|Home | About | Journals | Submit | Contact Us | Français|
Synaptic plasticity, the cellular basis of learning and memory, involves the dynamic trafficking of AMPA receptors (AMPARs) into and out of synapses. One of the remaining key unanswered aspects of AMPAR trafficking is the mechanism by which synaptic strength is preserved in spite of protein turnover. In particular, the identity of AMPAR scaffolding molecule(s) involved in the maintenance of GluA2-containing AMPARs is completely unknown. Here we report that Synaptic scaffolding molecule (S-SCAM, also called membrane-associated guanylate kinase inverted-2 and atrophin interacting protein-1) plays the critical role of maintaining synaptic strength. Increasing S-SCAM levels in rat hippocampal neurons led to specific increases in the surface AMPAR levels, enhanced AMPAR-mediated synaptic transmission, and enlargement of dendritic spines, without significantly effecting GluN levels or NMDAR EPSC. Conversely, decreasing S-SCAM levels by RNA interference-mediated knockdown caused the loss of synaptic AMPARs, which was followed by a severe reduction in the dendritic spine density. Importantly, S-SCAM regulated synaptic AMPAR levels in a manner, dependent on GluA2 not GluA1, sensitive to NSF interaction, and independent of activity. Further, S-SCAM increased surface AMPAR levels in the absence of PSD-95, while PSD-95 was dependent on S-SCAM to increase surface AMPAR levels. Finally, S-SCAM overexpression hampered NMDA-induced internalization of AMPARs and prevented the induction of long term depression, while S-SCAM knockdown did not affect long term depression. Together, these results suggest that S-SCAM is an essential AMPAR scaffolding molecule for the GluA2-containing pool of AMPARs, which are involved in the constitutive pathway of maintaining synaptic strength.
Efficient synaptic transmission relies on the precise organization of various proteins including neurotransmitter receptors, ion channels, signaling enzymes, and cytoskeletal elements (Okabe, 2007; Sheng and Hoogenraad, 2007). Furthermore, activity-dependent modification of synaptic strength, or synaptic plasticity, demands a fine-tuned orchestration of recruitment and removal of proteins at synapses. For example, long-term potentiation (LTP) and long-term depression (LTD) are mediated by activity-dependent trafficking of AMPARs to and out of excitatory synapses (Collingridge et al., 2004; Derkach et al., 2007; Shepherd and Huganir, 2007; Kessels and Malinow, 2009). Synaptic scaffolding proteins play important roles in the assembly of the macro-signaling complexes and trafficking of synaptic proteins. The membrane-associated guanylate kinase (MAGUK) family of proteins is one of the most well studied synaptic scaffolds at the postsynaptic density (PSD). Among the PSD-MAGUKs, PSD-95 (SAP-90) represents the prototypical member and is involved in the various aspects of excitatory synaptic transmission, including synapse maturation, AMPAR trafficking, and synaptic plasticity (reviewed in Elias and Nicoll, 2007; Xu, 2011).
S-SCAM was first identified as a protein interacting with SAPAP (also called GKAP; Kim et al., 1997) (Hirao et al., 1998). S-SCAM is also known as membrane-associated guanylate kinase inverted-2 (MAGI-2) (Wu et al., 2000) or atrophin interacting protein-1 (AIP-1) (Wood et al., 1998). The molecular organization of S-SCAM is in an inverse configuration to PSD-95, and is composed of six PDZ domains, one guanylate kinase (GK) domain, and two WW domains. Studies have shown that S-SCAM interacts with numerous PSD-95-binding proteins including NMDA receptor (NMDAR) (Hirao et al., 1998), ErbB4 (Buxbaum et al., 2008), Neuroligin (Iida et al., 2004), Kif1B (Mok et al., 2002), and transmembrane AMPAR regulatory proteins (TARPs) (Deng et al., 2006). S-SCAM also binds proteins that have no known interaction with PSD-95, which include Axin (Hirabayashi et al., 2004), β-catenin (Nishimura et al., 2002), β-dystroglycan (Sumita et al., 2007), and neuroligin 2 (Sumita et al., 2007). These overlapping protein-protein interaction profiles suggest that S-SCAM may play both similar and distinct roles, compared to PSD-95, in the molecular organization of PSDs and, in particular, AMPAR regulation. There are three S-SCAM isoforms of S-SCAM-α, -β, and -γ, which are generated by differential translational initiations from multiple sites (Hirao et al., 2000). Mice lacking the longest variant S-SCAM-α died within 24 h after birth (Iida et al., 2007), indicating that S-SCAM is an essential protein. Interestingly, hippocampal culture neurons prepared from these mutant mice showed abnormal elongated dendritic spines (Iida et al., 2007), suggesting that S-SCAM may function in the dendritic spine dynamics.
In addition to the probable role of S-SCAM in the molecular organization of PSDs and synaptic transmission, recent genetic studies uncovered S-SCAM gene mutations in patients with neurological diseases such as schizophrenia (Walsh et al., 2008) and infantile spasms (IS) (Marshall et al., 2008). Despite the potential importance of S-SCAM in synaptic biology and neurological diseases, little is currently known for the function of S-SCAM. Here, we report a novel and essential role of S-SCAM in the organization of PSDs and excitatory synaptic transmission.
Myc- or GFP-tagged S-SCAM expression constructs were obtained from Dr. Y. Hata (Tokyo Medical and Dental University, Tokyo, Japan). GluA1 and GluA2 shRNAs were as described previously (Lee et al., 2004). S-SCAM shRNA construct was created by cloning the following hairpin sequence GTACAGAACCTGAGCCATATTCAAGAGATATGGCTCAGGTTCTGTAC into pSUPER or pSUPER.neo+gfp vectors (OligoEngine). S-SCAM rescue construct was generated by introducing silent mutations in S-SCAM RNAi targeting sequence via site-directed mutagenesis, and confirmed by nucleotide sequencing. PSD-95 shRNA is as described previously (Nakagawa et al., 2004).
