Distinct Sets of Adjacent Heterogeneous Nuclear Ribonucleoprotein (hnRNP) A1/A2 Binding Sites Control 5′ Splice Site Selection in the hnRNP A1 mRNA Precursor*

In the heterogeneous nuclear ribonucleoprotein (hnRNP) A1 pre-mRNA, different regions in the introns flanking alternative exon 7B have been implicated in the production of the A1 and A1B mRNA splice isoforms. Among these, the CE1a and CE4 elements, located downstream of common exon 7 and alternative exon 7B, respectively, are bound by hnRNP A1 to promote skipping of exon 7Bin vivo and distal 5′ splice site selection in vitro. Here, we report that CE1a is flanked by an additional high affinity A1 binding site (CE1d). In a manner similar to CE1a, CE1d affects 5′ splice site selection in vitro. Consistent with a role for hnRNP A1 in the activity of CE1d, a mutation that abrogates A1 binding abolishes distal 5′ splice site activation. Moreover, the ability of CE1d to stimulate distal 5′ splice site usage is lost in an HeLa extract depleted of hnRNP A/B proteins, and the addition of recombinant A1 restores the activity of CE1d. Notably, distal 5′ splice site selection mediated by A1 binding sites is not compromised in an extract prepared from mouse cells that are severely deficient in hnRNP A1 proteins. In this case, we show that hnRNP A2 compensates for the A1 deficiency. Further studies with the CE4 element reveal that it also consists of two distinct portions (CE4m and CE4p), each one capable of promoting distal 5′ splice site use in an hnRNP A1-dependent manner. The presence of multiple A1/A2 binding sites downstream of common exon 7 and alternative exon 7B probably plays an important role in maximizing the activity of hnRNP A1/A2 proteins.

The alternative splicing of mRNA precursors (pre-mRNAs) 1 is a major contributor to the diversity of the mammalian pro-teome (1)(2)(3). The control of splice site selection therefore has profound implications in the production of protein isoforms with different functions. Recent progress in uncovering the molecular strategies that control alternative splicing has led to the identification of many types of sequence elements that influence either positively or negatively the selection of the alternative splice sites. Exonic splicing enhancers are bound by specific members of the SR protein family that can enforce the use of weak 5Ј and 3Ј splice sites (reviewed in Ref. 4). Enhancer elements have also been described in the introns flanking some alternative exons (5)(6)(7)(8)(9). Other types of proteins, including members of the hnRNP F/H family of proteins, can bind specifically to intron or exon control elements and hence can contribute to enhancer activity (10 -14).
Elements that reduce the use of a neighboring splice site are also important in the control of splice site selection. In many cases, the activity of splicing silencers can be mediated by proteins that inhibit specific steps of splice site recognition or spliceosome assembly. A frequent example of this kind of splicing control in mammals involves the polypyrimidine tract-binding protein, which binds to some 3Ј splice sequences and prevents U2AF binding (15)(16)(17)(18)(19). Other examples uncovered in mammalian pre-mRNAs include the binding of the SR proteins ASF/SF2 and SRp30c upstream of the branch site in the adenovirus IIIa pre-mRNA and the hnRNP A1 pre-mRNA, respectively (20,21). The mechanisms by which inhibition occurs in these cases appear different, since ASF/SF2 prevents U2 snRNP binding in the adenoviral pre-mRNA, whereas SRp30c does not prevent the assembly of a U2-containing complex on the downstream 3Ј splice site (21,22).
Members of the family of core hnRNP A/B proteins have also been identified as factors involved in the modulation of splice site selection. Using model pre-mRNAs carrying competing 5Ј splice sites, important shifts toward distal 5Ј splice sites can be obtained by the addition of purified or recombinant hnRNP A1 to a HeLa nuclear extract (23,24). The Drosophila hrp48 protein, which is similar to hnRNP A1, is required in collaboration with P-element somatic inhibitor and the U1 snRNP to repress splicing of the P-element pre-mRNA in somatic tissues (25). hnRNP A1 can elicit exon skipping, but not all pre-mRNAs are responsive to variations in the concentration of hnRNP A1 in vitro (26). A1 can also negatively affect the use of 3Ј splice sites in a variety of exons including the alternative exon of fibroblast growth factor receptor 2 gene (27), the tat and vpr exons of the human immunodeficiency virus (28 -30), and the V6 exon of the human CD44 gene (31,32). Consistent with the finding that A1 can recognize specific RNA elements (33), most of the functions that have been attributed to hnRNP A1 in natural pre-mRNAs are based on its ability to interact with specific sequences. In all cases examined to date, the hnRNP A1 splice isoform A1B, the A2 protein, and its splicing variant B1 can functionally replace A1 in 5Ј splice site and 3Ј splice site selection assays in vitro (24,28,29).
