Transcriptional Regulation of the GluR2 Gene: Neural-Specific Expression, Multiple Promoters, and Regulatory Elements

To understand how neurons control the expression of the AMPA receptor subunit GluR2, we cloned the 5' proximal region of the rat gene and investigated GluR2 promoter activity by transient transfection. RNase protection and primer extension of rat brain mRNA revealed multiple transcription initiation sites from -340 to -481 bases upstream of the GluR2 AUG codon. The relative use of 5' start sites was different in cortex and cerebellum, indicating complexity of GluR2 transcript expression among different sets of neurons. When GluR2 promoter activity was investigated by plasmid transfection into cultured cortical neurons, cortical glia, and C6 glioma cells, the promoter construct with the strongest activity, per transfected cell, was 29.4-fold (+/- 3.7) more active in neurons than in non-neural cells. Immunostaining of cortical cultures showed that >97% of the luciferase-positive cells also expressed the neuronal marker MAP-2. Evaluation of internal deletion and substitution mutations identified a functional repressor element I RE1-like silencer and functional Sp1 and nuclear respiratory factor-1 (NRF-1) elements within a GC-rich proximal GluR2 promoter region. The GluR2 silencer reduced promoter activity in glia and non-neuronal cell lines by two- to threefold, was without effect in cortical neurons, and could bind the RE1-silencing transcription factor (REST) because cotransfection of REST into neurons reduced GluR2 promoter activity in a silencer-dependent manner. Substitution of the GluR2 silencer by the homologous NaII RE1 silencer further reduced GluR2 promoter activity in non-neuronal cells by 30-47%. Maximal positive GluR2 promoter activity required both Sp1 and NRF-1 cis elements and an interelement nucleotide bridge sequence. These results indicate that GluR2 transcription initiates from multiple sites, is highly neuronal selective, and is regulated by three regulatory elements in the 5' proximal promoter region.

To understand how neurons control the expression of the AMPA receptor subunit GluR2, we cloned the 5Ј proximal region of the rat gene and investigated GluR2 promoter activity by transient transfection. RNase protection and primer extension of rat brain mRNA revealed multiple transcription initiation sites from Ϫ340 to Ϫ481 bases upstream of the GluR2 AUG codon. The relative use of 5Ј start sites was different in cortex and cerebellum, indicating complexity of GluR2 transcript expression among different sets of neurons. When GluR2 promoter activity was investigated by plasmid transfection into cultured cortical neurons, cortical glia, and C6 glioma cells, the promoter construct with the strongest activity, per transfected cell, was 29.4-fold (Ϯ 3.7) more active in neurons than in non-neural cells. Immunostaining of cortical cultures showed that Ͼ97% of the luciferase-positive cells also expressed the neuronal marker MAP-2. Evaluation of internal deletion and substitution mutations identified a functional repressor element I RE1-like silencer and functional Sp1 and nuclear respiratory factor-1 (NRF-1) elements within a GC-rich proximal GluR2 promoter region. The GluR2 silencer reduced promoter activity in glia and non-neuronal cell lines by two-to threefold, was without effect in cortical neurons, and could bind the RE1-silencing transcription factor (REST) because cotransfection of REST into neurons reduced GluR2 promoter activity in a silencer-dependent manner. Substitution of the GluR2 silencer by the homologous NaII RE1 silencer further reduced GluR2 promoter activity in nonneuronal cells by 30-47%. Maximal positive GluR2 promoter activity required both Sp1 and NRF-1 cis elements and an interelement nucleotide bridge sequence. These results indicate that GluR2 transcription initiates from multiple sites, is highly neuronal selective, and is regulated by three regulatory elements in the 5Ј proximal promoter region.
Key words: AMPA; glutamate receptor; transcription; REST; primary culture;transfection;luciferase;neurons;promoter;Sp1;silencer;neuronal expression;repressor The AM PA subtypes of glutamate receptors are assembled from combinations of GluR1, 2, 3 and 4 (GluR A, B, C, D) subunits and mediate a large fraction of the postsynaptic response at most fast excitatory synapses in the brain. Studies with recombinant receptors indicate that different subunit combinations produce functionally unique AM PA receptors (Boulter et al., 1990;Hollmann et al., 1991;L omeli et al., 1994). In particular, three AM PA receptor properties are strongly influenced by the level of edited GluR2 expression because of the positively charged arginine present in the Q / R site of the M2 channel-lining domain. The presence of edited GluR2 reduces C a 2ϩ permeability (Hollmann et al., 1991;Geiger et al., 1995), voltage-dependent channel block by cytosolic polyamines (Bowie and Mayer, 1995;Donevan and Rogawski, 1995;Kamboj et al., 1995;Koh et al., 1995), and single channel conductance (Swanson et al., 1997). Thus, EPSPs mediated by GluR2-deficient receptors may be larger than those mediated by GluR2-rich receptors, which should have significant physiological consequences.
The relative expression of AMPA receptor subunit mRNA and protein in certain populations of neurons is not static but may be remodeled during development (Pellegrini-Giampietro et al., 1992a), after seizures or ischemic insult (Pellegrini-Giampietro et al., 1992bPollard et al., 1993;Freidman et al., 1994;Kamphuis et al., 1994;Prince et al., 1995) and after administration of antipsychotics , drugs of abuse (Ortiz et al., 1995;Fitzgerald et al., 1996), or corticosteroids (Nair et al., 1998). After ischemia, a drop in relative GluR2 expression in hippocampal CA1 pyramidal cells is associated with changes in AMPA receptor properties consistent with an increase in GluR2deficient receptors (Gorter et al., 1997), although whether GluR2 selectively declines remains controversial (see Frank et al., 1995). Recent work from our laboratory indicates that the number of GluR2 subunits in a receptor is variable and that the relative abundance of GluR2 affects Ca 2ϩ permeability and rectification differentially (Washburn et al., 1997). Given that distinct populations of neurons are known to be GluR2-deficient or GluR2rich (Bochet et al., 1994;Jonas et al., 1994;Geiger et al., 1995;Washburn et al., 1997), the consequences of modulating GluR2 expression in these cells will depend on pre-existent GluR2 levels.
These studies indicate that mechanisms controlling the level of GluR2 expression may be important regulatory determinants of AMPA receptor phenotype. GluR2 expression in brain is pri-marily restricted to neurons (Monyer et al., 1991;Petralia and Wenthold, 1992;Sato et al., 1993), although AM PA receptor subunits are expressed by certain populations of glia Burnashev et al., 1992) and O-2A progenitor cells (Patneau et al., 1994). Functional promoter studies of NMDA and kainate receptor genes have been reported (Bai and Kusiak, 1995;Sasner and Buonanno, 1996;Huang and Gallo, 1997;Suchanek et al., 1997), and the organization of the mouse GluR2 gene has been described (Köhler et al., 1994), but no regulatory elements of AM PA receptor subunits have yet been identified. We show here that the GluR2 proximal promoter region contains a negative regulatory element and a positive regulatory region and directs the neural-specific expression of a luciferase reporter gene in transiently transfected primary cortical cultures.
Parts of this paper have been published previously in abstract form (Peters et al., 1995;Myers et al., 1996).
Pfu polymerase and the DASH II genomic library were from Stratagene (La Jolla, CA). Lipofectamine, Superscript II reverse transcriptase, and RNAsin were obtained from Life Technologies. MMLV reverse transcriptase was from Stratagene. Restriction enzymes were purchased from New England Biolabs (Beverly, M A) or Life Technologies. T4 DNA ligase, T4 DNA and K lenow polymerases, RNaseON E, luciferase vectors, and assay reagents were purchased from Promega. pcDNA3 and TA cloning vectors were from Invitrogen (San Diego, CA), and Sequenase 2.0 was purchased from United States Biochemicals (C leveland, OH). Geneclean kits were obtained from BIO 101 (La Jolla, CA). All oligonucleotides were synthesized in house on a MilliGen / Biosearch C yclone Plus DNA Synthesizer using E xpedite Chemistry (PerSeptive Biosystems). Radioisotopes were purchased from Amersham (Arlington Heights, IL). Nylon-backed nitrocellulose membranes were purchased from Bio-Rad (Hercules, CA). Other general salts and supplies were purchased from Sigma or Fisher Scientific (Houston, TX).
Isolation and characterization of genomic clones. Approximately 0.7 ϫ 10 6 plaques from a rat genomic library in DASH II were hybridized on nylon membranes with a 3505 bp EcoRI-XhoI restriction fragment containing the f ull coding sequence of the rat GluR2 cDNA, labeled with [␣-32 P]dC TP by random priming. Three positive plaques were purified and subjected to a second round of hybridization with an end-labeled oligonucleotide (R2-1; GAGGACAGAAATATGCATAATC TTTTG-CAT) complementary to the first 10 codons of the GluR2 coding sequence (Boulter et al., 1990). A single recombinant phage obtained from this screen had an 18 kb insert that included a long, interspersed repeat segment (LI N E-1) at its immediate 5Ј end, ϳ9 kb upstream of the 5Ј end of the GluR2 gene. A 1.64 kb XbaI-HindIII fragment, which contained the GluR2 proximal promoter region, exon 1, and 288 bp of the first intron, has been deposited in the GenBank database (accession number AF025917).
5Ј R ACE, R Nase protection, and primer e xtension and Southern blot. For 5Ј R ACE, poly(A ϩ ) mRNA isolated from adult male Sprague Dawley rat brains by the method of Chirgwin et al. (1979) was reverse transcribed by MMLV reverse transcriptase from the R2-1 primer. After addition of an oligo-dC tail with terminal deoxynucleotidyltransferase, PCR reactions were performed using the anchor primer provided by the Life Technologies 5Ј R ACE system, plus the R2-race1 primer (AGTC -CCGAGGAC TGGC TGAA) complementary to nucleotides 81-100 of the GluR2 cDNA sequence (Boulter et al., 1990). PCR products were subcloned into the TA cloning vector and selected by colony hybridization to the R2-20 oligonucleotide complementary to bp 64 -83 of the GluR2 cDNA (GAATGCAGTTTTTAGC TGAG). Ten clones were selected by size on agarose gels and sequenced through the anchor-priming site to locate 5Ј ends of the mRNA.
