Regulation of the NMDA Receptor Complex and Trafficking by Activity-Dependent Phosphorylation of the NR2B Subunit PDZ Ligand

Transcription

1 1248 The Journal of Neuroscience, November 1, 24 24(45): Cellular/Molecular Regulation of the NMDA Receptor Complex and Trafficking by Activity-Dependent Phosphorylation of the NR2B Subunit PDZ Ligand Hee Jung Chung, 1 Yan Hua Huang, 1 Lit-Fui Lau, 2 and Richard L. Huganir 1 1 Department of Neuroscience, Howard Hughes Medical Institute, Johns Hopkins University School of Medicine, Baltimore, Maryland 2125, and 2 CNS Discovery, Pfizer Global Research and Development, Groton, Connecticut 634 Interactions between NMDA receptors (NMDARs) and the PDZ [postsynaptic density-95 (PSD-95)/Discs large/zona occludens-1] domains of PSD-95/SAP9 (synapse-associated protein with a molecular weight of 9 kda) family proteins play important roles in the synaptic targeting and signaling of NMDARs. However, little is known about the mechanisms that regulate these PDZ domain-mediated interactions. Here we show that casein kinase II (CK2) phosphorylates the serine residue (Ser148) within the C-terminal PDZ ligand (IESDV) of the NR2B subunit of NMDAR in vitro and in vivo. Phosphorylation of Ser148 disrupts the interaction of NR2B with the PDZ domains of PSD-95 and SAP12 and decreases surface NR2B expression in neurons. Interestingly, activity of the NMDAR and Ca 2 / calmodulin-dependent protein kinase II regulates CK2 phosphorylation of Ser148. Furthermore, CK2 colocalizes with NR1 and PSD-95 at synaptic sites. These results indicate that activity-dependent CK2 phosphorylation of the NR2B PDZ ligand regulates the interaction of NMDAR with PSD-95/SAP9 family proteins as well as surface NMDAR expression and may be a critical mechanism for modulating excitatory synaptic function and plasticity. Key words: NR2B; PSD-95; CK2; PDZ ligand; phosphorylation; trafficking Introduction The NMDA-type ionotropic glutamate receptors (NMDARs) play critical roles in neuronal development, excitotoxicity, and synaptic plasticity (Bliss and Collingridge, 1993; Seeburg, 1993; Choi, 1994; Hollmann and Heinemann, 1994). Functional NMDARs are heteromultimers mainly consisting of NR1 and NR2 (NR2A NR2D) subunits (Wenthold et al., 23). The NR2 subunits determine synaptic localization and function of NMDARs because deletion of the C-terminal tail of NR2 perturbs synaptic NMDAR localization, impairs NMDAR-mediated synaptic activity, and alters synaptic plasticity (Mori et al., 1998; Sprengel et al., 1998). The C-terminal T/SXV motifs (where X represents any amino acid) of NR2A and NR2B can directly interact with the PDZ [postsynaptic density-95 (PSD-95)/Discs large/zona occludens-1] domains of PSD-95, SAP12 (synapseassociated protein with a molecular weight of 12 kda), and PSD- 93, and such interactions induce clustering of NMDARs in transfected heterologous cells (Kornau et al., 1995; Kim et al., 1996; Lau et al., 1996; Muller et al., 1996; Niethammer et al., 1996). PSD-95 inhibits NR2B-mediated internalization and enhances Received Feb. 16, 24; revised Oct. 4, 24; accepted Oct. 5, 24. This work was supported by the Howard Hughes Medical Institute and the National Institutes of Health (R.L.H.). WethankDr.D.ShugarforthespecificCK2inhibitorTBBandDr.J.BenovicforpurifiedGRK2,GRK5,andurea-treated rod outer segments. We also thank Dr. S. Vicini for prk5-gfp-nr2b cdna. Correspondence should be addressed to Dr. Richard L. Huganir, Department of Neuroscience, Howard Hughes MedicalInstitute, JohnsHopkinsUniversitySchoolofMedicine, 94APreclinicalTeachingBuilding, 725NorthWolfe Street, Baltimore, MD rhuganir@jhmi.edu DOI:1.1523/JNEUROSCI Copyright 24 Society for Neuroscience /4/ $15./ surface NMDAR clustering (Roche et al., 21), whereas targeted disruption of the PSD-93 gene reduces surface NR2A and NR2B expression and the NMDAR-mediated EPSC (Tao et al., 23). However, NMDAR clusters are recruited to new synapses without PSD-95 (Washbourne et al., 22), and disruption of PSD-95 clustering had no effect on synaptic localization of NMDARs (Passafaro et al., 1999). Similarly, PSD-95 mutant mice exhibit defects in synaptic plasticity and spatial learning despite the proper synaptic localization of NMDARs (Migaud et al., 1998). These results suggest that PDZ domain-mediated interactions between NMDARs and PSD-95/SAP9 family proteins are not required for synaptic targeting of NMDARs. Instead, these interactions are important for stabilizing and/or promoting surface NMDAR expression and linking NMDAR to cytoplasmic signaling pathways. Although our understanding of the functional roles of PDZ domain-mediated interactions between NMDARs and PSD-95/ SAP9 family proteins has increased, little is known about the mechanisms that dynamically regulate these interactions. For the AMPA receptor (AMPAR), PKC phosphorylation of the C-terminal PDZ ligand of the AMPA receptor subunit (GluR2) disrupts its interaction with the PDZ domains of glutamate receptor interacting protein (GRIP) but not PICK1 (protein interacting with C kinase-1) (Matsuda et al., 1999; Chung et al., 2) and regulates AMPAR trafficking (Chung et al., 2; Perez et al., 21; Braithwaite et al., 22) and synaptic plasticity (Daw et al., 2; Xia et al., 2; Kim et al., 21; Chung et al., 23; Seidenman et al., 23). Moreover, protein kinase A (PKA) phosphorylation of the C-terminal PDZ ligand (T/SXV motif) of the in-

2 Chung et al. Phosphorylation of the NR2B PDZ Ligand J. Neurosci., November 1, 24 24(45): ward rectifier potassium channel Kir2.3 disrupts its interaction with PSD-95 (Cohen et al., 1996). We hypothesized that phosphorylation of the C-terminal T/SXV motif of NR2B may regulate the interactions between NMDARs and PSD-95/SAP9 family proteins. We found that casein kinase II (CK2) directly phosphorylates the 2 serine residue (Ser148) within the C-terminal T/SXV motif of NR2B. Such phosphorylation disrupts the interaction of NMDARs with the PDZ domains of PSD-95 and SAP12 and reduces surface NMDAR expression in neurons. Furthermore, NMDAR activity regulates CK2 phosphorylation of Ser148 partially via Ca 2 /calmodulin-dependent protein kinase II (CaMKII). These results indicate that activitydependent phosphorylation of the PDZ ligand may be a critical regulator of NMDAR function and synaptic plasticity. Materials and Methods Generation and characterization of anti-nr2b-ps148 antibodies. Anti- NR2B-pS148 antibody was raised against the synthetic peptide KFNGSSNGHVYEKLSSIESDV corresponding to amino acids of NR2B with phosphoserine included at Ser148. Anti- NR2B-pS148 antibodies were affinity purified from sera by sequential chromatography on Affi-Gel (Bio-Rad, Hercules, CA) columns covalently linked to BSA-conjugated unphosphorylated and Ser148- phosphorylated NR2B peptides. Antibody characterization was performed on the immunoblots of rat hippocampal P2 membranes and human embryonic kidney 293T (HEK293T) cell crude membrane fractions. HEK293T cells were transfected with prk5, pgw1-nr2b (Kim et al., 1996), or pgw1-nr2bs148a using calcium phosphate coprecipitation (Chung et al., 2). Rat hippocampal P2 membranes were prepared as described previously (Luo et al., 1997). Crude membrane fractions of HEK293T cells were isolated as described previously (Chung et al., 2). Protein samples were loaded onto an SDS-PAGE gel, transferred to polyvinylidene difluoride (PVDF) membranes (Immobilon-P membrane; Millipore, Bedford, MA), and analyzed by immunoblotting with anti-nr2b-ps148 and anti-nr2b C-terminal antibodies. To test the specificity, anti-nr2b-ps148 antibodies were preincubated with either unphosphorylated or Ser148-phosphorylated NR2B peptides (2 g/ml), or PVDF membrane was treated with -phosphatase (12 U/ml; New England Biolabs, Beverly, MA) for 3 min at 3 C before immunoblotting. Site-directed mutagenesis. NR2B Ser148 in pgw1-nr2b was mutated to alanine using QuikChange Site-Directed Mutagenensis (Stratagene, La Jolla, CA) and the following synthetic oligonucleotides: sense 5 ctttctagtattgaggctgatgtctgagtgag3, antisense 5 ctcactcagacatcagcctcaatactagaaag3 ). NR2B Ser148 in prk5-gfp-nr2b (provided by Dr. S. Vicini, Georgetown University, Washington, DC) was mutated to a glutamate residue using the following oligonucleotides: sense 5 ctttctagtattgaggaggatgtctgagtgag3, antisense 5 ctcactcagacatcctcctcaatactagaaag3 ). Purification of crude membranes from cortical neurons. Neurons were homogenized by sonication in 1 ml of resuspension buffer [1 mm sodium phosphate, ph 7., 1 mm NaCl, 5 mm EGTA, 5 mm EDTA, 1 M okadaic acid, 5 mm NaF, 1 mm sodium pyrophosphate, 1 mm sodium orthovanadate, and protease inhibitor mixture (2 g/ml aprotinin, 1 g/ml leupeptin, 2 g/ml antipain, 1 g/ml benzamide, and 1 mm PMSF)]. The homogenates were centrifuged at 14, g to obtain crude membrane pellets, which were resuspended in SDS sample buffer. Immunoblot analysis. Qualitative immunoblots were visualized by enhanced chemiluminescence development using Renaissance substrate (PerkinElmer Life Sciences, Newton, MA). Quantitative immunoblots were visualized by enhanced chemifluorescence development (Amersham Biosciences, Piscataway, NJ) and quantified on a Storm Imaging system (Molecular Dynamics, Sunnyvale, CA) with Image Quant software (Amersham Biosciences). To quantify the relative degree of phosphorylation, we calculated the ratio of the intensity of the labeling with anti-nr2b-ps148 antibody to the intensity of the labeling with anti- NR2B C-terminal antibody. The phosphorylation ratio of control samples was taken as 1%, and the phosphorylation ratio of test samples was normalized to the ratio of control samples to obtain the percentage of phosphorylation. Activation of NMDARs in neurons. High-density cortical neuronal cultures from 18-d-old embryonic rats were prepared as described previously (Ghosh and Greenberg, 1995). DL-APV (2 M; Tocris, Ellisville, MO) was added to glia-conditioned culture medium 4 d after plating and maintained for 3 weeks as reported previously (Liao et al., 21). To activate NMDARs, these neurons were transferred to artificial CSF (ACSF) (in mm: 1 HEPES-free acid, 15 NaCl, 3 KCl, 1 glucose, and 2 CaCl 2, ph 7.4) containing 1 M picrotoxin, 5 M strychnine, and 1 M glycine [APV withdrawal (APV wd)] for 15 min at 37 C with or without 1 M TTX (Tocris). Control neurons were transferred to ACSF containing 1 mm MgCl 2 and 2 M DL-APV to continue to block NMDARs (APV control). In mm Ca 2 experiments, neurons were incubated in the APV withdrawal ACSF without calcium. Inhibition or activation of kinases in neurons. Three-week-old cortical neurons were incubated with DMSO (.1% v/v; as control) or various kinase inhibitors for 1 hr before APV withdrawal or APV control. Inhibitors were (in M): 5 1 KN93, 5 1 KN92, 5 chelerythrine, 1 RP-8-bromo-cAMP (RP-8-Br-cAMP), and 1 2 4,5,6,7-tetrabromobenzotriazole (TBB; provided by Dr. D. Shugar, Institute of Biochemistry and Biophysics, Warsaw, Poland). To activate PKC and PKA, respectively, neurons were incubated with.4 M phorbol 12-myristate 13-acetate (TPA) and 2 M forskolin for 15 min at 37 C. All kinase inhibitors and activators used here were purchased from Calbiochem (San Diego, CA), except RP-8-Br-cAMP (Sigma, St. Louis, MO). Surface biotinylation. Steady-state surface biotinylation of the neurons was performed with 1 mg/ml sulfo-nhs-lc-biotin (Pierce, Rockford, IL), and the biotinylated surface proteins were purified by incubating with 1 l of 5% NeutraAvidin agarose (Pierce) as described previously (Chung et al., 2). In vitro phosphorylation of NR2B fusion proteins. The vector pgex- NR2B containing the NR2B C-terminal tail (residues ) fused in frame with glutathione S-transferase (GST) was constructed. NR2B fusion proteins were purified from Escherichia coli BL21 cells after induction with isopropyl -D-1-thiogalactopyranoside (IPTG; Sigma) as described in the GST gene fusion system handbook (Amersham Biosciences). All in vitro phosphorylation reactions were performed with.2 g of NR2B fusion proteins and 1 M ATP or [ - 32 P]ATP (5 1 cpm/pmol) in a 5 l total volume for 2 min at 3 C and stopped by adding 25 l of3 SDS-PAGE sample buffer. Phosphorylation was visualized by autoradiography or by immunoblotting with anti-nr2bps148 antibodies. Phosphorylation of NR2B fusion proteins by CaMKII (5 U; Calbiochem) was performed in CaMKII reaction buffer (1 mm HEPES, ph 7, 1 mm MgCl 2,1mM CaCl 2, 1.2 M calmodulin; Calbiochem). Phosphorylation reactions by CK2 (2 5 U; New England Biolabs) were performed in CK2 reaction buffer (in mm: 2 Tris-HCl, ph 7.5, 1 MgCl 2, 15 NaCl, and 1 DTT). Phosphorylation reactions by G-protein-coupled receptor kinase 2 and 5 (GRK2 and GRK5) were performed in GRK reaction buffer (2 mm Tris-HCl, ph 7.5, 2 mm EDTA, 5 mm MgCl 2, and.4 g of GRK2 or GRK5; provided by Dr. J. Benovic, Thomas Jefferson University, Philadelphia, PA) with NR2B fusion proteins or.2 g of purified urea-treated bovine rod outer segments (provided by Dr. J. Benovic). For TBB inhibition, CK2, GRK2, and GRK5 were preincubated with 2 M TBB before the reaction. For in vitro phosphorylation of NR2B fusion proteins by CaMKII phosphorylated CK2, phosphorylation of CK2 (1 U) by rat brain CaMKII (2 ng) was first performed in a 25 l total volume of CaMKII reaction buffer with 1 M ATP for 3 min at 3 C. After the reaction mixture was cooled on ice with 1 M KN93 (Calbiochem) for 1 min, the mixture was added to 25 l of CK2 reaction buffer containing.2 g of GST fusion proteins of NR2B and 1 M ATP. The resulting 5 l reaction was incubated at 3 C for various time points ( 9 min). In vitro phosphorylation of CK2 by CaMKII. Phosphorylation of CK2 (2 g) was performed in a 5 l total volume of CaMKII reaction buffer containing 1 M [ - 32 P]ATP (5 1 cpm/pmol), 2 M TBB, and rat brain CaMKII (2 ng; Calbiochem) and analyzed by autoradiography. Immunoprecipitation. For immunoprecipitation (IP) of NR2B, neurons were harvested in IP buffer (25 mm Tris-HCl with 1 mm NaCl, 5

