|Home | About | Journals | Submit | Contact Us | Français|
The proteoloytic machinery comprising metalloproteases and γ-secretase, an intramembrane aspartyl protease involved in Alzheimer’s disease, cleaves several substrates besides the extensively studied amyloid precursor protein (APP). Some of these substrates, such as N-cadherin, are synaptic proteins involved in synapse remodeling and maintenance. Here we show, in rat and mice that metalloproteases and γ-secretase are physiologic regulators of synapses. Both proteases are synaptic, with γ-secretase tethered at the synapse by δ-catenin, a synaptic scafolding protein which also binds to N-cadherin and, through scaffolds, to α-amino-3-hydroxyl-5-methyl-4-isoxazole-propionate receptor (AMPAR) and a metalloprotease. Activity-dependent proteolysis by metalloproteases and γ-secretase takes place at both sides of the synapse, with the metalloprotease cleavage being N-methyl-D-aspartic acid receptor (NMDAR)-dependent. This proteolysis decreases levels of synaptic proteins and diminishes synaptic transmission. Our results suggest that activity-dependent substrate cleavage by synaptic metalloproteases and γ-secretase modifies synaptic transmission, providing a novel form of synaptic autoregulation.
Metalloproteases and γ-secretase act in succession to cleave single-pass transmembrane proteins. Metalloproteases and the closely related A Disintegrin And Metallopeptidase (ADAMs), or for some substrates, β-secretase (BACE1) initiate the proteolytic pathway by shedding the membrane protein substrate’s ectodomain (Thinakaran and Koo, 2008). γ-secretase, a multi-subunit aspartyl protease assembled from four proteins (presenilin 1 (PS1), nicastrin, anterior pharynx-defective 1 (Aph1) and presenilin enhancer 2 (Pen2)), cleaves the resulting C-terminal fragment (CTF) within its transmembrane domain (McCarthy et al., 2009). While this pathway has been studied most extensively for amyloid precursor protein (APP), whose cleavage yields the Aβ peptide, the metalloprotease/γ-secretase proteolytic machinery also cleaves other substrates, especially proteins implicated in synapse remodeling and maintenance, including EphRs, ephrins, and cadherins (Dalva et al., 2007; McCarthy et al., 2009). However, these physiological roles have received very little attention.
The precise location of the metalloprotease/γ-secretase proteolytic machinery is still unclear. ADAM10, ADAM17 and membrane type 5 matrix metalloproteinase (MT5-MMP), the three known metalloproteases that shed the ectodomains of γ-secretase substrates, have been localized at synapses (Monea et al., 2006; Marcello et al., 2007; Malinverno et al., 2010), whereas γ-secretase has been reported mainly in membranes of endosomes, though its presence at the plasma membrane has also been suggested (Lah et al., 1997; Georgakopoulos et al., 2000; Kaether et al., 2006). Interestingly, PS1, the catalytic subunit of γ-secretase, binds to a synaptic protein, δ-catenin, which regulates actin polymerization and cell adhesion (Kosik et al., 2005). We have previously shown that δ-catenin is present in the postynaptic density (PSD) (Silverman et al., 2007), where it links the N-cadherin intracellular domain via synaptic scaffolds to MT5-MMP, a metalloprotease that cleaves N-cadherin (Monea et al., 2006) and synaptic proteins particularly α-amino-3-hydroxyl-5-methyl-4-isoxazole-propionate receptor (AMPAR) (Silverman et al., 2007). Therefore, the proteolytic machinery formed by a metalloprotease and γ-secretase may be present in synaptic complexes with their substrates, where it could modulate synaptic function.
Synaptic dysfunction is a major early event of Alzheimer disease (Knobloch and Mansuy, 2008). Considering the putative role of these proteases in Alzheimer disease (Thinakaran and Koo, 2008), we here investigate possible physiologically-relevant synaptic functions of metalloproteases and γ-secretase. We find that the metalloprotease and γ-secretase proteolytic machinery operates on both sides of the synapse and cleaves synaptic N-cadherin. The metalloprotease-mediated cleavage is regulated by N-methyl-D-aspartic acid receptor (NMDAR)-dependent synaptic activity, whereas γ-secretase activity is insensitive to the activity of the synapse. This proteolytic machinery down-regulates synaptic function. We find a protease-dependent regulation of glutamatergic neurotransmission in the hippocampal neurons in which vesicle recycling and synaptic protein levels are modified. Because the pathway inhibits synaptic transmission in response to synaptic activity, we propose that this metalloprotease/γ-secretase pathway provides a novel form of activity-dependent synaptic autoregulation.
8 weeks old rats of either sex were used for fractionation experiments as indicated.