Dissociated hippocampal neuron culture was prepared from E18 embryos of either sex as described previously (Lee et al., 2002) and grown in Neurobasal media supplemented with B27. Neuron transfection was performed at DIV 14 using Lipofectamine 2000 reagent (Invitrogen) as described previously (Lee et al., 2004). Cold methanol (−20°C for 10 min) fixation was used for the staining of PSD-95, Shank, GluN1, GluN2B, and GKAP. For co-staining of β-Gal (or GFP) and S-SCAM, GKAP, or PSD-95, neurons were incubated first in 2% formaldehyde/4% sucrose/1×PBS for 2 min followed by cold methanol for 10 min. Surface AMPAR staining was performed as described previously (Lee et al., 2004). Rabbit polyclonal anti-S-SCAM antibodies were prepared by injecting rabbits with purified recombinant protein covering the WW domains of S-SCAM (aa 303–405) (Hirao et al., 1998), affinity purified, and used at 0.01 µg/ml concentration for immunocytochemistry or western blotting. We also tested a commercial rabbit polyclonal anti-MAGI-2 antibody (Sigma; 1:100 dilution) that is prepared from different antigen (aa 554–571), which is specific to S-SCAM/MAGI-2. Both antibodies produced very similar immunostaining patterns. However, to eliminate any potential cross-reactivity problem, immunostaining data obtained from the commercial antibody were shown. Primary antibodies and their dilution used for immunocytochemistry are: mouse anti-β-Gal (Promega; 1:1000), rabbit anti-β-Gal (Abcam, 1:5000), mouse anti-PSD-95 (clone K28/43, UC Davis/NIH NeuroMab; 1:500), rabbit anti-GKAP (1:250; Kim et al., 1997), mouse anti-NR2B (clone N59/36, UC Davis/NIH NeuroMab; 1:100), mouse anti-pan-SAPAP (clone N127/31, UC Davis/NIH NeuroMab; 1:250), mouse anti-NMDAR1 (BD Pharmingen; 1:100), mouse anti-Shank (1:250; Naisbitt et al., 1999), mouse anti-Bassoon (Stressgen; 1:200), mouse anti-synaptophysin (SVP-38, Sigma; 1:1000), rabbit anti-HA (Santa Cruz; 1:100), mouse anti-HA (clone 12CA5, Roche; 1:400), mouse anti-myc (clone 9E10, Santa Cruz; 1:100), rabbit anti-myc (Cell Signaling Technology; 1:100), rabbit anti-GluA1 (Oncogene; 5 µg/ml), and mouse anti-GluA2 (Chemicon, 5 µg/ml). Bound primary antibodies were visualized by Alexa Fluor 488 (Invitrogen; 1:250), Cy3-conjugated (Jackson Immunoresearch Laboratories; 1:500), and/or Alexa Fluor 647 (Invitrogen; 1:200) secondary antibodies.
AMPAR internalization assay was performed using GluA2 antibody recognizing extracellular epitope (MAB397, Millipore) and data analyses were done as described before (Lee et al., 2004). Briefly, after live-labeling surface GluA2 with the antibody (10 µg/ml) in the conditioned medium for 10 min at 37°C, neurons were incubated for 2 min with either conditioned medium (control), or conditioned medium plus NMDA (50 µM) or AMPA (100 µM). After brief wash in Neurobasal medium, neurons were further incubated for 8 min in the conditioned medium before fixation and staining.
Images acquisition was performed using a Nikon C1 plus laser scanning confocal microscope. Acquired z-series stacks images were converted to projection images (with maximal projection option) for analysis using either Metamorph software (Molecular Devices) or in-house custom software. Images were analyzed in a double-blind manner. To measure puncta number per given length of dendrites, per image, five dendritic segments (~5–10 µm in length each) were selected from transfected and neighboring non-transfected neurons, respectively. After applying threshold, only puncta with more than 4 pixel sizes were counted. All data collected were transferred to Microsoft Excel for computation.
All values represent means ± SEM, unless otherwise indicated. All transfection experiments were repeated at least three times. Statistical significance for pair was analyzed by the Student’s t test (unpaired, two tailed, assuming unequal variance), unless otherwise indicated in the figure legends. ANOVA with Tukey’s post hoc test were used for group comparisons. Cumulative plot data were analyzed by Kolmogrov-Smirnov test (K-S test). p < 0.05 was considered significant.
For mEPSC measurement, we used dissociated culture hippocampal neurons (plated at the density of 75 kcells/coverslip), which were transfected at div 14 with GFP alone (pEGFP-C1), GFP + S-SCAM, or pSUPER.neo+gfp plasmid expressing S-SCAM shRNA. At 3 days post-transfection, transfected pyramidal neurons were identified by GFP fluorescence and morphological inspection. All recordings were performed at 25°C. Whole cell patch recordings were performed by voltage-clamping neurons at −70 mV in bathing solution (in mM; 119 NaCl, 5 KCl, 2 CaCl2, 2 MgCl2, 30 glucose, 10 HEPES, pH7.4, ~300 mOsm) containing tetrodotoxin (1 µM; Tocris) and bicuculline (20 µM; Tocris), continuously perfused at the rate of ~0.5 ml/min. Internal solution was composed of 140 K·gluconate, 5 KCl, 2 MgCl2, 4 Mg·ATP, 0.3 Na2·GTP, 0.2 EGTA, 10 HEPES and adjusted to pH7.2 and ~290 mOsm. Micropipettes with tip resistance of 4–7 MΩ were used. mEPSCs were acquired through a MultiClamp 700B amplifier (Molecular Devices), filtered at 2 kHz, digitized at 10 kHz, utilizing the “gap-free” protocol. mEPSCs were detected and analyzed with MiniAnalyses software (Synaptosoft) by setting amplitude threshold to 5 pA (usually √RMS×3 values lower than 4), further filtered by selecting mini events of 10–90% rise time <3 msec. Cumulative probability plots were generated by combining mini events from all recorded neurons and analyzed for statistical significance by Kolmogrov-Smirnov test.