3 125 J. Neurosci., November 1, 24 24(45): Chung et al. Phosphorylation of the NR2B PDZ Ligand mm EDTA, 5 mm EGTA, 1 M okadaic acid, 5 mm NaF, 1 mm sodium vanadate, and protease inhibitor mixture) containing 1% SDS and then diluted with 5 vol of ice-cold 2% Triton X-1 in IP buffer. NR2B subunits were immunoprecipitated with 4 g of anti-nr2b C-terminal antibodies for 2 3 hr at 4 C. For coimmunoprecipitation, neurons were harvested in IP buffer containing 2% Triton X-1, and NMDAR complexes were immunoprecipitated with 4 g of anti-nr1 C-terminal antibodies. Immunoprecipitates were eluted with SDS-PAGE sample buffer and subjected to immunoblot analysis with anti-nr2b C-terminal, anti- NR2B-pS148, and anti-psd-95 antibodies. To quantify the coimmunoprecipitation, we calculated the ratio of the intensity of the labeling with anti-psd-95 antibody to the intensity of the labeling with anti-nr2b C-terminal antibody for coimmunoprecipitation and input. The PSD- 95/NR2B ratio of APV control samples was taken as 1%, and the ratio of APV withdrawal samples was normalized to the ratio of APV control samples to obtain the percentage ratio (PSD-95/NR2B). In vitro binding studies. Full-length SAP12 cdna was subcloned into the vector prk5 with an N-terminal myc tag. The HEK293T cells transfected with Myc-SAP12-pRK5 or PSD-95-pGW1 were harvested in IP buffer containing 1% Triton X-1 and then incubated with either unphosphorylated or Ser148-phosphorylated NR2B peptides conjugated to affinity columns at 4 C for 2 hr. The columns were prepared by coupling 2 mg/ml BSA-conjugated peptides with 2 ml of activated Affi- Gel-1 resin. After extensive washing with IP buffer, the bound proteins were eluted by SDS sample buffer and analyzed by immunoblot with anti-myc and anti-psd-95 antibodies. Yeast cotransformation assay for protein interaction. Yeast assays were performed as described previously (Dong et al., 1997) using the PJ69 strain harboring HIS3, ADE2, and -galactosidase as reporter genes. The C-terminal NR2B (residues ) or mutant NR2BS148E cdnas were subcloned into the vector ppc97, whereas PDZ domains (1 3) of PSD-95 cdnas were subcloned into the vector ppc86. The interactions between PSD-95 and NR2B were detected by growth on quadruple minus plates (Leu-, Trp-, His-, Ade-) and assayed for -galactosidase activity with 5-bromo-4-chloro-3 indolyl- -D-galactopyranoside as a substrate. Immunocytochemistry of hippocampal neurons. Low-density hippocampal cultures from 18-d-old embryonic rats were prepared on coverslips as described previously (Goslin, 1991). Permeabilized immunostaining of the hippocampal neurons [21 d in vitro (DIV)] was performed as described previously (Liao et al., 1999) using anti-ck2 (Santa Cruz Biotechnology, Santa Cruz, CA), anti-psd95 (Upstate, Charlottesville, VA), and anti-nr1 C-terminal antibodies (Liao et al., 1999). For surface immunostaining, hippocampal neurons (7 DIV) were transfected with the vector prk5-gfp-nr2b wild type or prk5-gfp-nr2bs148e using Lipofectamine2 (Invitrogen, Carlsbad, CA). After APV control or withdrawal, surface immunostaining on the transfected neurons (9 DIV) was performed using anti-green fluorescent protein (GFP) antibody (Chemicon, Temecula, CA) as described previously (Luo et al., 22). The surface GFP-NR2B subunits were stained with rhodamine Red X (RRX)-conjugated secondary antibodies, whereas the total GFP-NR2B subunits were visualized by GFP fluorescence. Images of the pyramidal neurons were taken with a digital camera of confocal microscope (Zeiss) using identical exposure times to visualize the difference of fluorescence intensity. The ratio of surface GFP intensity over total GFP intensity per unit area was measured using the MetaMorph Imaging System (Universal Imaging, Downingtown, PA). Data analysis. All data are reported as mean SE. Sample size n refers to the number of images processed in immunostaining and the number of dishes analyzed in phosphorylation, biotinylation, and in vitro phosphorylation assay. For phosphorylation, a group-paired t test was used to test the difference between the control and testing groups. For surface immunostaining, Student s t test was used (*p.5; **p.1; ***p.1). Results The C-terminal PDZ ligand of NR2B is phosphorylated at Ser148 in vivo We generated a phosphorylation site-specific antibody to recognize the in vivo phosphorylated serine residue (Ser148) within Figure 1. The C-terminal PDZ ligand of NR2B is phosphorylated at Ser148 in vivo. A C, Characterization of the phosphorylation site-specific anti-nr2b-ps148 antibody using immunoblots of HEK293T cell crude membranes expressing NR2B. After immunoblotting with anti- NR2B-pS148 antibody, the blot was stripped and reblotted with phosphorylationindependent anti-nr2b C-terminal antibody. A, Anti-NR2B-pS148 antibody recognized NR2B in cells transfected with wild-type NR2B (NR2B) but not empty vector (mock) or the mutant NR2B (NR2BS148), in which Ser148 was mutated to alanine. B, Anti-NR2B-pS148 antibody did not recognize NR2B when anti-nr2b-ps148 antibody was preincubated with Ser148- phosphorylated NR2B C-terminal peptide (p2b) but not unphosphorylated peptide (2B). C, Anti-NR2B-pS148 antibody no longer recognized NR2B when the blot was pretreated with -phosphatase before immunoblotting. D, E, Western blot for NR2B-pS148 in rat brain membrane homogenates. D, Anti-NR2B-pS148 antibody recognized NR2B in rat brain. Preincubation of anti-nr2b-ps148 antibody with Ser148-phosphorylated NR2B C-terminal peptide (p2b) but not unphosphorylated peptide (2B) blocked NR2B recognition by anti-nr2b-ps148 antibody. E, Anti-NR2B-pS148 antibody no longer recognized NR2B when the blot was pretreated with -phosphatase before immunoblotting. the NR2B PDZ ligand (T/SXV). The resulting antibody, anti- NR2B-pS148, detected a single protein of 18 kda, the predicted molecular weight of the NR2B subunit, in the HEK293T cells transfected with the wild-type NR2B but not empty prk5 vector (mock) (Fig. 1A). Mutation of Ser148 to alanine (NR2B- S148A) abolished recognition of NR2B by anti-nr2b-ps148 (Fig. 1A). Incubation of anti-nr2b-ps148 antibody with Ser148- phosphorylated NR2B C-terminal peptides (p2b) before immunoblotting blocked recognition of NR2B, whereas incubation with unphosphorylated peptides (2B) did not (Fig. 1B). Anti-NR2B-pS148 antibody did not recognize NR2B when NR2B was dephosphorylated by -phosphatase treatment before immunoblotting (Fig. 1C). These results demonstrate the specificity and phosphorylation dependence of anti-nr2b-ps148 antibody. In rat hippocampal membrane homogenates, anti-nr2bps148 antibody detected a protein of 185 kda, which comigrated with the NR2B subunit as detected by anti-nr2b C-terminal antibody, and two additional proteins of 9 kda (Fig. 1D). Incubation of anti-nr2b-ps148 antibody with p2b pep-

4 Chung et al. Phosphorylation of the NR2B PDZ Ligand J. Neurosci., November 1, 24 24(45): tides but not 2B peptides before immunoblotting abolished recognition of NR2B (Fig. 1D). Anti-NR2B-pS148 antibody did not recognize NR2B when the homogenates were dephosphorylated by -phosphatase treatment before immunoblotting (Fig. 1E). These results indicate that the PDZ ligand of NR2B is phosphorylated at Ser148 in vivo. Activity of NMDARs regulates phosphorylation of NR2B Ser148 To examine whether neuronal activity regulates phosphorylation of NR2B Ser148, the NMDAR antagonist (APV) was added to cultured cortical neurons at day 4 after plating, and the neurons were maintained in APV for 3 weeks as described previously (Liao et al., 21). This chronic blockade of NMDAR activity has been shown to increase NMDAR clusters in these neurons (Liao et al., 21). Quantitative immunoblot analysis with anti-nr2bps148 and anti-nr2b C-terminal antibodies revealed that chronic APV treatment significantly decreased phosphorylation of NR2B Ser148 compared with the APV-untreated control neurons (51 11% of control; n 4; p.5) (Fig. 2A), without affecting the level of NR2B expression (97 9% of control; n 4; p.5). To examine rapid activity-dependent changes in the phosphorylation of NR2B Ser148, chronic APV-treated cortical neurons were incubated with ACSF containing picrotoxin (GABA A receptor blocker), strychnine (glycine receptor antagonist), and glycine (NMDAR co-agonist) for 15 min as described previously (Liao et al., 21). This APV withdrawal treatment allows rapid activation of synaptic NMDARs by the spontaneous release of glutamate. Control neurons were maintained in ACSF containing APV (APV control). NR2B subunits were immunoprecipitated with anti-nr2b C-terminal antibodies from these cultures and immunoblotted with anti-nr2b-ps148 and anti-nr2b C-terminal antibodies. Rapid activation of NMDARs by APV withdrawal increased phosphorylation of NR2B Ser148 compared with APV control (271 43% APV control; n 4; p.5) (Fig. 2B, D). Similar levels of increased phosphorylation were detected for surface NR2B subunits (34 53% APV control; n 4; p.5) (Fig. 2D). Interestingly, activation of both synaptic and extrasynaptic NMDARs by bath application of glutamate (1 M) or NMDA (5 M) in APV-untreated neurons caused dephosphorylation of Ser148 [NMDA, 31 2% of control (n 3), p.5; glutamate, 3 5% of control (n 3), p.5; data are shown in supplemental material, available at High-density cortical neuronal cultures have high levels of spontaneous activity (Murphy et al., 1992). To test whether phosphorylation of NR2B Ser148 is attributable to NMDAR activation by spontaneous synaptic responses, APV withdrawal was performed in the presence of TTX, which blocks sodium channel-mediated action potentials. APV withdrawal increased the phosphorylation of NR2B Ser148 (489 64% of APV control; n 6; p.1). In contrast, APV withdrawal in the presence of TTX caused a much less dramatic increase in the phosphorylation of Ser148 (262 39% of APV control; n 4; p.1) (Fig. 2C). In addition, no increase in the phosphorylation of Ser148 was observed when we performed APV withdrawal in Ca 2 -free ACSF to block the release of presynaptic vesicles (92 16% of APV control; n 4; p.5) (Fig. 2C). These results demonstrate that synaptic NMDAR activation dynamically modulates phosphorylation of the NR2B PDZ ligand. Figure 2. NMDAR activity regulates phosphorylation of the NR2B PDZ ligand. Quantitative immunoblot analysis with anti-nr2b-ps148 and anti-nr2b C-terminal antibodies was performed on the crude membranes of cultured cortical neurons. A, Chronic APV treatment of culturedcorticalneuronsdecreasedphosphorylationofnr2bser148(apv;n 4;ttest;*p.5) compared with the control neurons maintained in APV-free medium(ctl), without affectingthelevelofnr2bexpression. B D, SynapticNMDARswereactivatedbyremovingAPVfor15 minfromthechronicapv-treatedneurons(apvwd). TheAPVcontrolneuronsweremaintained in APV (APV ctl). B, IP of NR2B subunits using anti-nr2b C-terminal antibody. Synaptic activation of NMDARs by APV withdrawal increased phosphorylation of NR2B Ser148. C, APV withdrawal significantly increased phosphorylation of NR2B Ser148 compared with the APV control (n 6; t test; ***p.1). This increase was partially blocked by 1 M TTX (APV wd TTX; n 4; t test; **p.1) and was abolished by APV withdrawal without calcium (APV wd mmca 2 ; n 4; t test; p.5). D, APV withdrawal significantly increased Ser148 phosphorylation of surface biotinylated NR2B subunits compared with APV control (APV ctl; n 4; t test; *p.5). CaMKII regulates phosphorylation of NR2B Ser148 Synaptic NMDAR activation has been shown to stimulate CaMKII activity (Fukunaga et al., 1992). To determine whether CaMKII can regulate phosphorylation of NR2B Ser148, chronic APV-treated cortical neurons were preincubated with DMSO (.1% v/v), KN93 (CaMKII inhibitor; 5 1 M), or KN92 (inac-

5 1252 J. Neurosci., November 1, 24 24(45): Chung et al. Phosphorylation of the NR2B PDZ Ligand tive analog of KN93; 5 1 M) for 1 hr before APV withdrawal. APV withdrawal in DMSO-treated neurons increased phosphorylation of NR2B Ser148 (33 49% of APV DMSO control; n 5; p.1) and activated CaMKII as detected by increased autophosphorylation of CaMKII at Thr286 (Fig. 3A). KN93 treatment completely inhibited the activation of CaMKII and partially blocked the APV withdrawal-induced phosphorylation of Ser148 (26 26% of APV DMSO control; n 4; p.5), whereas KN92 treatment did not (32 5% of APV DMSO control; n 4; p.5) (Fig. 3A). These results indicate that CaMKII is partially required for activity-induced phosphorylation of NR2B Ser148. Synaptic NMDAR activation has also been shown to stimulate PKA (Roberson and Sweatt, 1996) and PKC (Klann et al., 1993) activity. However, treatment of neurons with chelerythrine (PKC inhibitor; 5 M) or RP-8-Br-cAMP (PKA inhibitor; 1 M) did not inhibit APV withdrawal-induced phosphorylation of NR2B Ser148 [APV wd plus chelerythrine, % of APV DMSO control (n 3); APV wd plus RP-8-Br-cAMP, 29 36% of APV DMSO control (n 3); p.5] compared with DMSO treatment (APV wd plus DMSO, % of APV DMSO control; n 3; p.5) (Fig. 3B). Furthermore, treatment of neurons with forskolin (2 M) or TPA (.4 M) to stimulate PKA and PKC, respectively, had no effect on the phosphorylation of Ser148 (data not shown). To test whether CaMKII directly phosphorylates NR2B Ser148, we performed in vitro phosphorylation reactions using GST fusion proteins of the NR2B C-terminal tail (residues ) with the purified rat brain CaMKII. CaMKII did not phosphorylate Ser148 but robustly phosphorylated Ser831 of GluR1, a previously characterized CaMKII site (Mammen et al., 1997) (Fig. 3C). These results together suggest that it is not likely that CaMKII, PKA, or PKC phosphorylate NR2B at Ser148 in vivo and that a kinase(s) downstream of CaMKII phosphorylates this site after synaptic NMDAR activation. Figure 3. CaMKII, but not PKC or PKA, regulates Ser148 phosphorylation of NR2B during synaptic NMDAR activation. Quantitative immunoblot analysis with anti-nr2b-ps148 and anti-nr2b C-terminal antibodies was performed on the crude membranes of cultured cortical neurons. A, The chronic APV-treated cortical neurons were incubated with DMSO (.1% v/v), KN93 (5 1 M; CaMKII inhibitor), or KN92 (5 1 M; inactive analog of KN93) for 1 hr before APV withdrawal. Anti-CaMKII-pT286 antibody (CaMKII-pT286) detected autophosphorylation of CaMKII at Thr286. KN93 completely blocked CaMKII activation and partially blocked phosphorylation of NR2B Ser148 induced by APV withdrawal (APV wd plus DMSO and APV wd plus KN93; n 4; t test; *p.5). B, The chronic APV-treated cortical neurons were incubated with DMSO (.1% v/v), chelerythrine (5 M; PKC inhibitor), or RP-8-Br-cAMP (1 M; PKA inhibitor) for 1 hr before APV withdrawal. Chelerythrine and RP-8-Br-cAMP had no effect on APV withdrawal-induced phosphorylation of NR2B Ser148 (APV wd plus DMSO and APV wd plus chelerythrine or RP-8-Br-cAMP; n 3; t test; p.5). C, In vitro phosphorylation Synaptic NMDAR activation increases CK2 phosphorylation of NR2B PDZ ligand The amino acid sequence surrounding NR2B Ser148 (IESDV) is a strong consensus site for CK2 (Meggio et al., 1994). To determine whether CK2 is the relevant in vivo kinase for NR2B Ser148, chronic APV-treated cortical neurons were incubated with DMSO (.1% v/v) or TBB (1 2 M) for 1 hr before APV withdrawal. TBB is a cell-permeable, highly specific CK2 inhibitor that inhibits 85% of CK2 activity at this concentration in vitro (Sarno et al., 21). APV withdrawal of DMSO-treated neurons increased CaMKII activity and phosphorylation of NR2B Ser148 (348 5% of APV DMSO control; n 4; p.5) (Fig. 4A). In contrast, TBB treatment abolished APV withdrawalinduced phosphorylation of Ser148 (124 16% of APV DMSO control; n 4; p.5) but not autophosphorylation of CaMKII Thr286 (Fig. 4A). Moreover, CK2 directly phosphorylated GST fusion proteins of the NR2B C-terminal tail at Ser148 in vitro (Fig. 4B). These results suggest that CK2 acts downstream 4 reactions of GST fusion proteins of the NR2B C-terminal tail and GluR1 C-terminal tail were performed with CaMKII (5 U). Phosphorylation of NR2B Ser148 and GluR1 Ser831 (a previously characterized CaMKII site) was detected by immunoblotting with anti-nr2b-ps148 and anti-glur1-ps831 antibodies, respectively. CaMKII directly phosphorylated GluR1 Ser831 but not NR2B Ser148. The asterisk indicates the C-terminally truncated GluR1 fusion protein, which is recognized by anti-glur1-ps831 antibody but not by anti-glur1 C-terminal antibody.

