Heterogeneous Ribonucleoprotein A1 Is Part of an Exon-specific Splice-silencing Complex Controlled by Oncogenic Signaling Pathways*

Regulation of alternative pre-mRNA splicing, recognized as increasingly important in causing human disease, was studied using the CD44 gene, whose splice variants have been implicated in tumor progression. We identified heterogeneous ribonucleoprotein (hnRNP) A1 as a protein interacting in vitro and in vivo with regulatory splice elements in CD44 variant exon v5. Transient overexpression of hnRNP A1 prevented v5 exon inclusion, dependent on the exonic elements. HnRNP A1-dependent repression was exon-specific and could be relieved by coexpression of oncogenic forms of Ras and Cdc42. The results define hnRNP A1 as a decisive part of an oncogene-regulated splice-silencing complex, which can select between multiple alternatively spliced exons.

Regulation of alternative pre-mRNA splicing, recognized as increasingly important in causing human disease, was studied using the CD44 gene, whose splice variants have been implicated in tumor progression. We identified heterogeneous ribonucleoprotein (hnRNP) A1 as a protein interacting in vitro and in vivo with regulatory splice elements in CD44 variant exon v5. Transient overexpression of hnRNP A1 prevented v5 exon inclusion, dependent on the exonic elements. HnRNP A1-dependent repression was exon-specific and could be relieved by coexpression of oncogenic forms of Ras and Cdc42. The results define hnRNP A1 as a decisive part of an oncogene-regulated splice-silencing complex, which can select between multiple alternatively spliced exons.
Alternative pre-mRNA splicing, an important mechanism of differential gene expression in higher eukaryotes, gives rise to functionally distinct proteins encoded by a single gene. The generation of alternatively spliced mRNAs can be regulated, e.g. according to the developmental or physiological state of cells in an organism (for recent reviews see Refs. 1 and 2). Pre-mRNA splicing is accomplished by large ribonucleoprotein complexes called spliceosomes, which are assembled on pre-mRNA, directed by small conserved sequences localized at the intron ends, the splice sites (3). Several alternatively spliced pre-mRNAs have been shown to contain regulatory sequences, in addition to the splice sites, that can affect splice site selection dependent on developmental cues. Such sequences have been found in exons and introns, both in Drosophila and mammalian pre-mRNAs (reviewed in Refs. [2][3][4]. In Drosophila, sex-specific splice regulators have been cloned by genetic approaches (4 -6). In contrast, in mammalian cells, factors that govern alternative splicing during differentiation processes or in pathological conditions are largely unknown. Likewise, the mechanisms underlying the regulated generation of complex splice patterns due to selection between multiple alternatively spliced exons, a hallmark of many genes encoding highly variable proteins, are still elusive. Selection of splice sites in mammalian cells has been studied primarily by bio-chemical means. Two groups of proteins have been found that can affect splice site choice: serine/arginine-rich (SR) 1 proteins (for reviews on SR proteins and splicing, see Refs. 7-10) and heterogeneous nuclear ribonucleoproteins (hnRNPs) (for reviews, see Refs. [11][12][13]. Splice site selection plays an important role in human diseases (14 -18), and alternative splicing of several genes has been implicated in tumorigenesis and tumor progression (15,19 and references therein). The best known example of these genes encodes the cell surface molecule CD44 (for reviews on CD44 splicing and cancer, see Refs. 20 and 21). Alternative pre-mRNA splicing generates variant CD44 isoforms by the inclusion of up to 10 variant exons (v1-v10) during embryonic development, upon activation of lymphocytes (22)(23)(24)(25) and of dendritic cells (26) and during tumor progression (20,21,27). We showed previously by cell fusion experiments the existence of trans-acting factors regulating the expression of CD44 isoforms in different cell types and in tumor cells (28). In addition, using a minigene containing CD44 exon v5, an exon sequence frequently included in mRNA upon activation of immune cells and during tumor progression, we could identify splice regulatory RNA elements for this exon. These elements are located within the v5 exon coding sequence and are necessary for cell type-specific exon inclusion as well as for inclusion of the exon in response to phorbol-ester tumor promoters or oncogenic Ras (29).
Here we show that the splice regulatory elements of exon v5 are recognized by hnRNP A1 and that hnRNP A1, dependent on these elements, represses inclusion of the exon in vivo. Silencing of exon inclusion by hnRNP A1 is exon-specific and is relieved by oncogenic forms of the small G proteins p21 Ras and Cdc42. These findings define hnRNP A1 as a decisive component of an oncogene-controlled splice-silencing complex, which can select between multiple alternatively spliced exons.

