INTRODUCTION

Top Abstract Introduction Materials and Methods Results Discussion References

RNA-binding proteins participate in many pathways of gene expression and often function as part of large protein and RNA assemblies, such as the ribosome or spliceosome. To understand how such assemblies are formed and regulated, it is necessary to examine how individual RNA binding domains recognize their specific RNA sites and how their binding specificity may be modulated when placed in the context of a larger complex. Two of the most common types of eukaryotic RNA-binding domains are the RNP (or RRM) domain and the K homology (KH) domain (46). Structural studies of isolated RNP and KH domains and their complexes with RNA, as well as of other RNA-protein complexes, have elucidated many features important for specific RNA recognition, such as hydrogen bonding interactions and insertion of bases into hydrophobic protein pockets (21, 53). Recent studies with tethered RNP domains and multiprotein-RNA complexes have begun to define how the spatial organization of domains can contribute to RNA-binding specificity (2, 20, 26, 43, 57). Here we examine how a KH domain recognizes its RNA site in conjunction with other proteins that assemble at 3' splice sites.

KH domains are ~70 residues in length and, like RNP domains, often are arranged as tandem repeats (13, 49, 58). A subset of KH proteins contain only a single, larger (~100-amino-acid) domain, a "maxi-KH" domain, which contains additional amino acids within two loop regions. Maxi-KH domains also are flanked by two other conserved regions, QUA1 and QUA2, of unknown function (see reference 58 for a review). In addition, maxi-KH domain proteins contain sequences likely to be bound by SH3 or WW domains, suggesting potential roles in signal transduction, and therefore have been named STAR proteins (signal transduction and activation of RNA) (58). The structures of several isolated KH domains all display similar three-stranded antiparallel  sheets packed against three  helices but have relatively low sequence homology (5, 34, 39; G. Musco, A. Kharrat, G. Stier, F. Faternali, T. J. Gibson, M. Nilges, and A. Pastore, Letter, Nat. Struct. Biol. 4:712-716). Several largely conserved hydrophobic residues are interspersed throughout the domain, and all contain a GXXG motif in the loop connecting helices 1 and 2.

Relatively few physiological RNA sites have been identified for KH domains, but it already is clear that a wide range of RNA structures can be recognized by these domains. Currently known RNA targets range in size from 7 to 75 nucleotides, binding affinities range from 10⁶ to 10⁹ M, and in some cases one domain is used for recognition while in other cases multiple domains are used (9, 12, 27, 29). Recently, the cocrystal structure of a KH domain from the Nova-2 protein bound to an RNA hairpin was reported (35). This domain primarily recognizes four unpaired nucleotides, which bind in a hydrophobic pocket formed by the 1 and 2 helices and an edge of the 2 strand, with additional contacts made by the flanking GXXG loop, characteristic of all KH domains, as well as by a variable loop between 2 and 3. Given the wide range of RNA sites recognized by KH domains, it will be interesting to determine whether all KH domains, including the larger maxi-KH domains found in STAR proteins, use a similar binding mode.

Mammalian splicing factor 1 (SF1), also known as mammalian branch point binding protein (mBBP) and hereafter designated SF1/mBBP, is a member of the STAR family and participates in the assembly of the spliceosomal E complex (the early mammalian U1 snRNP complex or commitment complex) by binding to the seven-nucleotide branch point sequence (BPS) found at 3' splice sites (3, 9, 30). SF1/mBBP, together with splicing factor U2AF (which is composed of 65- and 35-kDa subunits), facilitates U2 snRNP binding to 3' splice sites (32). As SF1/mBBP and U2 snRNP both bind to the BPS, it has been postulated that SF1/mBBP is displaced upon U2 snRNP binding (19, 37, 47, 52). In one case, SF1/mBBP also has been shown to participate in exon definition by binding to a set of seven-nucleotide repeats located in a splicing enhancer of a microexon (15). In its canonical setting, the maxi-KH domain of SF1/mBBP specifically recognizes the BPS while an additional zinc knuckle domain interacts nonspecifically with RNA and raises the overall binding affinity (9, 10). SF1/mBBP also forms a cooperative complex with U2AF65, which binds to the polypyrimidine tract (PPT) just 3' to the BPS (8, 9, 44). In addition, U2AF65 forms a heterodimer with U2AF35, which recognizes the AG dinucleotide found at intron-exon boundaries 3' to the PPT (24, 38, 60, 61, 64). Mammalian BPSs show substantial variation (YNCURAY is the consensus sequence, where Y is pyrimidine, R is purine, and N is any nucleotide), and cooperative interactions with U2AF are believed to help SF1/mBBP recognize these diverse sequences. In contrast, the yeast BPS is highly conserved (UACUAAC) and yeast BBP (yBBP) binding appears less dependent on interactions with adjacent protein-RNA complexes (8, 9, 44). Additional protein-protein interactions between SF1/mBBP and the WW motifs of formin-binding proteins 11 and 21 and the SH3 domain of Abl also have been observed (6, 7), but their functional importance remains to be determined.