For primer extension, an oligonucleotide (PER2-3; TCCCAGTTG-TAGC TGGTGGC TGTTGATGCC) complementary to bp Ϫ163 to Ϫ192 from the coding AUG (see Fig. 1 A) was end-labeled with [␥-32 P]ATP (ϳ4.8 ϫ 10 5 cpm /pmol) and annealed to 30 g of adult rat cortical and cerebellar total RNA in 20 mM H EPES, pH 7.0, 350 mM NaC l, and 50% formamide at 50°C for 24 hr. Annealed samples were precipitated and resuspended in 50 l containing 50 mM Tris-HC l, pH 8.3, 75 mM KC l, 3 mM MgC l 2 , 2.5 mM dN TPs, 10 mM DTT, 50 g /ml actinomycin D, 20 units RNAsin, and 200 units Superscript II reverse transcriptase and were incubated at 42°C for 90 min; then 1 l of RNAsON E was added, and the incubation continued for an additional 35 min. Samples were recovered and loaded onto a sequencing gel. Yeast total RNA (48 g) and an in vitro synthesized f ull-length GluR2 sense RNA (10 ng) were used as templates in negative and positive control reactions, respectively. The in vitro GluR2 RNA includes 481 bases of GluR2 5Ј-untranslated region (UTR) sequence plus an additional 40 bp of polylinker sequence. To confirm the specificity of the primer extension reactions, we also prepared Southern blots of primer extension products. For these experiments the reaction conditions were identical to those described above except that the oligonucleotide primer was unlabeled and, after gel resolution, reaction products were electrotransferred to a nylon membrane. Southern blots were then hybridized with a 32 Pradiolabeled DNA probe (Ͼ10 8 cpm /g) made from bp Ϫ192 through Ϫ481 relative to the GluR2 AUG (see Fig. 2 A, probe B). Hybridized blots were washed under high-stringent conditions and exposed to a Molecular Dynamics PhosphorImager plate (Sunny vale, CA) for visualization.
RNase protection experiments were conducted using an Ambion RPA kit (Austin, TX) following the manufacturer's instructions. For identification of mRNA 5Ј ends, an antisense riboprobe complementary to nucleotides Ϫ192 through Ϫ481 relative to the GluR2 AUG (see Fig. 2 A, probe C) was radiolabeled and annealed (5 ϫ 10 5 cpm per reaction) to 30 g of adult rat cortical and cerebellar total RNA in 20 l of hybridization solution (Ambion kit) at 45°C overnight. Annealed samples were digested with RNase A and RNaseT1, resolved on an RNA-sequencing gel, and exposed to Kodak X-OM AT film for 12 hr. To quantif y the relative use of transcription initiation sites in cerebellum and cortex, an antisense RNA probe complementary to bases Ϫ136 through Ϫ361 relative to the GluR2 AUG (see Fig. 2 A, probe D) was annealed to total RNA as described above. This probe identifies the major short transcription initiation site at Ϫ340 and also summates all transcripts with 5Ј ends extending beyond Ϫ360 into a single band on the gel. Total RNA used in these experiments was prepared independently from four adult male Sprague Dawley rats.
Methylation sites in the promoter reg ion. Adult rat genomic DNA was isolated from lung, kidney, cortex, and cerebellum (40 -50 g) and digested with AccI to release a 1.3 kb promoter fragment (Ϫ731 to ϩ603 bp relative to the ϩ1 transcription initiation site) and then was digested again with one of the methylation-sensitive restriction enzymes BssHII, SacII, AvaI, or Eco47III. Southern blots of the genomic fragments were prepared and hybridized with an AccI-restricted 1.3 kb genomic fragment radiolabeled with [␣-32 P]dC TP by random priming (Ͼ10 8 cpm /g). All hybridizations and washes were conducted under high-stringent conditions.
Cell culture. Primary rat cortical neuronal cultures were prepared from embryonic day 18 fetal rat pups resected from pregnant Sprague Dawley rats. Briefly, cerebral cortices were minced in DM EM supplemented with 1.1 gm / l glucose, 2.1 gm / l NaHC O 3 , and 1.84 mM L-glutamine, pH 7.25 (F DM EM); 1 ml of fetal bovine serum (FBS) was added; and tissue was dissociated by trituration. Dissociated cells were rinsed in 8 ml of F DM EM media, pelleted at low speed for 5 min, and then resuspended in F DM EM supplemented with 100 g /ml human apotransferrin and (in M): insulin 0.5, putrescine 60, tri-iodothyronine 1 ϫ 10 Ϫ3 , ␤-estradiol 1 ϫ 10 Ϫ6 , progesterone 0.01, and sodium selenite 0.03 (termed TF DM EM). Cortical cells were plated at a density of 0.7-1.0 ϫ 10 6 cells/ well into Falcon 12-well culture dishes precoated overnight with 180 g /ml poly-D-lysine and coated with 20% FBS in F DM EM for 2-3 hr before plating. Cultures were grown for 3-6 d in vitro, without media change, in an humidified, 5% C O 2 , 37°C incubator. It should be noted that these cultures are a mixture of neurons and glia with ϳ60% of cells M AP-2 positive on days 4 -6 in culture.
Primary glial cultures were generated from the primary cortical neuronal cultures by repeated washing of cells with F DM EM plus 10% FBS over a period of 1-4 weeks after plating or by passage of cultures to facilitate the elimination of neurons. Cortical glia and rat C6 glioma cells were grown in F DM EM plus 10% FBS until use in transfections. HeLa cells were grown in DM EM supplemented with 10% FBS and 2 mM glutamine. The non-neuronal cultures were grown in the same incubator as the neurons.
Constructs and mutagenesis. To generate 5Ј-and 3Ј-GluR2 promoter deletion constructs, we amplified specific-length fragments of the GluR2 promoter by PCR with Pfu polymerase using oligonucleotides complementary to the GluR2 1.64 kb XbaI-HindIII GluR2 fragment. 5Ј-and 3Ј-oligonucleotides contained NheI and BglII restriction sites, respectively. Amplified DNA fragments were restricted and cloned into the NheI and BglII sites of the pGL2 basic vector upstream of the reporter gene firefly luciferase. Recombinants were analyzed by restriction and sequence analysis to confirm the location of the GluR2 promoter boundaries. Throughout this report, construct boundaries [i.e., R2(Ϫ822/ ϩ320)luc] are defined relative to the 5Ј-most major initiation site identified, Ϫ429 bases upstream from the recognized GluR2 AUG.
To generate internal GluR2 silencer mutations in the context of the R2(Ϫ302/ϩ320)luc construct, we used a two-round PCR mutagenesis protocol as described (Cormack, 1997). After the second-round PCR reaction, the mutagenized DNA fragment was restricted with NheI and BglII and cloned back into the pGL2 basic vector. To delete the GluR2 silencer, we replaced 25 bp spanning the silencer element with an NsiI restriction site. This construct is designated R2(Ϫ302/ϩ320)luc-⌬B. Substitution mutations were introduced by replacing different 6 bp segments of the silencer with either an EcoRI or an EcoRV restriction site and were designated R2(␦RI)luc and R2(␦RV)luc, respectively. Constructs that replaced the GluR2 silencer with the T ype II Na channel repressor element 1 (NaII-RE1) silencer in the context of the GluR2 promoter [R2(␦NaII)luc], or likewise replaced the NaII-RE1 element with the GluR2 silencer in the context of the NaII promoter [NaII(␦R2)luc], were also made using the two-round, oligonucleotide-directed PCR mutagenesis protocol. The NaII promoter construct without an RE1 element (pM B4) (Kraner et al., 1992) was excised with EcoRI, blunted with K lenow, cloned into the SmaI-digested pGL2 basic vector, and renamed NaII(⌬)luc. The NaII promoter construct retaining the RE1 silencer on a 118 bp fragment (Ϫ1051 to Ϫ933) f used 5Ј of the NaII minimal promoter region (Ϫ134 to ϩ177) as described (pSDK7) (Kraner et al., 1992) and was cloned upstream of firefly luciferase by shuttling it through the pBluescript HindIII-PstI polylinker sites and into the pGL2 basic vector KpnI and NheI sites. This construct was renamed NaIIluc. All constructs described were confirmed by restriction and sequence analysis and preserved the correct flanking sequences and silencer orientations of their natural promoter contexts; e.g., R2(␦NaII)luc contains type II sodium channel RE1 silencer with GluR2 flanking sequence and the GluR2 antisense orientation. The pairs of oligonucleotides used to introduce these mutations are the following: to delete the silencer in GluR2, R2(Ϫ302/ϩ320)luc-⌬B, sense TCCGAC TATGCATCGGAT-GCGCAACAC and antisense GCATCCGATGCATAGTCGGAGC -TC TGG; the missense mutation R2(␦RI)luc, sense AAAGCGC TGT-GAATTC TGC TAAAATCGGATGCG and antisense CCGATTT-TAGCAGAATTCACAGCGC TTTAGTCGG; the missense mutation R2(␦RV)luc, sense C TAAAGGATATCCC TCGGTGC TAAAATCGG and antisense CCGAGGGATATCC TTTAGTCGGAGC TC; to place the NaII RE1 silencer in the GluR2 promoter, R2(␦NaII)luc, sense TGC TGTCCGTGGTTC TGAAATCGGAT and antisense AGAAC -CACGGACAGCAC TTTAGTCG; to place the GluR2 silencer in the NaII promoter, NaII(␦R2)luc, sense GCACCGAGGACAGCGC TA-GAGTC TC TGAAT and antisense GC TGTCC TCGGTGC TAAAA-CCCAATTACAG.
