The zinc fingers of the SR-like protein ZRANB2 are single-stranded RNA-binding domains that recognize 5′ splice site-like sequences

Abstract

The alternative splicing of mRNA is a critical process in higher eukaryotes that generates substantial proteomic diversity. Many of the proteins that are essential to this process contain arginine/serine-rich (RS) domains. ZRANB2 is a widely-expressed and highly-conserved RS-domain protein that can regulate alternative splicing but lacks canonical RNA-binding domains. Instead, it contains 2 RanBP2-type zinc finger (ZnF) domains. We demonstrate that these ZnFs recognize ssRNA with high affinity and specificity. Each ZnF binds to a single AGGUAA motif and the 2 domains combine to recognize AGGUAA(N_x)AGGUAA double sites, suggesting that ZRANB2 regulates alternative splicing via a direct interaction with pre-mRNA at sites that resemble the consensus 5′ splice site. We show using X-ray crystallography that recognition of an AGGUAA motif by a single ZnF is dominated by side-chain hydrogen bonds to the bases and formation of a guanine-tryptophan-guanine “ladder.” A number of other human proteins that function in RNA processing also contain RanBP2 ZnFs in which the RNA-binding residues of ZRANB2 are conserved. The ZnFs of ZRANB2 therefore define another class of RNA-binding domain, advancing our understanding of RNA recognition and emphasizing the versatility of ZnF domains in molecular recognition.

Almost all human genes are thought to be alternatively spliced, and it has been estimated that at least 15% of human diseases are associated with changes in RNA processing (1). The selection of splice sites is influenced heavily by the binding of accessory splicing factors to regulatory sequences in the pre-mRNA. SR proteins are splicing factors that contain a C-terminal Arg/Ser-rich (RS) domain and either 1 or 2 N-terminal RNA recognition motifs (RRMs) (2). They play crucial roles in constitutive and alternative splicing, promoting recognition of splice sites by binding to exonic splicing enhancers (ESEs). RRM domains bind ssRNA with high affinity in a sequence-specific manner, whereas RS domains appear to facilitate both protein–protein and protein–RNA interactions (3, 4). Other RS domain-containing proteins that lack a canonical RRM, termed “SR-like” proteins (see, for example, ref. 5) are also known to play roles in splicing.

ZRANB2 (Zis, ZNF265) is an SR-like nuclear protein that is expressed in most tissues and is conserved between nematodes and mammals. It interacts with the spliceosomal proteins U1–70K and U2AF³⁵ and can alter the distribution of splice variants of GluR-B, SMN2, and Tra2β in minigene reporter assays (6, 7). As such, ZRANB2 appears to regulate splice site choice. However, in place of the canonical RNA-binding RRM domains, ZRANB2 displays 2 N-terminal RanBP2-type zinc fingers (ZnFs).

RanBP2-type ZnF domains are defined by the consensus sequence W-X-C-X_2–4-C-X₃-N-X₆-C-X₂-C. They occur multiple times in at least 21 human proteins, and the fold comprises 2 distorted β-hairpins sandwiching a central tryptophan residue and a single zinc ion (8, 9). RanBP2 ZnFs from Npl4 and Nup153 are protein recognition motifs, mediating interactions with ubiquitin (10) and the nuclear transport protein Ran (11), respectively. In contrast, we show here that the double RanBP2 ZnF domain of ZRANB2 recognizes ssRNA, binding tandem copies of an AGGUAA motif with high affinity. We have determined the structural basis for RNA recognition by ZRANB2, which reveals a large number of specific hydrogen bonds and a previously undescribed base-stacked “ladder” involving a tryptophan and 2 guanines. Our data define another class of ssRNA-binding protein and shed light on the biochemical functions of a number of human proteins that contain these domains.

Results

Identification of a High-Affinity RNA Ligand for ZRANB2-F12.

To test whether the ZnFs of ZRANB2 can recognize ssRNA, we carried out an in vitro site selection [systematic evolution of ligands by exponential enrichment (SELEX)] experiment. From a ssRNA pool containing a randomized 25-nt sequence, high-affinity RNA target sequences were selected by using a GST-fusion protein containing both ZnFs (GST-F12) immobilized on glutathione Sepharose beads. After 7 rounds of selection, we sequenced 26 unique clones; all contained a GGUA or a AGGU motif, and 15 contained the longer AGGUAA motif. Further selection enriched the RNA pool in sequences containing multiple AGGUA motifs, such that after 9 rounds of selection 15 of 33 unique clones contained 2 GGUA motifs. In 11 of these 15 clones, the 2 motifs occurred either in tandem or separated by 1 nt.

