Polypyrimidine tract-binding protein represses splicing of a fibroblast growth factor receptor-2 gene alternative exon through exon sequences.

The fibroblast growth factor receptor (FGFR)-2 gene contains two mutually exclusive exons, K-SAM and BEK. We made a cell line designed to become drug-resistant on repression of BEK exon splicing. One drug-resistant derivative of this line carried an insertion within the BEK exon of a sequence containing at least two independent splicing silencers. One silencer was a pyrimidine-rich sequence, which markedly increased binding of polypyrimidine tract-binding protein to the BEK exon. The BEK exon binds to polypyrimidine tract-binding protein even in the silencer's absence. Several exonic pyrimidine runs are required for this binding, and they are also required for overexpression of polypyrimidine tract-binding protein to repress BEK exon splicing. These results show that binding of polypyrimidine tract-binding protein to exon sequences can repress splicing. In epithelial cells, the K-SAM exon is spliced in preference to the BEK exon, whose splicing is repressed. Mutation of the BEK exon pyrimidine runs decreases this repression. If this mutation is combined with the deletion of a sequence in the intron upstream from the BEK exon, a complete switch from K-SAM to BEK exon splicing ensues. Binding of polypyrimidine tract binding protein to the BEK exon thus participates in the K-SAM/BEK alternative splicing choice.

Many eucaryotic genes are composed of exons and introns, and RNA splicing converts their primary transcripts into mRNAs suitable for translation (1). Some pre-mRNAs contain only constitutively spliced exons and yield a single mRNA. Others contain subsets of exons that undergo alternative splicing and generate several different mRNAs, depending on which combination of the available exons are actually spliced. Most often, the different mRNAs code for similar yet distinct forms of a protein. This means that alternative splicing can be used to ensure that a cell synthesizes the particular version of a protein adapted to its needs, which may differ from those of a neighboring cell of another type (2)(3)(4)(5). Controlling alternative splicing correctly is thus of fundamental importance.
One question posed by both constitutive and alternative splicing is how exons are recognized on pre-mRNAs. Specific sequences at or close to splice sites are important for this recognition and are known to interact with components necessary for spliceosome assembly (for reviews, see Refs. 1, 6, and 7). The 5Ј splice site sequence, for example, is important for U1 snRNP binding, and the polypyrimidine tract associated with the 3Ј splice site binds U2AF. In addition, bridging interactions between the proteins bound to the two splice sites are believed to exist (8). However, sequences distinct from those at intron/ exon junctions can also influence splicing. Frequently, exons contain exon splicing enhancers required for their splicing. Many exon splicing enhancers bind members of the SR protein family (7). SR proteins are known to interact with both U1 snRNP and U2AF, so they could clearly be involved in splicing activation (2,9,10). Certain exons, including some exons with an exon splicing enhancer, contain exon splicing silencers (ESSs); some of these bind hnRNP A1 (7). Intron sequences that activate or repress splicing of an adjacent exon have also been described, and in certain cases proteins interacting with them have been identified (for reviews, see Refs. 7 and 11). One such protein is polypyrimidine tract-binding protein (PTB). 1 Repression by PTB generally involves PTB binding to multiple intronic sites. Frequently, sites in both the introns flanking an exon are required, and it has been proposed that interactions between several bound PTB molecules may organize a zone of repression around a regulated exon (7,11). As an individual exon can be linked to several different control sequences, the decision to splice or not to splice an exon can be quite complex. For example, splicing of a short src gene exon in neuronal cells requires binding of a multiprotein complex including hnRNP F, hnRNP H, and the protein KSRP to a downstream intron splicing enhancer (12)(13)(14). In addition, a neural polypyrimidine tract-binding protein isoform may act to reduce the repression of the exon's splicing, which is normally exerted in non-neuronal cells by polypyrimidine tract-binding protein (14,15).
Some genes contain alternative exons that are mutually exclusive. This is the case, for example, for the FGFR-2 gene K-SAM (or IIIb) and BEK (or IIIc) exons (16). Epithelial cells splice the K-SAM exon, whereas cells of mesenchymal origin prefer the BEK exon (17). Abrogation of either specific receptor form in mice leads to severe developmental defects (18 -20), showing how important it is to get the splicing choice right. Furthermore, progression of prostate cancer has been shown to be accompanied by a change in FGFR-2 alternative exon choice in a rat model (21,22). The marked tissue-specific control of the K-SAM/BEK choice and its evident biological importance have encouraged work aiming to describe the underlying mechanism. Much of this work has been directed toward the K-SAM exon. It has been shown that multiple elements control splicing of this exon. Elements repressing splicing include an ESS that binds hnRNP A1 (23,24) and intron splicing silencers in the upstream and downstream introns that bind PTB (11,25). The downstream intron contains three sequences that activate K-SAM exon splicing (26 -28). One of these binds the recently identified splicing activator TIA-1 (29,30), which appears to assist U1 snRNP binding to the exon's 5Ј splice site. Proteins binding to the other two intron-activating sequences, which appear to act together (28), have not yet been identified. Compared with the K-SAM exon, little information is available about splicing control of the BEK exon. Splicing of this exon can be repressed (26,31,32), and among the sequences involved are some intron sequences also implicated in activation of K-SAM exon splicing. The proteins working to repress BEK exon splicing remain, however, to be identified. We set out to use a protocol based on gene transfer and drug selection (33) to clone cDNAs coding for BEK exon splicing repressors. This approach did not lead to the cloning of such cDNAs but did lead us indirectly to the observation that binding sites for polypyrimidine tract-binding protein within the BEK exon itself participate in repressing splicing of this exon.

