( Figure  S5C). Collectively, these results indicate that RalA and PSD-95 act together to bind and translocate RalBP1 to synapses.

NMDAR Activation by Low-Frequency Electrical Stimulation (LFS) Induces RalBP1 Dephosphorylation
The results described thus far are based on experiments using NMDA treatment to induce RalA activation and RalBP1 dephosphorylation. Bath application of NMDA leads to activation of both synaptic and extrasynaptic NMDARs, which can be coupled to different signal transduction pathways [43,44]; therefore, we attempted NMDAR activation by LFS (1 Hz, 900 pulses), which likely enhances activation of synaptic NMDARs [45]. The levels of RalBP1 phosphorylation at T645 were significantly decreased by LFS given to the CA1 region of hippocampal slices, an effect that was blocked by the NMDAR antagonist APV ( Figure 5A and 5B). RalBP1 phosphorylation levels returned to a normal range 60 min after LFS ( Figure 5C), a result similar to that obtained in NMDA-treated cultured neurons. In contrast, neither induction of LTP by theta-burst stimulation (TBS) nor induction of LTD by paired-pulse LFS (PP-LFS, 50 ms interstimulus interval) in the presence of APV induced RalBP1 dephosphorylation ( Figure 5D-5F). Although there was a tendency for LFS to increase coimmunoprecipitation of RalBP1 and PSD-95 compared with that of LFS and APV ( Figure 5G-5H), this difference did not reach statistical significance (p = 0.1; n = 6). This result is in contrast to the enhanced coprecipitation of RalBP1 and PSD-95 observed in NMDA-treated cultured neurons ( Figure 4D). This discrepancy might be attributable to the fact that PSD-95 proteins in slices are more difficult to extract than those in cultured neurons, leading to a decrease in the efficiency of coprecipitation between PSD-95 and RalBP1.

RalBP1 and RalA Are Required for NMDA-induced AMPAR Endocytosis and LTD
RalBP1, an endocytic adaptor, translocated to synapses by NMDAR activation might regulate AMPAR endocytosis. To test this hypothesis, we attempted knockdown of RalBP1 and RalA by shRNA constructs, which reduced expression of exogenous RalBP1 and RalA by 78% and 86%, respectively, in HEK293T cells, and by 90% and 77%, respectively, in cultured neurons ( Figure S6). Knockdown of endogenous proteins could not be observed due to the lack of suitable antibodies.

RalA, but not RalBP1, Inhibits Basal AMPAR Endocytosis in a GTP-independent Manner
We next tested whether RalA and RalBP1 regulate AMPAR endocytosis under basal conditions. Intriguingly, basal GluR2 endocytosis in the absence of NMDAR activation was enhanced by the knockdown of RalA, but not RalBP1 ( Figure 7A), suggesting that RalA, but not RalBP1, inhibits GluR2 endocytosis under basal conditions. Inhibition of RalBP1 by overexpression of RalBP1 TE and POB1 CC had no effect on basal GluR2 endocytosis ( Figure 7B and 7C), further suggesting that RalBP1 does not regulate basal GluR2 endocytosis. Intriguingly, basal GluR2 endocytosis was not affected by overexpression of RalBD ( Figure 7C), RalA S28N, or RalA GVDN ( Figure S8B). RalBD, RalA S28N, and RalA GVDN commonly interfere with GTPdependent actions of RalA by trapping activated RalA, blocking RalA activation, and suppressing the binding of activated RalA to RalBP1, respectively. Considering that RalA knockdown reduces total RalA (both active and inactive), these results suggest that RalA inhibits basal AMPAR endocytosis in a GTP-independent (or RalA activation-independent) manner.
RalA, but not RalBP1, Is Required for the Maintenance of Surface AMPAR Levels RalA inhibits basal AMPAR endocytosis, so we reasoned that RalA might regulate surface AMPAR levels. Indeed, knockdown of RalA, but not RalBP1, significantly reduced surface levels of GluR2 ( Figure 7D), suggesting that surface GluR2 levels are maintained by RalA but not RalBP1. Consistent with this, inhibition of RalBP1 by RalBP1 TE or POB1 CC had no effect on surface GluR2 levels ( Figure 7E and 7F).
Interestingly, surface GluR2 levels were reduced by overexpression of RalBD ( Figure 7F) or RalA S28N ( Figure S8C), which inhibits active RalA, indicating that active RalA is important for the maintenance of surface GluR2 levels. Collectively, these results indicate that both active and inactive RalA are involved in maintaining surface GluR2 levels. Inactive RalA may help maintain surface GluR2 levels by inhibiting basal GluR2 endocytosis ( Figures 7A, 7C, and S8B). How then might active RalA contribute to the maintenance of surface GluR2 levels? One possibility is that active RalA might help internalized GluR2 recycle back to the plasma membrane. However, inhibition of active RalA by overexpression of RalA S28N had no effect on GluR2 recycling ( Figure S9A). In addition, knockdown of RalA, which reduces total (active+inactive) RalA levels, did not affect GluR2 recycling ( Figure S9B), indicating that neither active nor inactive RalA regulate GluR2 recycling. An alternative possibility is that active RalA might regulate synaptic delivery of GluR2 from a cytoplasmic, non-recycling pool, perhaps via the interaction of RalA with the exocyst complex. In support of this possibility, two components of the exocyst complex (Sec8 and Exo70), which interact with active RalA, have been shown to promote surface insertion and synaptic targeting of AMPARs [46].