Antibodies used were: mouse anti-PSD-95 antibody (1:500; clone K28.43 or clone K28/86.2; NeuroMab, Davis, CA), rabbit anti-GluA2/3 antibody (1 µg/ml; Millipore), mouse anti-δ-catenin (1/200; BD Biosciences), mouse anti-N-cadherin (1/100; BD Biosciences), mouse anti-synaptophysin (1/1000; Sigma-Aldrich), mouse anti-synaptotagmin 1 (1/100: Synaptic Systems), rabbit anti-MT5-MMP (cytoMT5-MMP, 1.5 µg/ml; #2850,) made in our lab and described previously (Monea et al., 2006), rabbit Anti-EphB2 (1/1000 WB, Dr M. Greenberg), rabbit anti-ADAM 10 (1/200 for IF, 1/500 for WB, Sigma-Aldrich), Rabbit anti-ADAM 17 (1/200 for IF, 1/500 for WB, Abcam), mouse anti-BACE (1/100, 3D5, Dr Vassar), rabbit anti-presenilin N-terminal (1/500 IF and 1/1000 WB, PS1NT, Calbiochem), rabbit anti-presenilin loop domain (1/1000 WB, Calbiochem), rabbit anti-nicastrin (Dr. P. Fraser, 1/1000), anti-GluR1 (1/1000; Millipore), rabbit anti-tubulin (1/1500; Sigma-Aldrich), mouse anti-APP (clone C1/6.1 and m3.2, 1/1000, and for IF, 1/1000 Upstate).
Secondary antibodies were used at 1/1000, including monoclonal secondary antibodies (Alexa Fluor color; Invitrogen), isotype-specific monoclonal secondary antibodies IgG1 (Alexa Fluor color; Invitrogen), IgG2A (Alexa Fluor color; Invitrogen), and polyclonal secondary antibodies (Alexa Fluor color; Invitrogen).
Inhibitors used were: GM6001 (10 µM, Ilomastat; Chemicon), epoxomycin (1 µM, peptide International), L-685458 (1–10 µM γ-secretase inhibitor X, Calbiochem), DAPT (10 µM, γ-secretase inhibitor IX, Calbiochem), PP2 (10 µM,Calbiochem), KN-93 (CAMKII inhibitor 5 µM, Calbiochem), D-APV (100 µM Sigma-Aldrich), JNK inhibitor II (50 µM, Calbiochem), bicuculline (40 µM, Tocris), NMDA (40 µM, Sigma-Aldrich), glutamate (40 µM, Sigma-Aldrich), 4-aminopyridine (4AP, 100 µM, Tocris).
Cultures were prepared from E19 Sprague-Dawley rat embryonic brain tissue. Animals were killed by CO2 in compliance with New York University Medical Center's Institutional Animal Care and Use Committee. Hippocampal and cortical primary neurons were prepared as described previously (Osten et al., 1998). Neurons were plated at a density of 100,000 cells on poly-L-lysine-coated glass coverslips in a six-well plate for immunofluorescence or 3.5 millions on poly-L-lysine-coated 10cm dishes for biochemical experiments. Neurons were grown in Neurobasal medium with B27 (Invitrogen, San Diego, CA). At the first change of media, a one-time dose of the drug AraC (4 µM; Sigma-Aldrich) was added to inhibit growth of dividing cells, for immunofluorescence experiments (Macaskill et al., 2009).
At 18–21 days in vitro (DIV), neurons were fixed with 2% paraformaldehyde (Electron Microscopy Sciences, Fort Washington, PA) in 0.12 M sucrose in PBS (15 min at room temperature), permeabilized with 0.2% Triton X-100 (5 min at room temperature), blocked in 10% bovine serum albumin (BSA) in PBS, and then incubated with primary antibodies in 3% BSA. Cells were then washed three times in 1× PBS. Secondary antibodies conjugated to Alexa probes were incubated in 3% BSA for 1 h and mounted on slide. Images were captured with a Zeiss (Thornwood, NY) LSM 510 Meta laser-scanning confocal microscope using a Plan-Apochromat 63× objective and LSM 510 Meta software.
14 DIV neurons were treated overnight with inhibitors and incubated with synaptotagmin antibody for 5 min. Neurons were washed, fixed and permeabilized as described in the previous section. 10 neurons per coverslip were imaged and 9 separate experiments were performed. Each experiment was averaged. The 9 experiments were then averaged.
Pentobarbital-anesthetized adult Sprague-Dawley rats were perfused intracardially with mixed aldehyde fixatives (0.1% glutaraldehyde/4% depolymerized paraformaldehyde or 2% glutaraldehyde/2% paraformaldehyde) dissolved in 0.1M phosphate buffer (pH 7.4; PB) after a quick rinse with heparinized saline. After overnight postfixation in the same fixative at 4° C, 50 µm brain sections were cut on a Vibratome and collected in PB. Sections containing cortex and hippocampus were incubated 30 minutes on a shaker at room temperature with 1% sodium borohydride in phosphate-buffered saline (PBS) to block free aldehyde groups, then in 10% normal goat serum (to block nonspecific antibody binding). Sections were incubated overnight in primary antibody (rabbit anti-nicastrin, 1:1000–1:2000), rinsed, and then for two hours in goat anti-rabbit IgG (1:100), conjugated to ultrasmall gold particles (~1 nm; British BioCell). Sections were then rinsed in 0.01 M acetate buffer, and gold particles silver-intensified, using an IntenSE M kit (Amersham Pharmacia). Sections for electron microscopy were postfixed in 0.5–1% osmium tetroxide in PB for 1 hour, rinsed in 0.1 M maleate buffer pH 6, then stained en bloc with 1% uranyl acetate 1 hour. After dehydration in ascending ethanol series and propylene oxide, sections were infiltrated with Epon/Spurr resin, flat-mounted between sheets of Aclar plastic, and polymerized at 60° for 48 hours. Seventy nm sections from selected areas were cut with a diamond knife, collected on 300 mesh copper grids, stained with uranyl acetate and Sato's lead, and examined with a Tecnai 10 transmission electron microscope (Philips). Images were collected with a 12-bit 1024 × 1024 pixel CCD camera (Gatan); final photomicrographs were prepared using Adobe Photoshop to compose, crop, and adjust contrast.
