Involvement of an Upstream Stimulatory Factor as Well as cAMP-responsive Element-binding Protein in the Activation of Brain-derived Neurotrophic Factor Gene Promoter I*

The use of different brain-derived neurotrophic factor (BDNF) gene promoters results in the differential production of 5 (cid:1) -alternative transcripts, suggesting versatile functions of BDNF in neurons. Among four BDNF promoters I, II, III, and IV (BDNF-PI, -PII, -PIII, and -PIV), BDNF-PI was markedly activated, as well as BDNF-PIII, by Ca 2 (cid:1) signals evoked via neuronal activity. However, little is known about the mechanisms for the transcriptional activation of BDNF-PI. Using rat cortical neurons in culture, we assigned the promoter sequences responsible for the Ca 2 (cid:1) signal-mediated activation of BDNF-PI and found that the Ca 2 (cid:1) -responsive elements were located in two separate (distal and proximal) regions and that the DNA sequences in the proximal region containing cAMP-responsive element (CRE), which is overlapped by the upstream stimulatory factor (USF)-binding element, were largely responsible for the activation of BDNF-PI. CRE-binding protein (CREB) family transcription factors and USF1/USF2 bind co-transfected

neuronal survival, neuronal differentiation, and synaptic plasticity (1). BDNF is predominantly expressed in the central nervous system, and its mRNA expression is up-regulated by neuronal activity accompanying the influx of Ca 2ϩ into neurons (2,3). The BDNF gene consists of four short 5Ј-exons (exons I, II, III, and IV) and a common 3Ј-exon V encoding a prepro-BDNF protein (4). Four promoters, BDNF-PI, II, III, and IV, were mapped upstream of the 5Ј-exons, respectively, which are differentially regulated by kainic acid-induced seizure in distinct regions of the rat brain (5), suggesting versatile BDNF functions in the brain (6). Therefore, elucidating the regulatory mechanisms of BDNF gene transcription will provide a better understanding of BDNF functions in the brain.
We previously reported that BDNF-PI and -PIII were differentially activated by Ca 2ϩ signals evoked via two distinct Ca 2ϩ entry sites, the L-type voltage-dependent Ca 2ϩ channel (L-VDCC) and the N-methyl-D-aspartate glutamate receptor (NMDA-R) (7). BDNF-PI mainly responds to Ca 2ϩ signals evoked via L-VDCC, whereas BDNF-PIII reacts to those evoked via either NMDA-R or L-VDCC but less to those via NMDA-R, suggesting that the transcriptional mechanisms initiated in response to Ca 2ϩ signals evoked via Ca 2ϩ channels, at least in part, differ between BDNF-PI and -PIII. It has already been reported that the binding of cAMP-responsive element (CRE)-binding protein (CREB) to CRE on BDNF-PIII is required for the Ca 2ϩ responsiveness of BDNF-PIII (8,9). Quite recently, however, novel Ca 2ϩ -responsive elements (CaREs) of BDNF-PIII have been characterized and three CaREs, called CaRE1, CaRE2, and CaRE3 (or CRE), identified within a stretch of 170 bp upstream of exon III (10,11). In addition to CREB, which specifically binds to CaRE3 (or CRE), a novel calcium-responsive transcription factor, which specifically binds to CaRE1, has been found to drive the neuronal specific activation of BDNF-PIII in response to the Ca 2ϩ signals (11). However, it is still unknown whether the CREB or another transcription factor is involved in the activation of BDNF-PI.
In the present study, we assigned the Ca 2ϩ -responsive DNA elements of BDNF-PI and identified the transcription factors binding to these responsive elements. As a result, we found that not only the CREB family transcription factors but also the upstream stimulatory factor (USF) bound to the responsive site on which the CRE and the USF-binding element overlap each other. It is well established that CREB is a transcription factor, which conveys intracellular Ca 2ϩ signals to the gene through its phosphorylation at Ser-133 (12,13). On the other hand, USF transcription factors encoded by two distinct genes (USF1 and USF2) are basic helix-loop-helix/leucine zipper family members, which preferentially interact with the E-box (5Ј-CANNTG-3Ј) or with the consensus USF-binding element (5Ј-GGTCACGTGACC-3Ј) (14,15). USF-binding sites have been found in a number of cellular genes, which are recognized by the USF homodimer, the USF1/USF2 heterodimer, or some other heterodimer in combination with other transcription factors (16,17). However, it is still not clear whether USFs are involved in the activation of gene transcription in response to the Ca 2ϩ signals evoked via Ca 2ϩ influx into cells. Here we demonstrated a unique contribution of USFs to the Ca 2ϩ signal-mediated activation of BDNF-PI, which bind to the DNA sequences overlapped by CRE.