EXPERIMENTAL PROCEDURES
Plasmids-pN␤ (33) contains two NEO-coding exons separated by a globin intron under control of the SV40 early gene promoter. pN␤BEK was made by inserting into the N␤ intron a 2.1-kb fragment of the human FGFR-2 gene containing the BEK exon flanked by 1.0 kb of upstream and 0.95 kb of downstream intron flanking sequences. Vectors using the human cytomegalovirus immediate early gene promoter for expression of TIA-1 (pTIA-1), hnRNP C1 (phnRNP C1), and bacteriophage MS2 coat protein (pcoat) were based on pCI-neo (Promega) and have been described previously (29). An expression vector for PTB1 was a gift from Christopher Smith. RK3 is an FGFR-2 minigene containing the alternative K-SAM and BEK exons together with flanking intron sequences and has been described elsewhere (28).
Insert Identification-HeLa cells were cotransfected by the calcium phosphate technique as described previously (28) with pN␤BEK and a plasmid conferring resistance to puromycin. Puromycin-resistant colonies were selected, isolated, and expanded. Colonies were tested for their sensitivity to G418. G418-sensitive colonies were retained, and RNA was isolated from them. RT-PCR was carried out on the RNA, using primers ( Fig. 1) PSV (5Ј-TTCCAGAAGTAGTGAGGAGG-3Ј) from the SV40 promoter and neo2 (5Ј-AGGTGAGATGACAGGAGATC-3Ј) from the second NEO-coding exon. Use of these primers will yield RT-PCR products of 555 and 410 bp from cells splicing or skipping the BEK exon, respectively. One colony, H7, sensitive to G418 and in which pN␤BEK transcripts contain the BEK exon as judged by the RT-PCR analysis, was retained for further study. As described in the Results section, a G418-resistant derivative of H7 was isolated. DNA from this derivative was used for PCR with a primer (5Ј-CTGGGCAC-CATACTTTTGGAAAACC-3Ј) in FGFR-2 intron sequences upstream from the BEK exon and the neo2 primer. The product obtained was ϳ0.2 kb larger than expected. This product was cloned in pBluescript SKϩ (Stratagene). Restriction enzyme mapping localized the insert to within the BEK exon, and this insert was sequenced. The insert sequence (shown in Fig. 2) was compared with human genomic sequences and identified as a short fragment of chromosome 9 (working draft sequence segment NT 008457). pN␤BEKϩins, a version of pN␤BEK containing the insert, was made from pN␤BEK and the cloned PCR product obtained above by exchange of an SgrAI-EcoRV fragment (this fragment contains the insert in the PCR product).
pN␤BEK Derivatives-To search for an ESS, versions of pN␤BEKϩins from which various parts of the insert had been removed were prepared by restriction enzyme digestion, repair and religation. For finer mapping, oligonucleotides were introduced into either the BEK exon's EcoRV site (Figs. 2 and 4A) or into an Eco47III site introduced into the BEK exon 5 bp downstream from the original insert site by mutating the sequence gacgct to agcgct (Fig. 2) using the QuikChange mutagenesis protocol (Stratagene). To avoid eventual problems linked to nonsense-mediated decay (32), oligonucleotides did not contain any in frame stop codons, and their size was a multiple of 3, so as not to provoke a frameshift. Pyrimidine-rich stretches within the BEK exon of pN␤BEK were mutated as shown in Fig. 5A, and 5Ј splice site 1 was mutated from cag2gtatac to cag2atatac, and 5Ј splice site 2 was mutated from cgg2gtaatt to cgg2ataatt, using the QuikChange mutagenesis protocol (Stratagene). RK3 derivatives containing these various mutations were made by exchanging an SgrAI-EcoRV fragment (both sites cut within the BEK exon) of RK3 with corresponding fragments from the mutated pN␤BEK versions. ⌬IAS3 versions of these plasmids were made by deleting a 0.24-kb PstI fragment containing IAS3 from the intron upstream of the BEK exon. All mutations were verified by sequencing.
Transfections and RT-PCR-In general, pN␤BEK derivatives (2 g) and pcoat carrier (18 g) were cotransfected into 293-EBNA cells (Invitrogen) by calcium phosphate cotransfection (28). For cotransfections of pN␤BEK derivatives with expression vectors for TIA-1, hnRNP C1, or PTB1, the pN␤BEK derivative (2 g) was mixed with 2 or 18 g of expression vector as indicated in the figure legends, and pcoat was added where necessary to give a total of 20 g of DNA transfected. RNA was isolated from cells 48 h later. RT-PCR was carried out on transfected cell RNA as described previously (24,28), using primers PSV and neo2. Stable cell populations containing RK3 derivatives were prepared by calcium phosphate cotransfection of SVK14 cells with a vector conferring resistance to neomycin and the appropriate RK3 derivative. For each derivative, triplicate populations were prepared in parallel. Colonies resistant to neomycin were combined, cultured, and used for RNA extraction. RT-PCR was carried out on transfected cell RNA using primers P3 and P4 as described previously (29). Amplification cycles remained in the exponential range, and radioactive detection of amplification products was necessary. After migration on a 2 or 2.5% agarose gel, products were transferred to nylon filters (Hybond Nϩ; Amersham Pharmacia Biotech) and hybridized to a probe containing either neo exon 1 sequences (for analysis of N␤BEK transcripts), or FGFR-2 exon C2 sequences (for analysis of RK3 transcripts). For quantification, radioactivities present in bands were determined using a Molecular Dynamics PhosphorImager and used to calculate splicing percentages. For ⌬IAS3 and ⌬IAS3 Mut R, the relative contributions of the b1 and s products to the intensity of the b1 ϩ s band was determined using data from the AvaI digests, where the b1 band and the s* band (which is derived from the s band) are separated. All RT-PCR experiments were carried out at least in triplicate, and representative results are shown here.