Cell Culture and Transfections
CB3 erythroleukemia cells (30; kindly provided by Yaacov Ben-David) were cultured in RPMI 1640 medium supplemented with 10% fetal calf serum, 100 units/ml penicillin, 100 g/ml streptomycin, 2 mM L-glutamine, and 10 M ␤-mercaptoethanol. S194 mouse plasmacytoma cells (obtained from Thomas Wirth), KLN205 mouse carcinoma cells * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18  (ATCC CRL-1453), and NIH-3T3 cells (ECACC CB2435) were grown in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum, 100 units/ml penicillin, 100 g/ml streptomycin, and 2 mM L-glutamine. All cells were grown at 37°C and 6% CO 2 .
Transfections were performed using the polycationic reagent Super-Fect (Qiagen) according to the instructions of the manufacturer. Unless stated otherwise, in the cotransfection experiments analyzed by RT-PCR, 2 g of CD44 v5 minigene-plasmid DNA were cotransfected with 8 g of hnRNP A1 expression plasmid or empty expression plasmid. For analysis of hnRNP A1 expression levels in cotransfections, 8 g of the expression plasmid for myc-tagged hnRNP A1 was used. Where indicated, 2 g of expression constructs for activated Ras, Cdc42, MEKK1, or corresponding control constructs were transfected. Transfection efficiency of NIH-3T3 cells was assessed by transfecting cells with pCM-VSPORT-␤gal (Life Technologies, Inc.) followed by X-gal staining.
Other Expression Constructs-RSV-Ras expresses the activated Ha-Ras mutant leucine 61 (32). pCMV5-MEKK1 and pRK5mycCdcd42 express MEKK1 as described previously (33), with an additional Cterminal his tag (kindly provided by Peter Shaw) and an L61 mutant of Cdc42 carrying an N-terminal myc epitope (34; obtained from Alan Hall), respectively.
Yeast Three-hybrid Constructs-All constructs used were based on plasmids of the RNA-Protein Hybrid Hunter system from Invitrogen. pYESTrp2-A1 expresses a fusion of hnRNP A1 and an N-terminal B42 transactivation domain and was obtained by introducing the murine hnRNP A1 cDNA as a BamHI/XhoI fragment (after Pwo polymerase amplification) into the multiple cloning site of pYESTrp2. For the hybrid-RNA-expressing plasmid pRH5Ј-v5, a 180-bp fragment from pETv5, spanning CD44 exon v5 and approximately 30 bp of upstream and downstream intron sequences, was introduced as an AatII/KpnI (blunt) fragment into AatII/SmaI of pRH5Ј. pRH5Ј-blue, expressing a control hybrid-RNA in which the v5 exon sequence (except for the four 5Ј-most and the three 3Ј-most nucleotides) had been replaced for a pBluescript polylinker sequence, was obtained accordingly, using a corresponding AatII/KpnI fragment from pETv5blue (29). pRH3Ј-IRE and pYesTrp2-IRP were obtained from Invitrogen.

RT-PCR Analysis
Preparation of cytoplasmic RNA and RT-PCR analyses was carried out as described previously (29), comprising 35 PCR cycles. PCR reactions were in the linear phase (not in the plateau phase) under these conditions as shown by unchanged ratios between the amplification products for exon inclusion and exon skipping when using different amounts of cDNA. RT-PCR bands were quantified densitometrically after scanning using the Fuji Aida program. Three independent exper-iments were performed. Within an experiment, the differences observed in the ratios of exon inclusion between wild-type and mutant v5 exons were very similar (Ͻ10% variation). The same was true when comparing the on and off situation of oncogenic signaling.
For RNAP assays, nuclear extracts (120 g for silver staining, 60 g for Western blotting) were precleared with 0.5 l of Ultralink-NeutrAvidin beads (Pierce) per g of extract at 4°C for 45 min. Precleared extracts were incubated in binding buffer (10 mM HEPES, pH 7.9, 100 mM KCl, 0.025% Nonidet P-40, 10% glycerol, 1 mM dithiothreitol, 1 mM phenylmethylsulfonyl fluoride, 2 mM pepstatin, 2 mM aprotinin, 2 mM leupeptin) with 1.2 g of yeast tRNA and 4 units of RNase inhibitor (RNasin, Promega) per g of extract at 30°C for 10 min. 6.5 ng of biotinylated RNA oligonucleotide per g of extract was added, and the reactions were further incubated at 30°C for 15 min. After washing the beads four times with binding buffer without glycerol, precipitated complexes were resolved by SDS-polyacrylamide gel electrophoresis and either subjected to silver staining or transferred to a polyvinylidene difluoride membrane (Millipore) followed by Western blotting. Western blot analysis was performed according to standard procedures involving the enhanced chemiluminescence system (ECL, Amersham Pharmacia Biotech) using either 1:1-diluted tissue culture supernatant of mAB104 hybridoma (ATCC CRL-2067), to detect SR proteins, or 1:1000-diluted ascites fluid of monoclonal antibodies 4B10 and 4F4 (kindly provided by Gideon Dreyfuss) to detect hnRNP A1 and hnRNP C proteins, respectively.