We have shown previously that hnRNP A1 can modulate the alternative splicing of its own pre-mRNA through binding to sequences in the introns flanking alternative exon 7B. The 17-nt-long CE1a element downstream of exon 7 and the 24-ntlong CE4 sequence downstream of alternative exon 7B can promote distal 5Ј splice site selection in an A1-dependent manner (34,35). Both CE1a and CE4 contain the sequence UA-GAGU, which closely resembles the "winner" A1 binding site UAGGGU obtained by selection of amplified pools of RNA sequences (33). Whereas these high affinity A1 binding sites promote strong shifts in 5Ј splice site utilization in vitro, their effect on splicing is not associated with equivalent changes in U1 small nuclear ribonucleoprotein particle binding to the competing 5Ј splice sites (34). We have proposed that the mechanism by which hnRNP A1 controls 5Ј splice site selection in this system involves an interaction between bound A1 molecules, an event that would place the proximal 5Ј splice site in a loop and would bring in closer proximity the most distant pair of splice sites (35)(36)(37). Here, we have uncovered additional elements capable of promoting distal 5Ј splice site selection in vitro. Notably, we find that sequences flanking CE1a and different portions of CE4 can individually interact with hnRNP A1. Our results indicate that control elements promoting distal 5Ј splice site utilization in the hnRNP A1 pre-mRNA are organized in groups of adjacent A1 binding sites. This organization may facilitate the recruitment of hnRNP A1/A2 proteins and may stimulate or stabilize the proposed change in pre-mRNA conformation.
Transcription and Splicing Assays-Splicing substrates were produced from plasmids linearized with ScaI, except for p45-M3 which was obtained by cutting with EarI, and transcribed with T3 RNA polymerase (Amersham Biosciences) in the presence of cap analog and [␣-32 P]UTP (Amersham Biosciences). CE1e, CE1d, M6, and M7 RNAs were produced from pKCE1e, pKCE1d, pKCE1dM6, and pKCE1dM7 linearized with HindIII, and CE1a and CE4m were produced from pKCE1a and pKCE4m linearized with EcoRI and transcribed as above. RNA purification was performed as described in Ref. 38. HeLa, CB3C7, and CB3C7-20 nuclear extracts were prepared (39) and used in splicing reactions as described previously (34). Creatine kinase was added to HeLa, CB3C7, and CB3C7-20 extracts at a final concentration of 1 unit/12.5 l. In all splicing gels, we have used lariat molecules (intermediates and products) to monitor splicing at each 5Ј splice site. The fact that these molecules usually migrate above the pre-mRNA allows a more precise estimation of splicing efficiency, particularly when pre-mRNA degradation obscures mRNA production. Furthermore, adequate separation between lariat products requires longer runs or the use of lower percentage acrylamide gels, which cause the distal 5Ј exon and the distal mRNA to run out the gel.
Gel Shift Assays-The RNA samples were heated at 75°C for 5 min in a splicing mix (34) and snap-cooled on ice. The rA1, rA2, and rA1B proteins were then added to the reactions and allowed to incubate on ice for 10 min prior to the addition of heparin (0.74 mg/ml final concentration) and loading dye. The mixtures were run on 6% native acrylamide gels (29:1 acrylamide/bisacrylamide) in 1ϫ TBE running buffer.