Cross-linking and Immunoprecipitation-RNA probes were synthesized in vitro from pBluescript SK(ϩ)-based plasmids containing different versions of the BEK exon (including BEK exons carrying mutations or insertions of different oligonucleotides) downstream of the T7 promoter. For generation of the ␣-actinin probe, a plasmid containing the NM and SM exons plus the intron between them, with a 30-bp spacer between the NM 5Ј splice site and the SM branch point, was used (34). RNAs were uniformly labeled to high specific activity (2-3 ϫ 10 8 cpm/g for probes) or low specific activity (1-2 ϫ 10 4 cpm/g for competitors) using [␣-32 P]UTP with a MAXIscript in vitro transcription kit from Ambion using manufacturer's instructions. Unincorporated nucleotides were removed using a MicroSpin G-25 column (Amersham Pharmacia Biotech). For each cross-linking reaction, 3 l of HeLa nuclear extract (a gift from J. Stévenin) was preincubated for 15 min at 31°C in a final volume of 10 l containing 10 mM Hepes, pH 7.4, 50 mM KCl, 0.1 mM EDTA, 0.25 mM dithiothreitol, 0.76 mM ATP, 25 mM creatine phosphate, 1 mM MgCl 2 , and 0.25 g of tRNA. When appropriate, competitor RNA was added during the preincubation. 10 l of a buffer containing 10 mM Hepes, pH 7.4, 50 mM KCl, 0.1 mM EDTA, 0.76 mM ATP, 25 mM creatine phosphate, 1 mM MgCl 2 , 1.25 mM dithiothreitol, 44 units of RNasin (Ambion), 0.1% Nonidet P-40, 60 ng/l bovine serum albumin, and 10% glycerol was then added, together with 1 l of probe RNA (250,000 cpm). Samples were incubated for 15 min at 31°C and then irradiated on ice for 15 min, prior to the addition of RNase T1 (200 units) and incubation for 40 min at 37°C. Samples were either analyzed directly by electrophoresis on an SDS-8% polyacrylamide gel or immunoprecipitated first. In the latter case, 5 l of protein A-Sepharose beads (Amersham Pharmacia Biotech) was added, and samples were mixed at 4°C for 90 min. Beads were removed by centrifugation, and to 10 l of recovered supernatant was added 38.5 l of IPP buffer (50 mM Tris-HCl, pH 8, 150 mM NaCl) containing 0.1% Nonidet P-40, together with 3 l of a rabbit polyclonal antibody against polypyrimidine tract binding protein (gift from Christopher Smith). Samples were mixed gently at 4°C for 90 min, before the addition of 7.5 l of protein A-Sepharose beads and further gentle mixing at 4°C for 90 min. Beads were washed three times in IPP buffer containing 0.25% Nonidet P-40, and bound proteins were then eluted in 20 l of SDS loading buffer at 100°C for 5 min and loaded on an SDS-8% polyacrylamide gel.

An Insert in the BEK Exon Represses Its
Splicing-A gene transfer and selection strategy designed to detect proteins capable of changing specific splicing patterns has been described previously (33). We adapted this strategy to study BEK exon splicing. In the N␤BEK gene ( Fig. 1), sequences coding for the enzyme neomycin phosphotransferase (NEO) are interrupted by a hybrid intron into which the BEK exon and flanking intron sequences have been inserted. When introduced into cells that splice the BEK exon, the N␤BEK gene should induce synthesis of an inactive enzyme (NEO/BEK/NEO) containing an insert of BEK protein sequence. Such cells will be sensitive to the drug G418. Any event that leads to repression of BEK exon splicing in the cells (e.g. expression of a repressor following gene transfer) should lead to synthesis of an active enzyme (NEO/NEO), and cells will become G418-resistant.
The N␤BEK gene was stably cotransfected into HeLa cells, which normally splice the BEK exon, together with a marker plasmid conferring resistance to puromycin. Several puromycin-resistant colonies were obtained and analyzed. One of these, H7, was G418-sensitive and contained and expressed the N␤BEK gene. N␤BEK transcripts in H7 cells contained the BEK exon as expected (data not shown). H7 cells were transfected with a cDNA library made from SVK14 cells, which do not splice the BEK exon and should thus express the putative BEK exon splicing repressor. Cells taking up a repressor-encoding cDNA should become G418-resistant. A number of G418-resistant colonies were indeed isolated from the transfected cell population. In most of these, little detectable repression of BEK exon splicing was observed, and we did not investigate the reasons for their resistance to G418. In one of the colonies, BEK exon splicing was completely repressed (data not shown). However, this did not appear to be the result of expression of a repressor, since the endogenous FGFR-2 gene transcripts still contained BEK sequences. Analysis of DNA from this colony showed that its N␤BEK gene contained a 192-bp insert within the BEK exon (N␤BEKϩins, Fig. 1). The sequence of the BEK exon containing this insert is shown in Fig.  2. The insert itself (shown in capital letters) corresponds to a DNA sequence from a region of human chromosome 9 devoid of any known gene. Its insertion in the BEK exon has led to the duplication of three BEK exon base pairs (ttg) flanking the insertion site.
The BEK Insert Contains ESS Motifs-The insert is in some way responsible for blocking BEK exon splicing, so that N␤BEK transcripts no longer contain BEK exon sequences and thus code for a functional enzyme. We reasoned that the insert must contain an ESS. A plasmid containing an N␤BEK gene with the insert (pN␤BEKϩins) was prepared and transfected into 293-EBNA cells (which normally splice the BEK exon) in parallel with a plasmid containing the normal N␤BEK gene. An RT-PCR analysis was carried out on RNA from transfected cells using primers shown in Fig . Sequencing of cloned products showed that the larger of these corresponds to splicing of the BEK exon with residual insert sequences, whereas the smaller (marked by a star) corresponds to splicing of the BEK exon using a 5Ј splice site (aca2gttaag) created upon ligation of the repaired Tth111I and BbsI extremities. However, deletion of neither the left half (deletion 6, Figs. 2 and 3B) nor the right half (deletion 7, Figs. 2 and 3B) of the Tth111I-BbsI fragment led to relief of BEK exon splicing repression (Fig. 3B, lanes 8 and 9, respectively). This suggests that each half of the Tth111I-BbsI fragment contains an independent ESS.