For analysis of transiently expressed myc-tagged hnRNP A1 (in Fig.  6B) the monoclonal antibody 9E10 (Santa Cruz Biotechnologies) recognizing the human c-myc-epitope EQKLISEEDL was used, followed by ECL detection.

hnRNP A1 Binds to CD44 v5 Exon Sequences in Vitro and in
Vivo-We have previously identified exon sequences in CD44 variant exon 5 that account for exon inclusion or exon skipping in different cell lines and upon Ras signaling (29). The sequences form an exonic composite splice regulatory unit encompassing an exon recognition element and splice silencer elements in the 3Ј (or right) part of the exon, R, and splice silencer domains in the 5Ј (left) part, L, and in the middle portion, M, of the exon (29; see Fig. 1A). Both types of elements are necessary for regulated cell type-specific and -inducible inclusion of the exon. In a T-cell line equipped with a v5 exon-carrying minigene, concanavalin A, phorbol-ester tumor promoters, or expression of oncogenic p21 Ras induced exon inclusion (29).
To purify and identify nuclear proteins recognizing the exonic elements, we set up RNA-affinity precipitations (RNAP) using biotinylated RNA oligonucleotides corresponding to the subdomains L, M, and R of the CD44 v5 regulatory splice unit (see Fig. 1A). As a control for nonspecific binding to the matrix or for unspecific RNA binding, the oligonucleotide was omitted from the precipitation reaction, or precipitates were performed with a control oligonucleotide, "blue." The blue oligonucleotide equals the L, M, and R oligonucleotides in length (42-mer) but is derived from an unrelated polylinker sequence (from the pBluescript vector, Stratagene). The blue sequence in place of either one of the regulatory subdomains in the CD44-v5 exon abolishes their regulatory effects in vivo (29). Upon silver staining of the RNA-affinity precipitates separated by SDS-polyacrylamide gel electrophoresis (Fig. 1B), several protein bands were detectable that were specifically bound to the CD44 v5 exon sequences. Of these proteins, some bind to all three exon sequences (L, M, and R), whereas others bind only to one sequence (Fig. 1B).
In an approach to identify the proteins in the RNA-affinity precipitates, we screened the separated proteins by immunoblotting for the presence of two groups of proteins, which can affect splice site selection in mammalian cells: SR proteins and hnRNPs. Using an antibody recognizing a conserved phosphoepitope on all SR protein family members (mAb104 (37, 38)), we could not detect SR proteins in the precipitates (data not shown). The screen for hnRNP proteins, however, was successful. Although there were no hnRNP C proteins detectable in the precipitates (not shown), each of the three regulatory subdomains of the v5 exon, L, M, and R, were bound to hnRNP A1 (Fig. 1C, lanes 3-5). No hnRNP A1 binding was detectable with the matrix alone (lane 1) or with the unrelated blue oligonucleotide (lane 2), suggesting a specific association of hnRNP A1 with the CD44 v5 exon sequences.
To exclude that the interaction of hnRNP A1 with v5 sequences was merely an in vitro artifact, we employed a yeast three-hybrid system in which the binding of a bifunctional hybrid-RNA to each of two hybrid proteins activates transcription of reporter genes in yeast (35; see Fig. 2A). The yeast strain employed, L40uraMS2, contains a HIS3 gene and a lacZ gene under the control of lexA-operator sites. In addition, it expresses the first hybrid protein in which the bacterial LexA DNA-binding domain is fused to the coat protein of the bacteriophage MS2 that recognizes an RNA-stem loop structure in MS2 RNA. For the expression of the second hybrid protein we generated a construct (pYESTrp2-A1) that gives rise to an hnRNP A1 protein fused to the B42 transcriptional transactivation domain. As hybrid-RNA supposed to bridge the two fusion proteins, we used the entire CD44 v5 exon sequence plus approximately 30 bp of upstream and downstream intron sequences fused to two tandemly repeated bacteriophage-MS2 RNA recognition sites (pRH5Ј-v5). A construct in which the v5 exon sequence (except for the four 5Ј-most and the three 3Јmost nucleotides) had been replaced for a pBluescriptpolylinker sequence, served as a negative control (pRH5Ј-blue). Co-transformation of the MS2-v5 hybrid-RNA-encoding plasmid and the B42-hnRNP A1 expression construct (pRH5Ј-v5ϩpYESTrp2-A1) resulted in L40uraMS2 yeast cells, which could grow on histidine-deficient plates (Fig. 2B) and which expressed LacZ, as shown by X-gal staining (Fig. 2C). The result resembled that obtained using the positive control: MS2-IRE (iron responsive element) hybrid-RNA and B42-IRP (iron regulatory protein) fusion protein (pRH3Ј-IREϩpYESTrp2-IRP; Fig. 2, B and C). In contrast, co-transformation with the empty B42-expression vector and the MS2-v5 hybrid-RNA construct failed to give rise to cells that could grow in the absence of histidine and that expressed LacZ (pRH5Ј-v5ϩpYESTrp2; Fig. 2, B and C). Similarly, co-transformation of the B42-hnRNP A1 expression construct together with the control hybrid-RNA construct, in which the v5 sequence had been replaced by the Bluescript-polylinker sequence, failed to generate HIS prototrophy or LacZ-positive cells (pRH5Ј-blueϩpYESTrp2-A1; Fig. 2, B and C).