Like SF1/mBBP, other KH domain proteins bind RNA as part of larger complexes, and in some cases there is evidence that their RNA-binding properties are modulated by interactions with auxiliary proteins or protein-RNA complexes. The hnRNP E1 and E2 proteins contain three KH domains and have been implicated in stabilizing -globin mRNA and the translational silencing of 15-lipoxygenase mRNA by binding to 3' untranslated region elements in conjunction with different proteins (41). In the life cycle of poliovirus, these proteins also bind to a viral 5' cloverleaf structure important for replication and to an internal ribosomal entry site element, switching between the two sites based on the availability of the viral 3CD protein (23). Finally, the RNA-binding affinity of Sam68, a STAR protein that interacts with several cell signaling proteins, appears to be regulated by tyrosine phosphorylation (58).

To further examine how KH domains recognize RNA in the context of a larger assembly and to better characterize the binding properties of a maxi-KH domain, we have used a mammalian cell reporter system to examine the RNA-binding properties of SF1/mBBP. We demonstrate that the U2AF heterodimer is recruited along with SF1/mBBP to 3' splice site reporters, composed of the BPS, PPT, and AG dinucleotide, in mammalian cell nuclei and that binding to BPS variants correlates with previously measured splicing efficiencies, supporting a role for the SF1/mBBP complex in determining splice site usage. Using mutagenesis, we have defined an RNA-binding surface of SF1/mBBP that agrees well with the one used by Nova-2 (35), suggesting that the maxi-KH domains of STAR proteins and "simple" KH proteins utilize related binding mechanisms. We have generated a structural model of the SF1/mBBP KH domain and, based on the Nova-2 binding arrangement, propose an arrangement of the SF1/mBBP-U2AF complex assembled at 3' splice sites.

	MATERIALS AND METHODS

Top Abstract Introduction Materials and Methods Results Discussion References

Plasmid construction. BPS reporter plasmids were constructed by cloning synthetic oligonucleotide cassettes encoding the sequences shown in Fig. 1 into the AflII and HindIII restriction sites of an human immunodeficiency virus type 1 (HIV-1) long terminal repeat (LTR)-chloramphenicol acetyltransferase (CAT) reporter plasmid in place of the transcription activation response (TAR) site. This plasmid contains a previously modified HIV-1 LTR (51) in which additional restriction sites have been engineered between the start of transcription and the TAR element. The inserted sequence of the wild-type BPS reporter, beginning at the 5' end of the transcript and encompassing the BPS, PPT, and AG dinucleotide (shown in boldface) is 5'-GGTCTCTCTGGCTTAAGTTCGTACTAACCCTGTCCCTTTTTTTTCCACAGCAAGCTT, with the AflII and HindIII sites underlined. Plasmids encoding the Tat fusion proteins were constructed by cloning PCR amplification products into the SalI and SpeI restriction sites of a pSV2Tat expressor plasmid (33), creating C-terminal fusions following amino acid 72 of HIV-1 Tat and a linker of three glycines. Tat-fused SF1/mBBP was generated by amplifying a fragment encoding amino acids 2 to 307 from pGEX6P-SF1 (8), and a truncated version, Tat-fused SF1/mBBPN, was generated by amplifying a fragment encoding amino acids 69 to 307. Tat-fused U2AF65 was generated by amplifying a fragment encoding full-length U2AF65 (amino acids 1 to 475) from pGEX6P-U2AF65 (8). A variant with an internal deletion, Tat-fused U2AF6595-138, also was constructed. Alanine mutations were introduced into Tat-fused SF1 by PCR mutagenesis.