One ESS Motif Is a Pyrimidine-rich Sequence-We searched for the putative ESS remaining in pN␤BEKϩins after deletion 7. This ESS should be contained within a sequence made up of the nucleotides between the Tth111I site and the start of deletion 7 plus the nucleotides between the end of deletion 7 and the end of the insert. Oligo 2 (Fig. 4A) contains most of this sequence. Oligo 2 was inserted in the BEK exon of pN␤BEK five base pairs downstream of the original insert. Oligo 1, a sequence of the same size (Fig. 4A), was inserted in parallel. Oligo 1 is derived from insert sequences and partially overlaps oligo 2, but the overlapping sequences contain several point mutations (Fig. 4). 293-EBNA cells were transfected with the resulting plasmids, and RT-PCR analysis was carried out on RNA harvested from them. Insertion of oligo 2 repressed BEK exon splicing as effectively as the entire insert (Fig. 4B, compare lanes 4 and 2). In contrast, insertion of oligo 1 had a very limited effect (lane 3). We wished to determine whether oligo 2's action was position-dependent and decided to insert it into Different oligonucleotides (Fig. 4A, oligos 3-8) whose sequences were derived from that of oligo 2 were also tested for their repression activity following insertion into the EcoRV site of the BEK exon mutated for 5Ј splice site 2. Oligo 3 (Fig. 4C, lane 2) did not repress BEK exon splicing significantly better than the control oligo 1 (Fig. 4B, lane 3). Oligo 4, however, was more effective (lane 1). Oligo 5 was significantly less active than oligo 4 (compare lanes 3 and 1). One of the differences between these two oligonucleotides is that oligo 5 does not contain a tagg motif present in oligo 4 (Fig. 4A). We have previously shown this motif to have ESS activity in the K-SAM exon (23). However, oligo 6, a version of oligo 4 in which the tagg motif has been changed to tacg, represses BEK exon splicing about as effectively as oligo 4 (Fig. 4C, compare lanes 4 and 1). Oligo 7, a shorter version of oligo 4 with the tagg motif intact, does not repress BEK exon splicing significantly (lane 5). These results suggest that the tagg motif is not responsible for splicing repression here. Another difference between oligos 5 and 7, which do not repress BEK exon splicing efficiently, and oligo 4, which does, is that both of the former lack specific pyrimidine runs present in the latter (Fig. 4A). Oligo 5 lacks a tctt run, whereas oligo 7 lacks a ttct run. Both UCUU and UUCU in a pyrimidine-rich context have been identified previously as PTB binding sites (36,37). Mutating two pyrimidinerich stretches in oligo 4 to obtain oligo 8 (Fig. 4A) essentially abolishes its repressing activity (Fig. 4C, lane 6). These observations raise the possibility that PTB binding could be important for repression by oligo 4.
The Pyrimidine-rich ESS Represses Splicing of a Downstream, but Not an Upstream, 5Ј Splice Site-As mentioned above, the BEK exon contains a weak alternative 5Ј splice site, 5Ј splice site 2 (5'ss-2; Fig. 2). We mutated either the normal or the weak alternative 5Ј splice site in pN␤BEK and transfected resulting plasmids into 293-EBNA cells prior to an RT-PCR analysis. With pN␤BEK, use of the weak 5Ј splice site 2 is scarcely detectable (Fig. 4D, lane 1) and is eliminated by mutation of this site (lane 2). When the normal 5Ј splice site (Fig.  2, 5'ss-1) is inactivated, RT-PCR products represent a mixture of BEK inclusion products using 5Ј splice site 2 and BEK exclusion products (lane 3). A very similar pattern of products was obtained using a minigene in which both 5Ј splice sites were intact but which contained an insertion of oligo 4 in the EcoRV site (Fig. 2) between the competing 5Ј splice sites (Fig.  4D, lane 4). This can be explained if oligo 4 represses the downstream 5Ј splice site 1, as we have shown it can do (Fig.  4C, lane 1), but has much less effect on the upstream 5Ј splice site 2.
The Pyrimidine-rich ESS Binds PTB-The experiments described above show that oligo 2 contains an ESS. We used UV cross-linking to investigate proteins that bind to the ESS present in oligo 2. In vitro generated transcripts of the BEK exon,  Fig. 2 and A. ϪBEK, an RT-PCR product corresponding to BEK exon skipping; ϩBEK, a product corresponding to BEK exon splicing (taken to refer to splicing using the BEK exon's splice site). An asterisk (lane 7) refers to a product originating from use of a 5Ј splice site created by deletion 5 and discussed here. B, representation of the BEK exon with its insert. The insert is identified by diagonal shading. Cleavage sites for restriction enzymes used for creating some deletions are marked, and the extents of sequences deleted are indicated by arrows.

FIG. 4. Repression of BEK exon splicing by oligonucleotides.
A, map of the BEK exon showing the sites of oligonucleotide insertion. The sequences of oligonucleotides used are marked below the map. The sequence of oligo 2 is marked in heavy type, as are parts of other oligonucleotides that can be found in the oligo 2 sequence. The sequence of oligo 5 represents a contiguous stretch of nucleotides of the insert (see the insert sequence, Fig. 2). This is also the case for oligo 3, except for its last two nucleotides (cg). Residues mutated in individual oligonucleotides are shown as capital letters. For oligo 8, selected pyrimidines were mutated to purines. Other pyrimidines were interchanged (C to T or T to C) to avoid creating long purine-rich runs.  For lanes 2 and 4 in B, the ϩBEK product was below detectable levels. D, RT-PCR analysis of RNA from 293-EBNA cells transfected with the N␤BEK gene (BEK), the N␤BEK gene in which 5Ј splice site 2 has been inactivated by mutation (Mut 5'ss-2), the N␤BEK gene in which 5Ј splice site 1 has been inactivated by mutation (Mut 5'ss-1), or the N␤BEK gene containing oligo 4 in the unmutated BEK exon's EcoRV site. ϩBEK products use either 5Ј splice site 1 or 2 as marked.
the BEK exon containing oligo 1, and the BEK exon containing oligo 2 (Fig. 5A) were incubated in HeLa cell nuclear extract prior to UV cross-linking, ribonuclease digestion, and SDS-PAGE analysis. The BEK exon containing oligo 2 binds very strongly to an ϳ60-kDa protein (Fig. 5B, lane 2). This protein also binds to the BEK exon and the BEK exon containing oligo 1, although much more weakly (lanes 1 and 3, respectively).