RNA Affinity Chromatography-The depletion of hnRNP A1/A1B/ A2/B1 proteins from nuclear extracts was carried out essentially as described in Caputi et al. (28). Briefly, 50 nmol of synthetic RNA oligonucleotide corresponding to a sequence within CE1a (UACCU-UUAGAGUAGGC) (Dharmacon Research Inc.) were incubated for 1 h at 4°C, protected from light, in a 100-l reaction volume containing 100 mM Tris-HCl, pH 7.5, and 10 mM sodium periodate. The periodatetreated RNA was coupled to 0.5 ml of agarose adipic acid hydrazide resin following the manufacturer's protocol (Amersham Biosciences). The resin was washed twice with 10 ml of storage buffer (20 mM Hepes-KOH, pH 7.9, 100 mM KCl, 20% glycerol, 5.7 mM MgCl 2 , 1 mM dithiothreitol) and kept as a 50% slurry at 4°C. The coupling efficiency, which was typically higher than 95%, was measured by comparing the absorbance at 260 nm of 1% of the input periodate-treated RNA to 10% of the unbound material. 175 l of HeLa nuclear extract containing 5.7 mM MgCl 2 , 0.90 mM ATP, 36 mM phosphocreatine, 3.58 mM dithiothreitol, and 1.25 unit/ml RNAguard was incubated with 50 l of packed beads for 10 min at 30°C under agitation. The mixture was spun, and the supernatant was transferred to a second tube containing 50 l of the same packed beads. The beads were washed four times with 1 ml of 70% buffer D (20 mM Hepes-KOH, pH 7.9, 100 mM KCl, 20% glycerol, 1 mM dithiothreitol) containing 5 mM MgCl 2 . The bound proteins were eluted from the column with 100 l of loading dye (62.5 mM Tris-HCl, pH 6.8, 6 M urea, 10% glycerol, 2% SDS, 0.7 M mercaptoethanol, 0.003% bromphenol blue).
Western Analysis-Equivalent ratios of nuclear extract and proteins eluted from the CE1a RNA column were separated on SDS-10% polyacrylamide gels. These samples were transferred to nitrocellulose filters and probed for the presence of hnRNP A1/A1B/A2/B1 using a rabbit serum containing antibodies against a peptide sequence shared by these proteins (kindly provided by Telogene Inc., Sherbrooke, Québec, Canada). This serum was used at a dilution of 1:1000, and decorated proteins were revealed using the ECL detection kit (Amersham Biosciences) according to the manufacturer's instructions.

Additional Sequences within CE1 Modulate 5Ј Splice Site
Selection-The 150-nt-long CE1 element is located in the intron downstream of common exon 7 in the hnRNP A1 pre-mRNA. CE1 is required for efficient skipping of alternative exon 7B in vivo and stimulates distal 5Ј splice site use when inserted between the 5Ј splice sites of exon 7 and exon 7B in a model pre-mRNA in vitro (34). We have reported previously that a small 17-nt segment (CE1a) is responsible for a large portion of the effect of CE1 on 5Ј splice site selection (34). However, because CE1a does not fully restore the activity of the larger CE1 element, additional sequences within CE1 may also be promoting distal 5Ј splice site usage. To identify these sequences, we inserted the 3Ј end portion of CE1 (CE1z, 58 nt; Fig. 1A) between the two 5Ј splice sites in our model pre-mRNA. Typically, the level of proximal and distal 5Ј splice site use can be assessed by monitoring the production of lariat molecules following incubation of a labeled pre-mRNA in a nuclear extract and separating the spliced products on a denaturing gel. In comparison with the control S1 pre-mRNA, incubation of SCE1z RNA in a HeLa extract indicated that CE1z stimulated the use of the distal 5Ј splice site as observed by an increase in lariat products generated by the use of this site (Fig.  1B, compare lane 4 with lane 1). The amplitude of the stimulation was comparable with the effect of CE1a (Fig. 1B, compare lane 4 with lane 3). As observed previously, the full 150-nt CE1 element promoted a nearly complete loss of proximal 5Ј splice site use (Fig. 1B, lane 2). To further define the sequence within CE1z that was responsible for the shift, we fragmented CE1z into two smaller slightly overlapping units (CE1z35 and CE1z31). Whereas CE1z31 had a small effect on 5Ј splice site selection, CE1z35 was much more active in switching splicing to the distal 5Ј splice site (Fig. 1B, compare lanes 5 and 6 with  lane 4). A further dissection of CE1z35 into CE1e (16 nt) and CE1d (19 nt) showed that CE1d, but not CE1e, could shift splicing toward the distal 5Ј splice site (Fig. 1B, lanes 7 and 8,  respectively). These results indicate that the 19-nt-long CE1d element contains sequences that can promote distal 5Ј splice site selection in vitro as efficiently as the previously described CE1a element. When we tested a portion of CE1 that contained both CE1a and CE1d (CE1ad), we observed that the combination of the two elements shifted 5Ј splice site selection to a level that was comparable with the level obtained with the complete CE1 element (Fig. 1C, compare lane 3 with lane 2). Thus, the activity of CE1 in HeLa extracts apparently results from the combined effects of CE1a and CE1d.