View larger version (28K): [in this window] [in a new window]   FIG. 1.   (A) Schematic diagrams of the Tat fusion expressor and RNA reporter plasmids. The expressor plasmid encodes HIV-1 Tat residues 1 to 72, followed by a linker of three glycines and the relevant fusion, expressed from a simian virus 40 early promoter (PSV40). The reporter plasmid utilizes a modified HIV-1 LTR to drive CAT expression, with the BPS-PPT-AG sequence replacing the TAR site at the 5' end of the mRNA. (B) Domain organization of wild-type SF1/mBBP and U2AF65 (top) and the Tat fusion proteins (bottom). The maxi-KH domain with flanking QUA1 and QUA2 regions characteristic of STAR proteins, the Zn knuckle, and the proline-rich region of SF1/mBBP and the RS domain and three RNP domains of U2AF65 are indicated. Numbers refer to amino acid positions in the proteins. The Tat fusions contain amino acids 1 to 72 of Tat followed by the glycine linker. (C) The configuration of the BPS, PPT, and AG dinucleotide at canonical 3' splice sites (ss) (top) and the sequences of the BPS reporter and mutants (bottom) are shown. BPS variants indicate the single nucleotide substitutions used at each position.

Transient transfection and CAT assays. Levels of CAT activation were measured by cotransfecting 100 ng of BPS reporter plasmids or 25 ng of the TAR reporter plasmid and 5 ng of the Tat fusion expressor plasmid (except where indicated in the figure legends) into HeLa cells using 5 µl of Lipofectin (Life Technologies) in 3.8-cm² wells for 4 h. Total plasmid DNA was adjusted to 2 µg with pBluescript DNA. CAT activities were assayed after 44 h by using an appropriate amount of cell extract, as described previously (55). Activities were quantitated using a Molecular Dynamics phosphorimager, and fold activation relative to that for the reporter plasmid alone was calculated. For each experiment, CAT assays were performed in duplicate and percentages of activation for the different reporters or protein mutants relative to those for the wild-type combination were calculated. Percentages of activation were then averaged over three or four separate transfection experiments, and standard deviations of the means (shown in Fig. 3 to 5) were calculated. To control for possible differences in expression levels of the various fusion proteins, CAT activities also were measured on an HIV-1 LTR reporter containing the wild-type TAR element. All fusion proteins contained full-length Tat, including its own RNA-binding domain, and therefore could activate the TAR-containing reporter. Activation levels of the BPS reporters were normalized to the values with the TAR reporter, which showed less-than-twofold differences in activation for all fusion proteins. HeLa cells were grown in Dulbecco's modified Eagle's medium with 10% fetal bovine serum.

Molecular modeling. The LOOK algorithm (Molecular Applications Group, Palo Alto, Calif.) was used to construct a three-dimensional model of the SF1/mBBP KH domain based on the crystal structure of the Nova-2 KH domain bound to RNA (35) (PDB file 1EC6; chain A). The amino acid sequence alignment was based on those of Musco et al. (39) and Lewis et al. (35) but was modified to better align three residues within the variable loop between 2 and 3.