For whole-cell paired recording experiments, we used slice culture prepared from the hippocampi of 7-day-old Sprague-Dawley rats of either sex. Slices of 400 µm thickness were prepared using a tissue chopper, transferred onto MilliCell culture plate inserts (Millipore), and cultured for 4–6 days in MEM supplemented with 1 µg/ml insulin, 0.0012% ascorbic acid, 20% horse serum, 1 mM L-glutamine, 1 mM CaCl2 30 mM HEPES, 13 mM D-glucose, 5.2 mM NaHCO3 with Pen/Strep. Media was changed every other day. Sindbis virus expressing GFP-S-SCAM (a gift from Dr. Yutaka Hata; Nishimura et al, 2002) were injected into the CA1 region of hippocampal slices using a Toohey Spritzer microinjection system. Electrophysiological recordings were performed 15–24 hours post infection. A single slice was removed from the insert and placed in the recording chamber and continuously perfused with ACSF (in mM; 119 NaCl, 2.5 KCl, 1 NaH2PO4, 11 glucose, 26 NaHCO3, 4 MgCl2, 4 CaCl2, 290 mOsm) containing 50 µM picrotoxin and 2 µM chloroadenosine, bubbled with a mixture of 5% CO2 and 95% O2. Stimulating electrodes (2 conductor platinum/iridium cluster microelectrode, 25 µm diameter; FHC) were placed about 200 µm on either side of the recording cell. One pathway was used to induce LTD, the other served as a control pathway. The patch pipette was filled with internal solution (in mM, 115 CsMeSO3, 20 CsCl, 10 HEPES, 2.5 MgCl2, 4 Na2-ATP, 0.4 Na-GTP, 10 Na-phosphocreatine, 0.6 EGTA, 5 QX314, pH 7.2, 290 mOsm) and had tip resistance of 3–6 MΩ. For rectification experiments, spermine (0.1 mM) was added in the internal solution. AMPA EPSCs were recorded at −60 mV and measured as peak inward current within a 60 ms window after stimulation. NMDA EPSCs were recorded at +40 mV and measured 60–100 ms after the initiation of the EPSC. Rectification index was calculated by dividing peak AMPA amplitudes measured at −60 mV by those at +40 mV. Stimulation pulses were provided at 0.3 Hz. 60 traces were averaged for AMPA and NMDA EPSCs. For LTD experiments, at least 8 minutes of stable baseline responses were collected before LTD induction. LTD inducing stimulus consisted of 200 stimulation pulses at 1 Hz while holding cells at −40 mV. For paired pulse ratio (PPR), two consecutive stimuli were given 25 ms, 50 ms, 100 ms, and 200 ms apart with a 2 second interval between time points. At least 60 traces were averaged.
To study the function of S-SCAM in excitatory synaptic transmission, we first overexpressed S-SCAM in hippocampal neurons and examined the effect on the dendritic spines. We used the longest variant S-SCAM-α that produced mostly S-SCAM-α (> 80%) but also -β, and -γ isoforms when expressed in hippocampal neurons (data not shown). Transfected S-SCAM showed strong enrichment in all dendritic spines identified (Fig. 1A). Quantification of spine density indicated that S-SCAM overexpression reduced spine density to approximately 70% of control β-Gal-only transfected neurons (53.1 ± 2.2 vs 37.6 ± 1.6 spines per 100 µm, β-Gal vs S-SCAM; p < 0.001) (Fig. 1B).
Increasing S-SCAM levels also had influence on dendritic spine morphology. Following S-SCAM expression, as shown in Fig. 1C–E, there was a specific increase in spine head width by 24% (0.93 ± 0.03 vs 1.15 ± 0.05 µm, GFP control vs S-SCAM; p < 0.001) with no significant change in the spine length (1.26 ± 0.04 vs 1.29 ± 0.05 µm for GFP vs S-SCAM; p = 0.57). The cumulative plot of spine head width showed a uniform right shift, indicating that S-SCAM overexpression targeted all spines rather than a specific population (Fig. 1F; p <0.001). Cumulative plot analysis of dendritic spine length did not show a significant difference between GFP control and S-SCAM overexpressed neurons (Fig. 1G; p = 0.35). The increase of spine head width after S-SCAM overexpression was accompanied by the increased proportion of mushroom-type spines (46% for GFP vs 62% for S-SCAM; p < 0.05) while decreasing the proportion of thin type spines (37% for GFP vs 24% for S-SCAM; p < 0.05; Fig. 1H). There was no significant difference in stubby-type (16% GFP vs 12% S-SCAM; p = 0.33) and forked-type spines spines (1% GFP vs 2% S-SCAM; p = 0.17) following S-SCAM transfection. These results collectively indicate that increasing S-SCAM levels in hippocampal neurons promotes the maturation of dendritic spines.
S-SCAM binds to numerous proteins and can potentially influence the composition of synaptic proteins by selectively recruiting and/or stabilizing a set of interacting proteins at synapses. To study the scaffolding function of S-SCAM, we examined the effect of increasing S-SCAM levels on the synaptic accumulation of three key PSD scaffolding proteins. First we examined a direct S-SCAM-binding protein, GKAP. S-SCAM overexpression drastically increased GKAP puncta intensity at dendritic spines by > 2.5-fold (Fig. 2A,B), when compared to non-transfected (Non-txf) neighboring neurons (100 ± 3 vs 261 ± 27%, Non-txf vs S-SCAM; p < 0.001). The increase of synaptic GKAP levels after S-SCAM overexpression was accompanied by a small decrease (~14%) in GKAP puncta density (78 ± 3 vs 67 ± 4 per 100 µm dendrite, Non-txf vs S-SCAM; p < 0.05) (Fig. 2C), which was smaller than the decrease in spine density (~40% reduction; see Fig. 1B). Further analyses indicated that S-SCAM overexpression increased the population of dendritic spines containing multiple GKAP puncta at the expense of single puncta-spines (GFP vs S-SCAM, spines with 1 puncta, 84 ± 2 vs 73 ± 2%; p < 0.001; 2 puncta, 15 ± 2 vs 24 ± 2%; p < 0.001; 3 puncta, 2 ± 1 vs 3 ± 1%; p = 0.04; Fig. 2D,E). In contrast to GKAP, increased S-SCAM levels lead to only a modest (~1.2-fold) increase of PSD-95 intensity at dendritic spines (100 ± 2 vs 122 ± 5%, Non-txf vs S-SCAM; p < 0.001; Fig. 2A,B). Surprisingly, despite the large increase in GKAP levels at dendritic spines, the levels of Shank, a GKAP-binding scaffolding protein, were reduced by 17% (100 ± 4 vs 83 ± 4%, Non-txf vs S-SCAM; p < 0.01; Fig. 2A,B). Furthermore, the puncta densities of both PSD-95 and Shank were greatly reduced (by 44% for PSD-95; by 62% for Shank; p < 0.001 for both proteins; Fig. 2C), to a degree greater to the spine density reduction (by ~30%; Fig. 1B). Taken together, these results demonstrate that increasing S-SCAM levels led to drastic changes in the protein composition of PSD scaffolding proteins.