Cell Culture-Primary cultures of rat cortical neurons were prepared from the cerebral cortexes of 17-18-day-old rat (Sprague-Dawley) embryos as described previously (18). Briefly, small pieces of cerebral cortex were dissected by enzymatic (DNase I (Sigma) followed by trypsin (Sigma)) treatment and mechanical dissociation, and the cells were seeded at 5 ϫ 10 6 cells in a 60-mm culture dish (Iwaki). The cells were grown for 48 h in Dulbecco's modified Eagle's medium (Nissui) containing 10% fetal calf serum, and then the medium was replaced with serum-free Dulbecco's modified Eagle's medium containing glucose (4.5 mg/ml), transferrin (5 g/ml), insulin (5 g/ml), sodium selenite (5 g/ml), bovine serum albumin (1 mg/ml), and kanamycin sulfate (100 g/ml) (TIS medium). Cytosine arabinoside (Sigma) was also added at 2 M to prevent the proliferation of glial cells. The medium was replaced with fresh TIS medium, but devoid of cytosine arabinoside, 2 h before DNA transfection.
Reporter Vectors Used for Promoter Analysis and Overexpression Experiments-For the promoter analysis, the luciferase reporter vector, pGL3-BDNF promoter I (pBDNFpI), was constructed by ligation of pGL3-Basic vector (Promega) with the region (Ϫ528 to ϩ138) of rat BDNF gene promoter I. The procedure for plasmid construction was described previously (18). As an internal control vector against the luciferase experimental vector, we used Renilla luciferase vector carrying the human elongation factor 1␣ promoter (pRL-EF1␣) (7). For the overexpression experiments shown in Fig. 9, we used ␤-galactosidase vector carrying the human elongation factor 1␣ promoter (EF1␣-␤-gal) as an internal control because pRL-EF1␣ itself could interfere with the assessment of firefly luciferase gene induction (data not shown).
Expression vectors, A-CREB and A-USF, were generously provided by Dr. C. Vinson (National Cancer Institute, National Institutes of Health, Bethesda, MD).
For the dual (firefly and Renilla) luciferase assay, cell lysates were extracted with passive lysis buffer (Promega) and used as described previously (7). For the measurement of ␤-galactosidase activity as an internal control, cell lysates were extracted with 250 l of cell lysis buffer containing 1 mM potassium phosphate (pH 7.8), 1 mM dithiothreitol, and 0.5% Triton X-100; 30 l of the lysate was used for the chemiluminescence-based ␤-galactosidase assay (CLONTECH), and 20 l for the firefly luciferase assay.
Preparation of Nuclear Extracts and Electrophoretic Mobility Shift Assay (EMSA)-Forty hours after the medium exchange at 3 DIV, cells were stimulated with 25 mM KCl (25 mM KCl stimulation) or vehicle and incubated for 6 h. Then, nuclear extracts were prepared as described with minor modifications (20). EMSAs were carried out using probes corresponding to the BDNF-PI sequence (these are indicated in Fig. 4) as reported previously (21). Briefly, end-labeling of DNA probes was performed at 37°C for 20 min in 10 l of reaction mixture (6.6 mM Tris-HCl, pH 7.4, 50 mM NaCl, 6.6 mM MgCl 2 , 0.5 mM dATP, 0.5 mM dGTP, 0.5 mM dTTP, 1 mM dithiothreitol) containing 200 ng of DNA, 2 l of [ 32 P]dCTP, and 1 unit of Klenow fragments. Then, DNA probes were recovered with a Sephadex G-50 column. The DNA-protein binding reaction was carried out at 25°C for 15 min in 20 l of binding buffer (20 mM HEPES-NaOH, pH 7.9, 80 mM NaCl, 0.3 mM EDTA, 0.2 mM EGTA, 0.2 mM phenylmethanesulfonyl fluoride, 10% glycerol, 2 g of poly(dI-dC), 0.2-0.4 ng of 32 P-labeled DNA probes, and 5 g of nuclear extract). Then, DNA-protein complexes were separated on 4% polyacrylamide gel at 132 V for 2.5 h. The protein concentration was determined by the method of Lowry. For supershift EMSA, a series of antibodies (2 mg/ml) was used at a dilution of 1:20.

