Repression of Human Fibroblast Growth Factor 2 by a Novel Transcription Factor*

Here we describe the cloning of the regulator of fibroblast growth factor 2 (FGF-2) transcription (RFT) using a yeast one-hybrid screening with a defined motif in FGF-2 promoter as a target sequence. Overexpression of human RFT (RFT-A) reduces FGF-2 RNA and protein levels in both normal and tumor cell lines. Its splice variants, RFT-A′ and RFT-B, have deletions in the putative DNA binding domain and fail to bind FGF-2 promoter and repress FGF-2 gene expression. The ratios of RFT isoforms differ between normal and tumor cells, with the splice variants dominating in tumor cells. Overexpression of RFT-A induces glioma cell death. Our data suggest that regulation of FGF-2 by RFT is important for cellular functions and may be impaired in certain tumors.

Fibroblast growth factor 2 (FGF-2) 1 is a mitogen for neuroectoderm-and mesoderm-derived cells (1) as well as a potent angiogenic factor in vivo (2). FGF-2 acts in an autocrine or a paracrine manner, interacting with its specific receptors (3), and has been shown to play an important role in limb bud formation (4) and neural tissue formation (5). In addition, a role for FGF-2 has been suggested for brain diseases such as tumors (5-7), Alzheimer's disease (8), and Huntington's disease (9). Ninety percent of gliomas overexpress FGF-2 constitutively, and FGF-2 may play a crucial role in tumorigenesis and tumor progression of gliomas by acting through an autocrine mechanism (5)(6)(7). Although a clear case has been made for the involvement of FGF-2 in development and diseases, little is known about the regulation of FGF-2 gene expression at the transcriptional level (10 -15). Many factors have been shown to stimulate FGF-2 expression. In adrenal chromaffin cells, stimulation of acetylcholine nicotinic receptors, or angiotensin II receptors, or direct stimulation of adenylate cyclase (with forskolin) or protein kinase C (with phorbolester) increases the level of all FGF-2 isoforms (13). In primary astrocytes, endothelin-3 increases and natriuretic peptide decreases FGF-2 expression through modulating immediate early gene, egr-1 (10). FGF-2 has also been shown to function in both paracrine and autocrine manners, such that positive feedback may be the basis for neoplastic transformation (16).
FGF-2 promoter contains no TATA box but has multiple GC-rich regions for transcriptional initiation. The promoter contains putative sites for several transcription factors such as SpI and Egr-1 (10,15). The homeodomain protein, HOXB7, was shown to bind a ϩ130 to ϩ159 sequence and increase FGF-2 expression (11), and Egr-1 was shown to bind to human FGF-2 promoter at two sites (Ϫ160 and Ϫ60). Interestingly, the expression of FGF-2 in human astrocytes was shown to be inhibited by direct cell contact, and the promoter regions related to such density-dependent regulation have been localized to two regions (Ϫ650 to Ϫ513; Ϫ273 to ϩ314). A transactivator was shown to bind to the upstream site (Ϫ650 to Ϫ513) at low cell density but not at high density. In transformed glioma cells, such regulation is abolished, and the putative activator binds to the promoter in a density-independent manner (12). The density-regulated FGF-2 expression may explain some differing results obtained from different laboratories studying FGF-2 function and regulation. Because FGF-2 is a potent mitogen, the negative regulation of transcription could likely be an important regulatory mechanism for FGF-2 function; however, to date only the tumor suppressor, p53 protein, has been suggested as a repressor of FGF-2 expression (14).
Here we report the cloning and characterization of a novel transcription factor that represses FGF-2 expression at the transcriptional level. Our data indicate that loss of such regulation may be involved in FGF-2-dependent tumor progression.

EXPERIMENTAL PROCEDURES
Yeast One-hybrid Screening, Cloning Full-length cDNAs, and Northern Hybridization-Yeast one-hybrid screening was conducted as described in the manufacturer's manual (CLONTECH) using human testis cDNA library (17). The three "gccgaaccgccgaac" tandem repeat was subcloned into pHISi vector as a reporter plasmid. Full-length RFT cDNAs were cloned by screening human testis cDNA library (CLON-TECH) with the radiolabeled cDNAs discovered by yeast one-hybrid screening. All cloned cDNAs were sequenced by an automated sequencer in the DNA sequencing facility at the Salk Institute. Multiple tissue Northern blot membranes (CLONTECH) were hybridized with the same radiolabeled cDNA probes as above.