18 DIV cortical or hippocampal neurons in culture were stimulated with bicuculline and 4-aminopyridine for 1h or bicuculline for 24h, washed 3 times with PBS and lysed with 50mM Tris pH=7.4 and 2% SDS. Equal amount of total lysate (15 –20 µg) were loaded on 8 to 12% SDS-Page gels or precast gradient 4–20% SDS-Page gels (Thermo Scientific). Western blot was further performed with N-cadherin. Tubulin was used as loading control. For most of the experiments otherwise indicated, we quantified the ratio CTF1/full length protein. For NMDA and glutamate stimulation, cortical neurons were stimulated for the time indicated. Media was then exchanged for conditioned media for another hour. Cells were washed and lysed as before. Inhibitors were applied 15 min. prior to stimulation, during stimulation and for the hour after stimulation.
Synaptosomal and PSD fractions from rat and mouse brain and primary hippocampal cultures were prepared as described previously (Jordan et al., 2004). Equal amount of fractions (10–15 µg) were loaded on SDS-PAGE gel. Western blots were probed with different antibodies. Synaptophysin was used as loading control for whole brain (WB), total membrane (P2) and synaposomes (syn) fractions and PSD-95 for PSD fractions.
Pre and postsynaptic fractions were prepared from synaptosomes by extraction at differential buffer pH as described previously (Phillips et al., 2001). Equal amount of fractions (10–15 µg) were loaded on SDS-PAGE gel. Western blots were probed with different antibodies. Synaptophysin was used as loading control for WB, P2, syn and presynaptic fractions. PSD-95 was used as loading control for postsynaptic fractions.
For the biotinylation experiments, cortical neurons in culture were washed and incubated with 2mg/ml Sulfo-NHS-SS-Biotin (Pierce, Rockford, IL) in PBS for 15 min at 4°C. Non-reacted biotin was quenched with 50 mM glycine in PBS, and cells were rinsed twice with cold PBS. Total membrane (P2) and synaptosome fractions were prepared in RIPA immunoprecipitation buffer (50mM Tris pH 7.4,150mM NaCl, 5mM EDTA, 0.1% SDS, 1% NP-40, 0.5% sodium deoxycholate) and complete protease inhibitor cocktail (Roche Products). Biotinylated surface proteins were purified with Ultralink Immobilized NutrAvidin beads (Pierce) and analyzed by SDS-PAGE and western blot. Densitometry of the bands and quantification of surface proteins were performed using Image J software (NIH), and p values were determined by standard t test and ANOVA analysis when indicated. Equal amount of proteins were loaded on gels.
The synaptosomal fraction was also solubilized in immunoprecipitation buffer (50 mM Tris pH 7.4,150 mM NaCl, 5 mM EDTA and CHAPSO 1%) for 2 h at 4°C and cleared at 16,000 × g for 10 min at 4°C. Supernatants containing 1 mg of protein were incubated with the antibodies, rabbit anti-nicastrin, rabbit anti-PS1NT and a rabbit igG negative control (2 µg of antibody), overnight, then 20 µg of protein G or protein A beads (Santa Cruz Biotechnology) as determined by the genotype of the antibody were added at the reaction for an additional 2 h, followed by two washes with IP buffer containing 350 mM NaCl and once with IP buffer containing 150 mM NaCl and then eluted with loading buffer.
In vitro γ-secretase assays were performed as previously described (Farmery et al., 2003). Briefly, synaptosomal extracts from rat brain were solubilized in CHAPSO buffer and incubated with a fluorogenic peptide probe at 37°C overnight, in the presence of inhibitors. The probe contains a consensus aspartic acid cleavage site, which when proteolyzed by γ-secretase, generates a fluorescent peptide whose fluorescence is measured on a plate reader with excitation wavelength of 340 nm and emission wavelength of 490 nm. Results from 6 independent replicates.
mEPSCs were assayed in cultured hippocampal neurons at DIV14. Neurons were voltage-clamped with the whole cell ruptured patch technique throughout the experiment. The bath solution consisted of (in mM) NaCl (119), KCl (5), HEPES (20), CaCl2 (2), MgCl2 (2), glucose (30), glycine (0.001), bicuculline (0.01), pH 7.3, osmolarity adjusted to 330 mOsm with sucrose. The solution in the whole cell patch electrode consisted of (in mM) K-gluconate (130), KCl (10), MgCl2 (5), EGTA (0.6), HEPES (5), CaCl2 (0.06), Mg-ATP (2), GTP (0.2), leupeptin (0.2), phosphocreatine (20), and creatine-phosphokinase (50 U/ml). 1 µM tetrodotoxin was also added to the bath to suppress action potentials. Currents were recorded with a Warner amplifier (model PC-501A) (CT), and filtered at 1 kHz. In order to eliminate artifacts due to variation of the seal properties, the access resistance was monitored for constancy throughout all experiments. The recordings were digitized (Digidata 1440A, Molecular Devices) and analyzed with the minianalysis program (version 4.0) from Synaptosoft, Inc. (GA).