6 Chung et al. Phosphorylation of the NR2B PDZ Ligand J. Neurosci., November 1, 24 24(45): Figure 4. CK2 functions downstream of CaMKII during synaptic NMDAR activation and directly phosphorylates Ser148. A, The chronic APV-treated cortical neurons were incubated with DMSO (.1% v/v) or TBB (1 2 M; CK2 inhibitor) for 1 hr before APV withdrawal. Quantitative immunoblot analysis with anti-nr2b-ps148 and anti-nr2b C-terminal antibodies revealed that TBB completely blocked APV withdrawal-induced phosphorylation of NR2B Ser148 (APV wd plus DMSO and APV wd plus TBB; n 4; t test; *p.5) but not autophosphorylation of CaMKII at Thr286. B, C, Immunoblot analysis of in vitro phosphorylation reactions ofgstfusionproteinsofthenr2bc-terminaltailbyck2(2u;b)andck2,grk2,andgrk5(.4 g each; C). B, CK2 directly phosphorylated NR2B Ser148. C, GRK2 and GRK5 phosphorylated NR2B Ser148 very weakly compared with the same amount of CK2. D, Autoradiograph and CoomassiebluestainingofanSDS-PAGEgelcontaininginvitrophosphorylationreactionsofthe GST fusion proteins of the NR2B C-terminal tail by CK2 (.4 g) with or without TBB (2 M). TBB inhibited CK2 activity. E, Autoradiograph and Coomassie blue staining of an SDS-PAGE gel containing in vitro phosphorylation reactions of rhodopsin by GRK2 (.4 g) with or without of CaMKII and phosphorylates NR2B Ser148 after synaptic NMDAR activation. Because GRK2 and GRK5 also preferentially phosphorylate serine or threonine residues surrounded by acidic amino acids (Palczewski, 1997), we also performed an in vitro phosphorylation reactions with purified active GRK2 and GRK5. GRK2 and GRK5 phosphorylated NR2B Ser148 very weakly compared with the same amount of CK2 (Fig. 4C). To test whether the CK2 inhibitor TBB can block GRK2 and GRK5, we performed in vitro phosphorylation reaction with GRK2 and GRK5 in the presence of TBB (2 M) using rhodopsin as a substrate. TBB dramatically decreased CK2 phosphorylation of NR2B Ser148 (Fig. 4D) but had no effect on the phosphorylation of rhodopsin by GRK2 (Fig. 4E) and GRK5 (data not shown). These results suggest that GRK2 and GRK5 do not regulate phosphorylation of NR2B Ser148 after synaptic NMDAR activation. To examine the subcellular localization of CK2 in relation to the distribution of NMDARs, we performed immunostaining on 3-week-old cultured hippocampal neurons using anti-ck2 antibody and anti-nr1 C-terminal antibody. CK2 catalytic subunits were localized in soma and throughout dendrites as small clusters (Fig. 4F). CK2 colocalized with NR1 in dendritic shafts and some spines (Fig. 4F). To test whether excitatory synapses contain CK2, we performed immunostaining with anti-psd-95 antibody. CK2 subunits also colocalized with PSD-95 in some dendritic spines (Fig. 4G). These results suggest that CK2 exists in close proximity to NMDARs to phosphorylate NR2B Ser148. CK2 is cyclic nucleotide independent and cannot be directly activated by calcium (Litchfield, 23). Previous studies show that phosphorylation of its catalytic and regulatory subunits by CK2 itself or other kinases can regulate CK2 activity (Blanquet, 1998; Litchfield, 23). Because CaMKII was partially required for APV withdrawal-induced phosphorylation of NR2B Ser148 by CK2 (Figs. 3, 4), we investigated whether CaMKII can stimulate CK2 activity by directly phosphorylating its catalytic or regulatory subunits. We performed in vitro phosphorylation reactions of CK2 holoenzyme with CaMKII in the presence of TBB (2 M) to minimize the autophosphorylation of CK2 itself. CaMKII directly phosphorylated the but not the subunit of CK2 (Fig. 5A). To test whether CaMKII phosphorylation of CK2 can increase the catalytic activity of CK2, we first in vitro -phosphorylated CK2 with CaMKII and then added KN93 to inhibit CaMKII activity. This reaction mixture was incubated with GST fusion proteins of the NR2B C-terminal tail and CK2 reaction buffer at 3 C for 9 min. Preincubation of CaMKII with CK2 overall increased CK2 phosphorylation of NR2B at Ser148 compared with CK2 alone (Fig. 5B). However this increase was statistically significant at the 6 min time point (146 11% of CK2 alone; n 4; p.5) but not at the 3 and 9 min time points (n 4; p.5) (Fig. 5B). These results indicate that direct phosphorylation of CK2 by CaMKII has no effect on the catalytic activity of CK2 in vitro. 4 TBB (2 M). TBB did not inhibit GRK2 activity. F, Immunostaining of 3-week-old low-density cultured hippocampal neurons with anti-ck2 (CK2 ) and anti-nr1 C-terminal antibodies (NR1). CK2 subunits colocalized with NR1 in the dendritic shafts and some spines. G, Immunostaining of 3-week-old low-density cultured hippocampal neurons with anti-ck2 (CK2 ) and anti-psd-95 antibody (PSD-95). CK2 subunits also colocalized with PSD-95 in some dendritic spines.

7 1254 J. Neurosci., November 1, 24 24(45): Chung et al. Phosphorylation of the NR2B PDZ Ligand Figure 5. Preincubation of CaMKII with CK2 increases phosphorylation of NR2B Ser148 by CK2 in vitro. A, Autoradiograph and Coomassie blue staining of an SDS-PAGE gel containing in vitro phosphorylation reactions of purified CK2 holoenzyme (2 g) with TBB (2 M) by purified rat brain CaMKII (5 U). CaMKII directly phosphorylated CK2 -subunits. B, Immunoblot analysisofinvitrophosphorylationreactionsofthegstfusionproteinsofthenr2bc-terminaltailby CK2(2U), whichwasfirstphosphorylatedbycamkii(5u) invitro. AfterCaMKIIphosphorylationof CK2, KN93 was added to block CaMKII activity. Preincubation of CaMKII with CK2 overall does not significantlyincreaseck2phosphorylationofnr2bser148comparedwithck2alone. However, the increase was statistically significant at the 6 min time point (n 4; ttest; *p.5). Phosphorylation of NR2B PDZ ligand disrupts its interaction with SAP12 and PSD-95 To investigate whether phosphorylation of Ser148 in the NR2B C-terminal PDZ ligand can regulate the interaction of NR2B with PSD-95/SAP9 family proteins, we tested the in vitro binding of PSD-95 or SAP12 to Ser148-phosphorylated (p2b) or unphosphorylated NR2B C-terminal (2B) peptides. HEK293T cell lysates expressing PSD-95 and SAP12 were incubated with NR2B peptide-conjugated affinity columns, and bound proteins were eluted and analyzed by immunoblotting. PSD-95 and SAP12 interacted with the 2B peptide but not the p2b peptide (Fig. 6A). The interaction of PSD-95 and SAP12 with the p2b peptide was recovered when the p2b peptide was dephosphorylated by -phosphatase before the binding experiment (Fig. 6A). These Figure 6. Phosphorylation of the NR2B PDZ ligand disrupts its interaction with SAP12 and PSD-95. A, In vitro binding studies of PSD-95 (top) or SAP12 (bottom) with NR2B peptides immobilized on Affi-Gel resin. Extracts of HEK293T cells expressing PSD-95 or SAP12 were incubated with unphosphorylated (2B), Ser148-phosphorylated 2B (p2b), or the -phosphatase-treated p2b peptides. The bound PSD-95 or SAP12 proteins were detected by immunoblot with anti-psd-95 antibody or anti-sap12 antibody, respectively. Both PSD-95 and SAP12 failed to interact with Ser148-phosphorylated NR2B. B, IP of NMDAR complexes from the chronic APV-treated cortical neurons after APV control (APV ctl) or APV wd using anti-nr1 C-terminal antibodies. Coimmunoprecipitation of NR2B and PSD-95 was analyzed by immunoblot with anti-psd-95 (PSD-95), NR2B, and anti-nr2b-ps148 antibodies (NR2BpS148). Synaptic activation of NMDAR increased phosphorylation of NR2B Ser148 and decreased NMDAR interaction with PSD-95 (n 5; t test; *p.5). results indicate that phosphorylation of Ser148 disrupts the interaction of NR2B with the PDZ domains of PSD-95 and SAP12. Because APV withdrawal significantly increased phosphorylation of NR2B Ser148 (Fig. 2), synaptic NMDAR activation may regulate the interaction of NMDARs with the PSD-95/ SAP9 family proteins. To test this, NMDARs were immunoprecipitated with anti-nr1 C-terminal antibodies from cortical neurons treated with APV control or withdrawal. Compared with APV control, synaptic NMDAR activation increased phosphorylation of NR2B Ser148 but decreased the amount of PSD-95 that coimmunoprecipitated with NR2B subunits (Fig. 6B). This decrease was statistically significant as shown by the ratio of the

8 Chung et al. Phosphorylation of the NR2B PDZ Ligand J. Neurosci., November 1, 24 24(45): Figure 7. Mutation of Ser148 to glutamate decreases surface expression of NR2B in neurons. A, Surface immunostaining of GFP-NR2Bwild-typeandmutantsubunit(NR2BS148E) inwhichser148wasreplacedwithglutamatetomimicphosphorylated Ser148 (left; scale bar, 1 m). The right panels show higher magnification of the insets in the left panel. B, In the yeast two-hybrid system, PDZ domains of PSD-95 failed to interact with phosphorylation mimic mutant NR2B S148E and deletion mutant NR2B (NR2B 5aa), in which the PDZ ligand was deleted. C, Quantification of surface GFP staining of wild-type NR2B (n 19) and mutant NR2B S148E (n 26) as the ratio of surface GFP intensity over total GFP intensity per unit area (ttest; ***p.1). intensity of PSD-95 over NR2B [IP (APV wd), 45 14% of APV control (n 5); p.5] (Fig. 6B). In contrast, APV withdrawal did not change the overall amount of PSD-95 and NR2B in total lysate [input (APV wd), 84 12% of APV control (n 5); p.5] (Fig. 6B). These results demonstrate that activity-induced phosphorylation of NR2B Ser148 attenuates the interaction of NMDAR with PSD-95 in vivo. Phosphorylation of NR2B PDZ ligand decreases surface expression of NR2B Disruption of the interactions between NR2B and PSD-95/ SAP12 by phosphorylation of the NR2B PDZ ligand may regulate surface membrane trafficking of NMDARs. To test this, we first mutated Ser148 to glutamate, which mimics the negative charge of phosphorylated Ser148.In the yeast two-hybrid system, this phosphorylation mimic mutant, NR2B S148E C terminus, failed to interact with PDZ domains (1 3) of PSD-95, whereas the wild-type NR2B C terminus interacted well (Fig. 7B). Next, we transfected primary hippocampal neurons with wildtype subunits or the phosphorylation mimic mutant NR2B subunits, which were tagged extracellularly with GFP (GFP- NR2B and GFP-NR2B S148E). Surface GFP-NR2B subunits were labeled live with anti-gfp antibody and visualized by RRX-conjugated secondary antibody, whereas total GFP-NR2B was visualized by GFP fluorescence. Surface immunostaining showed surface expression of both wild-type GFP-NR2B and mutant GFP- NR2B S148E on the plasma membrane of dendrites and soma as distinct clusters in hippocampal pyramidal neurons (Fig. 7A). However, surface clusters of mutant GFP-NR2B S148E were less intense and smaller compared with wild-type NR2B (Fig. 7A). The ratio of surface GFP intensity over total GFP intensity per unit area of mutant NR2B S148E was significantly lower than that of wild-type NR2B in the dendrites [wild type, per unit area (n 19); S148E,.34.5 per unit area (n 26); p.1] (Fig. 7C). These data demonstrate that phosphorylation of NR2B Ser148 decreases the surface expression of NMDARs. Discussion The present study demonstrates that CK2 phosphorylation of Ser148 within the NR2B C-terminal PDZ ligand regulates PDZ domain-mediated interaction and surface expression of NMDAR. Interestingly, activity of NMDAR and CaMKII regulates CK2 phosphorylation of Ser148. Thus, activity-dependent phosphorylation of the PDZ ligand provides a powerful way to control NMDAR trafficking and excitatory synaptic function. Phosphorylation of the NR2B PDZ ligand by CK2 Various kinases and phosphatases modulate ion channel properties (Greengard et al., 1991; Durand et al., 1993; Lieberman and Mody, 1994; Wang and Salter, 1994; Lieberman and Mody, 1999) and forward trafficking of NMDARs (Lan et al., 21; Scott et al., 21), although only a few phosphorylation sites of NMDAR subunits have been identified so far (Hall and Soderling, 1997; Leonard and Hell, 1997; Tingley et al., 1997). NR2B Ser148 is the first CK2 phosphorylation site identified among the NMDAR subunits. CK2 is highly enriched in brain and phosphorylates various synaptic and nuclear substrates implicated in neuronal development and survival, neurite extension, synaptic transmission, and synaptic plasticity (Blanquet, 2). CK2 can regulate NMDAR channel gating (Lieberman and Mody, 1999), although the role of CK2 phosphorylation of NR2B Ser148 in NMDAR channel properties is not known. We also show that CK2 colocalizes with NMDAR in dendrites and at some excitatory synapses,

9 1256 J. Neurosci., November 1, 24 24(45): Chung et al. Phosphorylation of the NR2B PDZ Ligand consistent with studies demonstrating enrichment of CK2 activity in synaptosomes (Girault et al., 199). Interestingly, CK2 phosphorylation of NR2B Ser148 is regulated by activation of NMDAR itself. Suppression of NMDAR activity by chronic APV treatment decreases Ser148 phosphorylation, whereas rapid activation of synaptic NMDARs by APV withdrawal increases Ser148 phosphorylation. Similarly, induction of NMDAR-dependent long-term potentiation (LTP) in rat hippocampal slices leads to a calcium-dependent increase in CK2 activity (Charriaut-Marlangue et al., 1991). Chronic APV treatment of cultured cortical neurons increases the proportion of silent synapses containing only NMDAR (Liao et al., 1999), delaying synapse development. Chronic APV treatment may also affect the morphology of dendritic spines, because NMDAR activity regulates spine morphology (Engert and Bonhoeffer, 1999; Maletic-Savatic et al., 1999). However, APV withdrawal from the chronic APV-treated neurons allows activation of only synaptic NMDARs by synaptically released glutamate and stimulation of CK2 activity for which there is no known activator so far. Because APV withdrawal has been shown to convert silent synapses to conducting synapses containing both NMDAR and AMPAR (Liao et al., 1999; Liao et al., 21), this treatment may allow us to study whether CK2 phosphorylation of NR2B Ser148 plays a role in synapse maturation. In contrast to APV withdrawal, bath application of glutamate or NMDA activates both synaptic and extrasynaptic NMDARs and induces rapid internalization of AMPARs through activation of phosphatases (Beattie et al., 2; Ehlers, 2; Lee et al., 22). Not surprisingly, glutamate and NMDA treatment causes dephosphorylation of NR2B Ser148. It is unclear how synaptic NMDAR activation leads to calcium-dependent CK2 phosphorylation of NR2B Ser148 because calcium does not directly activate CK2 (Hathaway and Traugh, 1982). Because the CaMK inhibitor KN93 partially blocks APV withdrawal-induced phosphorylation of NR2B Ser148, calcium/ calmodulin-dependent protein kinases may regulate CK2 activity. However, we found that direct phosphorylation of CK2 by CaMKII does not increase CK2 activity in vitro, although the role of CaMKI and CaMKIV in CK2 activity remains to be tested. Alternatively, calmodulin may mediate synaptic targeting of CK2 by interacting with CK2 (Grein et al., 1999) and NR2B (Wyszynski et al., 1997). Recently, induction of NMDAR-dependent LTP in rat hippocampal slices has shown to increase both CaMKII activity and CaMKII association with NR2A and NR2B and decrease PSD-95 association with NR2A and NR2B (Gardoni et al., 21). Thus, an NMDAR-mediated rise in calcium may stimulate the binding of CaMKII to NR2B (Leonard et al., 1999; Bayer et al., 21) and/or CaMKII phosphorylation of NR2B Ser133 (Omkumar et al., 1996; Bayer et al., 21). These events may induce a conformational change in the NR2B tail, which attenuates PSD-95 binding and exposes the PDZ ligand for CK2 phosphorylation. Once Ser148 is phosphorylated, NMDAR may remain unbound by PSD-95. Figure 8. Model for regulation of NMDAR complex and trafficking by phosphorylation of the NR2B PDZ ligand. Synaptic activation of NMDAR stimulates CaMKII and CK2 activity. A, CK2 phosphorylates the PDZ ligand of surface NR2B at Ser148 and disrupts the interaction between surface NMDAR and the PDZ domains of PSD-95. This leads to destabilization and internalization of surface NMDAR and decoupling of NMDAR from signaling proteins at excitatory synapses. B, CK2 phosphorylates the PDZ ligand of intracellular NR2B at Ser148 and disrupts the interaction between intracellular NMDAR with the PDZ domains of SAP12. This leads to disruption of SAP12/Sec8-mediated forward trafficking of NMDAR to the cell surface. Regulation of PDZ domain-mediated interaction by phosphorylation of the NR2B PDZ ligand The inhibition of NR2B interaction with PSD-95 and SAP12 by phosphorylation of the PDZ ligand at Ser148 can be explained by the structure of type 1 PDZ domain T/SXV motif interaction (Doyle et al., 1996; Morais Cabral et al., 1996). The addition of a phosphate group to the 2 serine or threonine residue of the T/SXV motif will disrupt the critical hydrogen bond formed between the 2 residue of the T/SXV motif and the histidine within a hydrophobic groove in the PDZ domain (Doyle et al., 1996; Morais Cabral et al., 1996). Similar to our results, PKA phosphorylation of the 2 serine residue of inward rectifier potassium channel Kir2.3 abolishes binding of Kir2.3 to PSD-95 (Cohen et al., 1996). GRK5 phosphorylation of the 2 serine of the 2- adrenergic receptor disrupts binding of the receptor to the PDZ domain of NHERF (Na /H exchanger regulatory factor) protein (Cao et al., 1999). Moreover, the 3 serine of the AMPA receptor subunit GluR2 with its type 2 PDZ ligand ( SVKI) can be phosphorylated by PKC, and such phosphorylation disrupts binding of GluR2 to the PDZ domain of GRIP (Matsuda et al., 1999; Chung et al., 2). Thus, phosphorylation of the serine or threonine residue within the PDZ ligands may be a common mechanism for regulating PDZ domain-mediated interactions. Regulation of synaptic signaling by phosphorylation of the NR2B PDZ ligand PSD-95/SAP9 family proteins interact with many neuronal signaling proteins such as SynGAP (a neuronal RasGTPase activating protein) (Chen et al., 1998; Kim et al., 1998), neuronal nitric oxide synthase (Brenman et al., 1996; Christopherson et al.,