Production of BDNF Transcripts Containing Exons I, II, III, and IV Induced via Membrane
Depolarization-Although the complicated structure of the BDNF gene is suggested to give rise to at least eight species of BDNF transcript (4), it has already been reported that the exon III-containing transcript is the major form, the expression of which is controlled by BDNF-PIII in response to the Ca 2ϩ signals evoked via membrane depolarization in rat cortical neurons (8,9). In addition, we have also reported that BDNF-PI is activated by membrane depolarization, the basal promoter activity being ϳ8-fold less than that of BDNF-PIII (7,18). To confirm whether endogenous 5Ј-exon specific BDNF transcripts, particularly those containing exon I, are induced by membrane depolarization or not, we examined the increase in each transcript by RT-PCR. To distinguish each transcript, we designed a forward primer corresponding to exon I, II, III, or IV and a reverse primer corresponding to the common exon V (Fig. 1A). As shown in Fig. 1B, the expression of three transcripts containing exons I, II, and III was up-regulated by the exposure of cells to 25 mM KCl in medium (25 mM KCl stimulation), whereas the expression of the exon IV-containing transcript remained constant even after the stimulation. The minimum cycle numbers required for detecting transcripts by RT-PCR were 32 and 26 cycles for exon I and III, respectively (see the legend of Fig. 1), indicating that the expression level of the exon I-containing transcript is lower than that of the exon III-containing transcript. These findings indicate that BDNF-PI is activated by membrane depolarization in neurons although the expression level of the exon I-containing transcript is quite low, compared with that of the exon III-containing transcript.
Calcium-responsive Elements Located in Two Distinct Regions of BDNF Promoter I-To assign the calcium (Ca 2ϩ )-responsive elements in BDNF-PI, we constructed a series of firefly luciferase reporter vectors containing various lengths of the promoter I region with internal deletions (see Fig. 2) for the measurement of promoter activity. Each vector was transfected with Renilla luciferase control vector driven by EF1␣ promoter (pRL-EF1␣) into rat cortical neurons at 3 DIV. We previously reported that the luciferase activity derived from the vector carrying the region from Ϫ528 to ϩ138 increased in response to the Ca 2ϩ influx through L-VDCC, which was caused by 25 mM KCl stimulation, suggesting the presence of Ca 2ϩ -responsive elements within the region up to nucleotide position Ϫ528 on BDNF-PI (18). As shown in Fig. 2, the luciferase activities of plasmid vectors pBDNFpI(⌬378 -349), -(⌬357-331), -(⌬335-309), -(⌬288 -245), and -(⌬151-114) were nearly as inducible as those of pBDNFpI(full). Although the use of pBDNFpI(⌬488 -344) decreased the luciferase activity, the deletion of the 5Јflanking region up to Ϫ310 (pBDNFpI(Ϫ311 to ϩ138)) did not decrease the luciferase activity, compared with that of pBDN-FpI(full) (data not shown), suggesting that the upstream region EI, EII, EIII, and EIV are specific forward primers corresponding to exons I, II, III, and IV, respectively, and the reverse primer (EV) is located in the coding region. B, rat cortical neurons (5 DIV) were stimulated with 25 mM KCl (25 mM KCl stimulation), and total RNA was isolated at the times indicated (0, 6, and 12 h) and subjected to RT-PCR. The relative intensities of amplified DNA fragment containing each exon were analyzed with an Imaging scanner and plotted. The minimum cycle numbers required for detecting the amplified DNA were 32, 31, 26, and 29 for exons I, II, III and IV, respectively. The data represent the mean Ϯ S.E. from the experiment performed in triplicate, and the same tendency was obtained from at least two separate experiments. up to Ϫ310 is not so largely involved in the Ca 2ϩ signalmediated activation of BDNF-PI. In contrast, the use of plasmid vectors pBDNFpI(⌬345-184), -(⌬377-184), -(⌬377-205), and -(⌬377-223) resulted in a decrease in Ca 2ϩ responsiveness of ϳ40 -60%, suggesting that this region includes the Ca 2ϩresponsive elements. In addition, more marked decreases in Ca 2ϩ responsiveness were observed with the deletion mutants pBDNFpI(⌬107-31) and -(⌬107-65), also suggesting a localization of elements responsible for a strong Ca 2ϩ response in this region. We called these two regions the distal and proximal region, respectively. On the other hand, deletion of DNA sequences from Ϫ173 to Ϫ154 (pBDNFpI(⌬173-154)) rather increased the luciferase activity, indicating that DNA elements which function as silencers could be located within this region.
Critical Role of DNA Sequences Including the CRE-like Element and a Few Upstream Sequences in the Activation of BDNF-PI-As shown in Fig. 2, the proximal region from Ϫ107 to Ϫ65 was extremely effective in the activation of BDNF-PI. Therefore, we focused on the proximal region and investigated which DNA sequences within this region contribute to the activation of BDNF-PI. Using the TFSEARCH program (www. rwcp.or.jp/papia/) (22), we found that this region contained several candidate DNA-binding sites for transcription factors, such as CRE-and 12-O-tetradecanoylphorbol-13-acetate -responsive element (TRE)-like sequences (Fig. 3A). To investigate the role of these sites in the activation of BDNF-PI, we constructed a series of plasmid DNAs having internal mutations and deletions at the sites and transiently transfected them into rat cortical neurons. As shown in Fig. 3B, mutation and deletion of the downstream TRE-like sequence (TRE3m and TRE3⌬) did not affect the luciferase activity, whereas a series of mutants of the CRE-like sequence (CRE2m, CRE2⌬) showed a marked decrease in luciferase activity. Deletion of the middle TRE-like sequence which partially overlapped with CRE (TRE2⌬) resulted in the same decrease as with the CRE2 mutants (CRE2m, CRE2⌬). On the other hand, mutants of the upstream TRE-like sequence (TRE1m, TRE1⌬) were also effective in decreasing the transcriptional activity. However, the mutation combined with the deletion of CRE (TRE1/CRE2⌬) decreased the BDNF-PI activation to almost the same level as CRE2⌬, suggesting a minor contribution of upstream TRE to the activation of BDNF-PI. Among the mutant plasmid vectors constructed, plasmid TRE2/CRE2⌬ with an 11-base pair deletion containing middle TRE-and CRE-like sequences was most effective in reducing the level of BDNF-PI activity, indicating that the DNA containing the CRE-like sequence and a few bases upstream of this sequence are mainly responsible for the activation of BDNF-PI.