Although γ-secretase in neurons is found mainly in endosomes (Kaether et al., 2006), synaptic pools of both γ-secretase and metalloproteases have been suggested (Lah et al., 1997; Georgakopoulos et al., 2000; Peiretti et al., 2003; Monea et al., 2006; Marcello et al., 2007). Indeed, the synaptic scaffolding proteins glutamate receptor-interacting protein and AMPA receptor-binding protein (GRIP/ABP) and synapse-associated protein 97 (SAP97) associate respectively with MT5-MMP and ADAM10/17, proteases involved in the ectodomain shedding of γ-secretase substrates (Peiretti et al., 2003; Monea et al., 2006; Marcello et al., 2007). Moreover, γ-secretase interacts with δ-catenin, a scaffolding protein that also binds ABP/GRIP (Silverman et al., 2007). These interactions suggest the presence of the proteases in multiprotein complexes at synapses. To test this possibility, we performed immunofluorescence on 21 DIV dissociated hippocampal neurons. We found three metalloproteases that can shed ectodomains of γ-secretase subtrates (Reiss et al., 2005; Monea et al., 2006; Postina, 2008), ADAM10, ADAM17 and MT5-MMP (Fig. 1A, Arrows), and two components of γ-secretase, nicastrin and presenilin 1 (PS1) (Fig. 1B, Arrows) colocalized with the postsynaptic marker, postsynaptic density-95 (PSD-95).
Western blotting demonstrated that metalloproteases and two γ-secretase components, nicastrin and PS1 (including the PS1 N- and C-terminal fragments (PS1NTF and CTF)), were present in synaptosomes fractionated from adult rat brain (Fig. 1C and D). We used differential pH extraction to fractionate synaptosomal pre- and postsynaptic components (Phillips et al., 2001). Western blotting revealed the presence of the three sheddases, ADAM10, ADAM17, and MT5-MMP in pre- and postsynaptic fractions, with a postsynaptic enrichment (Fig. 1C). PS1-NTF and PS1-CTF were distributed equally between the pre- and postsynaptic fractions (Fig. 1D), while nicastrin showed presynaptic enrichment (Fig. 1D). However, the integrity of the γ-secretase complex is highly detergent-sensitive (Li et al., 2000); thus, nicastrin may be lost from the postsynaptic fraction, since it may detach from PS1, during the fractionation process. We therefore used pre-embedding immunogold electron microscopy to confirm that the γ-secretase complex may be both pre- and postsynaptic. We detected nicastrin silver-intensified label postsynaptically, at the edge of the PSD (Fig. 1Ei), presynaptically (Fig. 1Eii) and on endomembranes (Fig. 1Eiii). We also detected, the γ-secretase substrates N-cadherin and EphRB2 were detected at the synapse by immunofluorescence (Fig. 1F, arrows) (Dalva et al., 2007; McCarthy et al., 2009) and western blot revealed their presence in both pre- and postsynaptic fractions (Fig. 1G). Thus we conclude that the proteolytic machinery is present on both side of the synapse. These data raise the possibility that the metalloprotease and γ-secretase are tethered to the PSD through a direct interaction with synaptic scaffolding proteins.
We wondered whether the γ-secretase components, which are assembled in the secretory pathway (Kaether et al., 2006), reach the synapse as an active complex. PS1 co-immunoprecipitated with nicastrin from rat brain synaptosomes that were solubilized in CHAPSO to maintain γ-secretase integrity (Li et al., 2000), demonstrating that PS1 and nicastrin reach the vicinity of synapses as a physical complex (Fig. 2A). Furthermore, we detected γ-secretase enzymatic activity in solubilized synaptosomes, using a γ-secretase fluorogenic substrate (Farmery et al., 2003). This activity was partially but significantly inhibited by two γ-secretase inhibitors, DAPT and L-685458 (n=6, 1.5 fold; p<0.005, Fig. 2B), demonstrating the presence of enzymatically active γ-secretase at synapses. Total membrane (P2) was used as positive control (data not shown). Finally, biotinylation of surface proteins in live neurons, followed by synaptosomal fractionation, demonstrated both nicastrin and PS1NT at the synaptic plasma membrane (Fig. 2C). The GluR1 AMPAR subunit (a surface protein used as positive control) was also detected, whereas tubulin (an intracellular protein serving as negative control) was absent (Fig. 2C). These data suggest that γ-secretase components are present at the plasma membrane of synapses in an enzymatically-active complex.