S-SCAM was reported to bind TARPs (Deng et al., 2006) but its role in the regulation of AMPARs is unknown. We have found that increasing S-SCAM levels in hippocampal neurons led to a drastic increase in the amounts of all three major GluA subunits (GluA1, GluA2, and GluA3) at dendritic spines (Fig. 3A,B; 170 ± 9, 210 ± 9, and 140 ± 9% for GluA1, GluA2, and GluA3, respectively; compared to Non-txf, p < 0.001 for all three GluAs). Surface staining of these proteins indicated that S-SCAM increased the surface expression of GluA1 (sGluA1) and GluA2 (sGluA2) at the dendritic spines as well (Fig. 3C,D; 238 ± 14 and 229 ± 9% for sGluA1 and sGluA2, respectively; p < 0.001 for both sGluAs). However, the total expression levels of these proteins, as measured by the staining intensities in the soma and dendrites did not change significantly (Fig. 3A, quantified in 3B). This suggests that the increase of AMPARs in the dendritic spine came at the expense of AMPARs localized outside dendritic spines. Since S-SCAM overexpression promoted the maturation of dendritic spines, we also examined the number of dendritic spines lacking sGluA2, which measures silent synapses indirectly. Remarkably, as shown in a Fig. 3E, S-SCAM overexpression greatly reduced the percentage of the spines lacking sGluA2 (7 ± 2 vs 25 ± 7%, S-SCAM vs control GFP-transfected neurons, p < 0.001), consistent with synapse maturation. Blocking NMDAR activity with APV (100 µM) did not prevent the S-SCAM-induced increase of sGluA2 levels (Fig. 3F,G; p = 0.42), indicating that the increase of AMPAR by S-SCAM overexpression is an activity-independent process. In contrast to AMPAR, neither GluN1 nor GluN2B levels at dendritic spines were significantly affected by S-SCAM overexpression (Fig. 3H,I; p > 0.05), indicating that S-SCAM overexpression specifically increased AMPAR levels at synapses. Taken together, these data indicate that S-SCAM overexpression promoted the accumulation and/or stabilization of AMPARs at dendritic spines.
To examine the role of endogenous S-SCAM, we designed shRNA sequences to specifically knockdown S-SCAM levels in hippocampal neurons. When a plasmid expressing short hairpin RNA targeting S-SCAM (designated S-SCAM RNAi) was co-transfected into COS cells, as shown in Fig. 4A, it specifically reduced the protein levels of S-SCAM to < 5% of control levels (S-SCAM co-transfected with unrelated Zn-T3 shRNA) (Lee et al., 2004). The S-SCAM RNAi had no significant effect on the level of PSD-95 (Fig. 4A) or GKAP (data not shown), showing specificity of the knockdown. When transfected into hippocampal neurons, S-SCAM RNAi effectively reduced S-SCAM levels to 37% in dendrites (p < 0.001) as determined by immunofluorescent intensity (Fig. 4B,C).
We first examined the effect of S-SCAM RNAi on GluA2, a major subunit of AMPARs in the hippocampal neurons. S-SCAM RNAi reduced both the intensity and number of sGluA2 puncta significantly (Fig. 4D–F). The reduction of sGluA2 intensity started to show as early as 1 day post-transfection of S-SCAM RNAi (100 ± 7 vs 80 ± 6%, control vs S-SCAM RNAi; p < 0.05) and reached 60% of control levels 3 days post-transfection (Fig. 4E). The reduction of sGluA2 puncta density became statistically significant 2 days post-transfection (23 ± 2 vs 9 ± 1 per 100 µm dendrite, control vs S-SCAM RNAi; p < 0.001; Fig. 4F) and was reduced to 17% of control level after 3 days (23 ± 2 vs 4 ± 1 per 100 µm dendrite, control vs S-SCAM RNAi; p < 0.001). The reduction in sGluA2 was accompanied by a significant increase in the proportion of dendritic spines lacking sGluA2 staining (after 3 days, 40 ± 4 vs 76 ± 4%, control vs S-SCAM RNAi; p < 0.001; Fig. 4G). These data clearly indicate that S-SCAM RNAi removes sGluA2 from dendritic spines.
S-SCAM RNAi also had a dramatic effect on dendritic spines (see GFP channels in Fig. 4B,D). Knockdown of S-SCAM reduced the dendritic spine density (at 3 days post-transfection, 39 ± 2 vs 15 ± 1 per 100 µm dendrite, control vs S-SCAM RNAi; p < 0.001; Fig. 4H) and the size of dendritic spines (p < 0.001 for width and p < 0.05 for length; Fig. 4I). Furthermore, S-SCAM knockdown increased the proportion of stubby spines (13 vs 33%, control vs S-SCAM RNAi; p < 0.001) at the expense of mushroom-type mature spines (55 vs 39% control vs S-SCAM RNAi; p < 0.05; Fig. 4J). These data suggest that losing S-SCAM from synapses promotes the collapse of mushroom-type dendritic spines to stubby ones, which are eliminated eventually. Consistent with the reduction in the number of dendritic spines, S-SCAM RNAi greatly reduced the puncta density of PSD-95 (52 ± 5 vs 16 ± 2 per 100 µm dendrite, control vs S-SCAM RNAi; p < 0.001; Fig. 4K,L) and Bassoon (141 ± 9 vs 73 ± 8 per 100 µm dendrite, control vs S-SCAM RNAi; p < 0.001; Fig. 4L), indicating the reduction of overall synapse numbers.