The results indicate that activation of the yeast reporter genes was dependent on the presence of both the B42-hnRNP A1 fusion and the CD44 v5 exon sequences in the hybrid-RNA, strongly suggesting that hnRNP A1 interacted with v5 exon sequences in vivo.
HnRNP A1 Represses Inclusion of Exon v5 in Vivo-To examine whether the interaction of hnRNP A1 with CD44 exon FIG. 1. HnRNP A1 is part of protein complexes bound to CD44 v5 exon sequences in vitro. A, schematic representation of CD44 exon v5 and its regulatory subdomains. Subdomain L and M act as splice silencer elements (Ϫ) and prevent exon usage in cells or in conditions in which the exon is not used. Subdomain R contains splice enhancer elements (ϩ), necessary for exon recognition, and exon silencer elements (Ϫ) (29). B, silver staining of RNA-affinity precipitates from nuclear extracts of S194 mouse plasmacytoma cells separated by SDS-polyacrylamide gel electrophoresis. For precipitations, biotinylated 42-mer RNA oligonucleotides corresponding to the regulatory CD44 exon-v5 subdomains indicated (L, M, and R) and an avidin derivative coupled to a matrix, were employed. "blue" refers to an unrelated 42-mer RNA oligonucleotide derived from the polylinker sequence of pBluescript (Stratagene; see "Experimental Procedures"). Ϫ, oligonucleotide was omitted to control for unspecific protein binding to the matrix. Asterisks indicate matrix binding proteins. C, Western blot analysis of the RNA-affinity precipitates shown in B using the anti-hnRNP A1 monoclonal antibody 4B10. Numbers on the left indicate protein molecular weight standards (Benchmark prestained protein ladder, Life Technologies, Inc.) in kDa.
v5 was functionally relevant, we transiently overexpressed hnRNP A1 and determined the behavior of a CD44 v5-minigene splice reporter. The minigene construct contains the murine CD44-v5 exon and adjacent intron sequences between preproinsulin exons 2 and 3 and their corresponding intron sequences. This construct had previously been shown to recapitulate cell type-specific and -inducible alternative splicing of the endogenous CD44-v5 exon (29). Because hnRNP A1 is an abundant protein in most cell types (39), we reasoned that it could be difficult to achieve overexpression. We therefore used the mouse erythroleukemia cell line CB3, which, due to loss of one hnRNP A1 allele and a proviral integration 3Ј to the coding sequence of the second allele, shows no detectable expression of hnRNP A1 protein (30, 40; and Fig. 3A). After cotransfection of the v5 minigene (for scheme, see Fig. 3B) with the empty expression vector, the majority of the minigene-derived mRNA included CD44 exon v5 (indicated by the larger PCR product of 361 bp; Fig. 3C, lane 1). Cotransfection with increasing amounts of an hnRNP A1 expression vector, however, led to exon skipping in a dose-dependent manner (Fig. 3C, lanes 2  and 3). An expression construct carrying the hnRNP A1 cDNA in antisense orientation did not repress v5 exon inclusion (Fig.  3C, lane 4), suggesting that expression of hnRNP A1 protein is required for splice silencing. Expression of the antisense cDNA reproducibly enhanced v5 exon inclusion slightly. This result could be explained by the residual expression of hnRNP A1 mRNA in CB3 cells (40). Surprisingly, inhibition of CD44 v5 exon inclusion by hnRNP A1 was not confined to the hnRNP A1-deficient CB3 erythroleukemia cells but could also be shown in NIH-3T3 murine fibroblasts and in KLN205 mouse carcinoma cells. Although endogenous hnRNP A1 was easily detectable in both cell lines (see Fig. 3A), they showed predominantly inclusion of exon v5 in the minigene mRNA in the absence of additional hnRNP A1 (Fig. 3D, lanes 1 and 3). Transient overexpression of hnRNP A1 shifted the splice pattern to skipping of the v5 exon (Fig. 3D, lanes 2 and 4). To estimate the level of hnRNP A1 overexpression, we transfected NIH-3T3 cells with an epitope-tagged version of hnRNP A1. By Western blotting of cell extracts using an anti-hnRNP A1 antibody, we could determine the amount of epitope-tagged hnRNP A1 (which has a higher molecular weight due to the epitope tag) to be 35% of the endogenous hnRNP A1 level (data not shown). Taking into account the transfection efficiency of the cells under these conditions (13% cells transfected), it can be estimated that there is an approximately 3-fold overexpression of hnRNP A1 in the transfected cells. The repression of exon inclusion by overexpressed hnRNP A1 suggests that the endogenous levels of hnRNP A1 (and/or its functional state) in the two cell lines do not suffice for repression of v5 exon usage in the minigene pre-mRNA and that repression of CD44 v5 exon inclusion by overexpression of hnRNP A1 is not a peculiarity of the CB3 erythroleukemia cells.