10 Chung et al. Phosphorylation of the NR2B PDZ Ligand J. Neurosci., November 1, 24 24(45): ), and AKAP7/15 (a scaffolding protein for three signaling enzymes: PKA, PKC, and the calcium-dependent phosphatase calcineurin) (Colledge et al., 2). The targeted mutation of the PSD-95 gene in mice appears to disrupt downstream signaling from NMDAR without affecting synaptic localization of NMDARs (Migaud et al., 1998). In addition, overexpression of PSD-95 enhances synaptic clustering of AMPARs without altering the levels of synaptic NMDARs (El-Husseini et al., 2). Thus, phosphorylation of the NR2B PDZ ligand may regulate synaptic signaling downstream of NMDAR by modulating interaction of NMDAR with PSD-95/SAP9 family proteins. Regulation of NMDAR trafficking by phosphorylation of the NR2B PDZ ligand PDZ domain-mediated interactions facilitate synaptic clustering and surface expression of NMDARs by suppressing the endoplasmic reticulum retention signal in NR1 (Standley et al., 2) and the internalization signal in the NR2B C terminus, which binds to the adaptor protein-2 (Roche et al., 21; Lavezzari et al., 23), and by promoting formation of exocyst SAP12 NMDAR complexes (Sans et al., 23). Consistent with these findings, we found that mutation of NR2B Ser148 to glutamate, which mimics phosphorylated Ser148, disrupts NR2B interaction with PSD-95/SAP12 and decreases the size and intensity of surface NR2B clusters. Phosphorylation of NR2B Ser148 may regulate surface expression of NMDAR by multiple mechanisms. Because NMDAR can move laterally between synaptic and extrasynaptic domains of the plasma membrane (Tovar and Westbrook, 22), phosphorylation of surface NR2B at Ser148 would disrupt NMDAR interaction with PSD-95/PSD-93 and allow NMDAR to move from synaptic to extrasynaptic regions where they could be internalized by exposing an internalization signal in the NR2B C terminus (Fig. 8A). Alternatively, phosphorylation of intracellular NR2B at Ser148 would disrupt NMDAR interaction with SAP12 and block surface delivery of NMDAR mediated by SAP12/sec8 exocyst complex formation (Fig. 8B). Our studies suggest that NMDAR activity may regulate synaptic targeting of NMDAR by modulating phosphorylation of NR2B Ser148. We demonstrated that phosphorylation of Ser148 decreases with chronic APV treatment, whereas it increases after removal of APV. Interestingly, chronic APV treatment increases synaptic accumulation of NMDAR clusters with no preference between NR2A- and NR2B-containing receptors (Rao and Craig, 1997; Liao et al., 1999; Liao et al., 21; Tovar and Westbrook, 22), whereas removal of APV to induce spontaneous activity decreases synaptic NMDAR clusters (Rao and Craig, 1997). The increase in NR2A and NR2B and their association with PSD-95 has been suggested to be the potential mechanism of the APV-induced synaptic targeting of NMDAR (Rao and Craig, 1997), although we found no change in the level of NR2B by chronic APV blockade. Thus, the APV-induced synaptic targeting of NMDARs in our cortical cultures (Liao et al., 21) may be caused by the decrease in Ser148 phosphorylation and subsequent increase in the interaction between NR2B and PSD-95/ SAP12. In contrast to the above studies, APV blockade of NMDARs has been shown to induce trafficking of NR2B away from synapses and trafficking of NR2A subunits to synapses (Aoki et al., 23; Fujisawa and Aoki, 23). The differences in activity-dependent synaptic targeting of NR2B/A subunits may be explained by the differences in the duration of APV treatment and in vitro neuronal cultures versus in vivo mature cortical synapses in intact adult brain. Because synaptic targeting of NR2B- and NR2Acontaining receptors is regulated differently (Barria and Malinow, 22; Aoki et al., 23; Fujisawa and Aoki, 23) and determines the polarity of synaptic plasticity (Liu et al., 24), it will be interesting to study whether activity-dependent CK2 phosphorylation of NR2B Ser148 may affect the synaptic presence of NR2A-containing NMDARs. References Aoki C, Fujisawa S, Mahadomrongkul V, Shah PJ, Nader K, Erisir A (23) NMDA receptor blockade in intact adult cortex increases trafficking of NR2A subunits into spines, postsynaptic densities, and axon terminals. Brain Res 963: Barria A, Malinow R (22) Subunit-specific NMDA receptor trafficking to synapses. Neuron 35: Bayer KU, De Koninck P, Leonard AS, Hell JW, Schulman H (21) Interaction with the NMDA receptor locks CaMKII in an active conformation. Nature 411: Beattie EC, Carroll RC, Yu X, Morishita W, Yasuda H, von Zastrow M, Malenka RC (2) Regulation of AMPA receptor endocytosis by a signaling mechanism shared with LTD. Nat Neurosci 3: Blanquet PR (1998) Neurotrophin-induced activation of casein kinase 2 in rat hippocampal slices. Neuroscience 86: Blanquet PR (2) Casein kinase 2 as a potentially important enzyme in the nervous system. Prog Neurobiol 6: Bliss TV, Collingridge GL (1993) A synaptic model of memory: long-term potentiation in the hippocampus. Nature 361: Braithwaite SP, Xia H, Malenka RC (22) Differential roles for NSF and GRIP/ABP in AMPA receptor cycling. Proc Natl Acad Sci USA 99: Brenman JE, Chao DS, Gee SH, McGee AW, Craven SE, Santillano DR, Wu Z, Huang F, Xia H, Peters MF, Froehner SC, Bredt DS (1996) Interaction of nitric oxide synthase with the postsynaptic density protein PSD-95 and alpha1-syntrophin mediated by PDZ domains. Cell 84: Cao TT, Deacon HW, Reczek D, Bretscher A, von Zastrow M (1999) A kinase-regulated PDZ-domain interaction controls endocytic sorting of the beta2-adrenergic receptor. Nature 41: Charriaut-Marlangue C, Otani S, Creuzet C, Ben-Ari Y, Loeb J (1991) Rapid activation of hippocampal casein kinase II during long-term potentiation. Proc Natl Acad Sci USA 88: Chen HJ, Rojas-Soto M, Oguni A, Kennedy MB (1998) A synaptic Ras- GTPase activating protein (p135 SynGAP) inhibited by CaM kinase II. Neuron 2: Choi DW (1994) Glutamate receptors and the induction of excitotoxic neuronal death. Prog Brain Res 1: Christopherson KS, Hillier BJ, Lim WA, Bredt DS (1999) PSD-95 assembles a ternary complex with the N-methyl-D-aspartic acid receptor and a bivalent neuronal NO synthase PDZ domain. J Biol Chem 274: Chung HJ, Xia J, Scannevin RH, Zhang X, Huganir RL (2) Phosphorylation of the AMPA receptor subunit GluR2 differentially regulates its interaction with PDZ domain-containing proteins. J Neurosci 2: Chung HJ, Steinberg JP, Huganir RL, Linden DJ (23) Requirement of AMPA receptor GluR2 phosphorylation for cerebellar long-term depression. Science 3: Cohen NA, Brenman JE, Snyder SH, Bredt DS (1996) Binding of the inward rectifier K channel Kir 2.3 to PSD-95 is regulated by protein kinase A phosphorylation. Neuron 17: Colledge M, Dean RA, Scott GK, Langeberg LK, Huganir RL, Scott JD (2) Targeting of PKA to glutamate receptors through a MAGUK-AKAP complex. Neuron 27: Daw MI, Chittajallu R, Bortolotto ZA, Dev KK, Duprat F, Henley JM, Collingridge GL, Isaac JT (2) PDZ proteins interacting with C-terminal GluR2/3 are involved in a PKC-dependent regulation of AMPA receptors at hippocampal synapses. Neuron 28: Dong H, O Brien RJ, Fung ET, Lanahan AA, Worley PF, Huganir RL (1997) GRIP: a synaptic PDZ domain-containing protein that interacts with AMPA receptors. Nature 386: Doyle DA, Lee A, Lewis J, Kim E, Sheng M, MacKinnon R (1996) Crystal structures of a complexed and peptide-free membrane protein-binding domain: molecular basis of peptide recognition by PDZ. Cell 85: Durand GM, Bennett MV, Zukin RS (1993) Splice variants of the N-methyl-

11 1258 J. Neurosci., November 1, 24 24(45): Chung et al. Phosphorylation of the NR2B PDZ Ligand D-aspartate receptor NR1 identify domains involved in regulation by polyamines and protein kinase C. Proc Natl Acad Sci USA 9: Ehlers MD (2) Reinsertion or degradation of AMPA receptors determined by activity-dependent endocytic sorting. Neuron 28: El-Husseini AE, Schnell E, Chetkovich DM, Nicoll RA, Bredt DS (2) PSD-95 involvement in maturation of excitatory synapses. Science 29: Engert F, Bonhoeffer T (1999) Dendritic spine changes associated with hippocampal long-term synaptic plasticity. Nature 399:66 7. Fujisawa S, Aoki C (23) In vivo blockade of N-methyl-D-aspartate receptors induces rapid trafficking of NR2B subunits away from synapses and out of spines and terminals in adult cortex. Neuroscience 121: Fukunaga K, Soderling TR, Miyamoto E (1992) Activation of Ca 2 / calmodulin-dependent protein kinase II and protein kinase C by glutamate in cultured rat hippocampal neurons. J Biol Chem 267: Gardoni F, Schrama LH, Kamal A, Gispen WH, Cattabeni F, Di Luca M (21) Hippocampal synaptic plasticity involves competition between Ca 2 /calmodulin-dependent protein kinase II and postsynaptic density 95 for binding to the NR2A subunit of the NMDA receptor. J Neurosci 21: Ghosh A, Greenberg ME (1995) Distinct roles for bfgf and NT-3 in the regulation of cortical neurogenesis. Neuron 15: Girault JA, Hemmings Jr HC, Zorn SH, Gustafson EL, Greengard P (199) Characterization in mammalian brain of a DARPP-32 serine kinase identical to casein kinase II. J Neurochem 55: Goslin K, Banker G (1991) Culturing nerve cells. London: MIT. Greengard P, Jen J, Nairn AC, Stevens CF (1991) Enhancement of the glutamate response by camp-dependent protein kinase in hippocampal neurons. Science 253: Grein S, Raymond K, Cochet C, Pyerin W, Chambaz EM, Filhol O (1999) Searching interaction partners of protein kinase CK2beta subunit by twohybrid screening. Mol Cell Biochem 191: Hall RA, Soderling TR (1997) Differential surface expression and phosphorylation of the N-methyl-D-aspartate receptor subunits NR1 and NR2 in cultured hippocampal neurons. J Biol Chem 272: Hathaway GM, Traugh JA (1982) Casein kinases multipotential protein kinases. Curr Top Cell Regul 21: Hollmann M, Heinemann S (1994) Cloned glutamate receptors. Annu Rev Neurosci 17: Kim CH, Chung HJ, Lee HK, Huganir RL (21) Interaction of the AMPA receptor subunit GluR2/3 with PDZ domains regulates hippocampal long-term depression. Proc Natl Acad Sci USA 98: Kim E, Cho KO, Rothschild A, Sheng M (1996) Heteromultimerization and NMDA receptor-clustering activity of Chapsyn-11, a member of the PSD-95 family of proteins. Neuron 17: Kim JH, Liao D, Lau LF, Huganir RL (1998) SynGAP: a synaptic RasGAP that associates with the PSD-95/SAP9 protein family. Neuron 2: Klann E, Chen SJ, Sweatt JD (1993) Mechanism of protein kinase C activation during the induction and maintenance of long-term potentiation probed using a selective peptide substrate. Proc Natl Acad Sci USA 9: Kornau HC, Schenker LT, Kennedy MB, Seeburg PH (1995) Domain interaction between NMDA receptor subunits and the postsynaptic density protein PSD-95. Science 269: Lan JY, Skeberdis VA, Jover T, Grooms SY, Lin Y, Araneda RC, Zheng X, Bennett MV, Zukin RS (21) Protein kinase C modulates NMDA receptor trafficking and gating. Nat Neurosci 4: Lau LF, Mammen A, Ehlers MD, Kindler S, Chung WJ, Garner CC, Huganir RL (1996) Interaction of the N-methyl-D-aspartate receptor complex with a novel synapse-associated protein, SAP12. J Biol Chem 271: Lavezzari G, McCallum J, Lee R, Roche KW (23) Differential binding of the AP-2 adaptor complex and PSD-95 to the C-terminus of the NMDA receptor subunit NR2B regulates surface expression. Neuropharmacology 45: Lee SH, Liu L, Wang YT, Sheng M (22) Clathrin adaptor AP2 and NSF interact with overlapping sites of GluR2 and play distinct roles in AMPA receptor trafficking and hippocampal LTD. Neuron 36: Leonard AS, Hell JW (1997) Cyclic AMP-dependent protein kinase and protein kinase C phosphorylate N-methyl-D-aspartate receptors at different sites. J Biol Chem 272: Leonard AS, Lim IA, Hemsworth DE, Horne MC, Hell JW (1999) Calcium/ calmodulin-dependent protein kinase II is associated with the N-methyl- D-aspartate receptor. Proc Natl Acad Sci USA 96: Liao D, Zhang X, O Brien R, Ehlers MD, Huganir RL (1999) Regulation of morphological postsynaptic silent synapses in developing hippocampal neurons. Nat Neurosci 2: Liao D, Scannevin RH, Huganir R (21) Activation of silent synapses by rapid activity-dependent synaptic recruitment of AMPA receptors. J Neurosci 21: Lieberman DN, Mody I (1994) Regulation of NMDA channel function by endogenous Ca(2 )-dependent phosphatase. Nature 369: Lieberman DN, Mody I (1999) Casein kinase-ii regulates NMDA channel function in hippocampal neurons. Nat Neurosci 2: Litchfield DW (23) Protein kinase CK2: structure, regulation and role in cellular decisions of life and death. Biochem J 369:1 15. Liu L, Wong TP, Pozza MF, Lingenhoehl K, Wang Y, Sheng M, Auberson YP, Wang YT (24) Role of NMDA receptor subtypes in governing the direction of hippocampal synaptic plasticity. Science 34: Luo J, Wang Y, Yasuda RP, Dunah AW, Wolfe BB (1997) The majority of N-methyl-D-aspartate receptor complexes in adult rat cerebral cortex contain at least three different subunits (NR1/NR2A/NR2B). Mol Pharmacol 51: Luo JH, Fu ZY, Losi G, Kim BG, Prybylowski K, Vissel B, Vicini S (22) Functional expression of distinct NMDA channel subunits tagged with green fluorescent protein in hippocampal neurons in culture. Neuropharmacology 42: Maletic-Savatic M, Malinow R, Svoboda K (1999) Rapid dendritic morphogenesis in CA1 hippocampal dendrites induced by synaptic activity. Science 283: Mammen AL, Kameyama K, Roche KW, Huganir RL (1997) Phosphorylation of the alpha-amino-3-hydroxy-5-methylisoxazole4-propionic acid receptor GluR1 subunit by calcium/calmodulin-dependent kinase II. J Biol Chem 272: Matsuda S, Mikawa S, Hirai H (1999) Phosphorylation of serine-88 in GluR2 by protein kinase C prevents its C terminus from binding with glutamate receptor-interacting protein. J Neurochem 73: Meggio F, Marin O, Pinna LA (1994) Substrate specificity of protein kinase CK2. Cell Mol Biol Res 4: Migaud M, Charlesworth P, Dempster M, Webster LC, Watabe AM, Makhinson M, He Y, Ramsay MF, Morris RG, Morrison JH, O Dell TJ, Grant SG (1998) Enhanced long-term potentiation and impaired learning in mice with mutant postsynaptic density-95 protein. Nature 396: Morais Cabral JH, Petosa C, Sutcliffe MJ, Raza S, Byron O, Poy F, Marfatia SM, Chishti AH, Liddington RC (1996) Crystal structure of a PDZ domain. Nature 382: Mori H, Manabe T, Watanabe M, Satoh Y, Suzuki N, Toki S, Nakamura K, Yagi T, Kushiya E, Takahashi T, Inoue Y, Sakimura K, Mishina M (1998) Role of the carboxy-terminal region of the GluR epsilon2 subunit in synaptic localization of the NMDA receptor channel. Neuron 21: Muller BM, Kistner U, Kindler S, Chung WJ, Kuhlendahl S, Fenster SD, Lau LF, Veh RW, Huganir RL, Gundelfinger ED, Garner CC (1996) SAP12, a novel postsynaptic protein that interacts with NMDA receptor complexes in vivo. Neuron 17: Murphy TH, Blatter LA, Wier WG, Baraban JM (1992) Spontaneous synchronous synaptic calcium transients in cultured cortical neurons. J Neurosci 12: Niethammer M, Kim E, Sheng M (1996) Interaction between the C terminus of NMDA receptor subunits and multiple members of the PSD-95 family of membrane-associated guanylate kinases. J Neurosci 16: Omkumar RV, Kiely MJ, Rosenstein AJ, Min KT, Kennedy MB (1996) Identification of a phosphorylation site for calcium/calmodulin-dependent protein kinase II in the NR2B subunit of the N-methyl-D-aspartate receptor. J Biol Chem 271: Palczewski K (1997) GTP-binding-protein-coupled receptor kinases two mechanistic models. Eur J Biochem 248: Passafaro M, Sala C, Niethammer M, Sheng M (1999) Microtubule binding by CRIPT and its potential role in the synaptic clustering of PSD-95. Nat Neurosci 2: Perez JL, Khatri L, Chang C, Srivastava S, Osten P, Ziff EB (21) PICK1