Additionally, we constructed a plasmid vector, pBDNFpI (⌬distal/CRE2m), with both a deletion of the distal region (⌬277-223) and mutation of CRE2 (Fig. 3C). Compared with the luciferase activity of wild-type pBDNFpI(full), the activity of ⌬distal was low, which was consistent with the data shown in Fig. 2. On the other hand, ⌬distal/CRE2m showed a greater decrease in luciferase activity than ⌬distal, the level of which was almost the same as that of CRE2m (Fig. 3C). These results indicate that the proximal region, rather than the distal region, mainly contributes to the activation of BDNF-PI, induced by the Ca 2ϩ signals in rat cortical neurons.
Transcription Factors That Bind to the Proximal Region-We next investigated the transcription factors that could bind to the proximal region. To do this, we designed four kinds of native promoter I DNA probes (npIPb-1, -2, -3, and -4) to entirely cover the proximal region, and we performed EMSA (Fig. 4A, a). Among these four kinds of DNA probes, DNA binding activity was obvious with npIPb-2 and -3 probes (Fig.  4A, b). For npIPb-1 and -4 probes, however, shifted bands were only faint. An increase in DNA binding activity induced by 25 mM KCl was observed with npIPb-1, -2, and -3, but the increase obtained with npIPb-3 was very weak. Addition of APV, an antagonist for the N-methyl-D-aspartate receptor (NMDA-R), or nicardipine, an antagonist for L-VDCC, tended to reduce the DNA binding activity of npIPb-1, -2, and -3 probes, but the reduction was very weak with npIPb-3. The doublet of shifted bands were observed only with probe 3.
To investigate whether the binding to npIPb-3 probe involves the CRE-like sequence, we examined the competitive effect of a consensus CRE. As shown in Fig. 4B (b), both the upper and lower bands were gradually erased when an excess of not only the cold npIPb-3 but also consensus CRE competitors was added. The mutation of two bases introduced into the CRE-like sequence (npIPb-3m1) failed to affect the DNA binding (Fig.  4B, a and b). These results indicate that the CRE-like sequence in the proximal region functions as the CRE. The competitors containing a consensus TRE failed to compete with the binding to npIPb-3 probe, indicating that the TRE-like sequence on npIPb-3 was unable to function as the TRE. However, the binding to npIPb-2 probe was reduced by an excess of consensus TRE (data not shown), indicating that the TRE-like sequence on npIPb-2 probe functions as the TRE. These findings indicate that the CRE-like sequence in the proximal region works as the CRE and the proteins binding to npIPb-3 probe were different from those binding to the npIPb-2 probe.