HnRNP A1-dependent Exon Silencing Shows Exon Specificity-Using different combinations of variant exons, a multitude of variant CD44 isoforms can be generated under physiological conditions and in cancer (21). To assess whether hnRNP A1 is a general silencer of variant CD44 exons, or whether it regulates the usage of certain variant exons selectively, we used similar minigene constructs carrying either one of three other CD44 variant exons. When testing their splice patterns under the influence of hnRNP A1, inclusion of exon v6 was repressed similarly to exon v5 (Fig. 4, lanes 1 and 2, and lanes 5 and 6) whereas inclusion of exons v4 (Fig. 4, lanes 3 and 4) and v7 (Fig. 4, lanes 7 and 8) were not, indicating that hnRNP A1 does not cause general skipping of variant exon sequences. Rather, hnRNP A1-mediated splice silencing is exon-specific.
Repression of CD44 v5 Exon Inclusion Requires the Exonic Sequence Elements to Which hnRNP A1 Binds in Vitro-HnRNP A1 could be shown to associate with RNA sequences of CD44 exon v5 in vitro and in vivo (see Figs. 1 and 2). Moreover, Binding of the bifunctional hybrid-RNA to each of two hybrid proteins will activate transcription (indicated by arrow) of the reporter genes by recruiting the transcriptional activation domain to the LexA-operator sites in the promoter. B, L40uraMS2 yeast cells cotransformed with the expression constructs indicated plated on histidine-deficient plates. C, detection of lacZ expression in L40uraMS2 yeast cells cotransformed with the indicated expression constructs. Yeast cells were grown on histidine-deficient plates followed by transfer to nitrocellulose filters and X-gal staining. pRH5Ј-v5, expression vector for MS2-v5 hybrid-RNA; pRH5Ј-blue, expression vector for control hybrid-RNA in which the v5 exon sequence (except for the four 5Ј-most and the three 3Ј-most nucleotides) had been replaced for a pBluescript-polylinker sequence; pYESTrp2-A1, expression construct that gives rise to an hnRNP A1 protein fused to the B42 transcriptional transactivation domain; pYESTrp2, empty expression construct, expresses transactivation domain only. As a positive control to check the system in our hands, we used constructs expressing a B42-IRP (iron regulatory protein) fusion (pYESTrp2-IRP) and a MS2-IRE (iron-responsive element) hybrid-RNA (pRH3Ј-IRE), respectively. hnRNP A1 can be cross-linked by UV irradiation to these sequences (data not shown). Thus, exon specificity in hnRNP A1-dependent splice silencing could be the result of sequence preference in RNA binding of hnRNP A1 itself or of an hnRNP A1-containing protein complex. To verify that exon silencing of the v5 exon by hnRNP A1 depended on the exon sequences to which it binds in vitro, we made use of minigene constructs containing mutant v5 exons. In the mutants, either the L or M exonic silencer subdomains or both had been replaced by blue polylinker sequences of the same length (29; see Fig. 5A). HnRNP A1 did not bind to the polylinker sequences either in vitro or in vivo (see Figs. 1 and 2). When we cotransfected the minigene constructs into NIH-3T3 cells together with the empty expression plasmid, the mutant v5 exon sequences were included in the minigene mRNA to a slightly higher level than with the wild-type v5 exon (Fig. 5B, compare lanes 1, 3, 5, and  7). However, whereas cotransfection of the hnRNP A1 expression vector repressed inclusion of the wild-type v5 exon by 50.5% (Fig. 5B, lanes 1 and 2), inclusion of the L and M mutants was affected only by 9.4% and 22.4%, respectively (Fig. 5B, lanes 3-6) and remained virtually unaffected in the L/M double mutant v5bls7 (Fig. 5B, lanes 7 and 8). Thus, exon silencing by hnRNP A1 required the specific sequences in v5 subdomains L and M. As shown in Fig. 1B above, subdomain R also interacted with hnRNP A1. Because this domain, in addi-tion to negative regulatory elements, contains the exon recognition element without which the exon sequence cannot be included in mature mRNA at all, replacement of R by the blue sequence abolishes exon inclusion completely (29). Therefore, the role of subdomain R in hnRNP A1-dependent exon silencing FIG. 3. HnRNP A1 represses usage of CD44 exon v5 in vivo. A, Western blot analysis of hnRNP A1 expression in murine cell lines used for cotransfections. 