	RESULTS

Top Abstract Introduction Materials and Methods Results Discussion References

Reporter system to monitor SF1/mBBP-RNA interactions in vivo. We are interested in understanding how the maxi-KH domain of SF1/mBBP recognizes the seven-nucleotide BPS in the context of the complex it forms with other proteins that also assemble at 3' splice sites. We chose to test whether the Tat hybrid system (33, 54) might be suitable for studying these interactions in vivo. The HIV-1 Tat protein enhances the efficiency of transcriptional elongation from the HIV-1 LTR and binds to an RNA hairpin, known as TAR, located at the 5' end of the nascent transcript. Heterologous RNA-protein interactions can be used to deliver the Tat activation domain to the LTR (33), and therefore we asked whether SF1/mBBP fused to Tat could activate transcription from an LTR-CAT reporter containing the BPS in place of TAR (Fig. 1A). A fusion protein containing a fragment of SF1/mBBP (residues 1 to 307) linked to the C terminus of full-length Tat (residues 1 to 72) was constructed. This SF1/mBBP fragment includes the maxi-KH domain, surrounding QUA1 and QUA2 regions, and the zinc knuckle (Fig. 1B) and is sufficient for BPS recognition, its interaction with U2AF65, and splicing activity (8, 44). Because the Tat portion of the fusion protein also contains its own RNA-binding domain, it was possible to indirectly assess the expression levels of this and all other fusions described in this study by independently measuring activation levels on a HIV-1 TAR reporter. By this criterion, all fusion proteins were expressed at similar levels, showing less than a twofold variation in activity (Fig. 2C). All activities shown were normalized accordingly.

View larger version (23K): [in this window] [in a new window]   FIG. 2.   Activation of the BPS reporter by Tat-fused SF1/mBBP and Tat-fused U2AF65. (A) Titration of the Tat-fused SF1/mBBP expressor plasmid (filled circles) and unfused Tat (open circles) on the BPS reporter. Tat-expressing and BPS reporter plasmids were cotransfected into HeLa cells, and CAT activities were measured after 44 h. (Insets) CAT assays, with expressor plasmid amounts (nanograms) indicated. Fold activation was determined relative to the activity of the reporter alone. (B) Titration of the Tat-fused U2AF65 expressor plasmid (filled squares) and unfused Tat (open circles) on the BPS reporter. (C) Titration of the Tat-fused SF1/mBBP (filled circles), Tat-fused U2AF65 (filled squares), and unfused Tat (open circles) expressor plasmids on the HIV-1 TAR reporter. All fusions contain the RNA-binding domain of Tat and thus are able to activate transcription via the TAR element. Relative activities of the fusion proteins on the TAR reporter were used to normalize for fusion protein expression levels.

Cooperative binding of the SF1/mBBP-U2AF65-U2AF35 complex. To further test whether the endogenous U2AF subunits are recruited to the BPS reporter, we next measured activation of a mutant reporter, deficient for binding to U2AF65, in which several pyrimidines in the PPT were replaced by purines (Fig. 1C) (60). Activation of the PPT(m) reporter by both the Tat-fused SF1/mBBP and Tat-fused U2AF65 proteins was reduced 5- to 10-fold (Fig. 3), suggesting that the interaction of Tat-fused U2AF65 with the 3' splice site requires an intact PPT and that the U2AF65-PPT interaction stabilizes the SF1/mBBP-BPS interaction, as observed in vitro (8). The cooperative nature of the protein-protein and protein-RNA interactions is further supported by the observation that activation by the Tat-fused U2AF65 is reduced more than fivefold for the BPS(m) reporter (Fig. 3B).

View larger version (29K): [in this window] [in a new window]   FIG. 3.   Activation by the Tat fusion proteins requires RNA-protein interactions at the BPS, PPT, and AG dinucleotide. (A) Activation of the BPS reporter and mutants by Tat-fused SF1/mBBP (gray bars) and Tat-fused SF1/mBBPN (white bars). Tat-expressing (5 ng) and BPS reporter (100 ng) plasmids were cotransfected into HeLa cells, and CAT activities were measured after 44 h. Percent activation was calculated by normalizing to the activation level of the Tat-fused SF1/mBBP-BPS reporter interaction, and standard deviations (bars) were calculated as described in Materials and Methods. (B) Activation of the BPS reporter and mutants by Tat-fused U2AF65 (gray bars) and Tat-fused U2AF6595-138 (white bars). Activities were normalized to the activation level of the Tat-fused U2AF65-BPS reporter interaction.