The 60-kDa protein appeared as a doublet on shorter exposures (data not shown), suggesting that it might be PTB (38,39). Using a polyclonal antibody specific for PTB, the doublet could be immunoprecipitated from the BEK, BEK ϩ oligo 2, and BEK ϩ oligo 1 samples (Fig. 5C, lanes 1-3, respectively). In the experiment shown in Fig. 5, B and C, an ␣-actinin gene transcript previously shown to bind PTB (34) was used as a positive

FIG. 5. Polypyrimidine tract-binding protein cross-links to the BEK exon.
A, map of the BEK exon showing the site of insertion of oligo 1 or 2, and identifying two pairs (L and R) of BEK exon pyrimidine runs. The sequence of these runs is marked below the map (BEK). These runs have been mutated as marked in the Mut L, Mut R, and Mut L ϩ R series. B, 32 P-labeled transcripts of the BEK exon, the BEK exon containing oligo 2 or oligo 1, the BEK exon with the L ϩ R mutation, or a fragment of an ␣-actinin gene as marked were incubated in HeLa cell nuclear extract prior to UV cross-linking, RNase treatment, and SDS-PAGE analysis. Molecular weights (in thousands) are indicated on the left. All samples shown were migrated on the same gel but not on adjacent lanes and have been regrouped here to save space. C, samples as in B were subjected to SDS-PAGE after immunoprecipitation using rabbit polyclonal antibodies against PTB. All samples shown were migrated on the same gel but not on adjacent lanes and have been regrouped here to save space. D, 32 P-labeled transcripts of the BEK exon or the BEK exon with the L, R, or L ϩ R mutations as marked, were incubated in HeLa cell nuclear extract prior to UV cross-linking, RNase treatment, immunoprecipitation using rabbit polyclonal antibodies against PTB, and analysis by SDS-PAGE. E, 32 P-labeled transcripts (Probe) of the BEK exon or the BEK exon with oligo 2 were incubated in HeLa cell nuclear extract in the presence of various molar excesses of trace-labeled RNA transcripts (Competitor) as marked, prior to UV cross-linking, RNase treatment, and analysis by SDS-PAGE. Molecular weights (in thousands) are indicated. control. It cross-linked to a protein (Fig. 5B, lane 5) of similar size to that cross-linked to the BEK exon containing oligo 2 (lane 2), and both proteins were recovered with similar yields following immunoprecipitation with the anti-PTB antibodies (compare Fig. 5C, lanes 2 and 5). No cross-linked proteins were immunoprecipitated by monoclonal antibodies directed against U2AF65 (data not shown). Given the known splicing repression activity of PTB (7,11), these results suggest that oligo 2 represses BEK exon splicing by recruiting PTB. This is consistent with the data discussed above showing that oligos 5 and 7 repress splicing less well than oligo 4 (Fig. 4C); both oligos 5 and 7 lack one of the PTB-binding motifs present in oligo 4.
Binding of PTB to the BEK Exon-Whereas oligo 2 enhances PTB binding to the BEK exon, PTB binds somewhat to the BEK exon even in the absence of oligo 2 (Fig. 5, B and C, lanes 1). The BEK exon contains five runs of pyrimidines, which are underlined in Fig. 2. Four of these are represented schematically in Fig. 5A (marked as pairs L and R). In vitro generated transcripts of a BEK exon (Fig. 5A, Mut L ϩ R) in which these runs have been interrupted by the inclusion of purine residues no longer cross-link detectably to PTB (Fig. 5, B and C, lanes 4), showing that at least some of these pyrimidine runs are important for PTB binding. To determine which runs are important, we constructed BEK exons in which either the first two runs (Fig. 5A, Mut L) or the last two runs (Fig. 5A, Mut R) were interrupted. In vitro generated transcripts from the BEK exon or the BEK exon with the L, R, or L ϩ R mutations were incubated in HeLa cell nuclear extract prior to UV cross-linking, ribonuclease digestion, and immunoprecipitation with the polyclonal PTB-specific antibody. The results show that both the L (Fig. 5D, compare lane 2 with lane 1) and the R mutation (Fig. 5D, compare lane 3 with lane 1) diminish PTB crosslinking to the BEK exon, suggesting that at least one of the first two and at least one of the last two pyrimidine runs in the BEK exon are implicated in PTB binding.
The results described above suggest that the Mut L ϩ R BEK exon should not be able to compete with the BEK exon or the BEK exon containing oligo 2 for binding to PTB. This was confirmed by the experiments shown in Fig. 5E. A 150-fold excess of a very low specific activity transcript of the BEK exon containing oligo 2 competes effectively with the corresponding high specific activity transcript for cross-linking to PTB (compare lanes 3 and 1). In contrast, a 150-fold excess of very low specific activity transcript of the BEK exon with the L ϩ R mutation is not a competitor (compare lanes 5 and 1). A 150fold excess of very low specific activity transcript of the BEK exon competes effectively with the corresponding high specific activity transcript for PTB-cross-linking (compare lanes 11 and 9). However, a 150-fold excess of very low specific activity transcript of the BEK exon with the L ϩ R mutation has little effect on PTB cross-linking to the high specific activity BEK exon transcript, while significantly reducing cross-linking to other proteins (compare lanes 8 and 6).