Plasmid Construction, Cell Culture, DNA Transfection, Luciferase Assays, and ␤-Galactosidase Assays-The His-tagged RFT isoforms were generated by polymerase chain reactions and subcloned into pCDNA3 mammalian expression vector (Invitrogen). RFT-␦A was generated by digesting RFT-A with AvaI, which deleted the sequence encoding the first 111 amino acids. All constructs were sequenced. To generate FGF-2 promoter-luciferase reporter construct (pGL1.2F), 1.2 kb of FGF-2 promoter (18) was subcloned into pGL-2-basic vector (Promega). U87MG (ATCC) and 293 cells (19) were cultured in Dulbecco's modified Eagle's medium with 10% fetal calf serum in a 5% CO 2 incubator. For DNA transfection, the calcium phosphate method (14) was used for 293 cells and SuperFectant method (Qiagen) was used for U87MG cells. Briefly, 1 ϫ 10 5 cells were cotransfected with 0.9 g each of RFT expression vector, 0.9 g of pGL1.2F, and 0.2 g of H-RAS-␤galactosidase (20) as a transfection internal control. Luciferase assays and ␤-galactosidase assays were done as described previously (14). The assays were performed at least three times independently.
Immunocytochemistry-Immunocytochemistry was performed to detect overexpressed RFT isoforms and endogenous FGF-2 using anti-His tag monoclonal antibody (1:1000, Qiagen) and anti-human FGF-2 monoclonal antibody (1:20, R&D Systems), respectively. Cy3-conjugated donkey anti-rabbit IgG (1:250, Jackson Immunochemicals) for anti-His antibody and fluorescein isothiocyanate-conjugated donkey anti-goat IgG (1:250, Jackson Immunochemicals) for FGF-2 were used as secondary antibodies. The green fluorescent protein (GFP)-infected cells were visualized by green autofluorescence and stained with the anti-human FGF-2 primary antibody. In this case, Cy3-conjugated donkey anti-goat IgG was used as the secondary antibody to detect FGF-2. The image was analyzed by a confocal microscope and color modifications were made in Adobe Photoshop 5.0 as follows: His-tagged RFT shown in figures as blue, FGF-2 in red, and GFP in green.
In Vitro Translations and Gel Mobility Shift Assays-For in vitro translations, RFT expression vectors were incubated with [ 35 S]methionine in the rabbit reticulocyte lysate as described in the manufacturer's protocol (Promega). Translated products were separated on an 8% SDS-acrylamide gel and quantified by a PhosphorImager (Molecular Dynamics). The result was adjusted based on the numbers of methionine content in each of the RFT isoforms. For electrophoretic mobility shift assay, the same molar amount of each product was incubated with radiolabeled oligonucleotides (20,000 cpm) in the buffer containing 10 mM Hepes, pH 7.5, 50 mM KCl, 1 mM dithiothreitol, 0.1% Nonidet P-40, and 3.75% glycerol. The oligonucleotide containing RFT binding sequence is gccgaaccgccgaacgccgaaccgccgaacgccgaaccgccgaac. The same unlabeled oligo was used as a specific competitor in the assay. A agcttcgatcgcgataaggatttatccttatccccatcctcga oligo derived mainly from FGF-2 promoter (Ϫ574 to Ϫ597) was used as a nonspecific competitor.
Ribonuclease Protection Assay (RPA)-The partial RFT cDNA obtained through the yeast one-hybrid screening was subcloned into the pBluescript vector (Stratagene). In vitro transcription of the antisense probes and RPA was performed as described in the manufacturer's protocol (Ambion). Briefly, 100 ng of poly(A) ϩ RNA from human testis (CLONTECH) and 50 g total RNA from U87MG and 293 cells were hybridized with gel-purified probes (50,000 cpm). The protected bands were loaded on a 5% native gel as double-strand RNAs with the radiolabeled 123 bp DNA ladder (Life Technologies, Inc.). The results were analyzed by a phosphoimager, and RFT A/AЈ/B ratios were calculated with the adjustment of the numbers of [ 32 P]UTP in the protected bands.