hnRNP A1 Binds to CE1d-Because hnRNP A1 has been implicated in the activity of CE1a, we tested whether hnRNP A1 was also involved in the activity of CE1d. First, we assessed FIG. 1. Several elements within CE1 control 5 splice site selection. A, the structure of the pre-mRNA S1 is represented at the top. The SmaI site indicates the position where sequence elements were inserted between the two 5Ј splice sites. Below is a sequence alignment of the human and mouse CE1 elements. The underlined regions correspond to sequences inserted in pS1 and tested for their ability to influence 5Ј splice site selection in vitro. B and C, labeled S1 pre-mRNA derivatives carrying different insertions were incubated in a HeLa nuclear extract for 2 h, and splicing products were fractionated on an 11% acrylamide, 8 M urea gel. The relative frequencies of distal and proximal splice site utilization can be estimated by comparing the intensity of the bands derived from proximal and distal lariat intermediates and products that migrate above the pre-mRNAs. The position of the pre-mRNAs as well as of the distal and proximal lariat products is indicated. SCE1 and SCE1a have been described previously as S2 and S10, respectively by Chabot et al. (34). the ability of hnRNP A1 to interact with CE1d by performing a gel mobility shift assay using increasing amounts of a recombinant GST-A1 protein (rA1). The RNA used for this assay is 83 nt long and contains plasmid sequences followed by the CE1d element. Since CE1e had no effect on 5Ј splice site selection, it was used as a control in place of the CE1d element. Slower migrating complexes were observed with CE1d but not with CE1e (Fig. 2C, top panel), indicating a specific interaction of rA1 with CE1d.
The observation that recombinant A1 protein can specifically interact with CE1d suggests that this interaction may be required for the CE1d-dependent switch in 5Ј splice site selection observed in vitro. To test this possibility, we produced seven mutations spanning the CE1d element and assessed their effects on 5Ј splice site selection in a HeLa nuclear extract. The mutations are listed in Fig. 2A, and they were tested in a slightly shorter version of the S1 pre-mRNA (pre-mRNA 45) for cloning reasons. Inserting CE1d in pre-mRNA 45 (WCE1d) promotes distal 5Ј splice site utilization in a manner similar to what was observed in the S1 backbone (Fig. 2B, lane 2). The majority of the mutations in CE1d did not compromise the activity of the element (Fig. 2B, compare lanes 3-7 and 9 with lane 2). Mutant M6, however, failed to activate distal 5Ј splice site selection, yielding a splicing pattern that was similar to the profile obtained in the absence of CE1d (Fig. 2B, compare lane 8 with lane 1). We then tested the small 83-nt-long RNA containing the M6 sequence (M6*) instead of CE1d for binding by hnRNP A1. In comparison with a similar transcript containing the M7 sequence (M7*), which was bound by rA1 (Fig. 2C,  bottom panel, lanes 5-8), M6* RNA was not bound significantly by rA1 (lanes 1-4). The correlation between A1 binding and the efficiency of distal 5Ј splice site activation is consistent with the notion that A1 is involved in the activity of CE1d.
The Activity of CE1d Is Mediated by hnRNP A1-To demonstrate that hnRNP A1 mediates the activity of CE1d, we carried out the depletion of hnRNP A1 from a HeLa nuclear extract by affinity chromatography using RNA molecules carrying a high affinity binding site for hnRNP A1. The CE1a RNA was covalently linked to agarose-adipic acid beads and incubated in the presence of a HeLa nuclear extract. The flowthrough fraction was recovered and used as the depleted nuclear extract. Western analysis using a polyclonal antibody that recognizes the core hnRNP A1, A2, B1, and A1B proteins (Fig. 3A, lane 1) indicates that the majority of these hnRNP proteins had been removed from the nuclear extract (lane 5). hnRNP A1, A2, B1, and A1B proteins were found in the fractions bound to successive CE1a columns (Fig. 3A, lanes 6 and  7). A mock-depleted nuclear extract was also prepared by load- ing a HeLa nuclear extract on a column lacking RNA. The mock-depleted extract contains the core hnRNP A/B proteins (Fig. 3A, lane 2), and no signal was detected in the bound fractions (lanes 3 and 4).