Repression by PTB Overexpression Requires BEK Exon Pyrimidine-rich Sequences-As shown above, insertion of the pyrimidine-rich oligo 2 into the BEK exon leads to increased PTB binding to the exon and repression of its splicing. However, the BEK exon already contains several pyrimidine-rich stretches (Fig. 2), some of which we have shown to interact with PTB. We wondered if increasing PTB levels in 293-EBNA cells, which normally splice the BEK exon, would suffice to repress BEK exon splicing. To address this question, 293-EBNA cells were cotransfected with pN␤BEK together with expression vectors for bacteriophage MS2 coat protein (a negative control), PTB1, TIA-1, or hnRNP C1. Like PTB, the latter two proteins are known to bind to pyrimidine-rich sequences involved in splicing control (29,30,40). RT-PCR was carried out on RNA from transfected cells. As shown in Fig. 6A, PTB1 overexpression, but not overexpression of TIA-1 or hnRNP C1, led to efficient BEK exon repression (compare lane 2 with lanes 1, 3, and 4). Repression of BEK exon splicing by PTB overexpression requires the pyrimidinerich stretches present in the BEK exon, which bind PTB, since splicing of a BEK exon with the L ϩ R mutation (Fig.  5A) is not efficiently repressed by PTB1 overexpression (Fig.  6B, compare lane 4 with lane 2). Note that 18 g of PTB1 expression vector has little detectable effect on splicing of a BEK exon with the L ϩ R mutation (lane 4), whereas 2 g of PTB1 expression vector suffices to repress completely splicing of a normal BEK exon (lane 2).
BEK Exon Pyrimidine-rich Sequences Repress BEK Exon Splicing in Epithelial Cells-The above results show that the exonic polypyrimidine runs can be involved in repressing splicing of the BEK exon. Having established this, we wished to determine whether they are implicated in the mutually exclusive splicing choice between the K-SAM and BEK exons. The BEK exon is not normally spliced in epithelial cells, the alternative K-SAM exon being spliced instead. The RK3 minigene contains the alternative K-SAM and BEK exons together with flanking constitutive exons C1 and C2 (Fig. 7A). The RK3 minigene and a version thereof containing a BEK exon with the L ϩ R mutation that disrupts PTB binding (Fig. 5A) were stably transfected into the epithelial cell line SVK14. RT-PCR was carried out on RNA harvested from transfected cells using a minigene-specific primer pair. To distinguish between BEK and K-SAM exon splicing, RT-PCR products were analyzed by digestion with HpaI (there is one HpaI site in the BEK exon; Fig. 7A) or AvaI (there is one AvaI site in the K-SAM exon; Fig.  7A). As expected, RK3 pre-mRNA was spliced using the K-SAM exon (Fig. 7B, lanes 1-3; RT-PCR products are cleaved by AvaI and not by HpaI). RK3 Mut L ϩ R pre-mRNA was also mainly spliced using the K-SAM exon (Fig. 7B, lanes 4 -6), but there was a very modest increase in BEK exon splicing, as evidenced by the presence of some RT-PCR products resistant to AvaI digestion (lane 6) and some products cleavable by HpaI (lane 5).
This increase, albeit modest, encouraged us to investigate  7. Effect of mutating the BEK exon's pyrimidine runs on the splicing choice between the K-SAM exon and the BEK exon in epithelial cells. A, map of the RK3 minigene. RSV, the Rous sarcoma virus long terminal repeat promoter; BGH, the bovine growth hormone gene fragment providing the polyadenylation site. C1 and C2 are the FGFR-2 constitutive upstream and downstream exons, respectively. The normal BEK exon 5Ј splice site is marked 5'ss-1; alternative sites described are marked 5'ss-2 and 5'ss-3. Positions of primers used for the RT-PCR analysis shown in B and C are marked by arrows. Below the map are marked the structures of various RT-PCR products obtained, corresponding to skipping of both exons (c), splicing of the K-SAM exon (s), and splicing of the BEK exon using the normal 5Ј splice site (5'ss-1) (b1), or one of the alternative 5Ј splice sites (5'ss-2 (b2) or 5'ss-3 (b3)). Digestion products of these fragments detectable with the probe used in B and C are marked by stars. A, AvaI; H, HpaI. B and C, RT-PCR analysis using primers shown in A of RNA from SVK14 cells stably transfected with the RK3 minigene, an RK3 minigene carrying the L ϩ R mutation, an RK3 minigene from which IAS3 had been deleted (⌬IAS3), or versions of ⌬IAS3 carrying the L ϩ R, L, or R mutation. Products were left undigested (0), or were digested by HpaI or AvaI as marked before migration on a 2.5% agarose gel, transfer to a nylon filter, and hybridization with a probe corresponding to C2 sequences. RT-PCR products and their digestion products are identified by letters that correspond to structures shown in A. D, analysis of the RT-PCR data shown in C. For the four samples shown in C, the percentage of each RT-PCR product was determined by PhosphorImager analysis as described under "Experimental Procedures." Results are given as the averages of three independent experiments, and error bars represent the S.E. For lanes 4 -7 in C, the s and c products were below detectable levels. For lanes 1 and 10 in C, the b3 product was below detectable levels.
whether the exonic pyrimidine-rich sequences participate in BEK exon splicing repression, since even significant lifting of BEK exon repression might not suffice to render BEK exon splicing competitive with splicing of the preferred K-SAM exon in epithelial cells. Splicing of the K-SAM exon in SVK14 cells is under complex control. One control sequence is IAS3 (Fig. 7A), which activates K-SAM exon splicing (28). The equivalent of IAS3 in a rat FGFR-2 minigene also represses BEK exon splicing in a rat cell line in which the K-SAM exon is spliced normally (26). We set out to investigate the behavior of minigenes from which IAS3 has been deleted; K-SAM exon splicing should be less favored relative to BEK exon splicing on pre-mRNA from such minigenes. This did indeed prove to be the case when RNA from SVK14 cells stably transfected with an RK3 minigene lacking IAS3 (⌬IAS3) was analyzed by RT-PCR (Fig. 7C, lanes 1-3). Three RT-PCR products were observed and are identified by letters corresponding to structures shown in Fig. 7A. They correspond to a mixture (b1 ϩ s) of K-SAM exon splicing (s) and BEK exon splicing using 5Ј splice site 1 (b1); BEK exon splicing using 5Ј splice site 2 (b2); and skipping of both exons (c). Skipping of both exons is responsible for generation of the major RT-PCR product (Fig. 7D).