Adeno-associated Virus (AAV), Reverse Transcriptase Polymerase Chain Reaction (PCR), and TdT-mediated dUTP-X Nick End-labeling (TUNEL)-Each RFT isoform was subcloned into pBK1AAV, which is an AAV multiple cloning vector based on the cytomegalovirus immediate early enhancer and promoter (21). Recombinant AAV vectors were produced by a modified transient plasmid transfection protocol followed by two rounds of cesium chloride equilibrium density gradients and heat treatment at 56°C for 60 min to destroy residual adenovirus (21,22). The titer of the recombinant virus was determined by DNA dot blot and antibody staining against the His-tag region of the inserted protein.
As a control, AAV carrying GFP (CLONTECH) was produced. To analyze the level of endogenous FGF-2 message, cells were harvested 40 h after infection with RFT-AAV at the multiplicity of infection of 25. Total RNA were extracted using RNAzol B (TEL-TEST). One hundred ng of total RNA was reverse transcribed by avian myeloblastosis virus reverse transcriptase (Promega), and PCR reactions were done using Taq polymerase (Promega). Linear ranges of PCR were determined by sampling the PCR products at every two cycles. The data shown were sampled at 25 cycles for FGF-2 and 23 cycles for G3PDH. The primer sets for FGF-2 and G3PDH were described previously (11). To analyze the cellular levels of expressed RFT and endogenous FGF-2 protein, cells were fixed at 72 h postinfection and analyzed by immunocytochemistry. The TUNEL staining was performed as described in the manufacturer's protocol (Boehringer Mannheim).

RESULTS
Cloning RFT and Structural Analysis-Human FGF-2 gene has a TATA-less promoter, and its transcript is initiated at only one site (18). We further noticed that a gccgaac sequence is tandemly repeated at both sides of the transcription initiation site of the human FGF-2 gene (Fig. 1A). Using this DNA sequence as bait, we conducted a yeast one-hybrid screening (17). We screened 7 ϫ 10 6 cDNA clones generated from human testis poly(A) ϩ RNA. We retrieved two partial cDNA clones that were 447 and 405 bp. Both clones encode an identical cysteine (Cys)rich region that is a putative zinc finger domain (Fig. 1B).
Because this novel clone may encode a DNA binding protein that binds to FGF-2 promoter, we named this gene the regulator of FGF-2 transcription (RFT). We screened a human testis cDNA library (CLONTECH) using the partial clones as probes. From 12 positive clones isolated, we have obtained full-length clones (RFT-A), and two shorter isoforms, RFT-AЈ and RFT-B. RFT-AЈ has a single amino acid deletion at lysine-407 and RFT-B has a deletion of 58 amino acids in the putative DNA binding domain (Fig. 1, B and C). The open reading frame is 1893 nucleotides in RFT-A, 1890 in RFT-AЈ, and 1725 in RFT-B, which encode proteins that are 632, 621, and 574 amino acids, respectively. Based on the genomic DNA sequence (data not shown), RFT-AЈ is probably formed by a differential splicing mechanism and RFT-B is formed by an alternative splicing mechanism of this exon.
A structural analysis in Fig. 1C shows that the RFT-A and RFT-AЈ forms contain two Cys-rich regions. The N-terminal Cys-rich region shows 50% similarity to the human trithorax gene (HRX) (also called mixed lineage leukemia (MLL) (23) or acute lymphoblastic leukemia (ALL-1) (24) genes). The C-ter- minal Cys-rich region discovered by the yeast one-hybrid screening, a putative zinc finger domain, shows 50% similarity to the HRX protein and DNA methyltransferase genes (25). The function of the similar region in the HRX gene is not known. The electric charges of the two Cys-rich regions are different, with the N-terminal Cys-rich region being negatively charged and the C-terminal Cys-rich region being positively charged, implicating the C-terminal Cys-rich region as a DNA binding domain. Nine Ser/Thr-Pro-Xaa-Xaa (S/TPXX) motifs are present in all variants of RFT. S/TPXX motifs reportedly are to be found more frequently in gene regulatory proteins and are located on either side of DNA recognizing units (26). In addition, two putative nuclear localizing signals are present near the splicing region.