The mock-and A/B-depleted extracts were then tested for activity using a CE1a-containing pre-mRNA (SCE1a). We have shown previously that 5Ј splice site usage on this pre-mRNA can be displaced from the distal to the proximal donor site by the addition of an excess of DNA oligonucleotide carrying A1 binding sites and that supplementation with rA1 restores efficient distal 5Ј splice site use (35). Whereas distal 5Ј splice site use was predominantly observed in the mock-depleted extract (Fig. 3B, lane 1), only the proximal 5Ј splice site was used in the A/B-depleted extract (lane 2), consistent with a role for hnRNP A/B proteins in the activity of CE1a. SCE1d pre-mRNA was similarly spliced to the proximal 5Ј splice site in the A/Bdepleted extract, suggesting that these proteins may also be required for the activity of CE1d (Fig. 3B, lane 4). The addition of increasing amounts of rA1 to the A/B-depleted extract shifted splicing toward the distal donor site in a CE1d-dependent manner (Fig. 3C). At the highest concentration tested, the addition of rA1 to the SCE1d pre-mRNA nearly completely abrogated proximal 5Ј splice site use and activated the distal 5Ј splice site (Fig. 3C, lane 10). In contrast, at the same concentration of rA1, proximal 5Ј splice site use remained the predominant choice for the control S1 pre-mRNA (Fig. 3C, lane 6). Thus, the greater sensitivity of a pre-mRNA containing CE1d to the supplementation with rA1 indicates that hnRNP A1 can mediate the activity of CE1d.
hnRNP A2 Can Also Mediate the Activity of CE1a and CE1d-To determine whether hnRNP A1 is absolutely neces-sary for the activity of CE1a and CE1d, we monitored splicing in an extract prepared from a mouse erythroleukemic cell line previously shown to be severely deficient in hnRNP A1 protein expression. The CB3C7 cell line expresses at least 250-fold lower levels of A1 and A1B mRNAs, because one Hrnpa1 allele has been deleted while the other allele has suffered a retroviral insertion event (41,42). Similar to what was observed in a HeLa extract (Fig. 1B, lanes 1-3), both CE1 and CE1a improved distal 5Ј splice site utilization in a nuclear extract prepared from CB3C7 cells (Fig. 4A, compare lanes 2 and 3  with lane 1). We also tested splicing in a nuclear extract prepared from a derivative of the CB3C7 cell line, which is stably restored for hnRNP A1 expression (CB3C7-20; Fig. 4B, lane 3) (42). The relative levels of distal 5Ј splice site utilization of the SCE1a pre-mRNA were comparable in the CB3C7-20 and in the A1-compromised CB3C7 extracts (Fig. 4A, lane 5). These results suggest that CB3C7 cells contain a factor(s) that functionally replace hnRNP A1, allowing CB3C7 extracts to carry out CE1-and CE1a-dependent shifts in 5Ј splice site selection.
Several observations suggest that the hnRNP A2 protein can functionally replace hnRNP A1 in its ability to promote distal 5Ј splice site use in vitro (24). A2 and its splice variant hnRNP B1 can also replace A1 when repression of a 3Ј splice site is associated with nearby high affinity A1 binding sites (28 -30). Because CB3C7 cells express hnRNP A2 and B1 (Fig. 4B), these proteins may compensate for the loss of A1 and A1B in CB3C7 cells. We used RNA affinity chromatography to deplete hnRNP A2 and B1 proteins from the CB3C7 nuclear extract (data not shown). The removal of A2/B1 from the CB3C7 nuclear extract was associated with an incapacity to activate the distal 5Ј splice site upon incubation with the SCE1d pre-mRNA FIG. 3. hnRNP A1 mediates the activity of CE1d. A, depletion of hnRNP A1/A1B/A2/B1 proteins from a HeLa nuclear extract by RNA affinity chromatography. A HeLa extract was loaded onto an agarose-adipic acid column covalently linked to a portion of CE1a. A control mock depletion was run in parallel. Western analysis of the CE1a-and mock-depleted HeLa nuclear extracts was carried out with an antiserum raised against a peptide shared by hnRNP A1, A1B, A2, and B1 proteins. ⌬, a nuclear extract that has been depleted of hnRNP A/B proteins by chromatography on a CE1a column; M, a mock-depleted extract. Bd1 and Bd2, fractions eluted from two successive CE1a or mock columns. The input sample consisted of an equivalent fraction of the initial HeLa extract. The lower panel represents an overexposure of the upper panel.