60 g of nuclear extracts was resolved by SDS-polyacrylamide gel electrophoresis, blotted, and probed with a monoclonal antibody (4B10) against hnRNP A1. The band corresponds to the expected 34-kDa molecular mass of hnRNP A1. B, schematic representation of minigene organization and splicing alternatives; positions of PCR primers (arrows) and the predicted size of the PCR products are indicated. Black boxes denote insulin exons, hatched box represents CD44 exon v5. C, RT-PCR analysis of CB3 erythroleukemia cells transiently cotransfected with 2 g of the CD44 v5-minigene construct pETv5 (29) and either an empty expression plasmid (pcDNA3) or an expression construct carrying the murine hnRNP A1 cDNA in sense (pcDNA3-A1) or antisense (pcDNA3-A1as) orientation, in the amounts indicated. In the case where 2 g of pcDNA3-A1 was cotransfected, the total amount of pcDNA3-based DNA was brought to 8 g with the empty pcDNA3 plasmid. Numbers on the left indicate sizes (in base pairs) of DNA marker bands (lane M). D, RT-PCR analysis of CD44 v5-minigene RNA in NIH-3T3 murine fibroblasts and KLN205 mouse carcinoma cells after cotransfection with 2 g of pETv5-minigene plasmid and 8 g of either the hnRNP A1 expression construct (pcDNA3-A1) or the empty expression vector (pcDNA3). Results of a typical experiment are shown, values for exon inclusion (% exon incl; expressed as percentage of v5-containing RT-PCR products relative to total RT-PCR products) are boxed.

FIG. 4. HnRNP A1-dependent exon silencing is exon-specific.
RT-PCR analysis of minigene splicing patterns from cytoplasmic RNA of NIH-3T3 cells transiently cotransfected with 2 g of the pET-minigene constructs indicated (containing exons v5, v4, v6, or v7) and 8 g of either the hnRNP A1 expression construct pcDNA3-A1 (ϩ) or the empty expression vector pcDNA3 (Ϫ). Values for exon inclusion (% exon incl; expressed as percentage of variant exon-containing RT-PCR products relative to total RT-PCR products) are boxed.

FIG. 5. Repression of CD44 exon-v5 inclusion by hnRNP A1 is dependent on v5 exon sequences.
A, schematic illustration of mutant CD44-v5 exons in the minigenes used for cotransfections in B. Open boxes, regulatory subdomains L, M, and R of CD44 exon v5; gray boxes, unrelated sequence (blue) derived from the polylinker of the pBluescript plasmid (Stratagene), used to replace v5 exon sequences; hatched boxes, dimerized 9-bp sequence corresponding to purine-rich enhancer element from exon 5 of the cardiac troponin T gene (31). B, RT-PCR analysis of NIH-3T3 cells transiently cotransfected with minigenes containing the mutant CD44-v5 exons shown in A and either an empty expression construct (ϪA1) or a construct expressing hnRNP A1 (ϩA1). The figure shows the results of a typical experiment. Values for exon inclusion (% exon incl; expressed as percentage of v5-containing RT-PCR products relative to total RT-PCR products) are boxed. could not be examined by an R domain replacement mutant. We could ask, however, whether the silencer domains L and M were sufficient to mediate hnRNP A1 repression if subdomain R was replaced by a different splice enhancer, or whether sequences specific to the R subdomain contributed to repression. With this reasoning, we generated a minigene construct in which the R subdomain of exon v5 had been replaced by a duplicated 9-bp purine-rich exonic enhancer sequence from the cardiac troponin T (TnT) alternative exon 5 (31). Although the TnT sequences led to a lower level of exon inclusion (76.2%) than that of the wild-type v5 exon (90.7%), hnRNP A1 could inhibit exon inclusion of the chimeric exon only by 30.2% (Fig.  5B, lanes 9 and 10), compared with 50.5% in the case of the wild-type exon (lanes 1 and 2). This result argues in favor of a contribution of R sequences to repression of v5 exon usage by hnRNP A1.
The results obtained by the v5 exon mutants thus indicate that hnRNP A1-dependent repression of v5 exon inclusion depends on all three regulatory subdomains of the exon and strongly suggest that the binding of hnRNP A1 to these sequences is crucial for exon skipping.