Alignment of the sequences (Fig. 1A and Fig. S1A) suggests that each of the ZnFs recognizes an AGGUAA site. MFOLD (12) did not reveal any consistent secondary structure predictions for these sequences, suggesting that the binding is sequence-driven rather than structure-driven. To determine whether the AGGUAA motif is sufficient for binding, we tested the binding of the double finger construct F12 to a 17-nt RNA oligonucleotide containing a single AGGUAA motif. dsDNA, ssDNA, and dsRNA were also tested. As shown in Fig. 1B, the interaction is strongly selective for ssRNA, consistent with a role for ZRANB2 in mRNA processing. Mutation of the central GGU, which is the most highly-conserved element in the SELEX consensus, abrogated the interaction, showing that the interaction is sequence specific.

Fig. 1.

ZRANB2 binds to ssRNA containing AGGUAA repeats. (A) Sequence logo indicating the degree of sequence conservation within the 2 RNA sites from sequences obtained from SELEX (Fig. S1A). (B) Gel-shift showing the nucleic acid type specificity of ZRANB2. The same 17-nt sequence, encoded in ssDNA, ssRNA, dsDNA, and dsRNA, was electrophoresed in the presence of F12. In the sequence ssRNAmut, the central GGU is replaced by CUG. (C) Fluorescence anisotropy data showing the binding of GST-F2 to 17-nt ssRNA sequences containing point mutations. (D) Calculated association constants from D. Affinities are an average of 3 experiments and error bars indicate 1 SD. (E) Association constants, obtained by fluorescence anisotropy, for F12 binding to RNA sequences containing either a single AGGUAA site (with a scrambled second site) or double sites with spacings of −1, 0, 2, 5, and 8 adenines or the 5-nt sequence ACCCC (AC4). (F) RNA-binding affinity for single Ala point mutations of GST-F2, from fluorescence anisotropy data. Error bars indicate 1 SD from 3 measurements.

Sequence Specificity of the RNA-ZRANB2 Interaction.

The repetition of the AGGUAA motif in the SELEX sequence alignment suggested that each of the ZnFs can recognize a single site. We therefore expressed and purified the 2 individual domains (F1 and F2; Fig. 2) as GST fusions and tested their ability to bind ssRNA by fluorescence anisotropy, using a ssRNA oligonucleotide bearing a 5′fluorescein tag and containing a single AGGUAA site. F1 and F2 bound to this sequence with association constants of 3 × 10⁶ M⁻¹ and 1 × 10⁶ M⁻¹, respectively (Fig. 1 C and D and Fig. S1), demonstrating that both F1 and F2 can bind 1 AGGUAA site. We also showed that the central GG sequence is most important for binding by measuring the affinity of each finger for a series of ssRNA oligonucleotides that each contained a single purine ↔ pyrimidine mutation in the AGGUAA motif (Fig. 1D and Fig. S1).

Fig. 2.

Sequence data for ZRANB2. Sequences of F1 and F2 are shown. Cysteines are indicated by *, and side chains that directly contact RNA are marked by a gray box.

We next measured the binding of F12 to ssRNA containing 2 AGGUAA sites (Fig. 1E). The affinity of F12 for the sequence AGGUAAAGGUAA was 1.9 × 10⁷ M⁻¹ (Fig. 1E, lane 6). Randomizing one of the AGGUAA motifs (to give the sequence AGAAUGAGGUAA; Fig. 1E, lane 4) reduced the binding by a small, but reproducible, amount, although it was notable that binding was still significantly stronger than that of a single finger to a similar sequence (Fig. 1E, lanes 1 and 2), indicating that the additional finger in the F12 construct could still make contacts with the scrambled second site. Scrambling both sites effectively eliminated binding (Fig. 1E, lane 3).

We assessed the importance of the spacing between AGGUAA motifs by measuring binding of F12 to double sites that were separated by −1 to 8 nt (where −1 denotes the sequence AGGUAAGGUAA; Fig. 1E, lanes 5–10). Surprisingly, the affinity of the interaction increased monotonically as an increasing number of adenines were inserted between the sites. Further, a double site containing a cytosine-rich spacer (Fig. 1E, lane 10) had the same affinity for F12 as a penta-adenine spacer (Fig. 1E, lane 8), indicating that the identity of the intervening bases is unimportant.

Overall, these data demonstrate that the double ZnF domain of ZRANB2 recognizes ssRNA carrying the consensus sequence AGGUAA(X_−1–8)AGGUAA. Strikingly, AGGUAA is almost identical to the conserved consensus sequence for the 5′ splice site across metazoans (13, 14) and resembles the 3′ splice site consensus (CAGG).

Solution Analysis of the ZRANB2:RNA Interaction.

To provide structural insight into the ZRANB2:RNA interaction we first determined the structure of F2 by using NMR spectroscopy (Fig. 3A, and Table S1). The structure is well defined and comprises 2 short β-hairpins sandwiching a zinc ion that is ligated by the 4 conserved cysteines. The fold is consistent with that of F1 (8) and other structures from this class of ZnFs, including domains from HDM2 (15), Npl4 (9), and Nup153 (11).