Constitutive RalA Activation Combined with RalBP1
Binding to PSD-95 Reduces Surface AMPAR Levels and Occludes NMDA-Induced AMPAR Endocytosis The results described thus far suggest that two molecular mechanisms, RalA activation and RalBP1 binding to PSD-95, are important for NMDAR-dependent AMPAR endocytosis. We next reasoned that these two mechanisms might be sufficient to induce AMPAR endocytosis in the absence of NMDAR activation. To this end, we transfected cultured neurons with constitutively active RalA (G23V) and RalBP1 (YFP-tagged) and monitored surface levels of endogenous AMPARs, using surface GluR2 antibodies. Intriguingly, surface AMPAR levels in these neurons were significantly reduced in the absence of NMDA treatment, compared to those expressing RalA G23V alone (without RalBP1 coexpression) or those coexpressing WT RalA (not G23V) and RalBP1 ( Figure 8A). In contrast, coexpression of RalA G23V and a mutant RalBP1 (DC) that lacks PSD-95 binding did not induce a reduction in surface AMPAR levels, relative to the coexpression of RalA G23V and WT RalBP1. These results suggest that RalA activation combined with RalBP1 binding to PSD-95 are sufficient to induce a reduction in surface AMPAR levels in the absence of NMDAR activation, likely through a constitutive endocytosis of AMPARs.
The results described above ( Figure 8A) also suggest that a fraction of exogenously expressed RalBP1 proteins is basally dephosphorylated (in the absence of NMDAR activation), and the amount of dephosphorylated RalBP1 proteins is sufficient to bind to both RalA G23V and PSD-95 and induce significant AMPAR endocytosis. In support of this possibility, NMDA treatment of the neurons coexpressing RalA G23V and RalBP1 did not induce AMPAR endocytosis ( Figure 8B), suggesting NMDA-induced AMPAR endocytosis was occluded. In contrast, neurons coexpressing RalA G23V alone (without RalBP1 coexpression) showed an NMDA-induced reduction in surface AMPAR levels. It is conceivable that the amount of endogenous RalBP1 proteins, although a fraction of them is dephosphorylated, may not be sufficient to induce AMPAR endocytosis, unless a significant fraction of them is dephosphorylated by NMDAR activation.