Transfections. Primary neurons were transfected on days 3-6 after plating. Primary glia were transfected either as confluent, unpassaged cells or at 90% confluency 2-3 d after passage, and C6 glioma cells were transfected similarly at 80 -100% confluency. All transfections were conducted in Falcon 12-well dishes with Lipofectamine according to the manufacturer's instructions. Per well, 1 g of DNA, 50 l of 150 mM NaC l, and 5 l of Lipofectamine were combined in 0.5 ml of TF DM EM media and incubated with cells at 37°C for 5 hr. After transfection, the DNA and Lipofectamine mixture was replaced with fresh TF DM EM (0.5 ml) for primary neuronal cultures or with normal culture media for primary glia, C6 glioma, and HeLa cells. T wenty-four hours later, cells were rinsed with PBS and harvested by lysis. C ell lysates were cleared by centrif ugation and assayed for luciferase activity in a T urner TD-20e luminometer. A preliminary time course study with the GluR2, cytomegalovirus (C M V), and SV40 promoters in transfected cortical cultures indicated that luciferase expression was maximal 24 hr after transfection and decreased slowly by 48 and 72 hr.
For cotransfections, a similar protocol was used except the amount of Lipofectamine was increased to 7.5 l per well to accommodate the additional DNA. The amount of the GluR2 promoter-firefly reporter construct was varied from 0.1 to 1.0 g / well, and the amount of the RE1-repressor factor plasmid (RESTexpress) (Chong et al., 1995) or the control plasmid pcDNA3 was held fixed at 0.25 or 0.75 g / well, and pBS vector was added to adjust the total DNA for all transfections to a final concentration of 1.75 g / well. Under these cotransfection conditions, the amount of luciferase activity recovered in the cell lysates varied linearly with the amount of pGL2 reporter plasmid DNA transfected (data not shown). For nuclear respiratory factor-1 (N RF-1) cotransfections, 1 g of pN RF-1 plasmid (Virbasius et al., 1993) was mixed with 0.5 g of the GluR2 promoter-reporter construct .
All plasmid DNA preparations were prepared from 100 ml of 2xYT cultures using Qiagen Maxi-DNA Prep columns. At least three independent DNA preparations for each construct were transfected, in parallel, in each experiment. Each individual DNA plasmid was transfected into triplicate wells. For transfections into multiple cell types, a master mix of the plasmid DNA, NaC l, Lipofectamine, and TF DM EM media was prepared from which equal aliquots were delivered to the cells. In all experiments, an SV40-luciferase or a C M V-luciferase control plasmid was transfected into parallel wells to normalize GluR2 promoter activity across multiple experiments and primary culture preparations. Where indicated, luciferase activity was converted to moles of luciferase (ϳ2.2 T urner light units/10 Ϫ18 mole) using a standard curve generated with recombinant firefly luciferase under our assay conditions.
Immunofluorescence. Primary neurons prepared as described above were plated onto glass coverslips precoated with both poly-D-lysine (overnight) and then fibronectin (20 g /ml; 1-2 hr). Cultures were transfected with the R2(Ϫ302/ϩ320)luc or the R2(Ϫ134/ϩ147)luc construct and fixed 24 hr later with 4% paraformaldehyde in PBS for 15 min. C ells were rinsed three times with PBS, permeabilized with 0.1% w/ v Triton X-100 in PBS for 5 min, washed, and then incubated in staining buffer (2% horse serum in PBS) for 15 min. C ells were incubated for 2 hr in staining buffer with rabbit anti-luciferase (1:200) and either mouse anti-GFAP (1:800) or mouse anti-M AP-2 (1:600) and then rinsed four times for 5 min each with excess staining buffer before incubation with the secondary antibodies, Texas Red-conjugated donkey anti-mouse IgG (1:250) and FI TC -conjugated donkey anti-rabbit IgG (1:250), in staining buffer for 30 min. C ells were rinsed four times for 5 min each as described above and mounted onto glass slides. All washes and staining incubations were conducted at room temperature. Fluorescence was visualized with a Z eiss Axioplan microscope using a 40ϫ Plan-N EOFLUAR oil objective and appropriate fluorescent filters. C ells positively identified for firefly luciferase expression were counted and scored for positivity to either M AP-2 or GFAP expression.
Nuclear e xtract and gel mobilit y shif t assays. Nuclear extracts were prepared from cultured cells according to the method of Dignam et al. (1983). For gel mobility shift assays, complementary oligonucleotides were purified by denaturing PAGE and then annealed and 32 P-labeled by K lenow fill-in of the 5Ј overhangs for 25 min at room temperature. Labeled probes were extracted, purified through a spin column, precipitated, and stored in deionized water at 4°C until use.
Individual components of the binding reaction were combined and incubated on ice for 15-30 min before the addition of probe (ϳ0.3 nM, final concentration; 10 -30 ϫ 10 3 cpm) and then were incubated at 22°C for 15 min. Where indicated, a 30 -200-fold excess of cold competitor oligonucleotides was added on ice 15-30 min before addition of probe. After incubation, samples were separated on a 4% polyacrylamide, Trisglycine, glycerol gel. Dried gels were exposed to Kodak X-OM AT film or to a Molecular Dynamics PhosphorImager plate for visualization.
Statistical anal ysis. For comparison of the promoter activities between two or more constructs within a given cell type, at least three plasmid preparations for each construct were transfected in the same experiment. The order in which individual plasmid DNA preparations were added to the wells for transfection was randomized to reduce bias attributable to handling of the cells during the procedure. Raw luciferase activity was normalized to the activity of either the SV40 promoter or a defined control construct measured in parallel wells. Data from multiple experiments (culture preparations) were combined and analyzed by ANOVA followed by post hoc Dunnett's tests for significance between means or, when appropriate, Student's t tests. Comparisons of the promoter activities between constructs were made within, and not across, cell types.

Multiple tissue-specific transcription initiation sites and DNA methylation
Exon 1 and the proximal 5Ј-flanking sequence of the rat GluR2 gene are shown in Figure 1 A, along with a schematic identifying the salient features. To identif y transcription initiation sites, we first conducted 5Ј R ACE analysis of rat brain mRNA. Multiple potential initiation sites were identified from Ϫ286 to Ϫ481 bases upstream from the GluR2 translation AUG codon. Most RACE 5Ј ends mapped to one of two clusters, from Ϫ286 to Ϫ295 or from Ϫ419 to Ϫ427 bp, with one end residing at Ϫ481 bp ( Fig.  1 A, open triangles). The sequence of each R ACE PCR product matched that of the genomic DNA up through the anchorpriming site, reducing the likelihood that an intron resides in the GluR2 5Ј-UTR, in accordance with previous observations for the mouse GluR2 gene (Köhler et al., 1994).
To f urther identif y potential sites of transcription initiation in the GluR2 gene, we used both primer extension and RNase protection analysis of adult rat brain RNA (Fig. 2). Primer extension analysis on cortical RNA using an oligonucleotide, PER2-3 complementary to the GluR2 5Ј-UTR, revealed multiple reaction products extending from Ϫ312 to Ϫ452 nucleotides upstream from the GluR2 AUG ( Fig. 2 B). None of these cortical RNA primer extension products appeared in parallel reactions conducted with an in vitro synthesized genomic GluR2 RNA control (Ctl ) that includes putative 5Ј-UTR sequence from Ϫ1 through Ϫ481 bases; this positive control also includes an additional 40 bp of non-GluR2 polylinker sequence ( Fig. 2 B , left lane). This result indicates that the multiple extension products observed in cortical RNA reactions are not the result of RNA secondary structure preventing the processivity of reverse transcriptase. To determine which PER2-3 primed extension products from cortical RNA were derived from GluR2, a parallel primer extension reaction was run using unlabeled PER2-3 oligonucleotide as primer. Products from this reaction were then Southern blotted onto a nylon membrane and hybridized with probe B, DNA complementary to bases Ϫ192 through Ϫ481 5Ј of the GluR2 AUG (Fig. 2 A). The results of this Southern blot, depicted in the right lane of Figure 2 B, confirm the identity of nine GluR2 primer extension products. A single reaction product at Ϫ312 was not confirmed by the Southern blot and is presumably an artifact caused by extension from a non-GluR2 transcript.
We also conducted RNase protection assays on rat cortical and cerebellar RNA to confirm the primer extension findings. Using an antisense RNA probe (Ϫ192 through Ϫ481 bp of the GluR2 leader, probe C in Fig. 2 A), we again detected multiple bands predicting different GluR2 5Ј ends in the cortex and cerebellum, and these bands were in agreement with results obtained by primer extension (Fig. 2C). Major RNase-protected 5Ј ends mapped to Ϫ340, Ϫ368, Ϫ399, and Ϫ429 bp 5Ј of the GluR2 AUG codon, with additional minor ends mapping to Ϫ452 and Ϫ481 bp. Because the riboprobe in this experiment included GluR2 sequence only to position Ϫ481 bp, the Ϫ481 band observed at the top of the gel (Fig. 2C, Cb and Ctx lanes) is a summation of all natural GluR2 transcripts with 5Ј leaders at least this long. Despite this, however, the intensity of the Ϫ481 band was much weaker than that observed for shorter, more dominant mRNA species, indicating that long transcripts were few in number. From these analyses, it seems that GluR2 initiates transcription from multiple dominant sites between Ϫ340 and Ϫ429 bases upstream of the GluR2 AUG, with a minority of transcripts initiating at Ϫ452 and Ϫ481 bases. Therefore, depending on the site of transcription initiation, GluR2 mRNAs will contain a variable-length and unusually long 5Ј-UTR leader sequence (Kozak, 1987), a GC-rich 5Ј end, an imperfect GU repeat (40 nucleotides), and two to five upstream, out-of-frame AUG codons ( Fig. 1 A).