We next analyzed the physical interactions that maintain these proteases at the synapse. PS1 binds to the adherens junction protein, δ-catenin (Zhou et al., 1997; Levesque et al., 1999; Tanahashi and Tabira, 1999), which associates with synaptic N-cadherin (Kosik et al., 2005). We had previously shown that δ-catenin is a component of the PSD, where it is linked to the AMPA receptor subunit GluA2 by the scaffolding proteins GRIP/ABP (Silverman et al., 2007). To determine if δ-catenin is required for tethering PS1 at synapses, we compared PS1 levels in brain subcellular fractions (Jordan et al., 2004) from δ-catenin knockout and wild-type mice (Israely et al., 2004) (Fig. 2Di). PS1 was decreased in the crude postsynaptic density fraction (cPSD) in the knockout mice but not in other fractions (cPSD: 0.5, p<0.005; n=7), suggesting that for PS1 to associate with the PSD it requires the PS1-δ-catenin interaction, while PS1 outside the PSD does not depend on δ-catenin. This suggests that the PS1-δ-catenin interaction selectively tethers the synaptic fraction of γ-secretase. Unlike PS1, levels of full length N-cadherin were decreased in all fractions in δ-catenin knockout mice, especially in synaptosomes (syn) and the cPSD, suggesting that δ-catenin is necessary for the stabilization of synaptic N-cadherin [Fig. 2Dii; whole brain (WB): 0.79; total membrane (P2): 0.61, p<0.005; syn: 0.67, p<0.0005; and cPSD: 0.64, p<0.05; n=7]. Interestingly, GluA2 was also decreased in the total membrane, P2, synaptosome and cPSD fractions (P2: 0.58, P<0.005, syn: 0.9 and cPSD: 0.72, P<0.05; n=7 Fig 2Diii), consistent with previous evidence that δ-catenin tethers GluA2-containing AMPA receptors at synapses (Silverman et al., 2007). Taken together with previous reports of synaptic tethering of metalloproteases by PSD proteins (Peiretti et al., 2003; Monea et al., 2006; Marcello et al., 2007), our data suggest that synaptic scaffolds stabilize the metalloprotease/γ-secretase proteolytic machinery at synapses, positioning the proteases close to their synaptic substrates.
As shown previously (Marambaud et al., 2003; Reiss et al., 2005; Uemura et al., 2006), N-cadherin ectodomain shedding by a metalloprotease generates a membrane-anchored C-terminal fragment, CTF1, which is further processed by γ-secretase to yield the soluble CTF2 fragment (Fig. 3A). To test whether this processing occurs at synapses, we prepared synaptosomes from cultured neurons treated with a broad metalloprotease inhibitor (ilomastat), γ-secretase inhibitors (L-685458 or DAPT), or with epoxomycin, a proteosome inhibitor. While metalloprotease inhibition decreased the CTF1 levels in synaptosomes, γ-secretase inhibition significantly increased the CTF1 levels suggesting that CTF1 production by metalloprotease cleavage and its subsequent processing by γ-secretase take place at synapses (n=8, 1.5 fold, p<0.05; Fig. 3B, C). The CTF2 fragment generated by γ-secretase cleavage of CTF1 was only detected in the supernatant of total lysate after treatment with the proteosome inhibitor, epoxomycin, suggesting that normally the cleaved fragment is rapidly degraded by the proteosome (Fig. 3B, bottom panel). Thus, we conclude that N-cadherin is cleaved by metalloproteases and γ-secretase at synapses. Subcellular fractionation showed the presence of the CTF1 fragment in both pre- and postsynaptic compartments, suggesting that N-cadherin cleavage occurs at both sides of the synapses (Fig. 3D).
Discrepant findings have been reported regarding the effect of synaptic activity on cadherins. Synaptic activity can either stabilize or cleave cadherins (Tanaka et al., 2000; Marambaud et al., 2003; Reiss et al., 2005; Uemura et al., 2006; Tai et al., 2007). To analyze the effects of activity on the metalloprotease and γ-secretase pathway, we first assayed the effects of glutamate and NMDA stimulation on N-cadherin cleavage (Lee et al., 1998; Tai et al., 2007). We detected an increase in N-Cadherin CTF1 levels in total lysate following brief glutamate stimulation (1 min: n=4, 3.2 fold; 3 min: n=7, 3.8 fold; p<0.05; Fig. 3E). Similarly, stimulating cultured neurons with NMDA produced a significant increase in N-cadherin CTF1 level in total lysate (1 min: n=12, 2.2 fold; 3 min: n=12, 2.2 fold; 5 min: n=11, 3.2; 15 min: n=11, 2.5 fold; p<0.05; Fig. 3F). A similar increase was also observed in synaptosomes, though the effect was not significant (1 min: n=11, 1.4 fold; 3 min: n=11, 2 fold and 5 min: n=8, 1.7 fold; Fig. 3F). Similar kinetics were observed when the ratio of CTF1/tubulin was evaluated (data not shown).
Addition of glutamate or NMDA to the culture medium stimulates both synaptic and extrasynaptic receptors, which might result in non-specific effects. To determine the effects of selective synaptic stimulation, we enhanced synaptic transmission by simultaneous application of bicuculline, a gamma-aminobutyric acid A receptor (GABAAR) antagonist and 4-aminopyridine (4AP), a K+ channel inhibitor and observed an increase in N-cadherin CTF1 production (n=15, 1.7 fold and 1.4 fold, p<0.05; Fig. 3G) (Ehlers, 2000; Tai et al., 2007). This effect was blocked by inhibition of NMDARs (D-APV, n=5, P<0.05, Fig. 4A). We conclude that synaptic activation stimulates CTF1 production in an NMDA receptor-dependent manner. Likewise, N-cadherin CTF1 production was decreased when action potentials were blocked with the sodium channel blocker, tetrodotoxin (TTX, n=4, 0.5 fold p<0.05, Fig. 4A). Thus activation of synaptic NMDARs stimulates the cleavage of N-cadherin by proteolytic machinery composed of a metalloprotease and γ-secretase.