Generation and Characterization of RalBP12/2 Mice
To investigate the role of RalBP1 in AMPAR endocytosis and LTD in vivo, we generated RalBP12/2 mice using an ES cell line gene-trapped in the intron between exons 3 and 4 of the RalBP1 gene ( Figure 9A). PCR genotyping was used to identify WT and gene-trapped RalBP1 alleles ( Figure 9B). Expression levels of RalBP1 proteins in RalBP12/2 brain was ,18.1%63.1% (n = 8) of WT mice ( Figure 9C and 9D), likely due to incomplete gene trapping. The gene trapping generated a small amount of fusion proteins containing RalBP1 (first 235 aa) and b-geo ( Figure 9C), which were detected in various brain regions including hippocampus ( Figure S10). No abnormalities were observed in gross morphology of RalBP12/2 brain or in the cellular architecture of RalBP12/ 2 neurons, as determined by staining for NeuN and MAP2, respectively ( Figure 9E). There were no changes in expression levels of RalBP1-interacting proteins such as RalA, a-adaptin, and PSD-95, as well as subunits of AMPARs and NMDARs in RalBP12/2 brain ( Figure 9F). Interestingly, however, POB1 expression was significantly decreased by 43.8%610.9% (n = 8), suggesting that RalBP1 is important for the stability of POB1.
Hippocampal SC-CA1 synapses also exhibited mGluR-dependent LTD, which does not require protein phosphatase [1]. Bath application of DHPG (mGluR agonist) induced stable depression in both WT and RalBP12/2 slices (8 wk), with magnitude of depression essentially identical throughout the recording ( Figure 10B). These results suggest that the reduced expression of RalBP1 selectively impairs NMDAR-dependent LTD.  LTD deficits give rise to corresponding enhancement in potentiation, a metaplastic shift [47]. However, LTP induced by TBS in RalBP12/2 slices (4-7 wk) was not substantially different from that of WT littermates throughout the recording ( Figure 10C).
Homosynaptic LTD and depotentiation have many common properties [48]. To induce depotentiation, LFS (1 Hz, 900 stimulations) was delivered to slices (4-7 wk) 5 min after TBS. In contrast to de novo LTD, synaptic depression by LFS after TBS was not different in WT and RalBP12/2 mice ( Figure 10D). These results suggest that RalBP1 is involved selectively in NMDAR-dependent de novo LTD, but not in LTP or depotentiation.

Normal Excitatory Synaptic Transmission at RalBP12/2 CA1 Synapses
To test whether RalBP1 deficiency affects presynaptic functions at hippocampal SC-CA1 synapses, we examined paired-pulse facilitation (PPF), known to be inversely related to presynaptic release probability. PPF at all interstimulus intervals tested was not changed at RalBP12/2 SC-CA1 synapses ( Figure 11A and 11B). In addition, post-tetanic potentiation, another form of short-term plasticity, measured after TBS also appeared normal in RalBP12/ 2 mice ( Figure 10C and 10D).
We next examined the synaptic input-output relationship and spontaneous miniature EPSCs (mEPSCs) to test if RalBP1 deficiency affected excitatory synaptic functions. The relationship between the number of stimulated axons (presynaptic fiber volley) and the slope of postsynaptic fEPSPs at different stimulus intensities was not changed in RalBP12/2 slices ( Figure 11C and 11D). Furthermore, we observed no significant difference in the amplitudes or frequencies of mEPSCs between RalBP12/2 and WT mice ( Figure 11E-11H). Because mEPSCs and fEPSPs are mainly mediated by AMPARs, we examined NMDAR functionality by measuring the ratio of AMPAR and NMDAR currents (AMPA/NMDA ratio). The stimulation intensity was adjusted to achieve an AMPA-mediated current of ,150 pA at the holding potential of 270 mV. The AMPA/NMDA ratios measured in RalBP12/2 slices were not different from those of WT littermates ( Figure 11I and 11J). Pipette solutions used to measure mEPSCs and AMPA/NMDA ratios contained high concentrations of EGTA (10 mM) to inhibit possible contamination of recordings by currents activated by increases in intracellular Ca 2+ . This would not be expected to affect our interpretations because RalBP1 selectively regulates NMDAinduced AMPAR endocytosis and LTD but not surface AMPAR levels associated with mEPSCs and AMPA/NMDA ratios (Figures 6 and 7). Together, these results suggest that neither AMPA-nor NMDA-mediated excitatory synaptic transmission was affected by the reduced expression of RalBP1 under basal conditions.