To further confirm the specificity of S-SCAM RNAi effect, we performed a “rescue” experiment with S-SCAM RNAi-resistant S-SCAM (designated S-SCAM-r; see Fig. 4A). When rescuing S-SCAM-r was co-expressed with S-SCAM RNAi, the amount of sGluR2 was increased to a level similar to S-SCAM overexpression (257 ± 7 vs 252 ± 8%, S-SCAM RNAi + S-SCAM-r vs S-SCAM-r + Control RNAi; p = 0.67; Fig. 4M,N) and the number of dendritic spines was similarly restored (25 ± 1 vs 28 ± 2 per 100 µm dendrite, S-SCAM RNAi + S-SCAM- r vs S-SCAM-r + Control RNAi; p = 0.26; Fig. 4O). Thus, the effect of S-SCAM RNAi on AMPARs and dendritic spines was specifically related to the loss of S-SCAM proteins and not caused by non-specific effect of RNAi. Taken together, these data point out that S-SCAM is an essential scaffolding molecule for the stabilization/maintenance of AMPARs and dendritic spines.
To measure directly the effect of changing S-SCAM levels on AMPAR-mediated synaptic transmission, we measured AMPAR miniature excitatory postsynaptic currents (mEPSCs) after transfecting dissociated hippocampal culture neurons with either S-SCAM or S-SCAM RNAi constructs. Increasing S-SCAM levels enhanced AMPA mEPSC amplitudes (17.4 ± 0.9 vs 23.4 ± 1.0 pA, GFP control vs S-SCAM; p < 0.001; Fig. 5A–C), while S-SCAM RNAi reduces AMPA mEPSC amplitudes (14.5 ± 0.9 pA; p < 0.01 compared to GFP control; Fig. 5A–C). We did not observe significant changes in the AMPA mEPSC frequency after S-SCAM overexpression (3.9 ± 0.5 vs 4.6 ± 1.0 Hz, GFP control vs S-SCAM; p = 0.35; Fig. 5 D), despite that the reduction in dendritic spine density. This is likely due to the “unsilencing” of silent synapses by acquiring AMPAR (Fig. 3E). In contrast, S-SCAM RNAi greatly reduced mEPSC frequency (1.7 ± 0.4 Hz; p < 0.005), consistent with the reduction in synapse numbers after S-SCAM RNAi (Fig. 4 H). Therefore, S-SCAM levels directly influence synaptic AMPAR levels.
S-SCAM overexpression in CA1 pyramidal neurons of hippocampal slice culture by Sindbis virus infection drastically increased AMPAR-mediated synaptic transmission measured at −60 mV (26.5 ± 4.2 vs 51.0 ± 5.5 pA, uninfected neighboring neuron vs S-SCAM-infected; p < 0.001; Fig. 5E,F). In contrast, we did not detect significant changes in the NMDAR-mediated responses measured at +40 mV (16.5 ± 2.5 vs 17.4 ± 2.6 pA, uninfected vs S-SCAM-infected; p = 0.77; Fig. 5E,G), indicating that S-SCAM overexpression did not affect NMDAR-mediated synaptic transmission significantly. Importantly, S-SCAM-infected neurons showed a significant increase in the AMPA/NMDA ratio (1.9 ± 0.2 vs 3.3.0 ± 0.4, uninfected vs S-SCAM-infected; p < 0.005; Fig. 5H), indicating that S-SCAM overexpression specifically increased AMPA component of excitatory synaptic transmission. We did not find a significant difference in the paired-pulse ratio (PPR) of S-SCAM infected-CA1 neurons from uninfected neighboring neurons (1.7 ± 0.3 vs 1.5 ± 0.2 at 50 ms interval, 1.4 ± 0.1 vs 1.4 ± 0.2 at100 ms interval, uninfected vs S-SCAM infected; p > 0.4; Fig. 5I,J), indicating that S-SCAM overexpression did not change pre-synaptic function significantly.
Our data and literature indicate that both S-SCAM and PSD-95 can regulate synaptic AMPAR levels – do they play redundant or independent roles? To address this question, we performed “reciprocal rescuing” experiments. When S-SCAM or PSD-95 were overexpressed in hippocampal neurons (with a GFP txf marker), both proteins increased sGluA1 levels comparably (Fig. 6A,B; 100 ± 8 vs 235 ± 18 or 221 ± 43%, non-txf vs S-SCAM or PSD-95; p < 0.001). Surface GluA2 levels were also increased by both S-SCAM and PSD-95 overexpression, although S-SCAM was more effective than PSD-95 (100 ± 15 vs 184 ± 19 or 134 ± 11%, non-txf vs S-SCAM or PSD-95; p < 0.001, S-SCAM vs non-txf; p < 0.05, S-SCAM vs PSD-95). On the other hand, S-SCAM RNAi and PSD-95 RNAi reduced both sGluA levels drastically (for sGluA1, 41 ± 9 vs 57 ± 11%; for sGluA2, 23 ± 10 vs 46 ± 7%, S-SCAM RNAi vs PSD-95 RNAi; p < 0.001 compared to GFP control for both RNAi). Importantly, overexpression of PSD-95 in the presence of S-SCAM RNAi did neither increase nor restore either sGluA subunit levels at synapses (18 ± 5 % for GluA1, 6 ± 3 % for GluA2). In sharp contrast, overexpression of S-SCAM in the presence of PSD RNAi not only rescued the loss of PSD-95 but also increased both sGluA levels at dendritic spines comparable to S-SCAM overexpression alone (sGluA1, 196 ± 18 vs 235± 18%, sGluA2, 184 ± 19 vs 180 ± 22%, PSD-95 RNAi + S-SCAM vs S-SCAM; for both sGluAs, p < 0.001 and p > 0.1, compared to PSD-95 RNAi and to S-SCAM alone, respectively; Fig. 6 A,B). Quantification of relative intensities of PSD-95 and S-SCAM verified that both protein overexpression reached > 5-fold higher levels than the endogenous protein levels, regardless of the presence of RNAi (Fig. 6C). Therefore, S-SCAM increased surface AMPARs independently of PSD-95, while PSD-95 was dependent on S-SCAM to exert its effect on AMPARs. These data suggest that S-SCAM is an indispensable scaffolding molecule for AMPARs.