Although the banding pattern observed in cerebellar RNA was similar to that in cortical RNA in the RNase protection assays, longer transcripts appeared to predominate in the cerebellum (Fig. 2C). To address this further, an RNase protection probe, probe D ( Fig. 2 A), was designed that summates all transcripts Ϫ360 bases and longer into a single protected band. This probe is also partially protected by the major Ϫ340 transcript population. By hybridizing this RNA probe to total cortical and cerebellar RNA derived from each of four rats in separate reactions, we quantified the amount of long GluR2 transcripts (Ϫ360 and longer) relative to the amount of short GluR2 transcripts (Ϫ340) as shown in Figure 2 D. The ratio of long to short GluR2 transcripts was greater in the cerebellum than in the cortex in each of the four rats tested (Fig. 2 D; p Ͻ 0.05, paired t test). These results indicate that GluR2 initiation site usage differs between these two brain regions and in a strict sense supports the conclusion that the GluR2 gene contains multiple promoters.
We have designated the 5Ј-most dominant transcription initiation site, approximately Ϫ429 bases from the GluR2 AUG, the ϩ1 initiation site. All plasmid constructs used in evaluating GluR2 promoter activity by transient transfections are defined relative to this site. Our designated ϩ1 nucleotide for the rat gene is in agreement with the 5Ј-most start site identified for the mouse GluR2 gene (Köhler et al., 1994), but there is clear evidence of minor transcriptional initiation sites further upstream and additional major sites downstream; indeed, 5Ј initiation sites are dispersed over Ͼ141 bp of the GluR2 promoter region.
The GluR2 promoter region displays several areas of high (86%) and low (14%) GC content ( Fig. 1 B, right). The region of highest GC content spans ϳ150 bp adjacent to the ϩ1 site, includes some of the identified transcription initiation sites, and identifies a C pG island that could be subject to regulation by methylation in vivo. Indeed, when genomic DNA isolated from adult rat cortex, cerebellum, lung, and kidney was codigested with AccI and one of the methylation-sensitive enzymes SacII, Eco47III, or BssHII, only the SacII site at Ϫ568 bp was found methylated in a tissue-specific manner. This SacII site was fully methylated in kidney ( Fig. 1 B, 1.3 kb band) and lung (data not shown) genomic DNA but only partly methylated in DNA extracted from the cortex (Fig. 1 B) and cerebellum (data not shown). In contrast, Eco47III and BssHII restriction sites were unmethylated in all four tissues tested (data not shown). The SacII digestion pattern in cortex may be attributable to a combination of methylated glial DNA and unmethylated neuronal DNA in the sample, or it may reflect methylation of a single GluR2 allele. Although methylation of CpG dinucleotides increases the informational content of genes, the biological significance of the tissue-specific methylation of the GluR2 promoter remains to be addressed. The GC-rich proximal area also contains consensus recognition sequences for Sp1/Krox-24 and the NRF-1 transcription factors ( Fig. 1 A, box areas). The regulatory role of these cis Figure 1. Features of the rat GluR2 promoter and 5Ј-flanking region. A, Schematic showing the salient features of the GluR2 promoter region and partial nucleotide sequence of the GluR2 5Ј proximal promoter region, exon 1, and part of intron 1. The exon 1 sequence is in uppercase with the 5Ј promoter and intron 1 sequences in lowercase letters. mRNA 5Ј ends identified by 5Ј R ACE analysis are indicated by open triangles, and those identified by RNase protection and primer extension are indicated by solid or gray circles, respectively. The 5Ј-most prominent initiation site has been designated as the ϩ1 transcription initiation site, located 429 bases 5Ј of the GluR2 AUG (bent arrow). Consensus cis element sequences for the GluR2 silencer element and Sp1/Krox-24 and N RF-1 transcription factors are identified by name adjacent to the corresponding boxed sequence. The boundaries of the initiation site region and the location of the methylated C pG dinucleotide (C pG m ; filled diamond) are indicated in the schematic. The location of the antisense PER2-3 oligonucleotide used for primer extension reactions is indicated by double underlining below the nucleotide sequence. B, GC content of the GluR2 promoter and exon 1 and a methylation-sensitive Southern blot. The average GC content Ϯ 10 bp surrounding a central nucleotide was calculated and plotted along the length of the GluR2 promoter. The dashed line represents the average GC content (49.7%) for the whole 1.35 kb GluR2 promoter sequence shown. Open triangles, solid circles, and gray circles refer to R ACE, RNase protection, and primer extension sites, respectively, as described in A. The solid and open horizontal bars above the GC plot represent the location of the GluR2 silencer and the Sp1/ Krox-24 and NRF-1 regions, respectively. Also shown are the locations of C pG dinucleotides in the GluR2 promoter. The methylation state of the C pG dinucleotides was evaluated by sensitivity to SacII, Eco47III, and BssHII restriction digestion and represented by S, E, and B, respectively. The filled diamond represents a methylated CpG dinucleotide in kidney and lung but not in cortex and cerebellar DNA, whereas open diamonds indicate unmethylated CpGs in the DNA from all four tissues. The blot at lef t is an example of an AccI-and SacII-restricted Southern blot showing tissue-specific SacII digestion in cortex (Ctx) but not in kidney (Kid) DNA; lung and cerebellar Southern blots are not shown. The 1.3 kb fragment between the two AccI sites ( A) was radiolabeled as probe. elements is evaluated below. Regions more distal from the ϩ1 site exhibit a low GC content, minimal at consensus TATA box sequences at Ϫ760 and Ϫ361 bp and also at the GluR2 initiation AUG codon (Fig. 1). No TATA box element is near any of the transcriptional initiation sites identified. A weak consensus sequence for the pyrimidine-rich initiator element lies near the major 5Ј initiation site cluster at bp ϩ3 to ϩ11 (Javahery et al., 1994). The absence of a TATA box near transcription initiation sites is a common feature in mammalian genes, including glutamate receptor genes such as N R1, N R2B, N R2C, and KA-2 (Bai and Kusiak, 1993;Suchanek et al., 1995;Sasner and Buonanno, 1996;Huang and Gallo, 1997), and mouse GluR2 (Köhler et al., 1994). Overall, the nucleotide sequence similarity between the mouse and rat GluR2 promoter regions is high with a 90 -95% sequence identity over 1.2 kb of the proximal promoter region, including 0.8 kb 5Ј of the ϩ1 transcription initiation site.

Identification of a minimal promoter region
We generated a series of 5Ј-and 3Ј-GluR2 promoter deletion constructs and transfected these into primary cortical and primary glial cultures to define the boundaries of a minimal GluR2 promoter region and to identify cis elements that govern the cell-specific expression of GluR2. These constructs were also used to evaluate the role of consensus TATA and CAAT box sequences residing upstream in the GluR2 5Ј promoter. Whether deletions were made from the 5Ј or the 3Ј side, substantial ). The 32 P-labeled PER2-3 oligonucleotide-primed extension products from 30 g of cortical RNA (middle lane) are shown. Multiple bands are observed in cortical RNA (Ctx) reactions ranging from Ϫ312 to Ϫ452 bases 5Ј of the GluR2 AUG; none of these bands was observed in the control GluR2 in vitro synthesized RNA reactions (left lane), indicating that secondary structure does not account for the multiple bands observed in cortical RNA reactions. When the cortical primer extension products, derived from a separate reaction that used unlabeled PER2-3 oligonucleotide as primer, were Southern blotted and hybridized with probe B, all extension products, except the shortest, were positively identified as GluR2 (right lane). C, RNase protection of 30 g of adult cortical (Ctx) and cerebellar (Cb) RNA. Probe C detected multiple GluR2 5Ј ends both in cerebellar (middle lane) and cortical (right lane) RNA. The banding pattern shown here is nearly identical to that observed by primer extension. The 5Ј-most band at Ϫ481 relative to the AUG represents full-length probe sequence; therefore this band is a summation of all GluR2 transcripts Ϫ481 bases and longer. An RNase protection reaction to in vitro synthesized GluR2 that also extends to Ϫ481 bases is shown in the lef t lane (Ctl, Control). D, Regional distribution in brain of long and short GluR2 transcripts. The design of probe D, used for RNase protection experiments against 30 g of RNA isolated from the cerebellum and cortex of each of four rats, allows for the summation of all GluR2 transcripts Ϫ360 bases and longer into a single band on the gel as well as the identification of short GluR2 transcripts that initiate at Ϫ340 bases. Results for all four rats are shown at the top and are summarized in the graph at the bottom. Lines in the graph connect the ratios of long versus short transcripts in cerebellum (CB) and cortex (CX ) for each rat. Bar graphs are the mean Ϯ SEM of all data and are significantly different from each other ( p Ͻ 0.05, paired t test). luciferase expression was observed in primary cortical cultures when the region delimited by all transcription initiation sites (Ϫ52 to ϩ144; Fig. 3A, vertical dashed lines) was included in the promoter construct (Fig. 3B). A similar pattern of promoter activity was observed when these same constructs were transfected into primary glia (Fig. 3C) or HeLa cells (data not shown), with the exception that changes in promoter activity were generally less marked and, overall, promoter strength relative to SV40 was much lower than that observed in neurons. From these experiments it seems that GluR2 promoter activity is strongly depen-dent on cis elements positioned between Ϫ134 and ϩ320 bp (Fig. 3 A, B).