RalBP1 and PSD-95 in NMDAR-Dependent AMPAR Endocytosis and LTD
Our results suggest that RalBP1 is important for NMDARdependent AMPAR endocytosis and LTD. In support of this, NMDA treatment rapidly dephosphorylates RalBP1 via PP1 and enhances RalBP1 binding to PSD-95. In addition, RalBP1 knockdown in cultured neurons, or RalBP1 inhibition by overexpression of POB1 CC or phosphomimetic RalBP1 (TE), decreases NMDA-induced GluR2 endocytosis. RalBP1 knockdown in slice culture reduces LTD. Reduced RalBP1 expression in RalBP12/2 mice suppresses LFS-induced LTD. Quantitatively, reduced RalBP1 expression in transfected slices and mice reduced LTD magnitudes by ,50%-60%, despite that RalBP1 expression was not completely blocked; that is, RalBP12/2 mice have ,18% of residual RalBP1 expression. RalBP1 knockdown or inhibition in cultured neurons, however, decreased NMDAinduced GluR2 endocytosis by ,20%. This difference may arise from the use of exogenous GluR2, or chemical (not electrical) LTD induction, in cultured neurons.
How might the RalBP1-PSD-95 interaction promote AMPAR endocytosis during NMDAR-dependent LTD? RalBP1 directly binds AP2 and POB1 [34,35], which further associate with EH domain-containing endocytic proteins epsin and Eps15 [36]. Therefore, enhanced RalBP1 binding to PSD-95 may bring RalBP1-associated endocytic proteins such as AP2 and POB1 close to PSD-95, which is linked to the complex of TARPs and AMPARs [18]. In support of this possibility, we demonstrated that RalBP1 forms a complex with POB1 and AP2 (a-adaptin) in the brain. The association between RalBP1 and POB1 was particularly strong, to an extent that POB1 is destabilized in RalBP12/2 neurons.
Our data suggest that RalBP1 is rephosphorylated by PKA during the recovery phase (,1 h) after NMDA treatment. This would dissociate RalBP1 and RalBP1-associated endocytic proteins from PSD-95 and AMPARs, diminishing the drive for AMPAR endocytosis. Consistently, PKA activation markedly reduces NMDA (not AMPA)-induced AMPAR endocytosis, whereas PKA inhibition slightly increases NMDA-induced AMPAR endocytosis [11]. In addition, PKA activation prevents LTD induction and reverses previously established LTD, and PKA inhibition occludes LTD [49].

RalA in NMDAR-Dependent AMPAR Endocytosis and LTD
In support of the role for RalA in AMPAR endocytosis during NMDAR-dependent LTD, NMDAR activation rapidly induces RalA activation. Activated RalA binds and translocates RalBP1 to dendritic spines. NMDA-induced AMPAR endocytosis is suppressed by knockdown of RalA, and inhibition of RalA by overexpression of RalBD and RalA (S28N). In slice culture, RalBD overexpression suppresses LTD.
How does RalA activation contribute to NMDAR-dependent AMPAR endocytosis? A straightforward possibility is that activated RalA binds and translocates RalBP1 and RalBP1associated endocytic proteins to synapses, where target AMPARs are located. In addition, RalA-dependent translocation of RalBP1 to synapses might bring RalBP1 close to PSD-95, facilitating their predicted interaction and mediation of NMDAR-dependent AMPAR endocytosis. A previous study has reported that PP1 is recruited to synapses in response to NMDAR activation [50]. Thus, LTD-inducing NMDAR activation would seem to bring both enzyme (PP1) and substrate (RalBP1) together at synapses, enabling their functional interaction. RalA does not seem to affect other RalBP1 functions; in particular, the ability to interact with other proteins such as PSD-95 and POB1 is unchanged by RalA binding to RalBP1 (K.H., M.K., and E.K., unpublished data).
An important question for future study would be to determine how NMDAR activation leads to the activation of RalA. Ras, Rap, and Ca 2+ are known to act upstream of RalA [29,30]. Importantly, Rap1 regulates NMDAR-dependent AMPAR endocytosis during LTD via p38 MAPK [42]. In addition, a Drosophila study reported that Rap is more important than Ras for Ral activation [51]. These results suggest that NMDAR might activate RalA via Rap1.