Binding of CREB Family Transcription Factors to the CRE of BDNF Promoter I-To identify the DNA-binding proteins that can bind to the CRE-like sequence of BDNF-PI (CRE-PI), we performed EMSA using npIPb-3 probe, in which antibodies against transcription factors were added prior to the start of the DNA-protein binding reaction. When the anti-CREB family or anti-CREB specific antibody was added to the mixture containing npIPb-3 probe with nuclear extracts prepared from 25 mM KCl-stimulated cortical neurons, supershifted bands were observed but faintly (Fig. 5A, lanes 2 and 3). When anti-ATF1, anti-ATF2, anti-c-Fos, or anti-C/EBP␤ specific antibody was added to the mixture, no shifted band was observed (Fig. 5A,   lanes 4 -7). When the native promoter III DNA probe (npIIIPb) containing CRE-PIII that was previously identified as a calcium-responsive element (CaRE3) (8,9) was used, stronger shifted bands than those with npIPB-3 probe were observed in the presence of anti-CREB family or anti-CREB specific antibody (Fig. 5B, lanes 2 and 3). Anti-phosphorylated CREB specific antibody was also effective in shifting the bands upon EMSA (data not shown). This supershift of bands caused by anti-CREB family or anti-CREB specific antibody was clearly observed with consensus CRE probe (Fig. 5C, lanes 2 and 3). To test further whether the CRE-PI in the proximal region could solely function as the CRE, we used DNA probes, pICRE and pIIICRE, in which the CRE-like sequence of BDNF-PI or -PIII was surrounded by the same sequences as surrounded the consensus CRE. As shown in Fig. 5, D and E (lanes 2 and 3), supershifted bands were observed in the presence of anti-CREB family or anti-CREB specific antibody, in which the supershift of the bands was observed at almost the same level as that of pIIICRE probe. Thus, the CRE-PI in the proximal region of BDNF-PI functions as the CRE by itself, which is able to introduce CREB or CREB family proteins for binding, and its FIG. 5. Analysis of transcription factors binding to the proximal region using supershift EMSA. Nuclear extracts were prepared from the cells stimulated with 25 mM KCl and were subjected to supershift EMSA after DNA-protein binding reaction. Antibodies (0.08 g/ml) against the CREB family, CREB, ATF1, ATF2, c-Fos, C/EBP␤, and normal IgG were added to the reaction mixture before the DNA-protein binding reaction was started at 25°C (see "Experimental Procedures"). S indicates the positions of supershifted bands. The experiments shown in A and B were performed using the DNA probes derived from BDNF promoter I (npIPb-3) and from BDNF promoter III including the CRE of BDNF-PIII (from bp Ϫ36 to Ϫ29) (npIIPb), respectively. Experiments in C-E were performed using the DNA probes containing the consensus CRE (C), and the CRE-like sequence of BDNF-PI (CRE-PI) (D) and of BDNF-PIII (CRE-PIII) (E), surrounded by DNA sequences that were randomly chosen from the nucleotide sequences of pBR322. The bold letters in the DNA probes indicate the position of the consensus CRE, CRE-PI, and CRE-PIII. The experiments were done twice, and the same results were obtained. effectiveness seems to be equal to that of BDNF-PIII. As described above, however, the level of supershift of the bands formed on npIPb-3 probe in the presence of anti-CREB family or anti-CREB antibody appeared to be lower than that formed on native pIII DNA probe (Fig. 5, A and B).
Binding of USF to a Site Overlapping CRE-Searching other binding sites for known transcription factors in the proximal region of BDNF-PI using the data base TFSEARCH, we found that the USF-binding site (5Ј-TGGTCACGTAACTG-3Ј) was located and overlapped the CRE-PI (Fig. 7A). Therefore, we next investigated whether USF could be included in the DNA-protein complex formed on the npIPb-3 probe. As shown in Fig. 6B, addition of anti-USF1 or -2 specific antibody prior to the DNAprotein binding reaction resulted in a complete disappearance of the lower band of doublet (lanes 4 and 5). Furthermore, addition of anti-USF1 but not anti-USF2 antibody partially erased the upper band and slightly shifted its position downward (lane 4). The partial disappearance and downward shift of the upper band were also observed with the consensus CRE probe (Fig. 6C, lane 4), although only the upper band was obtained with this probe. When anti-CREB family antibody was added, supershifted bands appeared faint with npIPb-3 probe but were clear with consensus CRE probe (Fig. 6, B and  C, lanes 2), which was consistent with the results shown in Fig.  5 (A and C). Furthermore, using the npIPb-3 probe or consensus CRE, anti-c-Fos antibody did not give rise to the shifted band (Fig. 6, B and C, lane 3). Thus, the upper band of the doublet formed by npIPb-3 probe, at least in part, consists of CREB family transcription proteins or, at least, CREB and the lower band of the doublet consists of USF1/USF2 only.
Changes in DNA Binding Activities Induced by Mutations of CRE-PI and/or the USF-binding site-Because we found that the CRE-PI and USF-binding site overlapped in the proximal region, we next investigated which DNA sequences are preferentially or commonly used for the binding of CREB family proteins or USF. For this purpose, we introduced several substitutions of nucleotide bases in the CRE-PI and USF-binding site by constructing four mutant DNA probes for EMSA (Fig.  7A). The npIPb-3m1 and -3m4 probes were designed to destroy both CRE-PI and the USF-binding site. On the other hand, the npIPb-3m2 and -3m3 were designed for destruction of the USFbinding site only. Comparing the intensities of the upper and lower bands formed on the npIPb-3 probe, as shown in Fig. 7B, the upper band was found to disappear with npIPb-3m1 and -3m3 but remain with npIPb-3m2 and partially so with -3m4, whereas the lower band tended to remain with npIPb-3m1 and -3m3 but disappear with npIPb-3m2 and -3m4. Mutation of npIPb-3m3 probe, even though it is located outside the CRE-PI, markedly decreased the upper band, the DNA-protein complex of which should include the CREB family. Mutation of npIPb-3m2 probe, introduced at the 5Ј outside of the CRE-PI, selectively removed the lower band but not the upper band. Taken together, the formation of the upper band tended to be dependent upon the 3Ј side sequences, and that of the lower band upon the 5Ј side sequences within this region.