BPS binding specificity. As described above, mammalian BPSs typically are defined by the broad consensus YNCURAY sequence (14) and show rates of splicing that differ by more than 1 order of magnitude (63). Because the SF1/mBBP-BPS interaction is likely replaced later in spliceosome assembly by the base pairing of U2 snRNA, effects of BPS variation on splicing efficiency might reflect either one or both binding events. Our reporter system appears to accurately reflect assembly of SF1/mBBP-BPS complexes and therefore provides a reasonable tool to monitor the effect of BPS variation on the initial protein binding events, although we cannot exclude the influence of other factors (see Discussion). We constructed a series of BPS variant reporters with single nucleotide changes to the UACUAAC sequence (Fig. 1C) and measured activation by Tat-fused SF1/mBBP. A 20-fold range of activities was observed, with the yeast UACUAAC sequence and a conservative C-to-U change at the last position producing the highest activities and changes of the conserved branch point adenosine at the sixth position (UACUAGC) and of the conserved uridine at the fourth position (UACGAAC) producing the lowest activities (Fig. 4A). Mutations at the fourth and sixth positions also strongly decrease SF1/mBBP RNA-binding affinity in vitro (9). Changing the last position of the YNCURAY consensus sequence to a purine decreases activity about threefold, as does changing the fifth partially degenerate position (R) from A to G (Fig. 4A). Changing the third conserved C or the first partially degenerate position (Y) produces modest decreases of less than twofold. Changing the second nucleotide at the degenerate position (N) along with changing the fifth position, which creates the normal adenovirus major-late pre-mRNA 3' splice site region (encompassing the BPS, PPT, and AG), produces the same activity as changing the fifth position alone. It is interesting that in vitro binding assays were relatively insensitive to changes at the degenerate positions (9), suggesting that our reporter system, in which other proteins are recruited, may be more sensitive to small differences in RNA-binding affinity. Given the tight correspondence between SF1/mBBP complex formation and the known splicing efficiencies of BPS variants (see Discussion), our results support an important role for the SF1/mBBP-BPS interaction in determining 3' splice site utilization.

View larger version (44K): [in this window] [in a new window]   FIG. 4.   Activation of BPS variant reporters by Tat-fused SF1/mBBP (A) and Tat-fused U2AF65 (B). Percentages of activation and standard deviations (bars) were calculated as for Fig. 3, with activities normalized to the Tat-fused SF1/mBBP-BPS reporter and Tat-fused U2AF65-BPS reporter interactions, respectively. BPS substitutions are highlighted.

Mutagenesis and modeling of the KH domain. Relatively little is known about how the maxi-KH domains of STAR proteins recognize RNA, and we wished to use our reporter system to help define the RNA-binding surface of SF1/mBBP by mutagenesis. To identify amino acids potentially involved in BPS recognition, we chose the structures of several KH domains previously solved by nuclear magnetic resonance (NMR) or crystallography (34, 39; Musco et al., letter) and aligned their sequences with that of the KH domain of SF1/mBBP (Fig. 5A). Focusing largely on residues previously implicated in RNA binding (34), but prior to determination of the Nova-2 cocrystal structure, and focusing also on charged and conserved residues, we selected 23 positions in SF1/mBBP to generate alanine mutants (Fig. 5A and B). These positions were scattered throughout the different putative secondary structures and included no residues in the hydrophobic core. The expression levels of the mutants were similar, as assessed indirectly by measuring the activity of the Tat fusion on a HIV-1 TAR reporter as described above, and activities on the BPS reporter were normalized accordingly. Eleven mutations in Tat-fused SF1/mBBP reduced activity by at least twofold on the BPS reporter (Fig. 5C). All residues shown to be important, with the exception of Phe152, are highly conserved between SF1/mBBP and yBBP (Fig. 6A), and, conversely, almost all residues that do not appear to be important for binding are not conserved. Most of the important residues were grouped in adjacent structural elements, namely, the putative first and second  helices, the conserved GXXG loop between the two helices, and the second  strand.