RalA and RalBP1 Act in Concert to Mediate NMDAR-Dependent AMPAR Endocytosis
Our data indicate that RalA and RalBP1 act together to mediate NMDAR-dependent AMPAR endocytosis. NMDAR activation induces both RalA activation and RalBP1 dephosphorylation. Spine translocation of RalBP1 induced by RalA is further enhanced by NMDA treatment (Figure 4), which results in dephosphorylation of RalBP1. Therefore, synaptic localization of RalBP1 seems to be mediated by a dual mechanism involving the regulated binding of RalBP1 to both RalA and PSD-95; these processes require RalA activation and RalBP1 dephosphorylation, respectively.
Our data also indicate that both RalA and RalBP1 are necessary and sufficient to mediate NMDAR-dependent AMPAR endocytosis. In support of this, RalA activation combined with RalBP1 binding to PSD-95 is sufficient to reduce surface AMPAR levels in the absence of NMDAR activation; it also occludes the NMDA-induced reduction in surface AMPAR levels ( Figure 8). However, RalA alone (RalA G23V alone or RalA G23V cotransfected with RalBP1 DC) or RalBP1 alone (RalBP1 WT cotransfected with RalA WT) is not sufficient to reduce surface AMPAR levels. In addition, RalA G23V alone does not occlude the NMDA-induced reduction in surface AMPAR levels. The requirement for these two mechanisms-RalA activation and RalBP1 dephosphorylation-in NMDAR-dependent AMPAR endocytosis suggests that these two events may function as a dual-key mechanism that protects against AMPAR endocytosis under conditions in which only a single criterion is fulfilled.

Hippocalcin, RalBP1, and RalA
Hippocalcin, which binds calcium as well as AP2 (b2 adaptin), is translocated to the synaptic plasma membrane via the calciummyristoyl switch to mediate NMDAR-dependent AMPAR endocytosis during LTD [52]. RalBP1 is similar to hippocalcin in that it directly interacts with AP2 (m2 subunit), but it differs from hippocalcin in that it does not have a calcium-sensing activity. Another difference between hippocalcin and RalBP1 is that RalBP1 can be dephosphorylated by NMDAR activation.
An interesting question is whether and how these two pathways (hippocalcin and RalBP1), which are calcium-dependent and phosphatase (PP1)-dependent, respectively, act together to mediate NMDAR-dependent AMPAR endocytosis during LTD. Both hippocalcin and RalBP1 interact with AP2, so it is possible that AP2 may function as a point of crosstalk or convergence between the two pathways. It is conceivable that AP2 and AP2-associated endocytic proteins may be more efficiently translocated to the synaptic plasma membrane by interacting with both hippocalcin and RalBP1.
Lastly, our study has two general implications. Our study provides the first specific mechanism for the general question of how the interaction of activated RalA with RalBP1 is coupled to the endocytosis of target membrane proteins. That is, in our case, the direct interaction of dephosphorylated RalBP1 with a scaffolding protein that is coupled to target membrane proteins. Secondly, our results suggest that scaffolding proteins can switch their functions from the maintenance to regulated endocytosis of interacting membrane proteins. This principle may be applicable to diverse scaffolding proteins that are in association with membrane proteins including receptors, channels, transporters, and adhesion molecules.
In conclusion, our in vitro and in vivo results suggest that the dual and regulated binding of RalBP1 with RalA and PSD-95 mediates AMPAR endocytosis during NMDAR-dependent LTD. Possible directions for future studies include investigation of detailed upstream and downstream mechanisms of this regulated tripartite interaction.

Hippocampal Neuron Culture, Transfection, and Immunocytochemistry
Cultured hippocampal neurons were prepared from embryonic day 18 rat brain. Dissociated neurons on poly-L-lysine coated (1 mg/ml) coverslips were placed in Neurobasal medium supplemented with B27 (Invitrogen), 0.5 mM L-glutamine, and penicillinstreptomycin. Cultured neurons were transfected using mammalian transfection kit (Invitrogen) and fixed with 4% paraformaldehyde/ sucrose, permeabilized with 0.2% Triton X-100, and incubated with primary and dye-conjugated secondary antibodies.