We estimate that less than 5% of A1/ A1B/A2/B1 proteins remain in the A/Bdepleted extract. B, splicing in depleted extracts. Radiolabeled pre-mRNAs containing either CE1a or CE1d were spliced in the mock-depleted (M) and A/B-depleted (⌬) extracts. C, rA1 addback experiment. Distal 5Ј splice site selection was tested with S1 and SCE1d pre-mRNAs with increasing amounts of rA1 (0, 0.5, 1, and 2 g) added to an A/B-depleted HeLa nuclear extract. Both S1 and SCE1d were also spliced in the mock-depleted nuclear extract ( lanes  1 and 2). Splicing products in B and C were fractionated on an 11% acrylamide, 8 M urea gel. The positions of pre-mRNAs and proximal and distal lariat products are shown. (Fig. 4C, lanes 5 and 13). To confirm that hnRNP A2 could substitute for A1 in the activity mediated by CE1d, we added increasing amounts of recombinant GST-A2 (rA2) or GST-A1 (rA1) to the A2/B1-depleted CB3C7 nuclear extract. At the highest concentrations, the addition of rA1 or rA2 abolished proximal 5Ј splice site use on the control S1 pre-mRNA (Fig.  4C, lanes 3 and 4 and lanes 11 and 12, respectively). The reason for the repression of S1 pre-mRNA splicing by rA1 and rA2 in the mouse extract is unknown. Nevertheless, the addition of rA1 and rA2 had little or no effect on distal 5Ј splice site use on the S1 pre-mRNA. In contrast, the highest concentrations of rA1 and rA2 tested on the SCE1d pre-mRNA promoted both a reduction in proximal use and an increase in distal splice site selection (Fig. 4C, lanes 7 and 8, and lanes 15 and 16, respectively). These results indicate that hnRNP A2 can also mediate the CE1d-dependent activity in 5Ј splice site selection. Thus, hnRNP A2 is most likely responsible for the lack of a strong splicing defect in the A1-deficient CB3C7 cells.
The ability of hnRNP A2 to bind specifically to CE1d was confirmed by performing a gel shift assay (Fig. 4D). Recombinant hnRNP A2 bound efficiently to CE1d but only weakly to CE1e (Fig. 4D, top left panel). Moreover, rA2 bound efficiently to the CE1d derivative M7* RNA, but the binding to the M6 sequence was reduced considerably (Fig. 4D, bottom left panel). A binding assay performed with recombinant A1B also indicated specific binding to CE1d (Fig. 5D, top right panel) and stronger binding to M7* relative to M6* RNA (bottom right panel). Thus, CE1d can be bound by A1, A1B, or A2. The use of recombinant A1B also allowed a recovery of distal 5Ј splice site use in a CE1a-dependent manner in hnRNP A/B-depleted extracts (data not shown). Although we have not tested hnRNP B1, the splice variant of A2, we would expect this protein to display a similar binding behavior, because it is also depleted from a HeLa extract using a CE1a RNA column. Moreover, several groups have now reported that B1 also displays a function that is similar to hnRNP A1 in splice site selection (24,28).

CE4 Also Contains Several A1 Binding Sites That Influence
5Ј Splice Site Selection-We reported previously that the 24-nt-long CE4 element downstream of alternative exon 7B contains a high affinity A1 binding site. CE4 promotes distal 5Ј splice site selection in vitro and exon 7B skipping in vivo (35). CE4 contains the sequence UAGAGU (Fig. 5A), which is also present in CE1a and which was shown to be important for A1 binding and the activity of CE1a (34). Because CE1 contains adjacent A1-bound elements that can individually carry out distal 5Ј splice site selection, we wondered whether a similar organization existed in CE4. When we compared the in vitro splicing of pre-mRNAs carrying either CE4 or a shortened version lacking the CE4p portion that contains the UAGAGU sequence (see Fig. 5A), we noted a significant reduction in the efficiency of distal 5Ј splice site use (Fig. 5B, compare lane 2 with lane 4), consistent with the notion that the UAGAGU sequence contributes to the activity of CE4 on 5Ј splice site selection. On the other hand, the remaining portion of CE4 (CE4m) remained as active as CE1a at promoting distal 5Ј splice site selection (Fig. 5B, compare lane 4 with lane 3), indicating that CE4m can also promote distal 5Ј splice site selection in vitro.