PSD-95 regulates AMPARs mainly through GluA1 subunits that are inserted to synapses during high activity (Ehrlich and Malinow, 2004; Xu, 2011). Does S-SCAM also exhibit such AMPAR subunit-specific regulation? To examine the question, we utilized AMPAR subunit-specific RNAi (Lee et al., 2004). When neurons were transfected with GluA1 RNAi, as shown in Fig. 7A,B, the level of sGluA1 was greatly reduced regardless of S-SCAM overexpression (< 25% of β-Gal + Control RNAi ; p < 0.001), showing the effectiveness of the RNAi. GluA2 RNAi exhibited similar effectiveness (< 23% of β-Gal + Control RNAi; p < 0.001). In hippocampal neurons co-transfected with GluA1 RNAi and S-SCAM, sGluA2 levels were significantly increased to a level comparable to that of S-SCAM + Control RNAi (226 ± 18 vs 198 ± 18%, S-SCAM + GluA1 RNAi vs S-SCAM + Control RNAi, p = 0.28; p < 0.001 compared to β-Gal + Control RNAi). Therefore, GluA1 was not required for the S-SCAM-induced increase of surface AMPAR levels. On the other hand, surprisingly, S-SCAM failed to increased sGluA1 levels in neurons co-transfected with GluA2 RNAi (106 ± 9 vs 293± 14%, S-SCAM + GluA2 RNAi vs S-SCAM + Control RNAi; p < 0.001). These results suggest that S-SCAM increases surface AMPAR levels in a manner dependent on GluA2 subunit.
To further corroborate that S-SCAM exerts its influence through GluA2, we took advantage of NSF-interaction blocking peptide that targets GluA2 containing AMPARs and causes the rundown of AMPAR mEPSCs (Song et al., 1998; Luscher et al., 1999; Luthi et al., 1999; Lee et al., 2002; Evers et al., 2010). The NSF peptide-sensitive AMPARs are thought to represent the constitutively cycling pool of AMPARs involved in the maintenance of synaptic strength. We used peptides fused to EGFP for easier identification of transfection (Lee et al., 2002). Co-expressing pepR845A peptide that specifically blocks NSF interaction with GluA2 greatly attenuated S-SCAM-induced increase of sGluA2 (299 ± 13 vs 147 ± 11%, S-SCAM + GFP vs S-SCAM + pepR845A; p < 0.001; Fig. 7 C,D). In contrast, the negative control peptide pepK844A that interferes with neither NSF- nor AP2- interaction had no significant effect (299 ± 13 vs 270 ± 13%, S-SCAM + GFP vs S-SCAM + pepK844A; p = 0.14). Consistent with previous results (Lee et al., 2002), the expression of pepR845A alone did not change sGluA2 levels significantly (100 ± 8 vs 98 ± 11%, GFP vs pepR845A; p = 0.87). This is perhaps due to the fact that surface staining of AMPAR, unfortunately, cannot distinguish between synaptic and extra-synaptic AMPAR in the dendritic spines (Lee et al., 2002). Thus, pepR845A specifically targeted sGluA2 pool related to S-SCAM overexpression. The effect of pepR845 was specific for sGluA2, as it did not show significant effect on the sGluA1 levels increased by S-SCAM (247 ± 13 vs 239 ± 10%, S-SCAM + GFP vs S-SCAM + pepR845A; p = 0.61).
To further support GluA2-specific effect of S-SCAM, we performed electrophysiological experiments measuring rectification index (measured by peak AMPA amplitude ratios at −60/+40 mV). If S-SCAM can support the incorporation of AMPAR to synapse in a GluA1-dependent manner, S-SCAM co-expression with GluA1 should increase the rectification index (overexpressed GluA1 forms inward rectifying homomeric receptors; Shi et al., 2001). As shown in Fig. 7E, GluA1 co-expressed with S-SCAM did not change the rectification index of transfected neurons (1.44 ± 0.1 vs 1.43 ± 0.2, Non-txf vs S-SCAM + GluA1, p = 0.95), while GluA1 co-transfected with a truncated α-CaMKII (tCaMKII-α) clearly increased the rectification index as expected (1.48 ± 0.1 vs 2.0 ± 0.2, Non-txf vs GluA1 + tCaMKII-α, p < 0.01). These data further support our notion that S-SCAM is regulating AMPAR through GluA2, not via GluA1.
In C. elegans, MAGI-1L regulates AMPAR through the PDZ-0 domain (Emtage et al., 2009). However, the deletion of PDZ-0 of S-SCAM did not impair the S-SCAM’s ability to increase sGluA2 (372 ± 32 vs 395 ± 27%, S-SCAM WT vs ΔPDZ-0, p = 0.61; Fig. 7F).