Expression and Localization of RFT-Northern blot analysis of human multiple tissue shows that the gene is expressed ubiquitously (Fig. 2A). The size of the transcripts on the Northern blot analysis is about 3.0 kb in all tissues, with an approximately 2.6-kb transcript being present in many tissues ( Fig.  2A). We performed 3Ј-rapid amplification of cDNA ends method using RFT-specific primers and human testis poly(A) RNA. The results indicate that the difference in size of these transcripts was attributable to an alternative splicing in the 3Ј-untranslated region, suggesting a potential instability of mRNA (data not shown).
In vitro translated products of RFT-A, RFT-AЈ, and RFT-B forms migrate at the molecular weight of approximately 90, 90, and 75 kDa, respectively, on a denaturing gel (data not shown).
To determine the localization of each form of RFT in cells, we constructed His-tagged expression vectors of each form and transfected them into the human glioma cell line, U87MG (ATCC), and the human kidney tumor cell line, 293 (18). After staining by anti-His tag antibody, a confocal microscopic analysis study showed clear nuclear localization of all forms in each cell line (Fig. 2B).
Next, to determine the affinity specificity of each form to the gccgaac sequence, we performed gel mobility shift assays. The RFT-A form shows highly specific affinity to gccgaac sequence, and the RFT-AЈ and RFT-B forms show undetectable affinity to the same sequence (Fig. 3A). Thus, the structural examination, the localization study, and the DNA binding assay support the suggestion that the RFT gene product is a nuclear protein, and it may function as a transcription factor.
RFT-A Represses FGF-2 Expression-To investigate the function of the RFT gene, we performed co-transfection using the expression vectors for each RFT isoform and pGL1.2 F carrying 1.2-kb fragment of FGF-2 promoter fused with luciferase gene as a reporter construct into the U87MG and 293 cells. As an internal control, H-Ras ␤-galactosidase (19) was also transfected. After 40 h, luciferase and ␤-galactosidase assays revealed that, in the U87MG cell lines, the RFT-A form repressed the 1.2 kb of FGF-2 promoter activity more than 10-fold; in the 293 cell line, RFT-A nearly completely repressed the 1.2 kb of FGF-2 promoter activity. By contrast, RFT-AЈ and RFT-B did not repress but rather induced a small activation of the FGF-2 promoter in the 293 cell line and, to a lesser extent, in the U87MG cell line (Fig. 3B). In both cell lines, a dose-dependent effect of all forms of the RFT was observed (data not shown). To investigate if the transcriptional repression of RFT-A is specific for FGF-2 promoter, we did co-transfection using RFT expression vectors and epidermal growth factor receptor promoter-luciferase construct. RFT-A does not affect epidermal growth factor receptor promoter activity that does not have the gccgaac motif (data not shown).
To determine whether the RFT gene product could regulate the endogenous FGF-2 gene expression, reverse transcriptase PCR in the linear range was performed on the two cell lines after the cells were infected with an engineered AAV carrying each form of the RFT gene. The results were consistent with those of luciferase assay, demonstrating that RFT-A repressed FGF-2 expression, whereas neither RFT-AЈ nor RFT-B had any effect on FGF-2 (Fig. 3C).
To determine whether FGF-2 protein level is also decreased upon overexpression of RFT-A, we infected U87MG cells with AAV carrying His-tagged RFT-A and RFT-B (Fig. 4). The expression of RFT was detected by anti-His antibodies. The cellular FGF-2 level was shown by using anti-FGF-2 antibodies. RFT-A infected cells have undetectable levels of FGF-2, whereas the RFT-B infected cells can coexpress RFT-B and FGF-2 in the same cell. The absence of FGF-2 staining in some cells that stain for RFT-B may reflect the fact that not all U87MG cells express the same level of FGF-2. Control cells were also infected with AAV carrying green fluorescence protein (GFP-AAV), which does not have an obvious effect on endogenous FGF-2 expression. To test if RFT also represses endogenous FGF-2 protein level in untransformed cells, we did parallel experiments on primary human embryonic fibroblasts (HEF). The results from HEF cells are consistent with those from U87MG cells (Fig. 4).