View larger version (49K): [in this window] [in a new window]   FIG. 5.   Activation of the BPS reporter by Tat-fused SF1/mBBP mutants. (A) Sequence alignment of the maxi-KH domain of SF1/mBBP and selected KH domains from vigilin (vig; domain 6), Nova-2 (domain 3), Nova-1 (domain 3), and hFMR-1 (domain 1). Secondary structure elements were assigned based on the X-ray and NMR structures of these individual domains (34, 39; Musco et al., letter). B and H, -sheet and -helical residues, respectively, which are defined clearly in all structures; lowercase letters, residues that can be assigned to a secondary structure in only one or two structures. Residues considered to be part of the hydrophobic core and therefore not chosen for mutation are shaded. Numbers refer to the SF1/mBBP sequence, and residues marked with an asterisk were mutated to alanines. (B) Averaged NMR structure of the sixth KH domain from vigilin (PDB file 1VIH [39]), shown from two different faces. Circles, approximate locations of amino acids chosen for mutagenesis, assuming that SF1/mBBP adopts a similar fold; yellow circles, positions that decrease activity by at least twofold (see panel C); white circles, positions that have little or no effect. (C) Activities of the Tat-fused SF1/mBBP mutants normalized to the activity of the wild-type (wt) protein with standard deviations (bars) calculated as for Fig. 3. The line corresponds to a twofold decrease in activity. The predicted corresponding units of secondary structures are indicated.

View larger version (45K): [in this window] [in a new window]   FIG. 6.   (A) Sequence alignment of SF1/mBBP, BBP from Saccharomyces cerevisiae (yBBP), and KH domain 3 from Nova-2. Black boxes, identical residues; shaded boxes, conserved residues. The corresponding units of secondary structure are indicated. An extra C-terminal region is shown, compared to the sequence shown in Fig. 5A; this region corresponds to part of the Nova-2 structure. Asterisks and numbers, positions in SF1/mBBP important for binding based on the mutagenesis data (Fig. 5); minuses, positions where mutations have little or no effect. The numbering of Nova-2 residues discussed in the text (and not shown) corresponds to that described for the cocrystal structure (35). The region corresponding to the beginning of QUA2 is indicated. (B) Structure of the Nova-2 KH domain, taken from the cocrystal structure of the protein-RNA complex (35). Amino acids involved in RNA binding are shown in yellow. (C) Structure of the Nova-2 KH domain complexed to RNA (35), with the RNA tetranucleotide specifically recognized by the protein shown in yellow. The 5' and 3' ends of the RNA are indicated. (D) Homology model of SF1/mBBP maxi-KH domain, based on the structure of the RNA-bound Nova-2 domain and the alignment shown in panel A. The approximate locations of residues important for RNA binding, as defined by mutagenesis, are shown in yellow. The large 1/1 and 2/3 loops found in the SF1/mBBP maxi-KH domain were positioned arbitrarily by the homology modeling. (E) Proposed RNA-binding orientation of the SF1/mBBP-U2AF complex, based on similarities to the Nova-2 protein-RNA complex (see Discussion). The schematic drawing is not intended to indicate the relative orientations of the U2AF65 and U2AF35 subunits or of the RS domain of U2AF65, which also may contact the RNA (56).

	DISCUSSION

Top Abstract Introduction Materials and Methods Results Discussion References

Assembly of multiprotein complexes using the Tat hybrid system. We have described a reporter assay based on the Tat hybrid system that monitors the cooperative binding of SF1/mBBP, U2AF, and perhaps other splicing factors to 3' intronic sequences in vivo. An important advantage of this system is the ability to recruit additional endogenous nuclear proteins, in addition to the Tat fusion protein, to RNA sites engineered into the reporter. A related system utilizing the equine anemia virus Tat protein has been used to study a poliovirus protein-RNA interaction that also requires a host protein for RNA binding (11). With the BPS reporter, we have shown that Tat can activate transcription when fused to different components of the same multiprotein complex, SF1/mBBP or U2AF65, suggesting that Tat can act from multiple locations of a large RNA-protein complex, potentially even if tethered to proteins associated with RNA only indirectly via protein-protein interactions. This degree of flexibility is consistent with previous findings that a variety of RNA-binding domains can be functionally fused to Tat and many types of RNA sites can be accommodated in place of TAR (33). We do not know yet the limitations of the Tat hybrid system including, for example, possible steric restrictions imposed by the transcriptional machinery required for Tat activation. Nevertheless, the Tat hybrid system seems to be a useful tool for dissecting large and complex ribonucleoprotein complexes, such as the spliceosome, and for identifying novel interacting proteins from cDNA libraries fused to Tat (33, 54). Although we only have tested recruitment of SF1/mBBP and U2AF to the BPS, it is possible that other proteins such as UAP56, which interacts with U2AF (22), or even U2 snRNA or other snRNP components are stably or transiently bound. It also is possible that addition of a 5' splice site may allow assembly of higher-order complexes that bridge the 5' and 3' splice sites.