To define this region of the GluR2 promoter further, we designed shorter constructs that eliminated portions of this central promoter area (Ϫ134 to ϩ320 bp), from either the 5Ј or the 3Ј side, and transfected these constructs into primary cortical neurons and glia. Among these, R2(Ϫ134/ϩ147)luc, which spanned all identified initiation sites, possessed more than twice the promoter activity of any of the shorter constructs (Fig. 3 D, E). For example, when 44 bp were removed from the 3Ј end of Figure 3. 5Ј and 3Ј deletion analysis of the GluR2 promoter. A, A series of 5Ј and 3Ј deletion constructs was generated by PCR and cloned into the NheI and BglII sites of the pGL2 basic vector, upstream of firefly luciferase, for transfection into cultured cells. The boundaries of the promoter constructs are defined relative to the ϩ1 transcriptional initiation site (Fig. 1 A) and are indicated on the lef t, as are the relative locations of TATA and CAAT box sequences and the GluR2 AUG codon (A; bottom). The area spanning all identified transcription initiation sites is indicated by the vertical dashed lines, and the ϩ1 initiation site is marked with an arrow. B, Independent constructs were transfected into cortical neurons, and the promoter activity was normalized to that of SV40 to allow pooling of data from multiple experiments. When a deletion construct eliminated part of the transcription initiation site region, a significant drop in GluR2 promoter activity was observed. However, no single initiation site was essential for promoter activity. Note that these cortical neuronal cultures are a mixture of neurons (ϳ60%) and glia (ϳ40%). C, When constructs were transfected into pure cultures of primary glial cells, similar results were observed except that GluR2 promoter activity was very weak compared with that in neurons. Note the different scales of the x-axes in B and C. D, Short promoter fragments were made by PCR to identif y the minimal boundaries of the GluR2 promoter region necessary for strong neuronal expression. The locations of identified transcription initiation sites are shown with the same symbols used in Figure 1 A. E, Transfection of minimal constructs into cultured cortical neurons shows again that elimination of part of the transcription initiation site region (dashed lines in D) significantly reduced promoter activity. F, Similar results were observed in parallel transfections into cultured glia, and again, promoter activity was weak compared with that in the neurons. In these glial transfections, the SV40 construct had unusually low promoter activity, resulting in an artificially high ratio of GluR2 to SV40 activity on the x-axis. Results shown are the mean Ϯ SEM for 12-33 plasmid transfections in 47 neuronal cultures in B, for 3-9 plasmid transfections in 17 primary glial cultures in C, for 7-20 plasmid transfections in 20 neuronal cultures in E, and for 7-20 plasmid transfections in 9 glial cultures in F. *p Ͻ 0.05 from the R2(Ϫ134/ϩ147)luc construct by ANOVA and post hoc Dunnett's tests. Statistical analysis was not conducted on data in B and C. The average raw luciferase activity for the R2(Ϫ302/ϩ320)luc construct in neurons and glia was 1610 Ϯ 160 and 4.1 Ϯ 1.2 attomoles of luciferase per g of DNA per well, respectively. R2(Ϫ134/ϩ147)luc to generate R2(Ϫ134/ϩ103)luc, promoter activity was reduced by 53% in neurons, and likewise, when 124 bp were removed from the 5Ј end to make R2(Ϫ10/ϩ147)luc, promoter activity was reduced by 86% in neurons (Fig. 3E). Similar effects of these deletions on promoter strength were observed in primary glia (Fig. 3F ) and C6 glioma cells (data not shown), except promoter activity in glia was again much weaker than that in primary cortical cultures. The promoter activity of the R2(Ϫ169/ϩ103) construct was identical to that of R2(Ϫ98/ϩ103) (Fig. 3 D, E), indicating that the sequence between Ϫ169 and Ϫ98 bp plays no role in expression under our conditions. Judging from the activities of the constructs depicted in Figure 3, it appears that dual regions within the GluR2 central promoter area may be needed for optimum expression in both cortical neuronal and glial cells. The upstream region, from Ϫ98 to Ϫ10 bp, consists of a GC-rich sequence that includes the putative Sp1/ Krox-24 and NRF-1 elements and minor initiation sites. The downstream region, from ϩ103 to ϩ147 bp, includes a cluster of RACE 5Ј ends (Fig. 1 A) and a consensus sequence for metal element protein-1 (M EP-1) at bp ϩ127 to ϩ133 on the antisense strand. From these short constructs, we provisionally identif y the entire Ϫ98 to ϩ147 sequence as a minimal GluR2 promoter region that appears sufficient for strong expression in cortical cultures.
We also evaluated the f unctional role of upstream TATA box elements using constructs R2(Ϫ822/Ϫ743)luc and R2(Ϫ822/ Ϫ332)luc. Both of these constructs produced little or no activity in both neurons or glia (Fig. 3 B, C), eliminating a major role for the upstream TATA boxes. Similarly, inclusion of the tandem CAAT boxes at bp Ϫ317 and Ϫ324 in the R2(Ϫ822/Ϫ134)luc construct also produced very low activity in neurons and glia. These f unctional results support our R ACE, RNase protection, and primer extension findings that failed to demonstrate an association between transcription initiation sites and upstream TATA boxes and indicate that the GluR2 gene f unctions as a TATA-less promoter, as suggested by Köhler et al. (1994). A construct, R2(Ϫ822/ϩ832)luc, which contained 0.77 kb of 5Ј promoter sequence, all of exon 1, and part of the GluR2 intron 1, also showed no significant change in promoter activity in cortical cultures or primary glia (data not shown), indicating that in transiently transfected cortical cultures, the first GluR2 intron does not appear to harbor promoter regulatory elements.

Neural specificity of the GluR2 promoter
We transfected primary cortical neuronal and primary glial cultures, as well as C6 glioma and HeLa cell lines, with the R2(Ϫ302/ ϩ320)luc construct to evaluate the neural specificity of the GluR2 promoter. Judged by the percentage of stained cells after trans-fection with a Rous sarcoma virus (RSV)-␤-galactosidase expression plasmid, typical transfection efficiencies were 0.5-1% for primary neurons, 6 -8% for primary glia, and 10 -20% for the transformed cell lines (Table 1). Even though transfection efficiencies in the cultured neurons were low, the total raw firefly luciferase activity of the R2(Ϫ302/ϩ320)luc construct was quite strong in these cultures, on the order of 3000 Turner light units (TLU) per g of transfected DNA per well (background, ϳ0.10 TLU/well). This signal corresponds to 1.42 Ϯ 0.01 fmols of luciferase per g of DNA (n ϭ 80). We estimate that in each transfection up to 9000 neurons contribute to the total signal, based on transfection into triplicate wells, 1 ϫ 10 6 cells/well plated, an RSV-lacZ transfection efficiency of 0.5%, and a 60% fractional neuronal cell population in the culture.
When luciferase activities were normalized to SV40-or the CMV-promoter activity in each experiment and the average relative activity was calculated in neurons, primary glia, C6 glioma, and HeLa cells, the R2(Ϫ302/ϩ320)luc construct appeared to be 37-400-fold neuronal-selective. These values, however, are overestimates caused by the preferred activities of the SV40 and CMV viral promoters in the non-neuronal cells (Table 1). Nonetheless, when viral promoter cell preferences are accounted for, the R2(Ϫ302/ϩ320)luc construct still exhibited a neuronal selectivity of up to 72-fold.
We also analyzed our data on a per-transfected-cell basis by counting the actual number of RSV-lacZ transfected cells in parallel wells stained with 5-bromo-4-chloro-3-indolyl-␤galactopyranoside. Over four experiments, the average R2(Ϫ302/ ϩ320)luc promoter activity, per transfected cell, was 0.661 Ϯ 0.138 TLU in cortical cultures, 0.022 Ϯ 0.002 TLU/cell in cultured glia, and 0.034 Ϯ 0.019 TLU/cell in C6 glioma cells. The average ratio of cortical neuronal to glial and C6 glioma promoter activities, per cell, from all experiments, showed that GluR2 promoter activity was 32.5 Ϯ 3.9-and 26.6 Ϯ 8.6-fold more active in the cortical neurons than in cortical glia and C6 glioma cells, respectively.
To provide another estimate of cell selectivity, transfected cortical neuronal cultures were costained with an anti-firefly luciferase antibody and either anti-MAP-2 or anti-GFAP antibodies 1 d after R2(Ϫ302/ϩ320)luc transfection. Of 923 luciferasepositive cells counted, Ͼ97% also expressed the neuron-specific marker MAP-2 (Fig. 4 A, B). When immunostainings were conducted in parallel cultures to identify expression in glial cells, only 6 of 645 luciferase-positive cells costained for the glial specific marker GFAP (data not shown). In these six copositive cells, luciferase expression was marginally discernible above back- Cultures were transfected for 5 hr with 1 g of DNA/well, harvested 24 hr after transfection, and analyzed for luciferase activity. The GluR2 promoter is the R2(Ϫ302/ϩ320)luc construct, and the NaII promoter includes the RE1 silencer described by Kraner et al. (1992). Results shown are the mean Ϯ SEM for transfections in 27-43 independent culture preparations for neurons, 4 -15 for C6 glioma, 11-20 for glia, and 11-15 for HeLa cells for all promoters except NaII. Cortical neuronal cultures are a mixture of neurons (ϳ60%) and glia (ϳ40%). Transfection efficiencies were determined with an RSV-lacZ construct in three to five cultures. The numbers of cultures used for NaII promoter transfections were five for neurons and HeLa and seven for C6 glioma and glia.
ground fluorescence, indicating that even in these glial cells GluR2 promoter activity was weak. On the other hand, when cortical cultures were transfected with the C M V-luc construct, strong anti-luciferase staining in GFAP-positive cells was observed, indicating that transfected glial cells in the primary neuronal cultures were capable of strong luciferase expression (data not shown). These results together demonstrate that the GluR2 promoter is a strong, neuron-selective promoter that is ϳ30-fold stronger in primary neurons than in primary glia. Because the GluR2 minimal promoter region also showed strong activity in neurons, we were interested in determining, by immunofluorescence, whether a minimal promoter construct could regulate cell-type -selective GluR2 expression in the mixed neuron -glia cortical cultures. When cortical cultures were transfected with R2(Ϫ134/ϩ147)luc, Ͼ93% of 146 luciferase-positive cells costained positive for M AP-2 (Fig. 4C,D), indicating that elements in this region of the promoter appear sufficient to guide selective luciferase expression in transfected primary cortical neurons.