Transcriptional Activity of BDNF Promoter I Carrying the Mutated CRE/USF-binding site-Because substitutions of CRE-PI and/or the USF-binding site affected the binding activity of CREB family proteins and USF (Fig. 7), we constructed plasmid vectors in which the same substitutions were introduced into the proximal region of BDNF-PI and, then, measured transcriptional activity using transient DNA transfection. The luciferase activity of vector pBDNFpI-3m1 (also termed In npIPb-3m1 and -3m4 probes, both CRE-PI and USFbinding sequences were destroyed by mutations. In npIPb-3m2 and -3m3, the USF-binding site but not CRE-PI was lost. B, the cells were treated with 25 mM KCl (ϩ) or vehicle (Ϫ), and the nuclear extracts were prepared for EMSA after DNA-protein binding reaction. Arrows indicate the upper and the lower bands and asterisks nonspecific bands. F indicates free probes. The specific radioactivity of every mutant DNA probe was almost the same (data not shown). The experiments were done twice, and the same results were obtained. pBDNFpI-CRE2m in Fig. 3, A and B) decreased by approximately 50% in terms of Ca 2ϩ responsiveness (Fig. 8), which was consistent with the data shown in Fig. 3. Mutation of npIPb-3m2, which preferentially removed the lower band but not the upper band (Fig. 7B), resulted in a decrease of ϳ20% in Ca 2ϩ responsiveness. In contrast, mutation of npIPb-3m3, by which both the upper and the lower bands were erased (Fig. 7B), resulted in the greatest decrease in Ca 2ϩ responsiveness. The luciferase activity of pBDNFpI-3m4 was almost the same as that of pBDNFpI-3m1, whose mutation resulted in the disappearance of both the upper and lower bands but the upper band tended to be stronger than that for mutation of npIPb-3m3. Thus, the Ca 2ϩ responsiveness of four reporter mutant vectors showed a good correspondence with the preferential binding of DNA probes into which the same mutation was introduced (Figs. 7 and 8), indicating an involvement of the transcription factors that preferentially bind to CRE-PI and the USF-binding site in the activation of BDNF-PI.
Effect of Dominant Negative CREB or USF on the Ca 2ϩ Responsiveness of BDNF-PI-To address further the crucial role of CREB family proteins and USF in the Ca 2ϩ signalmediated activation of BDNF-PI, mutant plasmid vectors of dominant negative CREB and USF were introduced into rat cortical neurons with a reporter vector, pBDNFpI(full). Instead of the Renilla luciferase vector, we used another control vector carrying an EF1␣ promoter-driven Escherichia coli ␤-galactosidase gene. Overexpression of CREB M1, a mutant in which the serine-133 residue was converted to alanine (23), inhibited the activation of BDNF promoter I up to ϳ50% of the control transfected with empty vector (Fig. 9A). Overexpression of A-CREB, a mutant with an acidic extension at the N terminus of the CREB leucine zipper domain that could interfere with the binding of DNA and transcriptional activation of wild-type CREB (24), or K-CREB, a mutant with a substitution in the DNA-binding domain that could inhibit the binding of wildtype CREB (25), decreased the activation of BDNF-PI by ϳ20% (Fig. 9, B and C). On the other hand, overexpression of A-USF, a mutant with an acidic extension at the N terminus of USF (26), resulted in a marked reduction of BDNF-PI activity up to ϳ60% of the control (Fig. 9D). Combined DNA transfection of CREB M1 and A-USF, however, tended to induce the death of cortical neurons transfected (data not shown).