Antibody Feeding Assay
Live neurons expressing HA-GluR2 were incubated with mouse HA antibodies (10 mg/ml) for 10 min at 37uC. After DMEM washing, neurons were returned to conditioned medium containing 20 mM NMDA and incubated at 37uC for 3 min and in the same media without NMDA for 10 min. Neurons were incubated with Cy3 antibodies for surface GluR2, permeabilized, and labeled with Cy5 and FITC antibodies for internalized GluR2 and coexpressed proteins, respectively.

Surface and Internal GluR2 Labeling
HA-GluR2-expressing neurons were fixed and incubated with rabbit HA antibodies for surface GluR2, followed by permeabilization with 0.2% Triton X-100 and incubation with mouse HA antibodies for internal GluR2. Cy3-, Cy5-, and FITC-conjugated secondary antibodies visualized surface GluR2, internal GluR2, and other coexpressed proteins, respectively.

Preparation of Subcellular and Postsynaptic Density Fractions
Subcellular rat brain fractions were prepared as described [53]. Briefly, rat brains were homogenized in buffered sucrose (0.32 M sucrose, 4 mM HEPES, 1 mM MgCl 2 , 0.5 mM CaCl 2 , pH 7.3) with freshly added protease inhibitors (this homogenate fraction is H). The homogenate was centrifuged at 900 g for 10 min (the resulting pellet is P1). The resulting supernatant was centrifuged again at 12,000 g for 15 min (the supernatant after this centrifuge is S2). The pellet was resuspended in buffered sucrose and centrifuged at 13,000 g for 15 min (the resulting pellet is P2; crude synaptosome). The S2 fraction was centrifuged at 250,0006g for 2 h (the resulting supernatant is S3, and pellet is P3). The P2 fraction was resuspended in buffered sucrose and added of 9 volume of water]. After homogenization, the homogenate was centrifuged at 33,000 g for 20 min (the resulting pellet is LP1). The resulting supernatant was centrifuged at 250,0006g for 2 h (the resulting supernatant is LS2, and pellet is LP2). PSD fractions were purified as described [54,55]. To obtain PSD fractions, the synaptosomal fraction was extracted with detergents, once with Triton X-100 (PSD I), twice with Triton X-100 (PSD II), and once with Triton X-100 and once with sarcosyl (PSD III).

Western Blotting and Coimmunoprecipitation with Stimulated Slices
For Western blot analysis of RalBP1 phosphorylation, homogenates of hippocampal slices were prepared as described previously [12]. Briefly, hippocampal slices were sonicated in resuspension buffer (10 mM sodium phosphate [pH 7.0], 100 mM NaCl, 10 mM sodium pyrophosphate, 50 mM NaF, 1 mM sodium orthovanadate, 5 mM EDTA, 5 mM EGTA, 1 mM okadaic acid, and 10 U/ml aprotinin) and centrifuged at 14,000 g for 10 min at 4uC. The pellets were resuspended in SDS sample loading buffer. For coimmunoprecipitation, hippocampal slices were sonicated in resuspension buffer containing 1% Triton X-100 and 1% saponin, and additionally lyzed for 1 hr at 4uC. After centrifuge at 14,000 g for 10 min at 4uC, supernatants were incubated with antibodies for immunoprecipitation.

Image Acquisition and Quantification
Z-stack images were acquired using a confocal microscope (LSM510; Zeiss) under the same parameter settings for all scanning. All transfected neurons, with the exception of those with obvious morphological abnormalities, were imaged in an unbiased manner. Image analyses were performed by a researcher blinded to the experimental conditions. Morphometric measurements on randomly selected images were performed using MetaMorph (Universal Imaging). Neuronal areas for surface/ internal GluR2 analysis were manually selected.

Generation of RalBP1 Genetrap Mice
A mouse ES cell line (RRC077, strain 129/Ola) trapped in the RalBP1 gene was provided by Baygenomics. The gene-trap cassette (pGT1lxf) was integrated into a site 15 bp downstream of the 59 end of the intron 3. The ES cells were injected into blastocysts (C57BL/6J) to generate chimera. Heterozygotes (N1) were backcrossed to C57BL/6J for 4-6 generations. Littermates derived from heterozygous parents were used for all analyses.