During preparation of this manuscript, PCM-1 from human retinal cDNA library was computationally cloned (27). The sequence of this gene suggests it is a tissue-specific, alternatively spliced isoform of the RFT-A because we did not clone PCM-1 from human testis library. Both RFT and PCM-1 contain a methylated CpG binding domain (Fig. 1C) that can repress transcription of a methylated promoter in vitro. To determine whether the FGF-2 repressor function of RFT is independent of its CpG binding domain, we truncated the first 111 amino acids of RFT-A, which contain the homologous methylated CpG binding region of RFT-A (RFT-␦A), and co-transfected RFT-␦A with FGF-2 promoter construct into cells. The results reveal that truncated RFT-␦A repressed FGF-2 promoter activity to the same level as that of wild type RFT-A (data not shown). RFT-␦A also reduces FGF-2 protein level in both U87MG and HEF cells (Fig. 4). Therefore, the FGF-2 promoter repression function of RFT appears to be independent of its potential methylated-CpG binding function.
RFT-A Induces Glioma Cell Death-Because the majority of human gliomas overexpress FGF-2, we hypothesize that glioma cells may not have functional RFT. To determine the ratio of endogenous RFT isoforms in glioma cells, we conducted a ribonuclease protection assay using antisense riboprobes to the DNA binding region of RFT-A. In human testis, RFT-A form is dominant, and the RFT A/AЈ/B ratio is 20/1/1. In the 293 cell line, the RFT A/AЈ/B ratio is 1/2/4, and in the U87MG cell line, the RFT A/AЈ/B ratio is 1/9/44 (Fig. 5A). Quantitative comparison between each expression pattern may be difficult because the cells are derived from different tissue, but clearly we can tell that the ratio of RFT-A is much higher in normal tissue than in transformed tumor cell lines.
A neutralizing antibody against FGF-2 has been reported to induce apoptosis in glioma cell lines (28). If RFT-A is acting by suppressing FGF-2, and this suppression is lethal to gliomas, then overexpression of RFT-A should also kill tumor cells that overexpress FGF-2. Specifically, we hypothesized that infection of AAV carrying RFT-A (AAV-A) into the U87MG cells would repress FGF-2 gene expression, resulting in cell death. To test this hypothesis, we infected AAV carrying each form of the RFT gene (AAV-A, AAV-AЈ, and AAV-B) into U87MG cells. AAV carrying the green fluorescence protein gene (CLONTECH) (AAV-GFP) was infected as a control. Fig. 5B shows that only the RFT-A form actually induced U87MG cell death 96 h after infection. Interestingly, RFT-B overexpression did not cause cell death, but cells lost contact inhibition and started to grow in clumps. We subsequently checked two additional glioma cell lines (U251MG and U373MG) and obtained the same results (data not shown). We also infected cells with AAV carrying RFT-␦A (AAV-␦A) into glioma cells, and it also induced cell death, as did the wild type RFT-A (data not shown). This further proves that the reported functions of PCM-1 and RFT are independent.

DISCUSSION
In search of the mechanism that regulates FGF-2 expression, we have observed a tandem repeat sequence (gccgaac) around the transcription initiation site of FGF-2 promoter. Using this sequence as bait, we were able to clone a novel transcription factor, RFT, that represses FGF-2 expression in transient transfection assay and in cells. This is the first negative regulator for FGF-2 that has been shown to bind a definitive sequence in FGF-2 promoter. Given the significance of FGF-2 in development, neurological diseases, and tumor formation, dis- covery and understanding of such a factor bring us closer to the understanding of FGF-2-dependent cellular processes.
RFT Represses the Expression of FGF-2-Our data have shown that RFT can bind to the specific gccgaac sequence and reduces the promoter activity and endogenous RNA (Fig. 3) and protein (Fig. 4) levels of FGF-2. Taking these results together with the nuclear localization of each RFT form (Fig. 3A), we conclude that RFT-A is a transcription factor that can bind to the basal core promoter of the FGF-2 gene and repress the promoter activity. The working hypothesis is that RFT binds to the gccgaaccgccgaac sequence and blocks the binding of RNA polymerase II to the transcriptional initiation site. How such repression is regulated under physiological conditions is unknown.