DISCUSSION
Prior to the characterization of the molecular mechanisms for the Ca 2ϩ responsiveness of BDNF gene promoter I (BDNF-PI), we first demonstrated using RT-PCR that not only the expression of exon III-containing transcript but also that of the exon I-containing transcript was induced by membrane depolarization, although the basal expression level of the latter seems to be quite low, compared with that of the former (Fig. 1). This observation indicates that endogenous BDNF-PI in genome DNA responds to Ca 2ϩ signals to increase the expression of exon I-containing transcript in rat cortical neurons. By promoter assay using transient DNA transfection in cultures of rat cortical neurons, in addition, we demonstrated that Ca 2ϩ -responsive elements of BDNF-PI were located in two regions, distal and proximal (Fig. 2). Detailed analysis of the proximal region with deletions or mutations revealed that the CRE-like sequence in the proximal region of BDNF-PI (CRE-PI) functioned as the CRE (Figs. 4 and 5) and was potentially required for the activation of BDNF-PI (Figs. 2 and 3). Besides this CRE-PI, the TRE-like sequence can work as the TRE to bind the AP1 transcription factor because the DNA-protein complexes formed on native promoter I DNA probe-2 (npIPb-2) for EMSA, in which the TRE-like sequence was centered (Fig. 4A), included c-Fos (data not shown). Because this TRE-like sequence partially overlapped not only the CRE-PI but also the USF-binding site, however, the TRE-like sequence in native promoter I DNA probe-3 (npIPb-3) does not seem to work as a TRE but rather as a part of the CRE-PI and USF-binding site. On the other hand, the upstream TRE-like sequence seems to be responsible for the activation of BDNF-PI ( Fig. 3B) but would play a minor role because the mutant of the upstream TRE-like sequence combined with CRE-PI (TRE1/CRE2⌬) had the same effect as the mutant of CRE-PI alone (CRE2⌬) (Fig.  3C). The downstream TRE-like sequence was ineffective in the activation (Fig. 3B). With respect to the distal region, the elements involved in the Ca 2ϩ responsiveness of BDNF-PI have not been determined. A search for these elements using a series of plasmid vectors with a deletion or mutation in the distal region failed to specify their location because the Ca 2ϩ responsiveness resulting from the distal region was dispersed by every mutation and deletion (data not shown). As shown in Fig. 2, in addition, the mutant plasmid with both the deletion of the distal region and the mutation of CRE-PI (⌬distal/ CRE2m) showed the same level of activation as that having the mutation in CRE-PI (CRE2m) alone (Fig. 3C). These findings indicate that the proximal region could play a major role in activating BDNF-PI, whereas the distal region has a minor role.
Focusing on the proximal region, we have identified the transcription factors that could bind to CRE-PI. As shown in Figs. 4 -6, a doublet of shifted bands, upper and lower, were detected by EMSA using npIPb-3 probe, whereas only a single band corresponding to the upper band of the doublet was detected with the consensus CRE, pICRE, and pIIICRE probes (Figs. 5 and 6). The DNA binding activity responsible for the upper band was susceptible to consensus CRE (Fig. 4B), and FIG. 8. Effect of mutation within CRE-PI and/or the USF-binding site on the activation of BDNF promoter I. We constructed reporter vectors carrying the same substitutions in CRE-PI and the USF-binding site as were introduced into the DNA probes for EMSA (Fig. 7) and transfected them into rat cortical neurons for measuring transcriptional activation. Cells were stimulated with 25 mM KCl (ϩ) or vehicle (Ϫ). Transcriptional activation was indicated as a -fold increase of the control in which pBDNFpI(full) was transfected and the 25 mM KCl stimulation was omitted. The nucleotide sequences of pBDNFpI, pI-3m1, pI-3m2, pI-3m3, and pI-3m4 were shown in Fig. 7A. The data represent the mean Ϯ S.E. from experiments performed in triplicate. The same tendency was obtained from two independent experiments. the supershift of the upper band was caused by anti-CREB family or anti-CREB specific antibody, which was faintly observed (Figs. 5A and 6B). In addition, mutations introduced into CRE-PI in the proximal region (Fig. 8) and the overexpression of dominant negative CREB M1 decreased the level of activation of BDNF-PI (Fig. 9). Thus, it is evident that the binding of CREB to CRE-PI, at least in part, contributes to the activation of BDNF-PI. In addition, the fact that the three adjacent downstream bases outside the CRE-PI in the proximal region were required for the upper band to form on the npIPb-3 probe (Fig. 7) but not on the consensus CRE, pICRE, or pII-ICRE probe (Figs. 5 and 6), which were effective in the tran-scriptional activation of BDNF-PI (Fig. 8), may suggest that different transcription factors as well as CREB bind to npIPb-3 probe and the bases adjacent to the CRE-PI are required for their binding. Thus, the DNA sequences surrounding CRE-PI seem to affect the DNA binding of CREB and other transcription factors, which may confer some differences in the CREB binding to CRE between BDNF-PI and -PIII.
On the other hand, the lower band of the doublet was completely diminished by anti-USF1 or -USF2 antibody (Fig. 6B), indicating that it was formed only by the binding of USF1 and USF2 to the npIPb-3 probe. In support of this, the USF-binding site (5Ј-TGGTCACGTAACTG-3Ј) covered the CRE-PI (5Ј- , and A-USF (D) were co-transfected with firefly luciferase reporter vectors carrying BDNF promoter I and E. coli ␤-galactosidase internal control vectors carrying EF1␣ promoter into rat cortical neurons. Empty vectors were co-transfected as a control. Forty hours after transfection, cells were stimulated with 25 mM KCl (ϩ) or vehicle (Ϫ) and incubated for another 6 h. Cell lysates were then extracted for measuring luciferase activity. The value for the transcriptional activation of pBDNFpI(full) obtained with 25 mM KCl was standardized to 100%. The variability in the basal transcriptional activities is probably caused by an experimental flexibility, which could be caused mostly by culturing conditions. The data represent the mean Ϯ S.E. from experiments performed in triplicate. The same result was obtained from three independent experiments. TCACGTAA-3Ј) (Fig. 7A). These overlapping sequences are conserved in the rat, human, and mouse BDNF promoters I (data not shown). Some mutations within the USF-binding site preferentially diminished the lower band (Fig. 7), which also decreased the level of activation of BDNF-PI (Fig. 8). In addition, the overexpression of dominant negative A-USF was effective in reducing the transcriptional activation, the level of which was lower than that of CREB M1 (Fig. 9, A and D). These findings indicate that USF1 and/or USF2 are involved in the transcriptional activation of BDNF-PI through their bindings to the USF-binding site.