We have noticed earlier that CE4m is not bound efficiently by recombinant hnRNP A1 in a gel shift assay (35). However, when we repeated the binding assay with recombinant proteins that had been prepared using a high salt procedure (40), we observed that rA1 could bind to CE4m (Fig. 5C, lanes 6 -10). The fact that A1 does not strongly interact with the control CE1e RNA indicates that binding to CE4m is specific (Fig. 5C,  lanes 1-5). To assess the contribution of A1 in the activity of CE4m, we tested extracts that had been depleted of A/B proteins by RNA affinity chromatography. The control pre-mRNA C5Ј Ϫ/Ϫ was spliced almost exclusively to the proximal 5Ј splice site in a mock-depleted extract or in the A/B-depleted extract (Fig. 5D, lanes 1 and 2), and the addition of rA1 had no significant effect on 5Ј splice site selection at the concentrations used FIG. 4. The activity of CE1d can also be mediated by hnRNP A2. A, splicing of S1, SCE1a, and SCE1 pre-mRNAs in CB3C7 and CB3C7-20 nuclear extracts. CB3C7 are mouse erythroleukemic cells that are deficient in A1 expression. A1 expression was stably restored in CB3C7-20 cells. B, Western analysis of A1/A2 expression in CB3C7 and CB3C7-20 cells. *, an uncharacterized band detected only in the mouse erythroleukemic nuclear extracts. C, splicing of the S1 and SCE1d pre-mRNAs in a CB3C7 nuclear extract depleted by chromatography on a CE1a column and supplementation with increasing amounts of rA1 or rA2. To each set, 0, 1, 2, and 3 g of the recombinant protein was added. D, radiolabeled RNAs containing the CE1e, CE1d, M6, and M7 sequences were tested for their ability to interact with rA2 and rA1B. Each set consisted of 0, 0.125, 0.25, 0.5, and 0.75 g of rA1B or rA2. Mixtures were fractionated on native polyacrylamide gels. The positions of the RNA-protein complexes and free RNA are indicated. (lanes 3 and 4). In contrast, whereas the C5Ј 4m/4m pre-mRNA was spliced efficiently to the distal donor site in the mock extract (Fig. 5D, lane 5), distal splicing was abolished in the A/B-depleted extract (lane 6). Moreover, the addition of rA1 to the A/B-depleted extract improved the relative efficiency of distal 5Ј splice site use to an extent that was comparable with the level observed in the mock extract (Fig. 5D, compare lane 8 with lane 5). These results indicate that CE4m can also promote distal 5Ј splice site utilization in an A1-dependent manner. Thus, the strongest activity of CE4 relative to CE1a and CE4m probably reflects the fact that it contains several distinct A1 binding sites. DISCUSSION The activity of the previously identified CE1 and CE4 elements can be attributed in each case to adjacent units individually capable of promoting distal 5Ј splice site selection in vitro. CE1 is composed of CE1a and CE1d, each bound by members of the hnRNP A/B family of proteins. Likewise, the activity of CE4 on 5Ј splice site selection can be separated into CE4m and FIG. 5. CE4m also displays A1-dependent 5 splice site-modulating activity. A, schematic representation of the C5Ј Ϫ/Ϫ transcript. The sequence of CE1a, CE4, and CE4m are shown, and each one was inserted at both positions in C5Ј Ϫ/Ϫ indicated by the arrows. The high affinity A1 binding site UAGAGU is indicated by boldface and underlined letters. B, various pre-mRNAs were incubated in a HeLa extract, and splicing products were fractionated as described earlier. C, a radiolabeled RNA containing the CE4m sequence was compared with CE1e RNA for its interaction with rA1 in a gel shift assay. Each set was tested with 0, 0.125, 0.25, 0.50, and 0.75 g of recombinant protein. D, splicing reactions were performed in a mock-depleted (M) or an A/B-depleted (⌬) HeLa nuclear extract using the C5Ј Ϫ/Ϫ or the C5Ј 4m/4m pre-mRNA. Increasing amounts of rA1 (0.5 and 1 g) were also added to the ⌬ extract. Splicing products were separated on an 11% acrylamide, 8 M urea gel. Note that proximal lariat molecules from C5Ј 4m/4m, C5Ј 1a/1a, and C5Ј 4/4 migrate above the pre-mRNA, because their loop size is larger than the lariat derived from the C5Ј Ϫ/Ϫ transcript. CE4p, each bound by hnRNP A1 or A2. Recombinant hnRNP A1 and A2 can restore the CE1a-, CE1d-, or CE4m-dependent shift in 5Ј splice site selection in extracts that had previously been depleted of their endogenous hnRNP A/B proteins.