Splice Variants of RFT-Alternative splicing, which happens in 1 of 20 genes, plays important roles in diversifying cellular gene regulation. Splice variants often have different expression patterns and carry different physiological functions (29 -31). In several well characterized cases, splice variants either serve as dominant negative regulators by competing with the same effector or serve as traps by forming a dimer with the active form. For example, ITF-2, a basic helix-loop-helix protein, forms an active dimer with MyoD, but its splice variant ITF2B inhibits the function of ITF-2 by forming an inactive dimer with MyoD (32). ATFa, an ATF/CREM family transcription factor, has a splice variant, ATFa-O, that inhibits ATFa function by forming heterodimers (33).
The splice variants of RFT, RFT-AЈ, and RFT-B contain defective DNA binding domains and fail to bind the gccgaac sequence and repress FGF-2 transcription. This lack of function is attributable to the mutations in their DNA binding domain and is likely caused by alternative splicing but not to the disruption of the nuclear localization signal. So far, the functions of RFT-AЈ and B are not fully understood. Because the RFT binding sequence is a tandem repeat, RFT-A may form a dimer to bind to FGF-2 promoter. In such a case, RFT-AЈ and B may inhibit RFT-A function by forming a heterodimer, as in the case of ATFa. In Fig. 3, B and C, RFT-AЈ and B induce a slight activation in both U87MG and 293 cells, but there is a difference in the magnitude of responsiveness for RFT-AЈ and RFT-B between the U87MG and 293 cell lines. We hypothesize that this difference might be because of the difference in endogenous levels of each RFT isoform between the two cell lines. Our interpretation is that high endogenous RFT-AЈ and RFT-B levels in U87MG cells have a more dominant negative effect on cellular RFT-A; therefore, exogenous RFT-AЈ and B in U87MG cells are less effective than in 293 cells. When we measured the protein level of FGF-2 after overexpression of RFT, we saw that FGF-2 level is nondetectable in RFT-A-infected cells. However, in RFT-B-infected cells, some cells continue to show a high level expression of FGF-2 while they express high levels of RFT-B. These results indicate that, unlike RFT-A, RFT-B overexpression is not incompatible with FGF-2 expression.
It appears that, at least proportionally, transformed glial cells express more mutant forms of RFT (AЈ and B) than the functional form (A) (Fig. 5). We do not know if this is the cause of the transformation or a middle step or the result of tumor progression. Our data suggest that these splice variants could be either dominant negative regulators of RFT-A or by-products of mistaken alternative splicing. We are currently conducting experiments to investigate this issue.
RFT-A Causes Glioma Cell Death-Most human gliomas are characterized by high level expression of FGF-2. However, it is not known if overexpression of FGF-2 is the direct cause of tumorigenesis and progression. Previous data have shown that anti-FGF-2 neutralizing antibodies cause glioma cell apoptosis (28).
In our studies, introducing high level RFT-A carried by AAV into glioma cells also decreases intracellular FGF-2 level and causes cell death. The positive TUNEL staining suggests that the cells may undergo apoptosis, but at present we are exploring the mechanism of cell death further. Interestingly, overexpression of RFT-B, which is a dominant form of RFT in human glioma, not only did not cause cell death but also increased cell growth into proliferative aggregates of cells that piled upon each other (Fig. 5B). Such an effect may be because of the activating function of RFT-B as shown in transient transfection assay (Fig. 3B). Moffet et al. (12) observed that factors in human astrocytes bind to FGF-2 promoter in a cell density-dependent manner and induce contact inhibition. The function of RFT suggests that it may be one of the candidates for such factors. Therefore, our data suggest that even though FGF-2 may not be the original cause of tumor progression, it may be a key player in this process. An abnormally high level of FGF-2 may result from the lack of RFT-A expression relative to RFT-AЈ and B. If such a hypothesis is proven to be correct, we speculate that RFT-A could be a useful agent in treating FGF-2-dependent tumors.
In conclusion, FGF-2 gene expression is negatively regulated by RFT-A by transcriptional repression. Regulation of such function is partly achieved through the alternative splicing mechanism in the DNA binding domain of RFT, suggesting that the imbalance in the ratios of these splice variants may be responsible for the constitutive expression of FGF-2 gene, resulting in tumor progression in FGF-2-dependent tumors. Furthermore, overexpression of the normal transcriptional repressor, RFT-A, induces apoptosis when FGF-2 is deregulated in tumor cells.