USF1 and USF2 are members of the basic helix-loop-helix/ leucine zipper family of transcription factors and bind to DNA as homodimers or heterodimers (16). In addition, differential splicing has been shown to give rise to at least two isoforms, the 44-kDa USF2a and the 38-kDa USF2b (27,28). Therefore, the complicated formations of homodimer or heterodimer among these transcription factors are involved in the complex formation via the USF-binding site. USF1 has been reported to interact with other transcription factors, such as Fos-related antigen 1 (Fra1) (17) and transcription factor II-I (29). On the other hand, CREB is a member of the leucine zipper family that is also able to form a homodimer or heterodimer with ATF/ CREB family members or other factors at the leucine zipper motif (30,31). Although there are several cases in which the CRE and the USF-binding site (or E-box) neighbor each other (32,33), BDNF-PI is unique because the CRE and the USFbinding site overlap. Nevertheless, a tendency was observed for the 5Ј-side DNA sequences within the USF-binding site to be preferentially used for the formation of the lower band upon EMSA, whereas the 3Ј-side sequences were used for that of the upper band, i.e. a preferential usage of the DNA sequences within the USF-binding site for USF1/USF2 and CREB. These observations may also support the binding of a heterodimer between CREB and USF to this overlapping site, which should be investigated.
Overexpression of the dominant negative CREB and USF mutants, CREB M1 and A-USF, resulted in a marked reduction of BDNF-PI activation, indicating an involvement of CREB and USF in the activation of BDNF-PI (Fig. 9, A and D). Because CREB M1 binds to CRE but is devoid of phosphorylation at Ser-133, the dominant negative effect of CREB M1 seems to be caused by an interference in the interaction between CREB and coactivators like CREB-binding protein (CBP) on the promoter. All three types of dominant negative CREB mutants also reduced the activation of BDNF-PIII (data not shown).
As Ca 2ϩ signal-stimulated transcription factors other than CREB, MEF2 and NF-AT have already been found in neuronal as well as non-neuronal cells (34,35), and, quite recently, calcium-responsive transcription factor has been detected in neuronal cells as a calcium-and neuron-selective transcription factor (11). As described in the present study, it is highly possible that USF1/USF2 also functions as a Ca 2ϩ signal-stimulated transcription factor in neurons. It has been reported that the phosphorylation of USF1 enhanced its DNA binding activity (36); quite recently, Galibert et al. (37) reported that phosphorylation of USF1 at Thr-153 increased its transcriptional activity. In addition, the activation of gene transcription by USF1/USF2 could be mediated by CBP/p300 (38). Thus, it is very likely that USF1/USF2 is phosphorylated by Ca 2ϩ signaling pathways, such as extracellular signal-regulated kinase/ mitogen-activated protein kinase, CaM (calmodulin) kinase IV and adenylyl cyclase/cAMP-dependent protein kinase (39), and involved in the transcriptional activation through mediation of CBP. For full activation of the transcription of the mouse renin gene, co-operation between CREB/CREM and USF1/USF2 is required, in which the CRE-like element and E-box, for the binding of CREB and USF1/USF2, respectively, neighbor each other with an interval of ϳ10 bases (33). In BDNF-PI, however, the CRE-like sequence and the USF-binding site overlapped, suggesting a different mechanism for transcriptional activation of BDNF-PI. Rather than co-operation by these transcription factors for promoter activation, a competitive binding of CREB and USF1/USF2 to this overlapping site could be considered for controlling the BDNF-PI activation in accordance with the Ca 2ϩ signalings evoked in neurons. Differential usage of CREB or USF1/USF2 for promoter activation in response to Ca 2ϩ signals might be controlled by Ca 2ϩ signaling pathways among which extracellular signal-regulated kinase/mitogen-activated protein kinase, CaM kinase IV, and adenylyl cyclase/cAMP-dependent protein kinase are activated by the influx of Ca 2ϩ into neurons via the NMDA-R or L-VDCC. it is important to elucidate the molecular mechanisms of the CREB-and USF-mediated activation of BDNF-PI evoked via Ca 2ϩ signals in neurons for a better understanding of the intracellular mechanisms of neuronal activity-dependent gene expression, which might contribute to neuronal survival and plasticity.