Both S-SCAM overexpression and RNAi-mediated knockdown data suggest that S-SCAM maintains AMPARs at synapses. Does S-SCAM support activity-dependent changes in the AMPA levels at synapses? To examine the role of S-SCAM in synaptic plasticity, we first examined the effect of S-SCAM on AMPAR internalization, which is a molecular mechanism underlying hippocampal LTD (Lee et al., 2002). Both Non-txf and GFP-transfected hippocampal neurons briefly treated with NMDA (2 min, 50 µM NMDA) showed a great increase in the internalized AMPAR levels as visualized by anti-GluA2 antibody (>3–5 fold increase compared to untreated control condition; p < 0.001; Fig. 8A,B). In contrast, S-SCAM overexpressed neurons did not show such NMDA-induced internalization at all. Further, the basal level of AMPAR internalization (untreated) in S-SCAM overexpressed neurons was decreased (1.0 ± 0.09 vs 0.5 ± 0.05, Non-txf vs S-SCAM, p < 0.01). However, S-SCAM-transfected neurons showed normal AMPA-induced internalization (> 6.5 fold; p < 0.001), which is comparable to Non-txf and GFP-transfected neurons (> 4 fold and > 7 fold, respectively). Therefore, S-SCAM overexpression resulted in the failure of AMPAR internalization specifically related to NMDA receptor activation. These results suggested that S-SCAM overexpression impairs the formation of hippocampal LTD. Consistent with this idea, hippocampal CA1 neurons infected with S-SCAM-virus indeed showed no induction of LTD (Fig. 8C,D; 80 ± 7 vs 91 ± 14% of baseline after 25–30 min, LTD vs unpaired pathway; n = 10; p = 0.42). In contrast, neighboring uninfected CA1 neurons exhibited a pathway-specific robust and stable LTD (49 ± 7 vs 90 ± 6% of baseline after 25–30 min, LTD vs unpaired pathway; n = 10; p < 0.001). Inclusion of APV in the bathing medium completely abolished the LTD formation (95 ± 7 vs 89 ± 10% of baseline after 25–30 min, LTD vs unpaired pathway; n = 6; p = 0.6; Fig. 8E), indicating that the LTD is NMDAR-dependent. CA1 neurons transfected with S-SCAM RNAi showed normal LTD indistinguishable from Non-txf neighboring neurons (57 ± 5 vs 53 ± 6% of baseline after 25–30 min, S-SCAM RNAi vs Non-txf; n = 10, p = 0.61; Fig. 8E,F). Taken together, these results suggest that S-SCAM is not involved in the regulation of LTD-forming AMPAR pool.
In this study, we elucidate a novel and essential function of S-SCAM as a synaptic AMPAR scaffold. From overexpression experiments, we found that increasing S-SCAM levels promotes enlargement of dendritic spines, changes the molecular composition of PSD proteins, enhances AMPAR-mediated synaptic transmission, and blocks the induction of hippocampal LTD. Conversely, knockdown studies demonstrated that the loss of S-SCAM decreases synaptic AMPAR levels, reduces the size of dendritic spines, and severely reduces excitatory synapse numbers. Therefore, S-SCAM is an important scaffolding molecule involved in the synaptic protein organization and the control of synaptic AMPAR.
The most significant finding here is the novel and essential role of S-SCAM in the maintenance of AMPAR at synapses. One of the prevailing hypotheses in the AMPAR trafficking is that two separate pools of AMPARs contribute to differential aspects in AMPAR trafficking – constitutive and regulated pathways (Shi et al., 2001; Derkach et al., 2007; Shepherd and Huganir, 2007; Kessels and Malinow, 2009). The former is important for the maintenance of synaptic strength during protein turnover, while the latter is involved in changing synaptic AMPAR levels during synaptic plasticity. In this model, different AMPAR subunits play leading roles in the differential phases of AMPAR trafficking: GluA1 plays a dominant role in the activity-dependent insertion of AMPAR during LTP, while GluA2 is involved in the maintenance of synaptic AMPAR levels and activity-dependent removal of AMPAR from synapses during LTD. Accumulating evidence supports that PSD-95 is important for the regulated pathways. For example, PSD-95 overexpression strongly enhanced AMPA receptor - mediated synaptic transmission by molecular mechanisms similar to LTP, in that it drives GluA1-containing receptors to synapses, converts silent synapses to functional synapses, occludes LTP, and enhances LTD (El-Husseini et al., 2000; Schnell et al., 2002; Beique and Andrade, 2003; Stein et al., 2003; Ehrlich and Malinow, 2004; Nakagawa et al., 2004; Elias et al., 2006). On the other hand, knockdown of PSD-95 reduced synaptic AMPAR levels (Nakagawa et al., 2004; Prange et al., 2004), and impaired LTD (Ehrlich et al., 2007; Xu et al., 2008) without affecting synapse numbers significantly (Elias et al., 2006) (but see also Beique and Andrade, 2003).
In contrast to the regulated pathway, the identity of scaffolding protein(s) involved in the constitutive pathway has remained unclear. Our results suggest that S-SCAM is the scaffolding molecule important for the constitutive pathway. Multiple lines of evidence support this notion: First, S-SCAM increases synaptic AMPAR levels through a GluA2-dependent mechanism, as shown by GluA subunit specific knockdown and NSF-peptide experiments (Fig. 5). Second, the increase of AMPAR by S-SCAM is an activity-independent process (Fig. 3E,F), consistent with the delivery mechanism of GluA2 receptors replacing GluA1(Shi et al., 2001). Third, S-SCAM is required for maintaining AMPAR levels even in the presence of PSD-95 overexpression, as indicated by knockdown experiments (Fig. 4, ,6).6). Fourth, the knockdown of S-SCAM led to the loss of excitatory synapses (Fig.4H), while S-SCAM overexpression did not increase the number of excitatory synapses (cf. PSD-95 overexpression increases it). Fifth, S-SCAM does not support activity-dependent changes of synaptic AMPA receptor levels, as demonstrated by the blockade of LTD after overexpression and by normal LTD after knockdown (Fig. 8). Thus, a pool of synaptic AMPARs anchored by S-SCAM is expected to play a “house-keeping” role during synaptic plasticity. Taken together, we propose a model in which S-SCAM stabilizes a maintenance pool of AMPARs at synapses and PSD-95 (together with other PSD-MAGUKs) regulates a plasticity-related pool of AMPAR.