HnRNP A1-dependent Exon Silencing Is Relieved by Oncogenic Cell Signaling-We previously found that constitutively activated Ras is capable of activating inclusion of CD44 exon v5 in RNA of a T-lymphoma cell line (29). This link between oncogene action and up-regulation of CD44 variant exon inclusion may be related to the generation of CD44 variant isoforms during tumor progression. Concerning the action of Ras on splicing, an obvious question was whether repression of v5 exon inclusion by hnRNP A1 would be sensitive to Ras signaling. We performed transient three-factor cotransfections using the CD44-v5 minigene, the hnRNP A1 expression construct, and either an expression vector for activated Ha-Ras or a corresponding control plasmid. Cotransfection into CB3 erythroleukemia cells of the minigene with the hnRNP A1 expression construct and the control plasmid led, as expected, to skipping of the v5 exon sequence (Fig. 6A, compare lanes 1 and 3). In contrast, cotransfection with the Ras expression construct largely prevented the silencing effect of hnRNP A1 (see Fig. 6A,  lane 4). Surprisingly, similar experiments performed with NIH-3T3 and KLN205 cells showed only a marginal Ras effect (data not shown). One possibility is that Ras cannot efficiently induce the same effector pathways in NIH-3T3 and in KLN 205 cells as it does in CB3 cells, and it may not counteract the higher hnRNP A1 levels in these cells. Interestingly, however, constitutively active mitogen-activated protein/ERK kinase kinase 1 (MEKK1) (Fig. 6A, lanes 6 and 7) and an activated version of the small GTP-binding protein Cdc42 (Fig. 6A, lanes  8 and 9) reverted repression of v5 exon inclusion by hnRNP A1. A similar signaling-induced relief of hnRNP A1-dependent silencing was observed for the v6 exon (data not shown). MEKK1 and Cdc42 have in common their ability to activate the Junkinase (JNK) pathway (41,42), and both molecules can be activated by Ras signaling (43). To exclude an effect of the signaling molecules on the expression level of hnRNP A1, we cotransfected an epitope-tagged version of hnRNP A1. Coexpression of Ras, MEKK1, or Cdc42 did not down-regulate the amount of hnRNP A1 protein (data not shown for Ras, and Fig. 6B).
We conclude that oncogene-driven signaling pathways antagonize the activity of an exonic splice-silencing complex in which hnRNP A1 plays a decisive role. DISCUSSION Variant CD44 isoforms are generated by alternative pre-mRNA splicing and are involved in a variety of physiological and pathological processes, among which tumor progres-sion has received most attention. Despite this attention, the splice factors responsible for the regulation of alternative splicing have not been identified.
This paper describes a component, hnRNP A1, of an exonspecific splice-silencing complex, which is inactivated by signal transduction from Ras, MEKK1, or Cdc42. HnRNP A1 is shown here to bind to exonic regulatory sequences in CD44 exon v5 in vitro and in yeast cells in vivo. Transient overexpression of hnRNP A1 repressed recognition of the v5 exon in three different cell lines. The finding that exon v5 in both NIH-3T3 cells and KLN205 carcinoma cells was included despite abundant endogenous hnRNP A1, suggests that either the endogenous levels of hnRNP A1 do not suffice to efficiently repress exon inclusion from the v5-minigene pre-mRNA, or that the repressing activity of hnRNP A1 is down-regulated in these cells, e.g. The empty expression constructs, pRK5myc and pCMV5, served as controls. The figure shows the results of one typical experiment. Values for exon inclusion (% exon incl; expressed as percentage of v5-containing RT-PCR products relative to total RT-PCR products) are boxed. B, analysis of transiently overexpressed epitope-tagged hnRNP A1 in the absence or presence of coexpressed MEKK1 and Cdc42, respectively. NIH-3T3 cells were transiently cotransfected with a plasmid expressing a myc-tagged hnRNP A1 protein (pcDNA3-mycA1), or with the empty expression plasmid (pcDNA3), and either CMV expression constructs for MEKK1 or Cdc42, or an empty CMV expression construct as a control (con). 100 g of nuclear extract from the transfected cells was analyzed by Western blotting using the anti-myc monoclonal antibody 9E10. The band corresponds to the expected 35.5-kDa molecular mass of the myc-tagged hnRNP A1 protein.
through post-translational modification. Transient overexpression of hnRNP A1 could exhaust modifying enzyme systems in these cells and thus lead to a pool of hnRNP A1 active in splice silencing. The 3-fold overexpression determined in NIH-3T3 cells is well within the range of physiological differences in hnRNP A1 protein levels observed between different tissues (44).