To investigate the pathway by which NMDAR activation regulates N-cadherin cleavage, we stimulated cultured neurons with glutamate, with or without inhibitors of NMDARs or of specific kinases (Fig. 4B). Total lysate fractions were probed in western blots for the N-cadherin CTF1 fragment. As shown above (Fig. 3E), CTF1 fragment production increased after glutamate stimulation. The glutamate-induced increase in CTF1 fragment production was blocked by an NMDAR inhibitor (D-APV, n=4, 0.9 fold; p<0.05, Fig. 4B). The addition of a Src or JNK inhibitor (PP2 or JNK inhibitor II, respectively) also blocked the glutamate-induced increase in CTF1 fragment (n=6, respectively, 0.6 fold; p<0.005 and 0.75 fold; p<0.05, Fig. 4B). In contrast, KN93, an inhibitor of calmodulin-dependent Protein Kinase II (CaMKII) had no effect. These data suggest that glutamate stimulation of N-cadherin ectodomain shedding by a metalloprotease depends on the activation of the NMDA receptor and Src and JNK kinases.
We next examined whether glutamate stimulation differentially regulates the two N-cadherin cleavage steps. As shown above (Fig. 3F) NMDAR activation stimulates CTF1 production. A direct assay of CTF2 production, which reflects γ-secretase cleavage of CTF1, was unfeasible, because CTF2 was detected at very low levels only in the supernatants of total neuron lysates (Fig. 3B). As an alternative, we treated neurons with the metalloprotease inhibitor, ilomastat, to inhibit production of CTF1. In agreement with Fig. 3E, glutamate stimulation in the absence of any protease inhibitor increased the level of the CTF1 fragment (Fig 4Cb: n=16, 4 fold, p<0.05,). This increase was inhibited by ilomastat (Fig 4Cd compared to b, p<0.05,), suggesting that the metalloprotease is activated by glutamate stimulation. We reasoned that since ilomastat blocks the production of new CTF1, a decrease in CTF1 levels caused by glutamate stimulation in the presence of ilomastat would imply an activation of the γ-secretase cleavage step. We found that addition of ilomastat decreased the level of CTF1 compared to the control condition, as expected if γ-secretase cleaves CTF1 fragments that existed prior to glutamate stimulation (Fig. 4Cc: n=6, 0.56 fold,). However we did not observe a further decrease in CTF1 levels when neurons were first treated with ilomastat (to block CTF1 production) and then stimulated with glutamate (Fig. 4Cd compared to c,), suggesting that glutamate stimulation does not increase γ-secretase cleavage of CTF1. We conclude that glutamate stimulation regulates the metalloprotease cleavage of N-cadherin, but not the cleavage by γ-secretase.
To confirm this result, we assessed γ-secretase regulation by synaptic activity. By comparing the effects of a γ-secretase inhibitor (L-685458) on bicuculline-treated and untreated neurons, we could determine whether the bicuculline-dependant increase in synaptic transmission influences the rate of γ-secretase cleavage of CTF1. We analyzed two pairs of dishes of cultured neurons: one pair of unstimulated control dishes and a pair of bicuculline-treated dishes. In one dish of each pair, we inhibited γ-secretase for the duration of the bicuculline exposure. The N-cadherin CTF1 was quantified in the total lysates of the four dishes by western blots. In both the control and the L-685458 treated cultures, CTF1 fragment production increased with bicuculline treatment, as expected (Fig. 4Dc: 4.96 fold, p<0.05, and d: 4.84 fold; n=15). Inhibition of γ-secretase increased CTF1 fragment levels in both untreated and bicuculline-treated cultures (Fig. 4Db, control: n=15, 1.4 fold; d, stimulated: n=9, 4.84 fold). However, when we compared the effect of γ-secretase inhibition on bicuculline-treated and untreated control neurons, no significant difference was observed (Fig. 4D; ratio b/a=4.96 compared to d/c=4.03, p=0.53), confirming that synaptic activation did not regulate γ-secretase cleavage of CTF1.
We conclude that synaptic activation following bicuculline treatment or glutamate stimulation has little or no influence on the rate of γ-secretase cleavage of CTF1, suggesting that the increased rate of cleavage of N-cadherin induced by neuronal activity results specifically from activity-dependent stimulation of ectodomain shedding by a metalloprotease.
Cleavage of synaptic substrates by metalloproteases and γ-secretase might alter synaptic architecture or modify components of the transmission machinery; either effect could lead to functional changes. To determine whether metalloproteases or γ-secretase have an effect on synaptic function, we recorded mEPSCs from cultured hippocampal neurons treated overnight with the γ-secretase inhibitor, L-685458 (1 µM), or the metalloprotease inhibitor, ilomostat (10 µM) (Fig. 5). Changes in mEPSC frequency and amplitude are usually attributed to changes in pre- and postsynaptic functions, respectively (Wierenga et al., 2006). While we observed significant increases in both the frequency and the amplitude of mEPSCs in L-685458-treated neurons compared to vehicle-treated neurons (P<0.05), ilomostat-treated neurons showed a selective enhancement of mEPSC frequency (P<0.05) with no effect on mEPSC amplitude (P>0.05) [Fig. 5; Frequency: 2.3+/−0.2 (vehicle), 6.2+/−0.8 (ilomostat), 3.5+/−0.4 (L-685458); Amplitude: 13.9+/−0.4 (vehicle), 14+/−0.3 (ilomostat), 18.5+/−0.6 (L685458); N=17 (Vehicle), 15 (ilomostat), and 15 (L-685458)]. These results suggest that the activities of both γ-secretase and metalloproteases negatively regulate glutamatergic transmission in cultured hippocampal neurons; furthermore, γ-secretase activity appears to affect both pre- and postsynaptic functions, whereas metalloprotease activity selectively reduces presynaptic glutamate release. The decrease in mEPSC frequency could possibly also reflect a postsynaptic effect, for example one resulting from a reduction of spine or receptor numbers.