12 Chung et al. Phosphorylation of the NR2B PDZ Ligand J. Neurosci., November 1, 24 24(45): targets activated protein kinase C to AMPA receptor clusters in spines of hippocampal neurons and reduces surface levels of the AMPA-type glutamate receptor subunit 2. J Neurosci 21: Rao A, Craig AM (1997) Activity regulates the synaptic localization of the NMDA receptor in hippocampal neurons. Neuron 19: Roberson ED, Sweatt JD (1996) Transient activation of cyclic AMPdependent protein kinase during hippocampal long-term potentiation. J Biol Chem 271: Roche KW, Standley S, McCallum J, Dune Ly C, Ehlers MD, Wenthold RJ (21) Molecular determinants of NMDA receptor internalization. Nat Neurosci 4: Sans N, Prybylowski K, Petralia RS, Chang K, Wang YX, Racca C, Vicini S, Wenthold RJ (23) NMDA receptor trafficking through an interaction between PDZ proteins and the exocyst complex. Nat Cell Biol 5: Sarno S, Reddy H, Meggio F, Ruzzene M, Davies SP, Donella-Deana A, Shugar D, Pinna LA (21) Selectivity of 4,5,6,7-tetrabromobenzotriazole, an ATP site-directed inhibitor of protein kinase CK2 (casein kinase-2 ). FEBS Lett 496: Scott DB, Blanpied TA, Swanson GT, Zhang C, Ehlers MD (21) An NMDA receptor ER retention signal regulated by phosphorylation and alternative splicing. J Neurosci 21: Seeburg PH (1993) The TiPS/TINS lecture: the molecular biology of mammalian glutamate receptor channels. Trends Pharmacol Sci 14: Seidenman KJ, Steinberg JP, Huganir R, Malinow R (23) Glutamate receptor subunit 2 serine 88 phosphorylation modulates synaptic transmission and mediates plasticity in CA1 pyramidal cells. J Neurosci 23: Sprengel R, Suchanek B, Amico C, Brusa R, Burnashev N, Rozov A, Hvalby O, Jensen V, Paulsen O, Andersen P, Kim JJ, Thompson RF, Sun W, Webster LC, Grant SG, Eilers J, Konnerth A, Li J, McNamara JO, Seeburg PH (1998) Importance of the intracellular domain of NR2 subunits for NMDA receptor function in vivo. Cell 92: Standley S, Roche KW, McCallum J, Sans N, Wenthold RJ (2) PDZ domain suppression of an ER retention signal in NMDA receptor NR1 splice variants. Neuron 28: Tao YX, Rumbaugh G, Wang GD, Petralia RS, Zhao C, Kauer FW, Tao F, Zhuo M, Wenthold RJ, Raja SN, Huganir RL, Bredt DS, Johns RA (23) Impaired NMDA receptor-mediated postsynaptic function and blunted NMDA receptor-dependent persistent pain in mice lacking postsynaptic density-93 protein. J Neurosci 23: Tingley WG, Ehlers MD, Kameyama K, Doherty C, Ptak JB, Riley CT, Huganir RL (1997) Characterization of protein kinase A and protein kinase C phosphorylation of the N-methyl-D-aspartate receptor NR1 subunit using phosphorylation site-specific antibodies. J Biol Chem 272: Tovar KR, Westbrook GL (22) Mobile NMDA receptors at hippocampal synapses. Neuron 34: Wang YT, Salter MW (1994) Regulation of NMDA receptors by tyrosine kinases and phosphatases. Nature 369: Washbourne P, Bennett JE, McAllister AK (22) Rapid recruitment of NMDA receptor transport packets to nascent synapses. Nat Neurosci 5: Wenthold RJ, Prybylowski K, Standley S, Sans N, Petralia RS (23) Trafficking of NMDA receptors. Annu Rev Pharmacol Toxicol 43: Wyszynski M, Lin J, Rao A, Nigh E, Beggs AH, Craig AM, Sheng M (1997) Competitive binding of alpha-actinin and calmodulin to the NMDA receptor. Nature 385: Xia J, Chung HJ, Wihler C, Huganir RL, Linden DJ (2) Cerebellar longterm depression requires PKC-regulated interactions between GluR2/3 and PDZ domain-containing proteins. Neuron 28:

13 4. R. Lyczak, J. E. Gomes, B. Bowerman, Dev. Cell 3, 157 (22). 5. J. Pellettieri, G. Seydoux, Science 298, 1946 (22). 6. A. F. Severson, D. L. Baillie, B. Bowerman, Curr. Biol. 12, 266 (22). 7. D. Albertson, Dev. Biol. 11, 61 (1984). 8. S. W. Grill, P. Gönczy, E. H. K. Stelzer, A. A. Hyman, Nature 49, 63 (21). 9. D. J. Sharp, G. C. Rogers, J. M. Scholey, Nature 47,41 (2). 1. P. Gönczy, Trends Cell Biol. 12, 332 (22). 11. T. J. Mitchison, E. D. Salmon, Nature Cell Biol. 3, E17 (21). 12. S. C. Schuyler, D. Pellman, J. Cell Sci. 114, 247 (21). 13. J. Howard, A. A. Hyman, Nature 422, 753 (23). 14. See supporting data on Science Online. 15. M. F. Tsou, W. Ku, A. Hayashi, L. S. Rose, J. Cell Biol. 16, 845 (23). 16. F. J. Sigworth, Nature 27, 265 (1977). 17. F. J. Sigworth, J. Physiol. (London) 37, 97 (198). 18. D. Schmidt, S. Strome, unpublished data. 19. D. L. Dujardin, R. B. Vallee, Curr. Opin. Cell Biol. 14, 44 (22). 2. In the velocity-limited regime, a fragment moves at the maximum velocity whenever there are one or more force generators active. The more force generators per fragment, the higher the mean speed, because the probability of having at least one active force generator is increased. However, this model results in a mean-variance relationship identical to the force-limited case presented, with N 1. This is not consistent with our results. 21. With fewer than 12 aster fragments generated per spindle pole and one or two active force generators per fragment, we expect fewer than 5 active force generators throughout the embryo at any time during anaphase. 22. M. Gotta, J. Ahringer, Nature Cell Biol. 3, 297 (21). 23. Y. Cai, F. Yu, S. Lin, W. Chia, X. Yang, Cell 112, 51 (23). 24. P. Gönczy et al., Nature 48, 331 (2). 25. K. Colombo et al., Science 3, 1957 (23); published online 15 May 23 (1.1126/ science ). 26. M. Gotta, Y. Dong, Y. K. Peterson, S. M. Lanier, J. Ahringer, Curr. Biol. 13, 129 (23). 27. D. G. Srinivasan, R. M. Fisk, H. Xu, S. Van Den Heuvel, Genes Dev. 17, 1225 (23). 28. E. Hannak et al., J. Cell Biol. 157, 591 (22). 29. Because the laser was focused at regions where the pericentriolar material and -tubulin are localized, we expect a fraction of the centrosomal -tubulin to be destroyed. Consequently, not all aster fragments contained detectable levels of -tubulin. 3. L. Sachs, Applied Statistics:A Handbook of Techniques (Springer, New York, 1984). 31. Stray light from the pulsed ultraviolet (UV) laser used for OICD bleached the surrounding GFP tubulin. Therefore, UV laser power was reduced by about 7% to visualize fragment movement by GFP. With less laser power, we probably produced larger fragments with an increased drag coefficient. Thus, elementary speeds were smaller in the GFP assay than in the DIC assay. On the other hand, larger fragments have more force generators associated, consistent with the general increase in N for the GFP assay. 32. We thank C. Cowan, A. Desai, M. Diehl, P. Gönczy, M. Glotzer, E. Hannak, F. Jülicher, K. Oegema, L. Pelletier, A. Riedinger, G. Ritter, E. Sackmann, N. Salmon, M. Srayko, and E. Tanaka for useful discussions, experimental assistance, and helpful comments. Supported by Deutsche Forschungsgemeinschaft grant SPP 1111 (HY3/2-1). Supporting Online Material DC1 Materials and Methods SOM Text Figs. S1 to S7 References Movies S1 to S7 7 May 23; accepted 18 June 23 Role of Subplate Neurons in Functional Maturation of Visual Cortical Columns Patrick O. Kanold, Prakash Kara, R. Clay Reid, Carla J. Shatz* The subplate forms a transient circuit required for development of connections between the thalamus and the cerebral cortex. When subplate neurons are ablated, ocular dominance columns do not form in the visual cortex despite the robust presence of thalamic axons in layer 4. We show that subplate ablation also prevents formation of orientation columns. Visual responses are weak and poorly tuned to orientation. Furthermore, thalamocortical synaptic transmission fails to strengthen, whereas intracortical synapses are unaffected. Thus, subplate circuits are essential not only for the anatomical segregation of thalamic inputs but also for key steps in synaptic remodelingand maturation needed to establish the functional architecture of visual cortex. Subplate neurons, located in the developing white matter (WM), are among the first postmitotic cortical neurons (1). They are also first to receive functional synaptic inputs from the thalamus; their axons relay this input into the cortical plate well in advance of the invasion of thalamic axons into layer 4 (1). Once thalamic axons have arrived in layer 4, subplate neurons are gradually eliminated during the period of ocular dominance column (ODC) formation (1 3). Thus, subplate neurons are in a key, intermediate position to control the flow of information into the developing cortex when first spontaneous (prenatal) and then visual (postnatal) activity are present (3). Early subplate ablation prevents the invasion of thalamic axons into layer 4 (4), whereas later ablation blocks the anatomical segregation of thalamic axons from the lateral geniculate nucleus (LGN) into ODCs in primary visual cortex (5). The prevailing hypothesis about ODC formation and plasticity is that activity-dependent competition between LGN axons representing each eye leads to their selective growth or pruning (3). However, recent observations that ODCs emerge earlier than previously thought, even before the onset of patterned visual experience (2, 6), have required a revision of this hypothesis, and one suggestion is that subplate neurons are involved (2). To determine how the subplate might influence functional development and organization of visual cortex, subplate neurons were selectively ablated at P6 to P9 in cats (7). This time is just before the onset of ODC formation (6, 8, 9) and just after the Department of Neurobiology, Harvard Medical School, Boston, MA 2115, USA. *To whom correspondence should be addressed. E- mail: carla_shatz@hms.harvard.edu R EPORTS ingrowth of LGN axons to cortical layer 4 (1). To make localized and selective ablations, we either used kainic acid injections (4, 5) or an immunotoxin (11, 12) to p75, a neurotrophin receptor expressed in the neocortex only by subplate neurons at this age (1). Response properties of cortical neurons were evaluated after subplate ablation by optical imaging and microelectrode recordings at P24 to P49 (7), when well-organized orientation maps are normally present (6, 13, 14). Orientation maps in control hemispheres or far from the ablation were highly organized (Fig. 1A, left and center), in contrast to poor organization in the highly local region surrounding the immunotoxin-ablated area (Fig. 1A, center). Kainic acid injections created even larger regions of disrupted or absent orientation maps (Fig. 1A, right). Polar maps representing both the angle and strength of tuning revealed weaker orientation selectivity in the ablated region (Fig. 1B), indicating the existence of an additional severe functional deficit. ODC maps were also degraded inablated hemispheres (fig. S1), consistent with previous anatomical data (5). Degraded orientation maps can result from neurons well tuned to different orientations but not segregated into columns, from neurons poorly tuned for orientation, or simply from neurons that respond poorly to visual stimuli. To distinguish between these alternatives, ocular dominance and orientation tuning of single cortical neurons were measured with microelectrode recordings (7) in both ablated and control regions in each animal (Fig. 1A and fig. S2). Because LGN axons are abundant in layer 4 of subplate-ablated cortex (Fig. 2A) (5), we had expected to find strong visually driven responses. Instead, there were fewer visually responsive units in the ablated area (15), and they were less orientation-selec- SCIENCE VOL JULY

14 R EPORTS tive (Fig. 1, C and D). In addition, visually driven units in ablated cortex were only weakly driven by the ipsilateral eye, resulting in a strong response bias toward the contralateral eye (15). This bias is reminiscent of normal visual cortex at earlier developmental ages (6, 13), suggesting arrested development. Alterations in cortical organization, as measured by several anatomical and molecular correlates, were also similar after immunotoxin or kainate ablations. In the immunotoxin-ablated region, uniform transneuronal labeling showed that ODCs were absent despite a robust projection from the LGN to layer 4 (Fig. 2A), as with kainate ablations (5); ODCs were present in immediately adjacent unablated regions. Brainderived neurotrophic factor (BDNF) mrna (7) also increased in layers 2 to 4 of cortex located above the immunotoxinablated region (Fig. 2, B and D), as occurs after kainate ablations (16). Although immunotoxin ablation altered geniculocortical connectivity, gene expression, and columnar functional organization of cortex, the gross histology of cortical layers residing above the ablation site was indistinguishable from that of unablated adjacent regions (Fig. 2E). Cortical p75 is almost exclusively restricted to the subplate (1), but cholinergic fibers from basal forebrain are also known to express p75 (17). Therefore, we also verified that cholinergic fibers were abundant in immunotoxinablated regions (fig. S4). The only obvious histological effect seen after immunotoxin injection was the loss of neurons in WM, as expected because subplate neurons are normally located there (Fig. 2, F and G). In summary, both immunotoxin and kainate treatments have similar consequences for cortical development yet act through completely different mechanisms. Thus, observed changes are specifically because of removal of subplate neurons rather than nonspecific effects of the injections (18). One explanation for the functional deficits in visual cortex is that LGN neurons are affected by subplate ablation. When we recorded in the LGN, however, neurons projecting to the ablated site had responses and receptive field organization (Fig. 3A) appropriate for this age (19). Thus, observed defects in orientation tuning and visual responsiveness must occur beyond the LGN. To examine whether the cortical defect originates at thalamocortical synapses, we electrically stimulated the LGN while we recorded evoked field potentials at a retinotopically matching cortical location in vivo. Field potentials in subplate-ablated regions were smaller in all layers, particularly layer 4, than those in control regions (Fig. 3B), consistent with only weak activation of thalamocortical synapses. These results were confirmed by in vitro recordings from slices of visual cortex (7). Field potentials recorded in layers 2/3 and 4, evoked by WM stimulation, were vastly reduced in the ablated regions (Fig. 3C). The size of the monosynaptic response in layer 4 was significantly smaller, implying very weak thalamocortical synapses (Fig. 3D). In contrast, evoked potentials in layer 2/3 after layer 4 stimulation were close to normal in amplitude and waveform in ablated regions (Fig. 3E). Histological observations and in vivo recordings indicate that layer 4 neurons are present after subplate ablation. Whole-cell patch recordings performed on layer 4 neurons in slices (8) confirmed that neurons from ablated regions fired action potentials to current injections (Fig. 4A), had normal measures of intrinsic membrane properties (Fig. 4B), and appeared to have normal dendritic morphology (Fig. 4A, inset). The attenuated field potentials after subplate ablation could reflect a lack of spatial organization and neuronal synchrony in subplate-ablated cortex or an underlying attenuation of synaptic currents. Direct measurement of the size of excitatory synaptic inputs to layer 4 neurons showed that excitatory postsynaptic currents evoked by electrical WM stimulation (eepscs) were reduced by 8% in subplate-ablated slices compared to those of controls (Fig. 4C; see also fig. S5). Because thalamocortical axons are present in layer 4 (for example, Fig. 2A) and LGN neurons have normal visual responses, it is possible that the attenuation in thalamocortical transmission is because of a postsynaptic defect. In fact, mrna expression of the dominant AMPA receptor subunit, GluR1, is selectively reduced by 65% within layer 4 in the ablated area (Fig. 4D), implying that the maturation of layer 4 Fig. 1. Subplate ablation disrupts orientation maps and prevents sharpening of orientation tuning. (A) Invivo optical imaging of orientation maps in control and ablated visual cortex. Color indicates preferred orientation at each pixel. Scale bars indicate 1 mm. Each immunotoxin or kainate injection site is marked with an x. Compared with control hemisphere (left), which shows robust orientation domains and pinwheels, maps in subplateablated hemispheres at P48 after immunotoxin injections at P8 and at P37 after kainate at P9 are disorganized in regions receiving ablations. Immunotoxin treatments (n 2 animals) are more focal (area above dashed line) than with kainate (n 5 animals). Optical maps after sham injections (saline) were indistinguishable from that of the control [n 1 animal (29)]. (B) Polar maps showing both preferred orientation (color) and tuning strength (brightness). Strong orientation-selective responses (bright colors) and pinwheels are evident in unablated areas. Orientation maps are nonselective (dark) and unorganized in ablated cortex. (C and D) Summary histograms of orientation-tuning strength (OTS) (7) of single units in ablated (C) and control (D) cortex (, unselective; 1, perfectly selective). Compared to control (.48.25, n 31), OTSs of all units (including unresponsive units, in red) were reduced by 82% in ablated cortex (.9.15, n 62, P.1). Compared to those of controls (.48.25, n 31), OTSs in ablated cortex of just the visually driven units (black) were reduced by 44% (.27.24, n 22, P.5). Insets show typical OTS examples JULY 23 VOL 31 SCIENCE