HnRNP A1 has previously been described as a factor that can antagonize selection of alternative 5Ј-and 3Ј-splice sites by certain SR proteins in vitro and in vivo (40,(45)(46)(47)(48). Similarly to the hnRNP A1-dependent exon silencing of CD44 variant exons described in this paper, the antagonistic effect of transiently overexpressed hnRNP A1 on 5Ј-and 3Ј-splice site selection could be shown in cell lines expressing already high levels of endogenous hnRNP A1 protein (47,48). Very recently, two other examples of hnRNP A1 binding in vitro to negative regulatory exon sequences have been described: to the human fibroblast growth factor receptor 2 K-SAM exon and to human immunodeficiency virus tat exon 2 (49,50). In addition, intronic hnRNP A1-binding sites causing exon skipping have been found in pre-mRNA of hnRNP A1 (51). In the case of human immunodeficiency virus tat exon 2, hnRNP A1 was capable of inducing skipping of the exon in an in vitro splicing system (50). Recruitment of a MS2-hnRNP A1 fusion protein to a K-SAM exon carrying an MS2 recognition site resulted in exon skipping in transient cotransfection experiments (49). However, hnRNP A1 could not be shown to repress exon inclusion via the authentic hnRNP A1-binding exon element identified in vitro.
Interestingly, the silencing effect of hnRNP A1 could be detected using minigene constructs carrying CD44 exons v5 or v6, but not v4 or v7, thus indicating exon specificity of silencer action. The exon preference of the silencing effect seems to reflect special sequence requirements: inhibition of v5 exon inclusion by hnRNP A1 depends on the presence of exonic silencer sequences in subdomains L and M. The impaired inhibitory effect of hnRNP A1 on exon inclusion upon replacement of subdomain R by a heterologous splice enhancer from the TnT exon 5 could indicate additional sequence elements in subdomain R (harboring also the exon recognition element), contributing to repression of v5 exon inclusion by hnRNP A1. Alternatively, the TnT enhancer element could stimulate splicing through a molecular mechanism distinct from that of the enhancer in R and which could be not susceptible to hnRNP A1-mediated repression. The possibility of additional elements in subdomain R that add to the hnRNP A1 effect is compatible with linker-scan mutation analyses, which revealed silencing elements in the R subdomain of exon v5 (29). A consensus RNA-binding site (UAGGG(A/U)) had been obtained for hnRNP A1 by the SELEX procedure (52). However, there are no elements corresponding to this sequence in CD44 exon v5. One could envisage that hnRNP A1 binds, possibly in association with other partner proteins, through an exon-specific combination of distinct small sequence elements in a cooperative manner. These may be stretched out over the entire exon. Thereby, the silencing complex could displace positively acting factors binding to the exon recognition element within the 3Ј-part of the exon (subdomain R), or it might interfere with their potential to recruit spliceosomal components to the splice sites upstream and/or downstream of the exon. Sequence comparison of the CD44 variant exons v4 through v7 revealed numerous similarities. 2 These similarities could indicate a common ancestor during exon evolution. In addition to these general sim-ilarities, we found stretches of nucleotides that are very similar between exons v5 and v6. However, it is too early to speculate on their significance in exon selection.
The repressing effect of hnRNP A1 on CD44 v5 exon inclusion was abolished in the presence of activated Ras and by dominant active MEKK1 or Cdc42, two molecules that are part of Ras-effector pathways (43). Like Ras, activated forms of Cdc42 have been shown to have transforming potential (53,54). Furthermore, Cdc42 has been demonstrated to be necessary for Ras transformation in fibroblasts (53) and to induce invasive cell growth (55,56). Thus, oncogenic cell signaling can interfere with hnRNP A1-dependent exon silencing. This interference could be due to the targeting of hnRNP A1 itself, or of associated or antagonizing splice factors. Phosphorylation of hnRNP A1 has been previously shown to increase in response to platelet-derived growth factor via a pathway requiring protein kinase C (PKC) and has been suggested to control cytoplasmic localization of hnRNP A1 (57). This pathway is most likely not involved in relief of exon silencing by Cdc42 or MEKK1, because a constitutively active mutant of PKC could not release repression, and Cdc42 and MEKK1 did not induce an increase in hnRNP A1 phosphorylation. 2 Consistent with the possibility that hnRNP A1 itself is a target in signaling-induced relief of exon silencing, we have observed changes in methylation and in phosphorylation of hnRNP A1 in response to MEKK1 or Cdc42 overexpression (data not shown). Whether these changes in modification are relevant for hnRNP A1-silencing function still needs to be explored.
We thus can summarize: hnRNP A1 can function as a decisive component of an exon-selective splice silencer complex. Oncogene-derived signals may converge at this complex, inactivating its function and thus leading to the generation of mRNAs for variant CD44 isoforms in the course of tumor progression. Furthermore, such regulatable and exon-discriminatory complexes could be the basis for controlling the elaborate splice patterns observed in highly complex genes, like CD44.