Deletion scan of the GluR2 promoter identifies a silencer element and a positive regulatory region
To identif y elements more precisely within the proximal promoter that regulate GluR2 expression, we made internal deletions in the R2(Ϫ302/ϩ320)luc construct and analyzed these by transfection into cortical neuronal and glial cultures, in parallel, as above. Most constructs substituted a unique 6 bp restriction enzyme site for 25-30 bp of GluR2 promoter sequence; the deletions in ⌬K, ⌬L, and ⌬M were larger (Fig. 5B). For each experiment, three or more independent plasmid DNA preparations for each construct were transfected into the cells. By this analysis, we identified a 25 bp region that, when deleted in the R2(Ϫ302/ϩ320)luc-⌬B construct, resulted in a significant increase in expression in glial cells but not in neurons (Fig. 5B). R2(Ϫ302/ϩ320)luc-⌬D, which eliminated the Ϫ146 to Ϫ118 region, resulted in a 47% increase in promoter activity in neurons but not in glia, suggesting that this region of the GluR2 promoter may negatively regulate expression in neurons. Three other 25-30 bp deletion constructs, ⌬G, ⌬H, and ⌬I, caused a significant reduction in luciferase expression in both neurons and glial cells. In these cases, promoter activity in each cell type dropped by 35-65% compared with that of the parent R2(Ϫ302/ϩ320)luc construct, even though the basal promoter activity differed ϳ35fold in the two cultures (Fig. 5A). Results found in transfected C6 glioma cells were similar to those in glia (data not shown). The ⌬G, ⌬H, and ⌬I internal deletion constructs span the region from Ϫ64 to ϩ14 bp and include the consensus sequences for Sp1/ Krox-24 and NRF-1. The ⌬L deletion (ϩ71 to ϩ146) also resulted in a reduction in promoter activity that was marginally greater in glia than in the neuronal cultures. This deletion mutant eliminated the most 3Ј initiation site identified by primer extension and RNase protection (Fig. 2 A, B, Ϫ340 band), the 3Ј cluster of RACE 5Ј ends, and a putative MEP-1 element. In contrast, the deletion of other major transcription initiation sites in the ⌬J, ⌬K, and ⌬M constructs did not reduce luciferase expression.
The pattern of expression from these internal deletion constructs complements our findings from the external 5Ј and 3Ј deletion series (Fig. 3) and allowed us to provisionally eliminate up to 364 bases of the GluR2 promoter from further consideration. We focused our attention on the non-neural silencer element identified in deletion R2(Ϫ302/ϩ320)luc-⌬B and the positive regulatory region identified in deletions R2(Ϫ302/ ϩ320)luc-⌬G, -⌬H, and -⌬I.

The GluR2 silencer is a cell-specific modulator of expression, not a switch
Inspection of the GluR2 promoter sequence revealed a 21 base element with 71% identity to both the rat NaII-RE1 (Maue et al., 1990;Kraner et al., 1992) (Fig. 6 A) and the rat SCG10 gene neuron-restrictive silencer element (NRSE) (Mori et al., 1992). This RE1/NRSE-like element in the GluR2 promoter is located on the antisense strand at bp Ϫ174 to Ϫ194 (Fig. 1 A) and had been deleted entirely in the R2(Ϫ302/ϩ320)luc-⌬B construct (Fig. 5B). Given the sequence similarity, we investigated the role of the putative GluR2 silencer in regulating R2(Ϫ302/ϩ320)luc promoter activity in neuronal and non-neuronal cells. When the entire GluR2 silencer was replaced with an NsiI restriction site (Fig. 6A, ⌬B) or different 6 bp stretches were substituted by 6 bp EcoRI or EcoRV restriction sites (Fig. 6 A, ␦RI, ␦RV ), significant two-to threefold increases in R2(Ϫ302/ϩ320)luc activity were observed in glia and C6 gliomas (Fig. 6 B). Similar results were observed when these same mutations were introduced into the context of a larger R2(Ϫ822/ϩ320)luc promoter construct (data not shown). No changes in neuronal expression were observed by any of these mutations (Fig. 6 B), demonstrating that silencing activity was specific to non-neuronal cells alone. These results are consistent with known RE1/ N RSE silencers in other genes, in which short internal missense mutations are sufficient to disrupt silencing activity in non-neuronal cells (Kraner et al., 1992;Mori et al., 1992;Li et al., 1993;Mieda et al., 1997).
Because the GluR2 silencer was only 71% identical with the NaII-RE1 element and the effect of deleting this element was moderate (two-to threefold) when compared with homologous RE1/N RSE silencers (Kraner et al., 1992;Mori et al., 1992;Li et al., 1993;Wood et al., 1996;Mieda et al., 1997), we compared the ability of the GluR2 and the NaII-RE1 silencers to repress both GluR2 and NaII promoter (NaIIluc) activities in glia, C6 glioma, and neuronal cells. When the GluR2 silencer was mutated to that of the NaII-RE1 sequence in the context of the GluR2 promoter, this substitution resulted in a 30 -47% decrease in promoter activity in non-neuronal cells (Fig. 6C, black bars) relative to that of the unmutated R2(Ϫ302/ϩ320)luc control. Conversely, when the NaII silencer in the context of the NaII promoter was mutated to that of the GluR2 silencer sequence, promoter activity was increased by 54 -61% in non-neuronal cells (Fig. 6C, gray bars), indicating that repression by the NaII silencer was lessened by substitution with the GluR2 silencer. The effect of swapping these silencer elements was specific to the non-neuronal cells because no changes in neuronal expression were observed. For all of these constructs, the flanking nucleotide sequences and the orientation of the silencer element in each minimal promoter context were carefully preserved (Fig. 6C, top). Together, these results demonstrate that the NaII-RE1 silencer is more effective than the GluR2 silencer and indicate that their relative strengths are because of differences in their nucleotide sequence rather than differences in silencer orientation or in the context of the promoters in which they reside. This conclusion is further supported by the observation that both the NaII and the GluR2 minimal promoter constructs used for these experiments, in the absence of either silencer, had similar promoter activities in glia (2.0 and Figure 5. Mutation sensitive scan of the proximal GluR2 promoter identifies both positive and negative regulatory regions. A, The promoter activity of the base construct R2(Ϫ302/ϩ320)luc in cultured neurons (N; n ϭ 63) and glia (G; n ϭ 46), relative to SV40luc activity, is shown. Data are the mean Ϯ SEM combined from all experiments. The average raw luciferase activity recovered from neurons and glia for the unmutated R2(Ϫ302/ϩ320)luc control was 1320 Ϯ 120 and 31 Ϯ 14 attomoles per g of DNA per well, respectively. B, A series of short internal deletions (⌬A through ⌬N, typically 24 -31 bp long, except ⌬K, ⌬L, and ⌬M) were generated by replacement with unique 6 bp restriction sites in the R2(Ϫ302/ϩ320)luc GluR2 context. All deletions were generated by PCR and are shown schematically at the top. Dashed lines indicate the sequence deleted for the indicated construct (uppercase letters). Gray and solid circles and open triangles on the line indicate the locations of primer extension, RNase protection, and R ACE 5Ј ends, respectively. Open boxes identify the locations of Sp1/ Krox-24 and N RF-1 consensus sequences; the black box identifies the GluR2 silencer that is homologous with the RE1/NRSE silencers. The effect of each internal mutation on promoter activity in cultured cortical neurons and glia is shown. L uciferase activity was normalized to that of the unmutated R2(Ϫ302/ϩ320)luc control construct. Note the log scale on the y-axis. The R2(Ϫ302/ϩ320)luc-⌬B construct resulted in a significant twofold increase in promoter activity in glia. Likewise the R2(Ϫ302/ϩ320)luc-⌬D deletion resulted in a significant increase in promoter activity in neurons. Three other deletions, R2(Ϫ302/ϩ320)luc-⌬G, -⌬H, and -⌬I, resulted in significant decreases in promoter activity in both glia and neurons. This internal deletion scan identified both a silencer region that negatively regulates GluR2 promoter activity in glia and a larger 78 bp region that positively regulates promoter activity in both neurons and glia. Data shown are the mean Ϯ SEM for 6 -18 plasmid transfections in 21 cultured neuronal preparations and for 6 -8 transfections in 18 cultured glial preparations. lucif, L uciferase. *p Ͻ 0.05 and **p Ͻ 0.01 indicate significance from the control construct by ANOVA and post hoc Dunnett's tests.