15 R EPORTS synapses is impaired. In marked contrast to the decreased eepsc amplitudes after subplate ablation (Fig. 4C), the amplitudes of Fig. 2. Immunotoxin ablation of subplate neurons prevents segregation of LGN axons into ODCs and upregulates expression of BDNF in cortex. (A) Darkfield autoradiograph of visual cortex at P82 after four injections (indicated by asterisks) of immunotoxin at P9 (two posterior injections of.25 mg/ml and two anterior injections of.5 mg/ml). ODCs were visualized by transneuronal transport after monocular injection of 3 H-proline (7). (Silver grains, which label LGN axons and axon terminals, appear white in darkfield autoradiograph.) Ablation extended 1 mm from each injection site, consistent with reported data (11). Robust, continuous transneuronal labeling (between arrowheads) in layer 4 signifies absence of ODCs above the subplate ablation zone as compared to patchy labeling of ODCs in control regions. Control immunotoxin treatment does not disrupt ODCs (fig. S3). (B) In situ hybridization (darkfield autoradiograph) shows increased BDNF mrna expression in cortex at P16 after four A the spontaneous miniature EPSCs (mepscs) are unchanged in layer 4 neurons (Fig. 4E), and their frequency actually increased ODCs B BDNF C control D E F control G ablated ablated immunotoxin (1 mg/ml) injections (asterisks) into subplate at P9. Boxes indicate locations of (C) and (D). (C and D) BDNF mrna is abnormally increased in layers 2/3 and 4 in subplate-ablated (D) but not adjacent control (C) cortex. Normally, at this early age, BDNF mrna expression is restricted to the subplate layer 6 border (3). Cortical layers and WM indicated on right. (E) Histology of cortical plate (CP) at P28 (cresyl stain) appears normal after immunotoxin (1 mg/ml) ablations at P7. The presence of fluorescent microspheres (coinjected with immunotoxin) in the WM is indicated (above asterisks). (F and G) Immunohistochemistry with the use of NeuN [a neuron-specific marker (Chemicon, Temecula, CA)] confirms selective loss of subplate neurons at injection sites [indicated by asterisks in (G)]. Scale bars, 1 mm [(A) and (B)],.2 mm [(C), (D), and (E)], and 8 m [(F) and (G)]. CP WM 4 1 2/3 4 5/6 WM CP WM (Fig. 4F). The observed increase in spontaneous activity in the face of decreased thalamic inputs is consistent with the presence of a homeostatic mechanism controlling overall synaptic input (2). Taken together, recordings from layer 4 in vivo and in vitro indicate that layer 4 neurons are present and can receive intracortical synaptic input (as evident from the mepscs) but that there has been a selective loss of synaptic drive from the thalamus. The major finding of this study is that subplate ablation leads to an impairment in synaptic transmission of visually driven activity from the LGN into cortical circuits. Although LGN axons are present in layer 4, the cortex becomes essentially uncoupled from the thalamus, as assessed both in vivo and in vitro. In addition, functional ODCs and orientation maps do not develop [Supporting Online Material (SOM) Text]. Although the majority of neurons were unresponsive to visual stimuli, some cells did have robust visual responses but nevertheless were not orientation-selective. This global absence of functional tuning of cortical maps after subplate ablation is related to the failure of thalamocortical synapses to strengthen and the abnormally low expression in layer 4 neurons of key molecular components mediating glutamatergic transmission, such as GluR1. How can subplate ablation lead to the observed functional decoupling of thalamus from cortex? Although increased feed forward inhibition could mask thalamocortical EPSCs, our observation of increased mepsc frequency and absence of large inhibitory postsynaptic currents makes this hypothesis unlikely. We favor the hypothesis that depolarizing input from subplate Fig. 3. Subplate ablation alters thalamocortical but not intracortical field potentials. (A) Typical LGN receptive fields (P48) that retinotopically overlapped the visual cortex receiving immunotoxin ablations at P8. Receptive-field size and organization are within normal range for these ages (19). (B) In vivo cortical field potentials after electrical stimulation of LGN at matching retinotopic locations (vertical arrows mark stimulus onset). Fields are smaller in ablated compared with control regions throughout all layers (layer 4 is between green arrows). (C) Field recordings in vitro from ablated versus control hemispheres in cortical slices at P3 (ablated at P8) (7). WM stimulation evokes large field potentials in layer 2/3 and layer 4 in control slices (n 16), whereas only small field potentials are present in these layers in ablated slices (n 24). Stimulus artifacts are removed. (D) First-order current-source density (CSD) calculated from field potentials recorded in slices after WM stimulation. In control slices (left), a large monosynaptic sink is present at 1.5-ms latency in layer 4 (green arrow). Large, longer latency sinks are visible in layers 2/3 and 5/6. In contrast, no layer 4 sinks are present in ablated hemispheres (right) (CSD variance of V 2 /mm 4 versus V 2 /mm 4, P.2, Mann-Whitney). (E) Electrical stimulation of layer 4 in slices from ablated or control hemispheres produces monosynaptic field potentials in layer 2/3, demonstrating intact transmission from layer 4 to layer 2/3 even in ablated regions (n 15). SCIENCE VOL JULY

16 R EPORTS Fig. 4. Subplate ablation alters thalamocortical synaptic efficacy in layer 4 neurons. (A) Whole-cell recording of a layer 4 neuron in a subplate-ablated cortical slice (7). Traces show responses to multiple current injections. Inset shows a biocytin-filled layer 4 neuron with normal morphology located in ablated region. (B) Neurons from control ( c ) as compared to ablated ( a ) slices have similar resting potential ( mv for control, n 24, as compared to mv for ablated, n 2; P.1) and input resistance ( M for control, n 24, as compared to M for ablated, n 2; P.1). (C) Cumulative amplitude histograms show that WM stimulation evokes eepscs in ablated or control slices. The eepsc amplitude (inset) is 8% smaller in ablated slices ( pa for ablated, n 22, as compared to pa for control, n 22; P.1, double asterisks). Adjacent traces show mean eepscs of representative neurons from each population. eepsc amplitudes from saline and control-immunotoxin slices are indistinguishable from control (fig. S5). (D) Darkfield autoradiograph of in situ hybridization for GluR1 mrna in control (left) or ablated (right) cortex. Scale bar,.2 mm. Densitometry (histogram) shows that layer 4 GluR1 mrna levels in ablated regions are reduced to 35% of that of controls (P.5, double asterisks, n 4 animals). GluR1 mrna levels are unchanged after layer 4 kainate injections (n 2 animals) (fig. S6) or saline injections into the subplate [n 1 animal (29)]. (E) Cumulative histograms and bar graphs (inset) show that mepsc amplitudes are similar ( pa for ablated, n 13, compared to pa for control, n 13; P.1). (F) Cumulative mepsc frequency histograms and bar graph (inset) show that mepsc frequency is 3.6-fold higher in ablated slices ( Hz for ablated, n 13, compared to.65.8 Hz for control, n 13; P.1, double asterisks). neurons is needed for the activation of previously silent thalamocortical synapses (21) because subplate neurons supply excitatory input to cortical neurons (22, 23). In this case, some form of learning rule, known to be present at thalamocortical synapses (24, 25), is disrupted in the absence of subplate neurons, preventing the progressive strengthening of thalamic synapses. The observation here that GluR1 mrna is decreased after subplate ablation implies that excitatory drive on all cortical neurons, including inhibitory neurons, is also diminished. It is remarkable that eye segregation fails even though LGN receptive fields are normal and both LGN axons and layer 4 neurons are present in subplate-ablated cortex. Our observations constrain mechanisms of patterning of connectivity in cortical circuits. It has been suggested that thalamocortical connectivity during the critical period depends on cortical circuits outside of layer 4 (26). In contrast, our findings show that patterning of thalamocortical connectivity before onset of the critical period crucially depends on proper LGN input to the cortex, as regulated by subplate neurons. Furthermore, Crowley and Katz have proposed that eye-specific molecular cues set up the initial pattern of ODCs during development (2, 27), with activity being required only at later ages. If so, then our results argue that these hypothetical molecular cues in cortex reside not in layer 4 or LGN axon terminals, both of which are present in the subplate-ablated cortex, but instead in the subplate. Alternatively, subplate neurons could mediate very early activity-dependent interactions driven by correlated spontaneous activity relayed from the retina to the cortex by LGN neurons, which is also known to be present in corticogeniculate circuits (3, 28). Regardless of which alternative proves to be the case, our observations here demonstrate that circuits formed by subplate neurons are needed both for strengthening and for selective remodeling of thalamocortical excitatory synaptic connections that underlie the development of columnar organization in cortex. References and Notes 1. K. L. Allendoerfer, C. J. Shatz, Annu. Rev. Neurosci. 17, 185 (1994). 2. J. C. Crowley, L. C. Katz, Science 29, 1321 (2). 3. L. C. Katz, C. J. Shatz, Science 274, 1133 (1996). 4. A. Ghosh, A. Antonini, S. K. McConnell, C. J. Shatz, Nature 347, 179 (199). 5. A. Ghosh, C. J. Shatz, Science 255, 1441 (1992). 6. M. C. Crair, J. C. Horton, A. Antonini, M. P. Stryker, J. Comp. Neurol. 43, 235 (21). 7. Materials and methods are available as supporting material on Science Online. 8. In cats, ODCs form at about P14 (6) or later. Moreover, even at P19, many individual axons have broadly distributed processes that have not yet focused their branching to the width of an ODC (9). 9. A. Antonini, M. P. Stryker, J. Neurosci. 13, 3549 (1993). 1. C. J. Shatz, M. B. Luskin, J. Neurosci. 6, 3655 (1986). 11. L. Mrzljak, A. I. Levey, S. Belcher, P. S. Goldman-Rakic, J. Comp. Neurol. 39, 112 (1998). 12. A. Fine et al., Neuroscience 81, 331 (1997). 13. M. C. Crair, D. C. Gillespie, M. P. Stryker, Science 279, 566 (1998). 14. B. Chapman, M. P. Stryker, T. Bonhoeffer, J. Neurosci. 16, 6443 (1996). 15. Many neurons in ablated regions were not well driven by visual stimuli [average responsiveness (AR) ; 36 of 86 units were unresponsive and only 13 of 86 units were strongly driven], whereas neurons outside poorly tuned regions were strongly responsive to visual stimuli (AR ; 2 of 59 units were unresponsive and 43 of 59 units were strongly driven; P.1). Neurons in ablated areas showed increased spontaneous firing rate (a control rate of.66.9 Hz as compared to an ablated rate of Hz, P.7), whereas visually driven rates were similar (a control rate of Hz compared with an ablated rate of Hz, P.5). Visual responsiveness (VR) to ipsilateral eye stimulation is smaller in neurons from ablated regions (VR control , n 31; VR ablated.85.95, n 19; P.1). VR to contralateral eye stimulation was similar (VR control , n 31; VR ablated , n 15; P.1), resulting in response bias toward the contralateral eye for neurons in ablated regions but not control regions as measured by the Contralateral Bias Index (CBI) (CBI ablated.48.62, CBI control.1.39, P.5). 16. E. S. Lein, E. M. Finney, P. S. McQuillen, C. J. Shatz, Proc. Natl. Acad. Sci. U.S.A. 96, (1999). 17. E. P. Pioro, A. C. Cuello, Neuroscience 34, 57 (199). 18. Nonspecific inflammation after injections also cannot account for observed changes because control JULY 23 VOL 31 SCIENCE

17 immunotoxin (saporin linked to a nonspecific mouse immunoglobulin G) injections, saline injections into subplate, and kainate injections into layer 4 have no effect on ODCs (5, 16) (fig. S3). Furthermore, control immunotoxin and saline injections do not alter synaptic physiology as assessed in slices (fig. S5). Layer 4 injections of kainate or saline injections into subplate, although right at the site of the disconnection effects and causing mechanical damage from the injection needle, do not disrupt any known aspect of cortical structure or gene expression that have been monitored (5, 16) (fig. S6). 19. D. Cai, G. C. DeAngelis, R. D. Freeman, J. Neurophysiol. 78, 145 (1997). 2. G. G. Turrigiano, S. B. Nelson, Curr. Opin. Neurobiol. 1, 358 (2). 21. J. T. Isaac, M. C. Crair, R. A. Nicoll, R. C. Malenka, Neuron 18, 269 (1997). 22. E. M. Finney, J. R. Stone, C. J. Shatz, J. Comp. Neurol. 398, 15 (1998). 23. E. Friauf, C. J. Shatz, J. Neurophysiol. 66, 259 (1991). 24. M. C. Crair, R. C. Malenka, Nature 375, 325 (1995). 25. D. E. Feldman, R. A. Nicoll, R. C. Malenka, J. T. Isaac, Neuron 21, 347 (1998). 26. J. T. Trachtenberg, C. Trepel, M. P. Stryker, Science 287, 229 (2). 27. J. C. Crowley, L. C. Katz, Nature Neurosci. 2, 1125 (1999). 28. M. Weliky, L. C. Katz, Science 285, 599 (1999). 29. P. O. Kanold, P. Kara, R. C. Reid, C. J. Shatz, unpublished data. 3. E. S. Lein, A. Hohn, C. J. Shatz, J. Comp. Neurol. 42, 1 (2). 31. We thank R. Born for helpful discussions; M. Marcotrigiano, B. Printseva, and S. Yurgensen for surgical, histological, and technical assistance; and D. Butts, M. Majdan, J. Syken, and Y. Tagawa for helpful comments on the manuscript. Supported by NIH R1 EY2858 (C.J.S.), F32 EY1352 (P.O.K.), R1 EY1115 (R.C.R.), and P3 EY12196 (R.C.R.). Supporting Online Material DC1 Materials and Methods SOM Text Figs. S1 to S6 4 March 23; accepted 13 June 23 R EPORTS Melanopsin Is Required for Non Image-Forming Photic Responses in Blind Mice Satchidananda Panda, 1,2 * Ignacio Provencio, 4 * Daniel C. Tu, 5 * Susana S. Pires, 1 Mark D. Rollag, 4 Ana Maria Castrucci, 4,7 Mathew T. Pletcher, 1,2 Trey K. Sato, 1,2 Tim Wiltshire, 1 Mary Andahazy, 1 Steve A. Kay, 2 Russell N. Van Gelder, 5,6 John B. Hogenesch 1,3 Although mice lacking rod and cone photoreceptors are blind, they retain many eye-mediated responses to light, possibly through photosensitive retinal ganglion cells. These cells express melanopsin, a photopigment that confers this photosensitivity. Mice lackingmelanopsin still retain nonvisual photoreception, suggesting that rods and cones could operate in this capacity. We observed that mice with both outer-retinal degeneration and a deficiency in melanopsin exhibited complete loss of photoentrainment of the circadian oscillator, pupillary light responses, photic suppression of arylalkylamine-n-acetyltransferase transcript, and acute suppression of locomotor activity by light. This indicates the importance of both nonvisual and classical visual photoreceptor systems for nonvisual photic responses in mammals. 1 Genomics Institute of the Novartis Research Foundation, 1675 John J. Hopkins Drive, San Diego, CA 92121, USA. 2 Department of Cell Biology, 3 Department of Neuropharmacology, Scripps Research Institute, 155 North Torrey Pines Road, San Diego, CA 9237, USA. 4 Department of Anatomy, Physiology, and Genetics, Uniformed Services University, 431 Jones Bridge Road, Bethesda, MD 2814, USA. 5 Department of Ophthalmology and Visual Sciences, 6 Department of Molecular Biology and Pharmacology, Washington University School of Medicine, 66 South Euclid Avenue, St. Louis, MO 6311, USA. 7 Department of Physiology, Biosciences Institute, University of São Paulo, Rua do Matão, travessa 14, 558-9, São Paulo, Brazil. *These authors contributed equally to this work. To whom correspondence should be addressed. E- mail: hogenesch@gnf.org The eye is the principal mediator of light input to the central nervous system in mammals. In addition to vision, the eye mediates several nonvisual responses to light, including photoentrainment of the circadian oscillator, constriction of the pupil, acute suppression of pineal melatonin, acute suppression of activity (masking) in nocturnal mammals, and regulation of sleep latency. Many of these responses persist in mice that are visually blind from outer-retinal degeneration but are abolished by bilateral enucleation of the eyes (1). Here, we demonstrate the presence of inner-retinal, nonvisual ocular photoreceptors that specifically subserve these nonvisual photic responses. Intrinsically photosensitive retinal ganglion cells (iprgcs) (2, 3) project to brain sites that mediate many of these ocular, yet nonvisual, responses to light, including the suprachiasmatic nucleus (SCN), the intergeniculate leaflet, and the olivary pretectal nucleus, which mediates pupillary light reflexes (PLR) (1). The photosensitivity of these cells ex vivo depends on the presence of melanopsin (Opn4), a member of the opsin family of photopigment proteins (4). Whereas melanopsin-deficient (Opn4 / ) mice exhibit moderate attenuation in lightinduced phase resetting of the circadian oscillator (5, 6) and reduced PLR under high irradiance levels (4), most nonvisual photic responses in these mice remain largely intact. This suggests either the presence of additional inner-retinal photoreceptors, or contributions from the outerretinal classical photoreceptors to nonvisual photoresponses. To test the latter hypothesis, we generated mice that were deficient in both melanopsin and classical photoreceptors by breeding Opn4 / mice (5) with the C3H/HeJ mouse strain that carries the retinal degeneration (rd) mutation (7). Mice homozygous for the rd allele are visually blind as a result of a primary degeneration of the rods and a secondary loss of cones, but they retain melanopsin-containing RGCs (fig. S1). The Opn4 / ; rd/rd mice were healthy and viable with intact optic nerves. Outer retinal degeneration was indistinguishable between rd/ rd and Opn4 / ; rd/rd mice (fig. S1). To assess the circadian photoentrainment and acute light suppression of activity, we subjected the Opn4 / ; rd/rd mice, littermate wild-type, rd/rd, and Opn4 / mice to a 24-hour light:dark (LD) cycle (8L:16D) (7). Under conditions of constant darkness (DD), mice have a free-running circadian locomotor period of less than 24 hours. However, in a 24-hour LD cycle, photic input to the oscillator makes a small phase adjustment in each cycle and synchronizes the clock to an exact 24-hour period (photoentrainment). Wild-type mice and the single Opn4 / and rd/rd mutants entrained normally and consolidated their wheel-running activity to the dark period of the LD cycle (Fig. 1) as has been previously reported (5, 6, 8). In contrast, the Opn4 / ; rd/rd mice failed to entrain to the external lighting cycle and continued to exhibit free-running rhythms (Fig. 1 and Table 1). In addition, increasing the light intensity to 8 lux during the photoperiod and increasing the photoperiod to 12 hours failed to entrain these mice (Fig. 1 and Table 1; fig. S2). All four genotypes exhibited free-running DD periods of 24 hours (Table 1). Under constant light (LL) conditions, most nocturnal SCIENCE VOL JULY