How does S-SCAM serve a differential function from PSD-95 in the regulation of AMPA receptors and dendritic spines? We speculate that it arises from the overlapping but differential protein-protein interaction profile and/or post-translational modifications of S-SCAM protein. For example, S-SCAM does not have a palmitoylation site and lacks AKAP interaction, both of which were shown to be important for activity-dependent AMPAR internalization and LTD (El-Husseini Ael et al., 2002; Bhattacharyya et al., 2009). In addition, S-SCAM interacts with N-cadherins through β–catenin (Nishimura et al., 2002; Okabe et al., 2003), which may provide a mechanism for its stabilization at synapses during plastic changes. It is intriguing that S-SCAM promoted dendritic spine enlargements in the absence of increased synaptic levels of Shank (Fig. 2C) and SPAR (data not shown), critical molecules involved in dendritic spine enlargement/maturation (Pak et al., 2001; Sala et al., 2001). S-SCAM may increase dendritic spines through GluA2 subunits, whose extracellular interaction with N-cadherin mediates dendritic spine enlargement (Passafaro et al., 2003; Saglietti et al., 2007). Further studies are necessary to examine these possibilities.
Our finding that S-SCAM plays an important role in the regulation AMPAR is not without precedents. In C. elegans, the interaction of MAGI-1L (long form of MAGI-1) with the AMPA-like subunit GLR-2 has shown to be important for the synaptic localization of GLR-1/2 (Emtage et al., 2009). However, the molecular details of the action are different from the vertebrate system. While GLR-1/2 directly binds to MAGI-1L through PDZ interaction, vertebrate GluAs interact with S-SCAM indirectly via TARP (Deng et al., 2006), like PSD-95. Furthermore, the first PDZ domain (PDZ-0) is important for the synaptic GLR regulation by MAGI-1L in C. elegans, but PDZ-0 is not important for AMPAR regulation (Fig. 7F). Further detailed studies on the S-SCAM–TARP interaction are necessary to establish molecular mechanisms underlying S-SCAM-mediated regulation of AMPARs.
It is remarkable that S-SCAM did not increase sGluA1 levels in neurons with GluA2 knockdown, suggesting a GluA2-dependent mechanism. Since GluA2 plays an important role in the assembly and synaptic incorporation of heteromeric AMPAR (Wenthold et al., 1996; Lu et al., 2009), GluA2 knockdown may have caused a poor assembly and/or trafficking of GluA1-containing receptors to neuronal surface. However, this is unlikely the case in our condition, since GluA2 knockdown showed normal sGluA1 levels (Fig. 7B). Consistent with this, homomeric GluA1 receptors were detected at synapses shortly after GluA2 deletion (Lu et al., 2009). Furthermore, S-SCAM failed to accommodate GluA1 homomeric receptors at the synapses (Fig. 7E). Therefore, these data indicate that S-SCAM regulates AMPAR in a GluA2-dependent manner. At present, the molecular bases of GluA2-dependence of the S-SCAM action are unclear, as TARP binds all GluA subunits. However, one could imagine that GluA2-specific binding proteins such as GRIP, NSF, PICK1, and/or yet unidentified protein(s) may work together with S-SCAM to provide a GluA2-dependent mechanism at synapses. Another intriguing question is what role S-SCAM plays in the control of inhibitory synaptic transmission, since S-SCAM is also present in inhibitory synapses (Sumita et al., 2007).
Previously, mutant mice lacking the expression of S-SCAM α did not show drastic abnormalities in dendritic spine density, albeit they showed elongated spine length (Iida et al., 2007). However, the S-SCAM α knockout mice still express other S-SCAM isoforms of β and γ. S-SCAM β is the most abundant form expressed in the forebrain (Hirao et al. 2000; Deng et al., 2006). The main difference between the α and β isoform is the presence of PDZ-0 domain. As ΔPDZ-0 mutant increased sGluR2 levels as effectively as WT (Fig. 7F), PDZ-0 domain is not required for the S-SCAM-mediated regulation of AMPAR. Thus, it is highly likely that the remaining S-SCAM β (and possibly γ) in the S-SCAM α knockout mice are sufficient to control AMPAR and dendritic spine.
In addition to uncovering the role of S-SCAM in excitatory synaptic transmission, our study also contributes to the understanding pathobiology of neurological diseases, since S-SCAM gene duplication and deletion were found in individuals with schizophrenia and IS, respectively (Marshall et al., 2008; Walsh et al., 2008). Although these discoveries do not establish a causal relationship between S-SCAM and schizophrenia and/or IS per se, out results provide valuable insight on the pathological bases of these diseases. Here we have demonstrated that changing levels of S-SCAM profoundly affect synaptic transmission and synaptic plasticity. For example, increasing levels of S-SCAM enhance AMPAR-mediated synaptic transmission, but, at the same time, renders neurons less adaptable for at least one type of synaptic plasticity, LTD. Conversely, decreased S-SCAM levels profoundly reduce the number of excitatory synapses and weaken synaptic transmission. Incidentally, elevated levels of S-SCAM, a highly likely situation under S-SCAM gene duplication, caused the reduction in the number of dendritic spines, which is consistent with postmortem studies on the brains of patients with schizophrenia (Garey et al., 1998; Glantz and Lewis, 2000; Sweet et al., 2009; Penzes et al., 2011). Furthermore, aberrant Neuregulin-1 (NRG-1)–ErbB4 receptor signaling is one of the prevailing hypotheses in schizophrenia research (Mei and Xiong, 2008). Since S-SCAM also binds to ErbB4 receptors through PDZ interaction (Buxbaum et al., 2008), further studies on the role of S-SCAM in the NRG-1/ErbB4 signaling will provide important clues on better understanding pathological bases of the disorder.
This work was supported by the Whitehall foundation (S.H.L.) and US National Institute of Health grants, MH078135 (S.H.L.) and AG032320 (N.G.). We thank Dr. Yutaka Hata (Tokyo Medical and Dental University, Tokyo, Japan) for providing S-SCAM expression plasmids and Dr. Qing-song Liu for advice on electrophysiology and data analyses.