When the analysis was carried out on cells transfected with a version of ⌬IAS3 carrying a BEK exon with the L ϩ R mutation, a different pattern of bands was observed (Fig. 7C,  lanes 4 -6). The L ϩ R mutation derepresses BEK exon splicing, since now essentially no K-SAM exon splicing (s) or skipping of both exons (c) was observed (Fig. 7D). The major RT-PCR product corresponded to splicing of the BEK exon using 5Ј splice site 1 (b1), although some products corresponded to use of 5Ј splice site 2 (b2), or 5Ј splice site 3 (b3), an additional BEK exon 5Ј splice site hitherto undescribed. (The identity of b3 was established by sequencing cloned b3 products). 5Ј splice site 3 lies between the two polypyrimidine-rich sequences making up the L runs (Fig. 2), so it seemed likely that its use was a consequence of mutating these runs. Indeed, b3 (reflecting 5Ј splice site 3 use) was the major RT-PCR product observed when the analysis was carried out on cells transfected with a version of ⌬IAS3 carrying a BEK exon with just the Mut L mutation (Fig. 7, C (lanes 7-9) and D). Note that this mutation changes the cryptic 5Ј splice site sequence from tg2gtaact to tg2gtaaca, actually rendering it a worse match to the 5Ј splice site consensus (ag2gtaagt). Mutation of the L runs stops PTB binding to them, and it is likely that this renders the cryptic site accessible. The Mut L mutation also derepresses 5Ј splice site 1 and 2, since no K-SAM exon splicing or exon skipping was observed, whereas the amounts of products b1 and b2 (reflecting BEK exon 5Ј splice site 1 and 2 use, respectively) increased relative to the ⌬IAS3 sample (Fig. 7D). Consistent with our previous observation that a pyrimidine-rich sequence is most effective at repressing a downstream 5Ј splice site, the Mut R mutation appears to derepress mainly 5Ј splice site 1. This effect on 5Ј splice site 1 use can be seen by comparing 5Ј splice site 1 use in the ⌬IAS3 sample (20% of products; see Fig. 7, C (lane 1) and D) with 5Ј splice site 1 use in the ⌬IAS3 Mut R sample (43% of products; Fig. 7, C (lane 10) and D). It can also be observed by comparing 5Ј splice site 1 use in the ⌬IAS3 Mut L sample (25% of products; see Fig. 7, C (lane 7) and D) to 5Ј splice site 1 use in the ⌬IAS3 Mut L ϩ R sample (48% of products; Fig. 7, C (lane 4) and D).
Note that in the ⌬IAS3 Mut L ϩ R sample from epithelial cells, use of the alternative 5Ј splice sites 2 and 3 is significantly more marked than in cells that normally splice the BEK exon (see Fig. 4D, lane 1, for example). DISCUSSION We describe the results of work with a cell line that normally splices the FGFR-2 gene alternative exon BEK but was designed to become drug-resistant if splicing of this exon was repressed. We hoped to use this cell line in an expression screening to isolate cDNA clones coding for a specific splicing repressor. This approach did not work. Perhaps an appropriate full-length cDNA was not present in the library screened, or it was not possible to maintain durably a sufficiently high level of repressor expression. However, during the course of the work, we identified a drug-resistant cell line that no longer spliced the BEK exon as a result of the insertion, within this exon, of a 192-bp fragment from human chromosome 9. This fragment contains at least two independent exon splicing silencers. We localized one of these to a pyrimidine-rich fragment, which represses BEK exon splicing by increasing PTB binding to the exon. The BEK exon itself binds to PTB even in the absence of the silencer, and this binding depends on some exonic pyrimidine runs. Overexpression of PTB1 blocks BEK exon splicing, and this again depends on the exon's pyrimidine runs, suggesting that PTB overexpression represses splicing by increasing PTB binding to them. Our results show that PTB molecules bound to exonic pyrimidine runs can act as splicing silencers. This is of interest; splicing repression by PTB binding to intron sequences has been well documented (7,11), but this is not the case for repression by PTB binding to exon sequences.
PTB binding to exon sequences has been described before but not shown to correlate with splicing repression. The bovine papilloma virus type 1 ESS (41) contains a U-rich part, which binds U2AF and PTB, and a C-rich part, which binds SR proteins. However, deletion or mutation of just the U-rich region did not significantly affect ESS activity. This was in marked contrast to the dramatic loss of ESS activity when the C-rich core was mutated, and it was concluded that the role, if any, of PTB binding in splicing repression was unclear in this case (41). Several PTB binding sites have been characterized in the intron upstream from a neuron-specific 24 nucleotide exon of the GABA A receptor ␥2 gene (36). Most of these sites are grouped around the exon's branch point. Adding an RNA competitor containing the 3Ј splice site region upstream of the exon to in vitro splicing reactions derepressed exon splicing (36), demonstrating the role exerted by PTB in repression. There is also a PTB binding site in the exon itself. However, it was not shown that the exon site is implicated in splicing repression. Some other work has linked pyrimidine-rich exon sequences to splicing repression but led to the suggestion that U2AF binding rather than PTB binding might be involved. Thus, several pyrimidine-rich sequences were identified in a search for human genomic sequences capable of inhibiting splicing of a constitutively spliced exon following insertion therein (42). Whereas the proteins interacting with these sequences were not identified, it was shown that inhibition could be achieved by insertion of a U2AF65, but not a PTB, binding site consensus sequence.