18 articles 2 Nature America Inc. Large-scale oscillatory calcium waves in the immature cortex Olga Garaschuk 1,2, Jennifer Linn 2, Jens Eilers 2,3 and Arthur Konnerth 1,2 1 Institut für Physiologie, Technische Universität München, 882 München, Germany 2 Physiologisches Institut, Universität des Saarlandes, Homburg, Germany 3 Present address: Department of Neurobiology, Duke University Medical Center, Durham, North Carolina, 2771, USA Correspondence should be addressed to A.K. (konnerth@physiol.med.tu-muenchen.de) 2 Nature America Inc. Two-photon imaging of large neuronal networks in cortical slices of newborn rats revealed synchronized oscillations in intracellular Ca 2+ concentration. These spontaneous Ca 2+ waves usually started in the posterior cortex and propagated slowly (2.1 mm per second) toward its anterior end. Ca 2+ waves were associated with field-potential changes and required activation of AMPA and NMDA receptors. Although GABA A receptors were not involved in wave initiation, the developmental transition of GABAergic transmission from depolarizing to hyperpolarizing (around postnatal day 7) stopped the oscillatory activity. Thus we identified a type of large-scale Ca 2+ wave that may regulate long-distance wiring in the immature cortex. Spontaneous correlated neuronal activity represents a hallmark of the developing central nervous system (see refs. 1 4 for review). The mechanisms underlying these spontaneous oscillations are highly distinct and include communication through gap junctions and Ca 2+ release from intracellular Ca 2+ stores in the cortex 5,6, activation of acetylcholine receptors in the retina 7 and the depolarizing action of GABA A receptors in the hippocampus 8,9. The oscillations differ not only in the underlying mechanisms, but also in their specific patterns. For example, retinal waves are initiated at random locations and travel in different directions through a subset of retinal cells 7,1. The same retinal cell can sequentially participate in different waves of activity that travel in various directions. By contrast, the hippocampal early network oscillations represent stereotypical activity patterns that involve the whole neuronal population and recur with regular frequency 9. The region specificity of the spontaneous oscillatory activity patterns is believed to promote the establishment of region-specific connections 11. So far, spontaneous oscillatory activity in the immature neocortex is detected only in individual cells 12,13, in cell pairs 14 or in rather small groups of 5 5 neurons, so-called neuronal domains 5. These spontaneous cortical activities are also very different mechanistically. They are mediated either by activation of metabotropic glutamate 13 or GABA A 12 receptors or by passage of IP 3 through gap junctions 6. The restricted amount of synchronous oscillatory activity in the neonatal cortex and the small size of spontaneously active neuronal domains seem to be consistent with the very low density of synaptic contacts, as demonstrated both morphologically 15,16 and functionally 17. However, it remains unclear how such local types of neuronal activities can promote the establishment of long-range neuronal connections, which are typical for both excitatory and inhibitory neurons in the adult cortex 18 and are found in the developing visual cortex already at the time of eye opening, before extensive visual experience 19,2. These long-range connections are thought to represent the morphological basis for binding signals oscillatory electrical activity underlying the reintegration or binding of information acquired by different receptive fields into a single perceptual entity 21. Here we identified a form of large-scale spontaneous oscillatory activity using two-photon imaging to simultaneously monitor intracellular calcium concentration ([Ca 2+ ] i ) in thousands of cortical neurons in newborn rat slices. In marked contrast to previously detected activities, these cortical early network oscillations involved the entire network of neurons, were highly correlated across the cortical slice and consisted of Ca 2+ waves propagating along the longitudinal axis of the cortex, crossing anatomical boundaries between different cortical subregions. RESULTS Our approach enabled real-time functional analyses of fairly large cortical networks in the immature rodent brain (Fig. 1). We prepared acute brain slices of the desired brain region (Fig. 1a and c) and then stained the cells with a fluorescent, membrane-permeable ion indicator dye 12,22. For example, the Ca 2+ indicator dye fura-2 AM routinely stained all neurons within the top 5 15 µm of the slices efficiently. When imaging this preparation with a two-photon laser scanning microscope 23,24, we could simultaneously monitor thousands of neurons in a thin optical section within the slice 24. The improved depth penetration and the high sensitivity of two-photon imaging 25,26 allowed us to record at a depth at which we would expect cells to be minimally, if at all, affected by the slicing procedure. Images were obtained at various magnifications (Fig. 1c e). At high magnifications, individual neurons and their dendrites could be resolved (Fig. 1e) and functionally analyzed. Recordings with a good signal-to-noise ratio were obtained at relatively low excitation intensities. Under these conditions, no significant photobleaching or photodamage was observed; thus, it was possible to measure fluorescence continuously for long periods (up to eight hours). We used two-photon imaging to investigate the cellular activities in brain slices from newborn rats. Surprisingly, instead 452 nature neuroscience volume 3 no 5 may 2

19 2 Nature America Inc. articles 2 Nature America Inc. a b Mode-locked laser light 79 nm, 8 Hz, 1 fs Two-photon excitation Focal plane Brain slice Scan mirrors Lens Brain slice stained with fura 2-AM Fig. 1. Two-photon Ca 2+ imaging in large neuronal assemblies. (a) Schematic drawing of the neonatal rat brain. Dotted lines delineate the region from which cortical slices were obtained. (b) Illustration of the experimental arrangement. Note that fluorescence, generated by two-photon excitation, is restricted to the focal plane. (c) Two-photon fluorescence image of a cortical slice from a 2-day-old rat taken at a low magnification (1.9 objective). Scale bar, 1 mm. In this and the following figures, nylon threads used to keep the brain slice in place appear as dark transverse lines. The cortical region delimited by the white box is shown at a higher magnification (2 objective; d). Scale bar, 5 µm. (e) At the highest magnification (6 objective), individual cells and their dendritic process are well resolved. Scale bar, 1 µm. of small active domains 5, prolonged examination of slices from one- or two-day-old (P1 P2) rats (Fig. 2a) revealed prominent spontaneous, transient elevations in the intracellular Ca 2+ concentration that occurred globally throughout the cortex (Fig. 2b). These Ca 2+ transients recurred at very low rates (approximately once per 1 12 minutes; Fig. 2e and f) and were correlated throughout the entire cortical slice over distances of 8 millimeters or more. Each trace in Fig. 2b shows population responses from about 5 neurons recorded at low magnification. These spontaneous network oscillations of intracellular Ca 2+ levels were a robust phenomenon that persisted for many hours with little alteration, even after over seven hours of continuous fluorometric recordings (Fig. 2d). Individual waves occurred either as bursts of Ca 2+ transients, mostly in the posterior (entorhinal) and the anterior (perirhinal/insular) region, or as single transients in the middle portion (temporal/perirhinal) of the cortex (Fig. 2c). In recordings from seven slices, the activity was analyzed in the different anatomical layers of the temporal/perirhinal cortex. We found that the activity was always present in all cortical layers, further indicating that the whole cortical network was involved in this large-scale spontaneous neuronal activity. Therefore, we termed this new form of spontaneous activity cortical early network oscillations (ceno). c d e 1mm In view of the weakly developed synaptic connectivity in the cortex of neonates 3,15 17,27, it is important to know how many neurons participate in the cenos. Comparison of the population activity recorded in a region covering 38 neurons (Fig. 3a c) to the activity registered in 2 individual neurons of this group (Fig. 3d) demonstrated a tight correlation of most signals. Asynchronous, small Ca 2+ signals were sometimes detected in the individual neurons (Fig. 3d). These Ca 2+ transients may be associated with action potentials firing out of phase. The ceno-associated Ca 2+ transients were not restricted to the cell bodies but were always present in the main apical dendrite of each pyramidal neuron (Fig. 3d). Interestingly, ceno-associated Ca 2+ transients were detected in almost all neurons (32 of 38) within the cortical region shown in Fig. 3b. In another experiment, more than 1 cells were analyzed (Fig. 3e). Again, most neurons of this group responded during almost every wave of the cenos with a high degree of synchrony. Taken together, such analyses of 771 cells in 11 slices (groups of 3 2 cells) revealed that 87 ± 3% of the neurons examined participated in the cenos. The ceno-associated Ca 2+ transients were closely correlated with changes in the field potential (Fig. 3f) recorded with extracellular electrodes. Similar ceno-associated changes in field potential were also observed in non-stained and non-imaged slices (Fig. 3g), indicating that neither the staining procedure nor the imaging-associated illumination significantly affected the oscillatory activity. Field potential transients had comparable amplitudes and durations in both cases (Fig. 3h). Notably, the amplitudes of the field potentials were very small both in stained (51 ± 4 µv, n = 47 events, 4 slices, P3 P5 rats) and in non-stained slices (63 ± 2 µv, n = 34 events, 5 slices, P4 rats), with a signal-to-noise ratio that was far inferior to that obtained by Ca 2+ imaging. When examined at higher temporal resolution by using linescan recordings (sampling rate 16 Hz), each oscillatory event represented a spreading wave of activity (Fig. 4). A Ca 2+ wave (Fig. 4a) was monitored at three points over a distance of 4.7 mm to yield posterior, intermediate and anterior time traces (Fig. 4b). At higher magnification (Fig. 4c e), the propagation could be easily resolved at the level of individual neurons as a series of sequential responses. In the majority of cases (84 ± 7%, n = 18 slices), the waves of activity propagated from posterior to anterior within the slice (Fig. 4a f). However, in 7 of 18 slices, we also observed waves traveling in the opposite direction (Fig. 4e and f). Propagation speeds measured on the level of single cells (2. ±.4 mm per s, n = 8 slices) and for long-range propagation (2.1 ±.5 mm per s, n = 5 slices; Fig. 4g) were similar. The speed of propagation in the opposite, anterior-to-posterior direction (1.6 ±.3 mm per s; n = 4 slices) was slightly, but not significantly (p =.45, Student s two-tailed t-test), smaller then the speed of propagation in the posterior-to-anterior direction. When the line-scan measurements were conducted in the vertical direction, across cortical layers (n = 8 slices, 16 locations in entorhinal, temporal and perirhinal cortices), the cenos were detected in all layers almost simultaneously, with only small deviations along the vertical axis toward the cortical surface. nature neuroscience volume 3 no 5 may 2 453

20 articles 2 Nature America Inc. 2 Nature America Inc. a 1 mm b e Number of events 2 1 * Interburst interval (min) * 8.5 F/F 2 min f Interburst interval (min) 12 c d 1 2 first hour third hour seventh hour.1 F/F 1 s.5 F/F 1 min Experiment number Fig. 2. Spontaneous oscillatory Ca 2+ waves in the immature cortex. (a) The box in the schematic drawing of a horizontal section of the brain (top) delimits the cortical region shown in the photomicrograph (bottom). Slice preparation from a two-day-old rat. Black dots indicate the position of nonoverlapping regions of interest (covering about 1 mm 2 ) from which the activitydependent fluorescence was recorded (see b). Scale bar, 1 mm. (b) Fluorometric Ca 2+ recordings from the sites indicated in (a). In this and in the following figures, fluorescence data are expressed as F/F (background-corrected decrease in fluorescence divided by the resting fluorescence). (c) Ca 2+ transients labeled with 1 and 2 shown at a higher temporal resolution. To monitor activity in the whole preparation, the recording was made in two steps using overlapping frames. Asterisks mark traces obtained during the second recording step. Note that Ca 2+ bursts marked with diamonds did not occur at all sites. (c) Ca 2+ transients (corresponding to events marked in b) on an expanded time scale. Note different response patterns in different cortical areas. (d) Ca 2+ oscillations in the entorhinal cortex of a one-day-old rat recorded continuously for seven hours. (e) Interval histogram (bin width, 18 s) of 122 intervals between the Ca 2+ waves recorded in the experiment in (d). Interburst intervals are measured between the starts of two consecutive Ca 2+ bursts. (f) Box plot of interburst intervals from ten experiments using one- to two-day-old rats. Data are expressed as median (line), interquartile (box), outliers (filled circles) and the farthest points that are not outliers (extended lines). Experiments illustrated in (a c) and (d, e) refer to slice numbers 4 and 6, respectively. We next examined the mechanisms underlying the initiation of cenos. In contrast to gap junction-mediated activity in cortical domains, the cenos were completely and reversibly blocked by tetrodotoxin (TTX, 25 or 5 nm; n = 7 slices; Fig. 5a and c). Furthermore, the frequency of the cenos was reduced by lowering the temperature from 32 C to 2 C (78 ± 19% of control, n = 5; p =.22, Student s two-tailed t-test; Fig. 5b and d), and not enhanced, as found for the activity in cortical domains 28. In addition, the number of Ca 2+ transients within the ceno-associated Ca 2+ burst decreased at 2 C to 59 ± 9% of the control value (Fig. 5e). The cenos were also completely and reversibly blocked by a mixture of ionotropic GABA- and glutamate receptor antagonists (1 µm CNQX + 5 µm D,L-APV + 2 µm bicuculline, n = 5) and thus they seem to require active chemical synapses. When the effects of these antagonists were studied separately, we found that CNQX completely and reversibly blocked cenos even at concentrations as low as 2 µm (n = 6, Fig. 6b and e), whereas even 2 µm bicuculline failed to block cenos (n = 9, Fig. 6d and e). The cenos were also completely and reversibly blocked by 5 µm APV (n = 5). The effects of blockers were independent of the age of the rat (one to five days) from which the slices were obtained. Both CNQX and APV blocked cenos in all cortical regions throughout the slice, suggesting that the activation of ionotropic glutamate receptors is essential for initiation of cenos. The experiments also demonstrated some distinctive functional properties of cenos (Fig. 6). Besides the much higher rates of repetition of hippocampal early oscillations as compared with those detected in cortex, antagonists of AMPA and of GABA A receptor channels had opposite, reciprocal effects in these two regions (Fig. 6b and d) in all seven slices tested. Even at P1, the cenos critically depended on activation of AMPA receptors and made no use of the depolarizing GABAergic transmission (Fig. 6b and d). An additional distinctive feature of cenos was their rapid, region-specific developmental changes in the frequency of oscillation during the first postnatal days. The network oscillations emerged after birth as a rather coherent activity in the majority of neurons of the entire cortex (92 ± 3% coherency, n = 347 waves, 11 slices, 8 rats; Fig. 2b). The median interval between individual waves was around four minutes in all cortical subregions (Figs. 2f and 7b). During the following days of development, the frequencies of activity recorded in the posterior (entorhinal cortex) 454 nature neuroscience volume 3 no 5 may 2