We also investigated whether the GluR2 silencer could interact with a known repressor factor, REST (Chong et al., 1995), also known as neuron-restrictive silencer factor (N RSF) (Schoenherr and Anderson, 1995) or X2 box repressor (X BR) (Scholl et al., 1996), by cotransfecting the pC M V-RESTexpress plasmid with GluR2 promoter constructs into cultured neurons (Fig. 6 D). There are three important considerations in the design of this cotransfection experiment. First, it is important to compare the effect of REST on normal and mutated GluR2 silencers to eliminate indirect effects of REST on other regions of the GluR2 promoter or on the expression of other endogenous genes. Second, it is necessary to cotransfect with an empty CMV vector, pcDNA3, as a control for indirect effects of the CMV promoter on luciferase expression. Third, it is necessary to demonstrate that simple inclusion of additional DNA in the transfection mix does not change transfection efficiency. Regarding the latter, luciferase activity was found to be linearly related to the amount of input R2luc DNA between 0.1 and 1.5 g/well, and luciferase Figure 6. A weak GluR2 silencer interacts with the repressor REST. A, Schematic showing the location and antisense orientation of the silencer in the GluR2 promoter. The native (GluR2) and mutagenized silencer elements introduced into the R2(Ϫ302/ϩ320)luc promoter construct are listed. Dashes represent deleted sequence, and lowercase, boldface t ype represents substitution mutations. The ⌬B mutation is the same mutation as the internal deletion construct depicted in Figure 5. The NaII-RE1 sequence shown is the T ype II Na channel RE1 silencer (Kraner et al., 1992), which differs from GluR2 in six positions. B, Functional analysis of silencer mutations on GluR2 promoter activity in primary cortical neurons, primary cortical glia, and C6 glioma cells. In neurons, no mutation resulted in a significant change in promoter activity from that of the R2(Ϫ302/ϩ320)luc control. In primary glia and C6 glioma cells, all three silencer mutations resulted in two-to threefold increases in GluR2 promoter activity. Data shown are the mean Ϯ SEM for 12-24 plasmid transfections in neurons, 14 -22 plasmid transfections in glia, and 6 -34 plasmid transfections in C6 gliomas from 10, 9, and 7 culture preparations, respectively. *p Ͻ 0.05 from the control by ANOVA and post hoc Dunnett's test. The error bar for the control (R2) represents the average SE among control plasmids. C, Promoter context independence of the GluR2 silencer. The GluR2 silencer was replaced by the NaII-RE1 silencer (see A) in the context of the R2(Ϫ302/ϩ320)luc promoter. Likewise, the NaII-RE1 silencer was replaced by the GluR2 silencer in the context of the NaII promoter construct. These constructs were transfected into primary neurons, glia, and C6 glioma cells. The NaII-RE1 silencer in the context of the GluR2 promoter (black bars) reduced promoter activity by 30 -47% relative to that of the unmutated R2(Ϫ302/ϩ320)luc control (open bars defined as 100%) in glia and C6 glioma cells. In contrast, the GluR2 silencer in the context of the NaII promoter ( gray bars) increased promoter activity by 54 -61% relative to that of the unmutated NaII minimal promoter control (open bars also) in glia and C6 glioma cells. The effect of the alternative silencer in each of the promoters was significant from their respective controls; *p Ͻ 0.05, Student's t test. Data shown are the mean Ϯ SEM for 8 -9 plasmid transfections in neurons and 8 -15 transfections in glia and C6 gliomas, all from three to five culture preparations for each cell type. D, Cotransfection of REST represses GluR2 promoter activity in neurons. The GluR2 silencer in the context of R2(Ϫ302/ϩ320)luc was either unchanged ( gray circles), deleted (open circles), replaced by the NaII silencer (black circles), or replaced by tandem GluR2 silencers (hatched circles) and then cotransfected into cultured cortical neurons with 0.75 g of pCMV-RESTexpress per well or the pcDNA3 empty vector. The amount of R2(Ϫ302/ϩ320)luc reporter plasmid was varied from 0.1 to 1.0 g/well, and all tubes were balanced to 1.75 g of total DNA / well with carrier DNA (pBluescript). REST cotransfection reduced GluR2 promoter activity by 20 -42% from that of the pcDNA3 control in neurons in a silencer-dependent manner. This effect by REST was observed at all concentrations of input GluR2 promoter DNA tested (D, lef t). When luciferase activities were normalized to those of the silencer-deficient construct R2(Ϫ302/ ϩ320)luc-⌬B and pooled (right, open bar), REST coexpression reduced GluR2 promoter activity in neurons by an average of 37% (right, gray bar). Results are the mean Ϯ SEM for 3-14 cotransfections in 14 culture preparations. Identical results were obtained using two independent preparations of pCMV-RESTexpress, pcDNA, and pBluescript plasmid DNA. *p Ͻ 0.05 or better from all other groups at the designated amount of input DNA (left, ANOVA and post hoc Dunnett's) or from the silencer-deficient construct (right, Student's t test). activity was not influenced by adding carrier pBluescript DNA (data not shown). The cotransfection experiments are summarized in Figure 6 D. The RESTexpress plasmid in the transfection cocktail was fixed at 0.75 g/well, the GluR2 promoter construct was varied from 0.1 to 1.0 g/well, and all tubes were balanced with carrier DNA (pBluescript) to a final mass of 1.75 g of DNA per transfected well. Compared with the pcDNA3 control, the pCMV-RESTexpress cotransfections resulted in a consistent 20 -42% reduction in neuronal R2(Ϫ302/ϩ320)luc promoter activity when the reporter construct contained an intact GluR2 or NaII-RE1/N RSE silencer element (Fig. 6 D, lef t). REST coexpression actually increased GluR2 promoter activity in neurons transfected with the GluR2 construct lacking a silencer, presumably because of an action of REST on other genes. Thus, when the data across all input DNA amounts were pooled and normalized to the activity of the construct lacking the GluR2 silencer, a significant 37% REST-mediated reduction in GluR2 promoter activity was observed (Fig. 6 D, right). The addition of an extra GluR2 silencer element, or the substitution of the GluR2 silencer by the NaII-RE1 silencer, did not significantly enhance the effect of REST. Presumably the NaII-RE1 element was no more effective than the GluR2 silencer because 0.75 g of pCMV-RESTexpress was saturating in both cotransfections. The consistent ability of REST to reduce GluR2 promoter activity in neurons in a silencer-dependent manner indicates that in intact cells the REST repressor can bind the GluR2 silencer to modify GluR2 promoter activity.

Functional NRF-1 and Sp1 cis elements in the GluR2positive regulatory region
Based on the internal deletion scan, it was concluded that elements within the R2(Ϫ302/ϩ320)luc-⌬G, -⌬H, and -⌬I deletions must contribute to GluR2 promoter activity. As shown in Figures  1 A and 5B, these regions harbor consensus sequences for Sp1/ Krox-24 and N RF-1 elements and, within the ⌬H deletion, an element with 75% identity with the partial dyad (Patankar et al., 1997). We introduced substitution mutations into or near these elements to determine specifically whether these elements were necessary for GluR2 promoter activity in neurons. When six bases of the N RF-1 element were substituted with a BglII site (AGATC T) to generate the R2(Ϫ302/ϩ320)luc-␦i1 construct (Fig. 7A), a 46 and 26% drop in promoter activity was observed in cortical neurons and cortical glia, respectively (Fig. 7B, gray bars). The effect of this mutation approaches the effect observed for the larger R2(Ϫ302/ϩ320)luc-⌬I deletion in neurons and is consistent with the effect of mutations introduced into other NRF-1 elements (Virbasius et al., 1993;Seelan et al., 1996). The R2(Ϫ302/ϩ320)luc-⌬K deletion, generated by excising the sequence between two natural BssHII sites, eliminated all of the sequence in the ⌬I deletion except for the N RF-1 consensus element, and yet this construct did not decrease but, instead, slightly elevated promoter activity in neurons and glia (Fig. 7B,  black bars). This result strongly suggests that the consensus NRF-1 sequence is a f unctional regulatory element in the GluR2 promoter. However, because the N RF-1 element is GC-rich and might therefore interact with factors other than N RF-1 (e.g., Sp1), we cotransfected the N RF-1 transcription factor (under the CMV promoter) with the unmutated R2(Ϫ302/ϩ320)luc control construct and the mutated R2(Ϫ302/ϩ320)luc-⌬I and -␦i1 constructs into neurons (Fig. 7C). N RF-1 cotransfection increased GluR2 promoter activity approximately twofold, but only when the GluR2 promoter construct contained an unmutated NRF-1 cis element (Fig. 7C). The lack of a larger effect by the cotransfected NRF-1 may be attributable to the presence of endogenous NRF-1 or some other transcription factor in the cortical neurons. These results indicate that the NRF-1 cis element sequence supports positive GluR2 promoter activity in neurons and suggests, but does not prove, that the NRF-1 transcription factor is a component of the transcriptional complex in vivo.
The NRF-1 element partially overlapped the R2(Ϫ302/ ϩ320)luc-⌬H deletion construct (Fig. 7A), and also because the nucleotide sequence within the ⌬H deletion resembled the partial dyad motif described previously (Patankar et al., 1997), we introduced additional short substitution mutations into this portion of the GluR2 promoter, again in the context of R2(Ϫ302/ϩ320)luc (Fig. 7A). All three substitution mutations, R2(Ϫ302/ϩ320)luc-␦h1 (TCTAGAT), -␦h2 (CGAGT), and -␦h3 (AGATC), were without significant effect on GluR2 promoter activity in both neurons and glia (Fig. 7B, hatched bars), indicating that neither the partial dyad nor another unidentified element in this region of the promoter contributed to GluR2 promoter activity. The strong effect of the large ⌬H deletion may be caused by disruption of a single base of the NRF-1 element or, more likely, caused by disruption of an appropriate architectural distance between the functional NRF-1 and Sp1/Krox-24 (see below) elements.
A third consensus element within the GluR2-positive regulatory region that we investigated by mutation analysis was the consensus Sp1 and Krox-24 sequence at Ϫ47 to Ϫ60 bp. This sequence was entirely deleted in the internal deletion R2(Ϫ302/ ϩ320)luc-⌬G construct, resulting in a significant 35-50% reduction in promoter activity in both neurons and glia (Fig. 5B). When four bases of the consensus element were mutated (TCTA) by base substitution to generate R2(Ϫ302/ϩ320)luc-␦g1, a nearly identical reduction in GluR2 promoter activity was observed ( Fig.  8 A, B), suggesting that it is the consensus Sp1/Krox-24 sequence, and not the immediate flanking sequence within the larger ⌬G deletion construct, that is important for GluR2 promoter activity. To differentiate between an Sp1 and Krox-24 component for GluR2 expression in neurons, we prepared nuclear extracts from cultured cortical neurons and combined these extracts with 32 Pend-labeled double-stranded oligonucleotide probes in vitro. Oligonucleotide probes consisting of the ⌬G and ⌬H sequences shown in Figure 8 A produced different protein-DNA complexes in nondenaturing gel mobility shift experiments (Fig. 8C). Probe G formed three major complexes that were effectively competed away by 200-fold molar excess of cold probe (compare Fig. 8C,  lanes 2-3). Excess Sp1 oligonucleotide competed away bands 1a and 1b but was less effective on band 1c (Fig. 8C, lane 4), and addition of an anti-Sp1 antibody to the incubations supershifted the complex, indicating that Sp1 is part of the protein complex recruited by probe G (Fig. 8C, arrow, lane 5). The addition of anti-Krox-24 antibody (Fig. 8C, lane 6) or competing Krox-24 oligonucleotides (Fig. 8C, lane 7) to the incubations had no effect on complex formation with probe G. The mutant ␦g1 oligonucleotide was a poor competitor of Sp1 binding in vitro because 30 -200-fold excess oligonucleotide only weakly decreased protein binding to probe G (Fig. 8C, lanes 18 -20), compared with competition by 30 -200-fold excess cold probe G DNA (Fig. 8C, lanes  15-17), consistent with the conclusion that the Sp1 element in Figure 8 A is the Sp1-binding site required for functional promoter activity.