Assembly of SF1/mBBP-RNA complexes. Spliceosome assembly is highly dynamic and involves an ordered set of binding and rearrangement events utilizing different protein and RNA components. Recognition of the BPS by SF1/mBBP is an early step in the assembly pathway, and SF1/mBBP forms part of a complex that includes U1 snRNP bound near the 5' splice site and U2AF bound to the PPT near the 3' splice site. Later, SF1/mBBP is thought to be displaced through an unidentified ATP-dependent mechanism resulting in U2 snRNA base pairing to the BPS (40, 52). BPS recognition is likely to be regulated, and indeed it is known that splicing efficiency is strongly influenced by variations in the BPS (45, 63). Our results support a direct role for the SF1/mBBP-BPS interaction in determining the efficiency of 3' splice site usage. We observed a 20-fold range of activities on a set of BPS variant reporters, with the UACUAAC conserved yeast sequence and UACUAAU (changed residues are in boldface) having the highest activities, GACUAAC, UAGUAAC, UACUGAC, and UACUAAG having intermediate activities, and UACUAGC and UACGAAC having the lowest activities. These preferences correlate well with in vitro SF1/mBBP binding experiments (8) and splicing assays that show a >10-fold range in splicing efficiency (UACUAAC > UACUGAC UACGGAC) (63). Additional in vitro splicing assays have shown that a BPS with a U-to-A change at the fourth position strongly decreases splicing efficiency and also alters 3' splice site selection in vivo (45). Our corresponding U-to-G mutation shows the weakest activity (Fig. 4). In contrast, changes to the last position have relatively little effect on splicing efficiency (45) and either had no effect (C to U) or produced a modest decrease (C to G) in our assays. Compensatory mutations between the BPS and U2 snRNA have shown that the differences in splicing efficiency can only be partially explained by effects on base pairing (42). Thus it seems likely that the strength of the SF1/mBBP-BPS interaction contributes directly to 3' splice site selection.

RNA recognition by the maxi-KH domain. Based on our mutagenesis of the maxi-KH domain of SF1/mBBP, a presumptive RNA-binding surface emerged (Fig. 6D), which is consistent with structural studies of non-STAR protein KH domains (4, 35). The protein-RNA interface seen in the Nova-2 KH domain-RNA complex (35) shows extensive van der Waals contacts from an aliphatic platform of the KH domain and hydrogen bonds between hydrophilic side chains and the Watson-Crick faces of single-stranded bases. We identified a number of charged and hydrophilic side chains within 1, 2, 2, and the GXXG loop of SF1 that appear to contribute to RNA binding (Fig. 5), and these correspond to regions of Nova-2 that contact the RNA (compare Fig. 6B and D). We did not examine residues in the aliphatic platform because many also help form the hydrophobic core of the domain. The 2/3 loop of Nova-2 also contacts the RNA, and it is interesting that the large 1/1 and 2/3 loops found in maxi-KH domains seem positioned to form additional interactions (Fig. 6D). These large loops also have been implicated in protein-protein interactions in maxi-KH domain proteins (17, 39). Residues from the flanking QUA1 and QUA2 regions seem positioned to further extend the binding surface and might enable recognition of the longer seven-nucleotide BPS (versus the four-nucleotide recognition site for the Nova-2 KH domain). Deletion of QUA1 of the Qk1 protein was found to abolish RNA binding, and deletion of QUA2 had a modest effect (18). In yBBP, a mutation in QUA2 contributes to a mutant phenotype deficient in forming commitment complexes (48). Although not conserved in sequence, the regions flanking simple KH domains also may be important for RNA recognition. In Nova-2, an arginine located in an extended C-terminal helix and corresponding to the beginning of QUA2 (Fig. 6A) is required for high-affinity RNA binding (35). Additional mutagenesis of Tat-fused SF1/mBBP may help test whether these other regions directly contact the RNA, particularly if mutants with altered RNA-binding specificities can be found, or whether they mediate protein-protein interactions.