To explore the mechanistic basis for these electrophysiological changes, we assayed whether inhibition of metalloproteases or γ-secretase affects synaptic protein levels or functions. Inhibition of metalloprotease (ilomastat) or γ-secretase (L685458) activity did not change the level of the presynaptic marker synaptophysin, as assayed by immunofluorescence (total intensity, respectively 1.08 and 0.99, n=9, Fig. 6Ai and ii). However, an increase in the rate of synaptic vesicle cycling, as reflected in the level of synaptotagmin reuptake measured by an antibody-feeding assay (Matteoli et al., 1992), could be detected after metalloprotease, but not γ-secretase inhibition (total intensity, 1.2 and 0.97, n=5, Fig. 6Bi and ii). This effect is consistent with the increase in mEPSC frequency recorded after metalloprotease inhibition (Fig. 5), suggesting that a metalloprotease decreases the rate of synaptic vesicle recycling.
We next assayed changes in postsynaptic proteins after inhibition of the proteolytic machinery. Inhibition of metalloproteases or γ-secretase increased the level of PSD-95, as measured by immunofluorescence (total intensity, respectively 1.14 and 1.19, p<0.05, n=9, Fig. 6Ci and ii), in agreement with previous reports (Parent et al., 2005). Moreover, a significant increase of surface GluA2, but not N-cadherin, was detected by a live neuron biotinylation assay after inhibition of metalloproteases or γ-secretase (respectively, GluA2, 3.24 and 2.94, p<0.05 and N-cadherin 1.25 and 1.28; Fig. 6Di and ii). We conclude that the activity of the proteolytic machinery decreases levels of a postsynaptic scaffold and an AMPA receptor subunit, consistent with a role for these enzymes in attenuating synaptic transmission.
Proteolytic enzymes can regulate synaptic efficacy (Cartier et al., 2009; Mabb and Ehlers, 2010; Malinverno et al., 2010), but this regulatory mechanism is still largely uncharacterized. Here we have identified synaptic functions of metalloproteases and γ-secretase, two proteases that act in succession to cleave single-pass transmembrane proteins. Our work indicates that the proteolytic machinery might be part of multiprotein complexes containing the proteases and their substrates and tethered to the PSD by synaptic scafold proteins. Synaptic activity stimulates proteolysis through enhancement of the metalloprotease cleavage, leading to protease-dependent decreases in synaptic efficacy through both pre- and postsynaptic mechanisms, providing a novel form of synaptic autoregulation.
We find that the proteolytic machinery processes N-cadherin on both sides of the synapse, consistent with the homophilic nature of the N-cadherin interaction and with the observed distribution of the metalloproteases, ADAM10, ADAM17, and MT5-MMP, and the presence of the γ-secretase in both pre- and postsynaptic compartments. The postsynaptic localization of the metalloproteases may be a consequence of the tethering of ADAM10, ADAM17, and MT5-MMP by the postsynaptic scaffolds SAP97 (Peiretti et al., 2003; Marcello et al., 2007) and ABP/GRIP (Monea et al., 2006). Likewise, the association of the γ-secretase catalytic subunit PS1 and of the ADAMs with the presynaptic scaffold X11 and with the postsynaptic protein δ-catenin (Leonoudakis et al., 2004; Kosik et al., 2005; Miller et al., 2006; McCarthy et al., 2009) may enable γ-secretase to function at both sides of the synapse.
Several metalloproteases can shed γ-secretase substrates (Reiss et al., 2005; Monea et al., 2006; Uemura et al., 2006; Postina, 2008). A number of them are tethered to synaptic scaffold proteins and could thus cleave synaptic substrates (Peiretti et al., 2003; Monea et al., 2006; Marcello et al., 2007). Indeed in this study, we identified three metalloproteases, MT5MMP, ADAM 10 and ADAM 17, at synapses. We have then focused our study on understanding the general mechanism of metalloprotease cleavage regulation. Each metalloprotease might have a substrate or location preference, but little is yet known regarding possible differences (Postina, 2008).