Similar in vitro binding experiments with probe H resulted in the formation of a single complex that was competed away with excess unlabeled probe H DNA (Fig. 8C, lane 10). This protein-DNA complex was not disrupted by competing unlabeled probe G (Fig. 8C, lane 11) or Krox-24 DNA (data not shown), nor was it part of the complex supershifted by coincubation with anti-Sp1 (Fig. 8C, lane 12) or anti-Krox-24 antibodies (data not shown). The different banding patterns observed for the G and H probes as well as the nonoverlapping competition between the two probes indicate that the protein complex recruited by sequence contained within ⌬G of the GluR2 minimal promoter region includes Sp1.

DISCUSSION
We cloned the proximal promoter region of the rat GluR2 gene and identified three regulatory elements that govern GluR2 promoter activity in transfected cortical cultures, and we confirm the location of a major transcription initiation site Ϫ429/Ϫ430 nucleotides from the AUG codon, as described previously (Köhler et al., 1994). The new conclusions of this study are (1) GluR2 transcription initiates from multiple major sites spread over Ͼ141 bp, (2) initiation sites are used differentially in adult cortical and cerebellar neurons, (3) GluR2 promoter activity is 30-fold stronger in cultured cortical neurons than in primary glia or C6 glioma, (4) promoter activity is regulated by both an RE1/NRSE-like silencer element and two positive regulatory elements (NRF-1 and Sp1) separated by a nucleotide bridge sequence, and (5) the GluR2 silencer is a weak silencer relative to the NaII RE1 silencer but still contributes to the neuronal specificity of the GluR2 promoter. These findings identify the first functional regulatory elements in the GluR2 promoter and point to a complex transcriptional control region for GluR2 expression in neurons.

Functional implications of multiple GluR2 promoters
Our primer extension and RNase protection results identified multiple transcription initiation sites resulting in lengthy (340 -481 bases) GluR2 mRNA leader sequences. These, and additional results (Fig. 2), suggest the existence of multiple GluR2 Figure 7. Contribution of the N RF-1 element to GluR2 promoter activity. A, Schematic of the positive regulatory domain, Ϫ64 to ϩ38 bp, of the GluR2 promoter region is shown with regulatory cis elements as shaded bars above the sequence; gray and solid circles and triangles signify some of the transcription initiation sites described in the Figure 1 A legend. Internal deletion and substitution mutations introduced into the R2(Ϫ302/ϩ320)luc construct are shown by open bars below the sequence and identified. Deletion mutants are signified by "⌬," and base-for-base substitution mutations are signified by "␦." Deletion constructs are the same as those in Figure 5. B, Multiple plasmid DNA preparations for each construct were transfected into cultured cortical neurons or cortical glia, and luciferase activity was measured 1 d later. Promoter activity of the mutated constructs was compared with that of the R2(Ϫ302/ϩ320)luc control construct (Con). The ␦h1, ␦h2, and ␦h3 substitution mutations had no effect on GluR2 promoter activity in neurons and glia and did not replicate the effect of the larger ⌬H deletion construct, indicating that the effect of the ⌬H deletion is not dependent on nucleotide sequence (i.e., the partial dyad). In contrast, both the ⌬I and ␦i1 mutations resulted in a reduction in promoter activity in neurons and glia, whereas the ⌬K and ␦h3 constructs that modif y the sequence immediately adjacent to the N RF-1 element were without effect. Together these results identify the consensus NRF-1 as a positive regulator of GluR2 promoter activity. Results shown are the mean Ϯ SEM for 8 -12 plasmid transfections into four to five cultures. *p Ͻ 0.05 from the control group by ANOVA and post hoc Dunnett's tests. C, One microgram of a plasmid expressing the NRF-1 transcription factor (black bars) or an empty pcDNA3 vector (cross-hatched bars) was cotransfected with 0.5 g of R2(Ϫ302/ϩ320)luc (Con) or the ␦i1 or the ⌬I mutation constructs into neurons. Data were normalized to the activity of a parallel SV40-luc transfection in each experiment. The NRF-1 cotransfection increased promoter activity in the control construct but not in those constructs with a compromised N RF-1 cis element. Results are the mean Ϯ SEM in five neuronal cultures. *p Ͻ 0.05 from the pcDNA3-(no N RF-1) transfected group, Student's t test.
promoters. First, the observed wide distribution of transcriptional start sites (Ͼ141 bp) is uncommon even for TATA-less promoters but is shared by other C NS genes in which multiple promoters choreograph appropriate developmental or tissue-specific gene expression (Timmusk et al., 1993;Pathak et al., 1994) or increase the complexity of mRNA transcripts within the same tissue type (Liu and Fischer, 1996). Second, we found that the number and relative use of major GluR2 transcription initiation sites in cortex were different from that in cerebellum (Fig. 2C, D). These regional differences in initiation site usage perhaps reflect the greater variety of neuronal cell types found in cerebral cortex versus cerebellum and that distinct subpopulations of cortical neurons may transcribe GluR2 from different initiation sites. Alternatively, transcription initiation could occur at multiple sites in each cortical neuron, thereby increasing the variety of GluR2 transcripts expressed per cell. In either case, the usage of the upstream transcriptional start sites is cell-selective, consistent with multiple promoters. Thirdly, no single initiation site was completely essential for GluR2 promoter activity based on our internal deletion scan (Fig. 5), indicating that RNA polymerase can initiate GluR2 transcription from multiple, independent sites. Depending on the site of transcription initiation, GluR2 mRNAs will contain variably long GC-rich 5Ј-UTR leaders, an imperfect GU repeat (40 nucleotides), and from two to five upstream out-of-frame AUG codons. Unusually long 5Ј-UTR sequences (Kozak, 1987) with high GC content and upstream AUG codons can interfere with appropriate mRNA translation (see Geballe and Morris, 1994). Indeed, we have found that sequences in the 5Ј Figure 8. Identification of an Sp1 element required for GluR2 promoter activity. A, Schematic of the positive regulatory domain, Ϫ64 to ϩ38 bp, of the GluR2 promoter is shown with identified cis elements as shaded bars above the sequence; gray and solid circles and triangles signify some of the transcription initiation sites described in the Figure 1 A legend. Internal deletion and substitution mutations introduced into the R2(Ϫ302/ϩ320)luc construct are identified by open bars below the sequence. Deletion mutants are signified by "⌬," and substitution mutations are signified by "␦." Deletion constructs are the same as those in Figure 5. B, Multiple plasmid transfections were made into cortical neurons or cortical glia, and promoter activity was measured 1 d later. Both the ⌬G deletion and the ␦g1 substitution mutation resulted in significant reductions in GluR2 promoter activity in both cell types. These results suggest that the Sp1/ Krox-24 consensus element from Ϫ47 to Ϫ60 positively regulates GluR2 promoter activity. Results shown are the mean Ϯ SEM for eight to nine plasmid transfections in three cultures. *p Ͻ 0.05 from the control group by ANOVA and post hoc Dunnett's tests. C, Nuclear extracts (8 g/lane; Extr) were prepared from cultured cortical neurons and incubated with 32 P-end-labeled double-stranded oligonucleotide probes (ϳ0.3 nM) at room temperature for 15 min. Probe G (lanes 1-7 and 13-20; same sequence as ⌬G in A) resulted in the formation of three DNA-protein complexes (1a, 1b, and 1c), which were effectively competed away with 200-fold excess (Comp) unlabeled probe G oligonucleotide (lane 3). Excess competitor Sp1 oligonucleotides also competed for probe G binding (lane 4), and an anti-Sp1 antibody supershifted part of the complex (lane 5). A Krox-24 oligonucleotide (Kx; lane 7) and an anti-Krox-24 antibody (lane 6) had no effect. When the ⌬H sequence was used as probe H (lanes 8 -12), a different protein-DNA complex was resolved. This complex was competed away with excess cold probe H DNA (lane 10) but not excess cold probe G (lane 11), nor was it supershifted by anti-Sp1 (lane 12) antibodies. In a separate experiment, 30 -200-fold excess competitor oligonucleotides were added to similar probe G-binding incubations (lanes 13-20). Here, excess cold probe G (lanes 15-17) was an effective competitor at displacing the protein from the complex, whereas excess ␦g1 (lanes 18 -20) was not. These results demonstrate that the Sp1/ Krox-24 consensus element identified in A is important for GluR2 promoter activity and for Sp1 binding. lacking specific 5Ј initiation sites still showed neuronal preference ( Fig. 3D-F). Finally, a neural-specific element in the GluR2 promoter may exist but reside outside the region analyzed by our internal deletion scan.
Findings from this study provide a necessary foundation for understanding the transcriptional regulation of GluR2 expression in neurons. In particular, it will be of interest to determine the biological roles of the identified Sp1, N RF-1, and silencer regulatory elements in mediating changes in GluR2 levels after ischemia and status epilepticus. These results also suggest a strategy to determine why some C NS interneurons express low levels of GluR2, affecting the properties of AM PA receptor-mediated synaptic transmission. One approach would be to determine whether appropriate neuronal expression of the N RF-1, Sp1, Sp3, Sp4, and /or REST transcription factors correlates with the GluR2 expression level in different cell populations.