How could PTB binding to the BEK exon repress its splicing? Most of the available data concerning PTB's action on splicing describes the effects of its binding to intron sequences (7,11,43). In some cases, PTB may repress splicing via intron binding sites by a simple competition with U2AF (44,45). In many cases, however, there is no overlap between PTB binding sites and the U2AF binding site, and PTB binding sites often exist in both of the introns flanking an exon. In one model, PTB oligomerizes between the upstream and downstream binding sites and thus across the exon (for a review, see Ref. 11). Clearly, molecules of PTB bound to exon sequences would be particularly well placed for participation in this latter type of mechanism. Binding sites for PTB have been characterized in the intron upstream from the BEK exon in the rat FGFR-2 pre-mRNA (25). It is not known if PTB binding sites exist in the downstream intron, although this intron does contain several candidate binding sites. It is thus possible, although we have not tested this here, that PTB molecules bound to the BEK exon interact with other PTB molecules bound to flanking intron sequences to repress BEK exon splicing. Other mechanisms for PTB repression via exon sequences must of course be taken into consideration. It is interesting to note that an ESS in the IgM exon M2, which forms an ATP-dependent complex containing U2 snRNP, contains several pyrimidine runs (46). PTB could thus conceivably be involved in its action and facilitate formation of this complex (46). Perhaps PTB binding to the BEK exon leads to formation of a similar complex on this exon.
The BEK exon and the K-SAM exon are a pair of mutually exclusive FGFR-2 gene alternative exons. Cells of mesenchymal origin splice the BEK exon, whereas epithelial cells splice the K-SAM exon. We have shown that PTB binding to the BEK exon represses its splicing in epithelial cells. Splicing of the BEK exon is thus repressed in epithelial cells in at least two different ways. A sequence in the upstream intron termed IAS3 in the human gene (or ISAR in the rat gene) represses BEK exon splicing but also activates K-SAM exon splicing (26,27). Deletion of the IAS3 sequence leads to a decrease in K-SAM exon splicing in epithelial cells and an increase in BEK exon splicing. However, skipping of both the BEK and the K-SAM exon is the major splicing choice. If in addition to the IAS3 deletion the BEK exon's pyrimidine runs are mutated to stop PTB binding to the BEK exon, a complete switch to BEK exon splicing results. The pyrimidine runs appear to repress downstream 5Ј splice site use, and runs toward the beginning of an exon may also be implicated in repressing cryptic splice site use. This repression may not be required in the epithelial cell line used here (mutating the pyrimidine runs leads to only a minimal increase in BEK exon splicing in epithelial cells when IAS3 is intact), but it could be of more importance in certain of the wide variety of other cells in which the FGFR-2 gene is expressed and in which BEK exon repression by IAS3 might be less effective. It is interesting to note that splicing of the rat equivalent of the K-SAM exon (exon IIIb) is repressed by a sequence in the upstream intron (ISS1) that binds PTB (25). Deletion of ISS1 or a mutation of ISS1 that abolishes PTB binding increases IIIb exon splicing from 14% to around 30% when pre-mRNA from a minigene carrying the IIIb exon but lacking the IIIc exon (the equivalent of our BEK exon) is transfected into cells that normally splice the IIIc exon (25). It is not known, however, if the ISS1 mutations would suffice to induce significant IIIb exon splicing if the IIIc exon were available as a competitor. PTB may thus play comparable roles in repression of the K-SAM exon in cells that normally splice the BEK exon and in repression of the BEK exon in cells that normally splice the K-SAM exon.
What role does PTB play in the cell-specific choice between BEK and K-SAM exon splicing? Studies on the human and the rat FGFR-2 genes have unveiled multiple sequences controlling these exons (24 -29). Some of these sequences are probably common to the human and rat systems, but this has not been investigated in detail. A map of the human FGFR-2 pre-mRNA region containing the K-SAM and BEK exons and identified human control elements is shown in Fig. 8. Splicing of the K-SAM exon is repressed by an ESS that binds hnRNP A1 (24). K-SAM exon splicing is activated by three intron sequences, IAS1, -2, and -3. IAS1 binds TIA-1, which may help U1 snRNPbinding to the exon's 5Ј splice site (29). IAS2 and part of IAS3 form a secondary structure required for efficient K-SAM exon splicing (28). Splicing of the BEK exon is repressed by IAS3 and by pyrimidine runs within the BEK exon that bind PTB. It is not clear how these various sequence elements are used to control the cell-specific splicing of the K-SAM and BEK exons. There is no evidence that hnRNP A1 or TIA-1 can influence splicing of the K-SAM exon in a cell-specific manner. We have not detected any difference in PTB expression or PTB crosslinking to the BEK exon between the cells used here that splice the K-SAM exon and those that splice the BEK exon. 2 It is thus possible that these proteins are used to set the stage for additional, yet to be discovered, proteins needed to impose the cell-specific splicing patterns. These proteins could conceivably act through IAS2/IAS3 in epithelial cells to activate K-SAM exon splicing and to repress BEK exon splicing. Our results demonstrate, however, that stopping such proteins from acting is not sufficient to obtain efficient BEK exon splicing; the BEK exon is also repressed by PTB, and this repression needs to be overcome too. There is some indirect evidence that the BEK exon's normal 5Ј splice site is under repression exerted by yet another system in epithelial cells. Thus, when both IAS3 and the BEK exon pyrimidine runs are mutated, a complete switch from K-SAM exon splicing to BEK exon splicing occurs, but use of two alternative 5Ј splice sites for BEK exon splicing is more marked in epithelial cells than it is in cells that normally splice the BEK exon, where use of these sites is minimal.
Finally, it is interesting to note that repression of BEK exon splicing by the de novo insertion of additional sequences within the exon has been described before. The basis of one case of Apert syndrome is repression of BEK exon splicing by insertion of an Alu element within the exon (47). This element is associated with a long stretch of pyrimidines, which our results suggest may be implicated in the observed splicing repression. These pyrimidines presumably reflect the element's origin as a reverse transcript. There is no evidence in favor of an RNA intermediate in the process that led to the insertion of the 192-bp fragment from human chromosome 9 into the BEK exon, and the underlying mechanism for this insertion is not clear. However, our results show that such insertions can take 2 C. Le Guiner and R. Breathnach, unpublished results.
place. Inhibitory sequences may be frequent in the human genome (42), and the type of event we have described here, which results in silencing of an exon, may well have played a role in the evolution of our genes.