The role of δ-catenin, which anchors PS1 (Zhou et al., 1997; Levesque et al., 1999; Tanahashi and Tabira, 1999), is of particular interest. δ-catenin can also bind to N-cadherin (Kosik et al., 2005) to PSD-95, and to ABP/GRIP (Silverman et al., 2007). The ABP/GRIP scaffolds in turn can bind to the metalloprotease MT5-MMP, which cleaves N-cadherin to release a CTF that is a substrate of γ-secretase (Monea et al., 2006). The ABP/GRIP scaffolds also bind to Eph receptors, ephrins and LAR (Bruckner et al., 1999; Dunah et al., 2005; McCarthy et al., 2009), all potential synaptic substrates of the proteolytic pathway (Georgakopoulos et al., 2006; Haapasalo et al., 2007). This suggests the existence of large, supramolecular synaptic complexes containing metalloproteases and γ-secretase, lying close to cadherins and other synaptic substrates of these proteases. The synaptic tethering of AMPAR complexes by ABP/GRIP and PSD-95 (Silverman et al., 2007) raises the possibility that disruption of these complexes could release AMPARs from the synapse. δ-catenin may serve as a central organizer of these supramolecular complexes, as has been described in non-neuronal cells (Kouchi et al., 2009). In support of this model, we observed that synaptic levels of PS1, N-cadherin and GluA2 are markedly decreased in δ-catenin knockout mice.
Application of glutamate or NMDA increased the cleavage of N-cadherin by metalloproteases, as did enhancement of synaptic activity by bicuculline. This action on N-cadherin was NMDAR-dependent and enhanced the metalloprotease cleavage, but not the γ-secretase cleavage. Different members of the metalloprotease family may be localized or regulated differently with respect to particular substrates. We suggest that the presence of the NMDA receptor adjacent to the postsynaptic δ-catenin-cadherin complexes may account for rapid N-cadherin cleavage. Previous work showing that metalloproteases can be activated by kinases via G-protein coupled receptors (Postina, 2008), and by Src phosphorylation (Zhang et al., 2006), is consistent with the present observation that NMDAR stimulation of the metalloprotease cleavage of N-cadherin depended on Src and JNK kinases.
Our work demonstrates that both metalloprotease and γ-secretase activities reduced the efficacy of synaptic transmission. Consistent with an action on both sides of the synapse, inhibition of protease activity increased both mEPSC frequency and mEPSC amplitude. Furthermore, protease activity reduced the rate of synaptic vesicle reuptake (consistent with a decrease in mEPSC frequency) while also reducing levels of PSD-95 and surface expression of the AMPA receptor subunit, GluA2, (consistent with a decrease in mEPSC amplitude). This latter effect suggests that postsynaptic cleavage can disrupt a supramolecular complex that tethers AMPA receptors at synapses, as implied in previous studies of PS1 knockout mice (Parent et al., 2005; Pratt et al., 2011).
The level of full-length N-cadherin showed little change with synaptic activity, and only a small fraction of full-length N-cadherin underwent metalloprotease/γ-secretase cleavage. Although the subpopulation of N-cadherin that is degraded remains to be identified, it is likely to include the synaptic pool, since substantial levels of N-cadherin CTF1 were found at synapses. Cleavage of the CTF1 pool by γ-secretase might play a structural role, but of special interest is its possible role in synaptic regulation. Besides releasing δ-catenin and associated components, cleavage of CTF1 releases an intracellular signaling peptide, as for other peptides released by γ-secretase from adhesion molecules and related proteins (Marambaud et al., 2003; Bao et al., 2004; Hass et al., 2009; McCarthy et al., 2009). Indeed, cleavage of N-cadherin CTF1 by γ-secretase releases the CTF2 fragment, which represses CBP/CREB-dependent transcription (Marambaud et al., 2003). This dual effect might explain discrepancies between previous reports suggesting that synaptic activity can either stabilize or degrade cadherins at synapses (Tanaka et al., 2000; Marambaud et al., 2003).
Our work demonstrates a synaptic autoregulatory loop involving proteases. The finding that metalloproteases are induced by synaptic activity, and that the cleavage machinery in turn negatively regulates synaptic function, suggests a negative autoregulatory loop at the synapse reminiscent of homeostatic plasticity (Pozo and Goda, 2010). This specialized proteolytic machinery may provide novel pathways for modification of the number, shape or function of synapses, modifications that may contribute to the cellular basis of memory (Segal, 2005). Indeed, changes in synaptic plasticity and spine morphology have been observed in PS1 knockout and Familial Alzheimer’s Disease (FAD) mutant mice (Saura et al., 2004; Auffret et al., 2009). Pathological disruption of synapses through deregulation of the metalloprotease and γ-secretase cleavage pathways could play a role in early synaptic deficits which subsequently lead to cognitive dysfunction and neurodegenerative changes in AD (Knobloch and Mansuy, 2008).
We thank Sebastien Thuault, Nicholas J Cowan, Randy Nixon and Gordon Fishell for critical readings of the manuscript and discussions, Pierre Trifilieff for help with statistical analysis and Yan Deng for help with image analysis. We thank Bryen Jordan, Vivien Chevaleyre and Pierre Trifilieff for discussion, Kristen Phend for the electron microscopy experiments. SR was supported by a Levine Fellowship and the work was supported by Alzheimer Association grants (NIRG-06-25401 to SR and NIRG-08-90001 to IN) and NIH grants, respectively, R01 AG13620 to EBZ, R01 NS35527 to RJW and P01 AG017617 to PMM.
Author ContributionsS.R. and E. B. Z. designed the experiments and wrote the paper. S. R. performed the immunocytochemical, cell biological and biochemical experiments. L. K. performed the western blots. I.N. performed the electrophysiology experiments. X.L. provided the δ-catenin knockout mice. R.J.W. performed the electron microscopy.