The A1 binding sites in CE1a and CE4p are identical and correspond to UAGAGU, a close match to the optimal A1 binding site UAGGGU sequence obtained by selection from a pool of randomized RNA sequences (33). In contrast, the sequence responsible for the binding of A1 and A2 to CE1d and CE4m is less clear. Although CE4m contains the sequences UAGAUU and UAGACU, we have shown previously that mutating the CE1a UAGAGU into UAGACU compromises A1 binding and abrogates its activity (34). Thus, a pyrimidine at position ϩ5 may be incompatible with efficient A1 binding. CE1d also contains a variety of sequences that matches or resembles the A1 binding consensus UAGRR(A/U) (where R represents purine). However, none of the dinucleotide mutations that hit one of these sites substantially reduced CE1d activity (see Fig. 2). Notably, CE1d and CE4m share the sequence RRGCUAG and ARACU. Moreover, a mutation (M7) that targets these two regions simultaneously affects splicing. This result may indicate the existence of additional nonclassical A1/A2 binding sites in CE1d and CE4m.
The existence of several hnRNP A/B binding sites in close proximity could be advantageous for several reasons. First, cooperativity effects due to an increase in the local concentration of A/B proteins may stimulate the apparent K d of hnRNP A/B proteins. Improving the overall efficiency of A1/A2 binding to these sites could be important when the concentration of A1/A2 proteins becomes limiting (e.g. when a cell is actively transcribing many genes). An additional advantage of this organization of A1 binding sites emerges when we consider the model in which A1 molecules bound to CE1 and CE4 interact with one another to loop out and repress the internal 5Ј splice site in a manner that is similar to the effect of duplex-forming RNA sequences (35). The existence of elements (CE1 and CE4) made up of several adjacent A1/A2 binding sites may be similar to having a longer stretch of complementary sequences that improves the rate of duplex formation and offers more stability once the duplex has formed. In the presence of multiple adjacent A1 binding sites, the postulated A1/A1 interactions between CE1 and CE4 could therefore be established more efficiently, and the resulting complex would be more stable. A further gain in stability could be obtained by "cross-strand" interactions, which have been observed in the crystal of a shortened version of A1 (UP1) bound to telomeric DNA sequences (43).
This repetitive arrangement of binding sites is a feature that is found in other elements controlling splice site selection. In the Drosophila dsx enhancer, six consecutive 13-nt-long Tra binding sites contribute to the assembly of a highly efficient enhancer complex (44 -46). It has been proposed that this organization most efficiently allows the looping of RNA sequence between the enhancer and the 3Ј splice site region (47). The introns upstream and downstream of the alternative exon in the c-src, ␣-actinin, fibroblast growth factor receptor 2, and ␣-tropomyosin genes also contain several adjacent binding sites for the hnRNP I/polypyrimidine tract-binding protein (reviewed in Ref. 48).
Several of the functions that have been attributed to hnRNP A/B proteins are associated with the presence of high affinity binding sites. In addition to a role in the selection of 5Ј splice sites, many reports have now implicated A/B binding sites in the negative control of 3Ј splice site choice (28 -32). Although the mechanism responsible for this repression has not been examined in all cases, a recent study indicates that an exonic high affinity A1 binding site can selectively interfere with the binding of some SR proteins to a nearby exon splicing enhancer (49). In this case, the interaction of A1 with the silencer element was associated with the binding of adjacent A1 molecules (49). Thus, although the mechanisms by which exonic and intronic A1 binding sites modulate splice site selection may be different, both situations are apparently mediated by the binding of multiple adjacent A1 molecules.