21 2 Nature America Inc. DISCUSSION Two-photon Ca 2+ imaging revealed a previously undescribed type of spontaneous activity in the developing cortex. The TTX sensitivity of these cenos and their dependence on activation of ionotropic glutamate receptors suggest that they require synaparticles 2 Nature America Inc. Fig. 3. Cortical early network oscillations involve the majority of cortical neurons. (a) Twophoton fluorescence image from the temporal cortex of a 3-day-old rat acquired 6 µm below the surface of the slice with a 1x objective. The region delimited by the shaded box is shown at a higher magnification in (b). Scale bar, 1 µm. (b) High-resolution fluorescence image of layer II/III showing individual cells and their neurites. The image represents a x y projection of the stack of images taken from µm below the surface of the slice (z-step,.5 µm). Ca 2+ signals from cell 1 and 2 are shown in (d). Arrowheads point toward the dendrites of these two cells. Scale bar, 2 µm. (c) Population Ca 2+ recordings obtained by averaging cenoassociated fluorescence transients obtained from 38 individual neurons that can be unambiguously distinguished in a c d Cell 1 Soma Dendr. Cell 2 Soma Dendr. Population recording (n = 38 cells) Single-cell recording (b). (d) Ca 2+ recordings from the somata and the dendrites of two cortical cells (corresponding to the sites marked in b). Note that ceno-associated Ca 2+ bursts detected in single cells are synchronous with those observed in the population recording (c). The diamonds indicate asynchronous Ca 2+ transients. (e) Ca 2+ transients recorded simultaneously in 11 cells. Data were reduced to a binary form by setting a threshold of four times the root mean square value of the resting fluorescence of recordings from individual cells. (f) Population Ca 2+ response (upper trace) and field potential changes (lower trace) recorded simultaneously in the same region of the temporal cortex of a five-day-old rat. (g) Field potential recordings in the temporal cortex of a four-day-old rat obtained in a non-stained slice. (h) Bar graphs illustrating the mean amplitudes (left) and durations (right) of ceno-associated changes in field potential measured in non-stained (n = 5, 34 events) and fura-2 AM-loaded (n = 4, 47 events) slices. b 1 µm 2 µm.5 F/F 1 min.1 F/F 1 min f g h Amplitude (µv) non-stained slices Duration (s) stained slices.1 F/F 1 s 1 V 1 s or in the more anterior cortical regions diverged rapidly (Fig. 7a). Whereas the rate remained almost constant until P4 and then steeply increased at P5 P6 in the entorhinal cortex, it continuously dropped in the temporal/perirhinal/insular cortex, reaching values of 1 wave per 2 5 minutes at P4 (Fig. 7a and b). Note that the majority (74 ± 13%, n = 38 waves, 7 slices) of these Ca 2+ waves were correlated with one of the more frequently occurring waves in the posterior cortex (Fig. 7a), and only 26% were more localized and restricted to the perirhinal/insular cortex. Furthermore, the Ca 2+ waves maintained their region-specific patterns at least until P5. Thus the oscillatory activity in the entorhinal cortex was characterized by bursts of Ca 2+ transients (on average, 2.5 ±.3 Ca 2+ transients per burst; n = 26 slices), whereas in all other areas it consisted of a mixture of single and paired Ca 2+ transients. In all areas studied, we occasionally observed large bursts comprising up to 4 6 Ca 2+ transients. For better quantitation, the baseline-subtracted area under the measured fluorescence trace was taken as an integral measure of the strength of cenos 9. In all cortical subregions, the area progressively decreased with development (Fig. 7c) and, finally, no cenos were detected in the cortex after P6. This decrease in the overall activity, however, showed region specificity. In the entorhinal cortex, the developmental increase in frequency was paralleled by a significant decrease in the amplitude, whereas the amplitude of the ceno-associated Ca 2+ transients in the anterior cortex remained constant but their frequency decreased considerably (Fig. 7a and b; note different scaling of the traces in a). No cenos were observed in the anterior cortex beyond P5, even during long periods (two to three hours) of continuous recording. A final set of results provided evidence that the developmental disappearance of cenos may be determined by the switch from GABAergic excitation to inhibition. Consistent with this assumption, we found that the mean amplitude of GABA-mediated Ca 2+ transients gradually decreased during postnatal development (Fig. 8). By using the ratio of GABA- against KCl-evoked Ca 2+ transients, we found that the amplitudes of the responding neurons reached a minimal value around P5 (Fig. 8e). From birth until P5/P6, during the time of the most intensive ceno activity, more than 9% of all examined cortical neurons produced GABA-mediated Ca 2+ transients (Fig. 8a, b and f). Around P7/P8, the time at which cenos are no longer detected, however, the proportion of the responding neurons dropped to 41% (Fig. 8f). Furthermore, bicuculline, which was ineffective in blocking the initiation of cenos (Fig. 6d and e) induced oscillatory cenolike Ca 2+ waves in slices from rats 7 or more days old (n = 4; Fig. 7d). Taken together, these findings clearly indicate that, whereas GABAergic excitation was unnecessary for initiation of cenos early after birth, overall GABAergic inhibition beyond P7/P8 was sufficiently strong to suppress cenos. e Cell number Time (min) nature neuroscience volume 3 no 5 may 2 455

22 articles 2 Nature America Inc. 2 Nature America Inc. Fig. 4. Transcortical propagation of cenoassociated Ca 2+ waves. (a) Top, fluorescence image of the temporal cortex of a one-dayold rat. Here and in (c), the dashed line indicates the site of line-scan recordings. Bottom, line-scan recording of a Ca 2+ wave. (b) Normalized fluorescence signals from three cortical regions as indicated in (a, bottom). The distance relative to the posterior region is indicated next to each trace. (c) Fluorescence image of cells of layer II/III in the posterior part of the temporal cortex (6 µm below the slice surface) from a 3-day-old rat. (d) Latency of the onset of a Ca 2+ wave in cells crossed by the scan line in panel (c) plotted against their distance to the leftmost cell in (c). The Ca 2+ wave propagated from the posterior to the anterior cortex. The solid line represents a linear least-square fit with a slope of 1.8 mm per s. (e) Same analysis as in (d) in an experiment in which Ca 2+ waves propagated in both directions, posterior-toanterior (filled symbols) and anterior-to-posterior (open symbols). The solid and dashed lines represent linear least-square fits to the respective sets of data with slopes of 1.7 mm per s and 1.3 mm per s. (f) Plot of the probability of Ca 2+ oscillations propagating in either direction (n = 18 slices). (g) Comparison of the speed of propagation of the Ca 2+ waves between single cells (short range, n = 8 slices) and large cortical regions (long range, n = 5 slices). tic transmission. This contrasts with other previously described forms of spontaneous correlated activity in the immature cortex 5,14. It remains to be investigated whether the action of glutamate during cenos is mediated solely through conventional synapses, or whether it also involves spillover of transmitter from conventional synapses to extrasynaptic receptors 29 and/or nonsynaptic release of transmitter, as is suggested for taurine 3. The importance of AMPA receptors for the initiation of cenos was unexpected in view of GABA s widespread depolarizing action in a b c Normalized frequency (%) Control complete block TTX d Normalized frequency (%) a Time b Distance 5 ms Posterior Posterior,. mm Intermediate, 2.1 mm Anterior, 4.2 mm Intermediate 5 nm TTX.5 F/F 2 C Control 32 C 2 C e Normalized number of transients/burst (%) Control 32 C 2 min.5 F/F 1 min 2 C.5 mm F/F.4.2 Anterior 1 s c d Probability (%) Distance (mm) f Latency (ms) Latency (ms) the immature cortex 12,22 and its importance in generation of hippocampal ENOs 9,31. However, the cortical early network oscillations, unlike the hippocampal ones, persisted in bicuculline. Even at P1, the cenos were extremely sensitive to low concentrations of CNQX (2 µm), which only partially blocked AMPA receptors 32. This suggests that in the developing cortex, synapses with functional AMPA receptors have a much more pronounced impact than in the hippocampus at the same age, where 8% of all glutamatergic synapses are silent 33. Thus, the maturation of excitatory glutamatergic transmission in the cortex seems to occur earlier than in the hippocampus, in line with findings that only 34% of all thalamocortical synapses are silent at P2 (ref. 34). A characteristic feature of cenos was their propagation along the longitudinal axis of the cortex. The average speed of propagation of ceno-associated Ca 2+ waves was 2 mm per second, different from the speeds of propagation found within cortical domains (1 µm per second; ref. 28), in the retina (2 4 µm per second; ref. 35) or in the hippocampal formation (8 mm per second; ref. 36). Although the usual cortical pacemaker seemed to be located in the entorhinal cortex, cenos were also sometimes found to emerge from anterior cortical regions. This suggests that more than one region is a potential pacemaker of cenos. Remarkably, the speed of propagation of cenos is most similar to the mean velocity found for the spread of non-synaptic epileptiform activity in the adult hippocampus (1.7 mm per Fig. 5. The cenos are preserved at room temperature and blocked by tetrodotoxin. (a) cenos recorded in the temporal cortex of a 1-dayold rat in control (left), in the presence of 5 nm TTX (middle) and after washout of TTX (right). (b) cenos recorded in the temporal cortex of a 2-day-old rat in control conditions (left), at 2ºC (middle) and after return to 32ºC (right). (c e) Plots of the relative (normalized to control) frequency of Ca 2+ oscillations (c, d) and of the relative number of Ca 2+ transients within the ceno-associated Ca 2+ burst (e) during the application of 25 or 5 nm TTX (c) and during the change in temperature (d, e). Each bar shows an average of seven experiments. 2 e Distance (mm) g Speed of propagation (mm/s) µm. p a a p Short range Long range (4 6 µm) (3 5 mm) p a a p 456 nature neuroscience volume 3 no 5 may 2

23 2 Nature America Inc. articles 2 Nature America Inc. Fig. 6. Striking differences between cortical and hippocampal early network oscillations. (a) Microphotograph of a horizontal brain slice from a 1-day-old rat taken at low magnification (4 objective). Regions of interest from which the recordings shown in (b) were obtained are indicated. (b) Simultaneous fluorometric recordings from cortical (upper panel) and hippocampal (lower panel) regions in a control (left), in the presence of 2 µm 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX, middle) and after washout of CNQX (right). (c, d) Experiment similar to the one shown in (a, b) illustrating the effect of 2 µm bicuculline. The slice was obtained from a one-day-old rat. Note that whereas the reduction in frequency of cortical ENOs was robust (see panel e), increased amplitude was observed only in five of nine slices. (e) Plot of the mean relative (normalized to control) frequency of Ca 2+ oscillations in the cortex (filled bars) and in the hippocampus (open bars) in the presence of 2 µm CNQX (n = 6), 2 µm bicuculline (n = 9) and 2 µm bicuculline (n = 6). second), which involves the sequential recruitment of inactive neurons through a wave of extracellular K + released by active e neurons 37. What might be the function of cenos? In the immature central nervous system, transient elevations of the intracellular Ca 2+ concentration are important in many aspects of growth and differentiation. These include proliferation 38, gene expression 39,4, neuronal migration 41, motility of axonal and dendritic growth cones 42, clustering of the postsynaptic receptor-channels 43 and activity-dependent maturation of glutamatergic synapses 33. Many of these Ca 2+ -dependent processes require repeated, cyclic changes of the intracellular Ca 2+ level: in the developing spinal cord 44, for example, the speed of growth cone migration is controlled by the frequency of spontaneously occurring Ca 2+ transients. Therefore, growth cones experiencing a high frequency of Ca 2+ transients (1 12 per h) migrate slowly or retract, whereas growth cones experiencing a low frequency of Ca 2+ transients (.4 per h) migrate rapidly. As the cenos are comparable in frequency and involve large populations of neurons, they represent a prime candidate to promote neuronal development and to synchronize development over large cortical regions. In conclusion, our results indicate that the spontaneous patterned activity in the developing cortex is far more complex than has been previously anticipated. It seems to consist of large-scale cenos acting in concert with more localized, mechanistically distinct oscillations 5,14. Although all types of synchronous activity may sculpt neuronal circuits by reinforcing connections among coactive cells 3, their specific assignments may differ. Because of the radial organization and distinct borders of the local domains, correlated activity within these structures would promote the columnar organization of the cortex 5, whereas cenos would be particularly important for the establishment of long-range neuronal connections. Thus, during the earliest postnatal stages, the intrinsic cenos might serve as a pacemaker that a c b d Normalized frequency (%) Cortex Hippocampus Cortex Hippocampus µm CNQX 2 µm bicuculline complete block in hippocampus 2 µm CNQX 2 µm bicuculline 2 µm bicuculline Cortex Hippocampus.5 F/F.4 F/F.5 F/F determines a coherent and coordinated development of the weakly interconnected cortical subregions. METHODS All experimental procedures were approved by local government authorities. Horizontal cortical slices (4 5 µm) were prepared from 116 Wistar rats aged from P (the day of birth) to P12, as previously described 45,46. The extracellular solution was continuously bubbled with 95% O 2 and 5% CO 2 at 32 C and contained 125 mm NaCl, 4.5 mm KCl, 2 mm CaCl 2, 1 mm MgCl 2, 1.25 mm NaH 2 PO 4, 26 mm NaHCO 3 and 2 mm glucose at ph 7.4. For fluorometric Ca 2+ measurements, slices were loaded with the membrane-permeable form of fura-2 (fura-2 AM, Molecular Probes, Eugene, Oregon or fura-pe3 AM, TefLabs, Austin, Texas) using standard procedures 9. Fluorometric recordings were made using a custom-built two-photon laser-scanning microscope 23 based on a Ti:sapphire laser system ( Tsunami and Millennia, both from Spectra-Physics, Mountain View, California) that provided mode-locked laser light (pulse width <1 fs; repetition rate, 8 MHz, center wavelength, 79 nm) and a laser-scanning system (MRC 124, Bio-Rad, Herts, UK) attached to an upright microscope (BX 5 WI, Olympus, Tokyo, Japan). Either water-immersion objectives (1,.3 NA; 4,.8 NA; 6,.9 NA; all from Olympus) or dry objectives (1.9,.14 NA, Olympus; 5,.25 NA, Zeiss, Oberkochen, Germany) were used. For recordings with dry objectives, the recording chamber was sealed with a.13 mm-thick cover glass (Fisher Scientific, Pittsburgh, Pennsylvania) to prevent image distortion due to fluctuations in the bath solution level. Image analyses were 2 min 2 min nature neuroscience volume 3 no 5 may 2 457

24 articles 2 Nature America Inc. 2 Nature America Inc. a b Posterior cortex, P3 Anterior cortex, P3 Interburst interval (min) F/F.1 F/F Postnatal day 1 min Anterior Posterior performed off-line with routines written in Labview and Igor Pro software (National Instruments, Austin, Texas and Wavemetrics, Eugene, Oregon, respectively). Extracellular field potentials were low-pass filtered at 1 khz, recorded with a custom-made DC amplifier and digitized at 5 khz with an A/D converter (ITC 16, Instrutech, Port Washington, New York) controlled by Pulse software (HEKA, Lambrecht, Germany). The recording microelectrodes (resistances of MΩ) were filled with 3 M NaCl and positioned in layer II/III of the cortex. GABA- and K + -containing solutions were pressure-applied from fine pipets (resistances of 6 12 MΩ a c e GABA/ KCl µm 2 µm Postnatal day no cenos no cenos no cenos no cenos b d f Cell1 Cell2 Cell3 Cell4 Cell1 Cell2 Cell3 Cell4 Number of responding cells (%) c d Normalized area Anterior cortex, P7 - KCl application - GABA application Postnatal 1 µm bicuculline no cenos no cenos Anterior Posterior no cenos no cenos.5 F/F 5 min Fig. 7. Developmental confinement of early network oscillations to the first few days after birth. (a) Twohour-long continuous fluorometric recordings from the entorhinal (posterior) and the perirhinal/insular (anterior) cortices in a slice from a three-day-old rat. (b) Averaged data showing the interburst intervals in the posterior and the anterior cortical regions plotted against postnatal age. Data are expressed as the mean value of the median interburst interval from 4 12 experiments per data point. (c) Averaged data showing the mean baseline-subtracted areas under the fluorescence traces (see Methods) as a function of the postnatal age. The area measurements were normalized with respect to the mean value obtained at P1. (d) Fluorometric recordings from the anterior cortex at P7 in control (left), in the presence of 1 µm bicuculline (middle) and after washout of bicuculline (right). when filled with extracellular solution) using a Picospritzer II (General Valve, Fairfield, New Jersey). To obtain saline containing 8 mm K +, 8 mm NaCl was substituted for KCl in the standard extracellular solution. The tips of both drug application pipets were placed near ( 5 µm) the target neuron. Brief (5 ms) applications of the drugs usually evoked a Ca 2+ -response in this neuron and in adjacent secondary neurons within a radius of 5 µm. We used only primary target neurons for the amplitude analysis, whereas we also included the immediately adjacent secondary neurons in the calculation of the total number of responding cells. We determined the baseline-subtracted areas under fluorescence.1 F/F 3 s.1 F/F 3 s Postnatal day Fig. 8. GABA-activated Ca 2+ transients in the immature cortex. (a) Left, high-resolution fluorescence image of cells in layer II/III of the temporal cortex of a 1-day-old rat taken with a 6 objective. Right, the same image transformed into black and white mode by using a threshold function. Here and in (c), cells to which GABA was pressure-applied are color coded. Cells that responded to GABA application with a Ca 2+ transient are shown in red. (b) Fluoromeric recordings from neurons marked with corresponding numbers in (a). Ca 2+ transients were activated either by 5 ms applications of 8 mm K + (open triangles) or by 5 ms applications of 1 µm GABA (filled triangles). (c, d) Analyses similar to those shown in (a, b) in the temporal cortex of an eight-day-old rat. Cells that did not show Ca 2+ elevations in response to GABA application (for example, cells 1 and 2 in d) are shown in blue. (e) The ratio of the amplitudes of GABA- versus K + -evoked Ca 2+ transients plotted against postnatal age. Analyses were conducted on 1 15 cells per age. Only the primary target cells (see Methods) were chosen for these analyses. (f) Percentage of cells showing GABAactivated Ca 2+ transients plotted against postnatal age. The plot summarizes experiments conducted in cells per data point. 458 nature neuroscience volume 3 no 5 may 2