Fig. 3.

NMR analysis of the ZRANB2:ssRNA interaction. (A) The structure of ZRANB2-F2. An overlay of the 20 lowest energy structures (residues 67–95) is shown. The zinc ligands C71, C74, C85, and C88 are shown in gold, and the zinc ion is shown in red. (B) Space-filling representation of F2 with residues showing significant chemical-shift changes (>1 SD from the mean) colored red (Fig. S2A). The protein is rotated ≈90° counterclockwise about the vertical axis, compared with A. (C) Overlay of ¹⁵N-HSQC spectra of free F2 (black) and F2 in the presence of 1.2 molar equivalents of CCAGGUAAAG (red). Arrows show the shifting of selected resonances.

We next carried out chemical-shift perturbation experiments, titrating short (6 or 10 nt) RNA oligonucleotides containing a single AGGUAA motif into ¹⁵N-labeled F2. Significant chemical exchange broadening was observed for a subset of signals during the titration, but all signals reappeared after the addition of 1 molar equivalent of RNA (Fig. 3C), and no further changes were observed if more RNA was added, consistent with the formation of a well-defined 1:1 complex.

Mapping the most significant chemical-shift changes onto the structure of F2 (Fig. 3B and Fig. S2A) reveals a single contiguous surface. The binding surface comprises a mixture of amino acid types, including aromatic (W79), aliphatic (V77, A80, M87), polar (N76, N78, N86), and charged (R81, R82) residues. These residues are almost completely conserved in F1, suggesting that each ZnF recognizes RNA in the same manner.

We corroborated our structural data by examining the effect of alanine point mutations on the RNA-binding ability of F2. Fig. 1F shows that alanine substitutions of W79, R81, R82, N86, and M87 significantly reduced the association constant (the correct folding of each mutant was confirmed by NMR; Fig. S2B). In contrast, much smaller changes in affinity were observed for K72A and T73A, which are oriented away from the RNA contact surface.

¹⁵N-HSQC titrations were also carried out by using the double finger construct (F12; residues 1–95) and the double site oligonucleotide AGGUAAAGGUAA. Chemical shifts for the free F12 protein were essentially unchanged from those in the 2 individual finger constructs, and no signals were observed in the ¹⁵N-HSQC of F12 for residues in the 25-residue linker. This titration (Fig. S2D) gave rise to the same pattern of chemical-shift changes as the 2 single-finger experiments. Only 3–4 new signals appeared, most likely from linker residues. However, most of the linker residues remained unobservable, indicating that the linker does not become ordered upon RNA binding.

Structural Basis for ssRNA Recognition by ZRANB2.

Only a small number (<10) of intermolecular NOEs were observed between F2 and RNA; intermediate chemical exchange broadened signals from key residues at the protein:RNA interface. We therefore crystallized F2 in complex with a 6-nt RNA with the sequence AGGUAA and determined the structure of the complex to a resolution of 1.4 Å. The overall quality of the structure, as judged by Molprobity (16), is excellent. All F2 residues lie in the favored region of the Ramachandran plot, and Molprobity scores the structure in the top 1% of all structures determined to comparable resolution. All RNA nucleotides have acceptable sugar puckers and residues Ade1–Ura4 have acceptable backbone conformations.

In the structure (Fig. 4, Table S2, and Fig. S3), F2 adopts the same backbone fold as it does free in solution. Electron density for Gua2, Gua3, and Ura4 is unambiguous and clearly reveals the mode of recognition of these bases (Fig. 4). Most prominently, a tryptophan side chain (W79) stacks between Gua2 and Gua3; the plane of the indole side chain is parallel with those of the 2 purines and makes extensive contacts with both bases (Fig. 4 A and B). Although this purine-Trp-purine ladder appears well-suited to direct recognition of single-stranded nucleic acids, a motif of this type has not previously been observed in any protein–nucleic acid structure to our knowledge.

Fig. 4.

Structure of the ZRANB2:RNA complex. (A) Electron density at the protein/RNA interface. The binding interface is shown with unbiased F_o − F_c density (green) at 2.5 σ calculated by omitting the RNA portion of the model. The final model for the RNA is shown to illustrate the fit with the difference density. 2F_o − F_c electron density (blue) calculated by using the final model is shown at 1.4 σ on the protein portion of the model only. (B) Overview of the ZRANB2:RNA structure. F2 is shown as a ribbon diagram. The zinc ligands are shown as sticks and the zinc ion is shown as a sphere. The first 4 bases (AGGU) are shown as sticks. Protein side chains that interact with RNA are shown in yellow. (C) (Left) Hydrogen bonds between F2 and the 2 guanines (green dashed lines). Water molecules are shown as red crosses. (Right) Summary of hydrogen bonds to each guanine. (D) (Left) The hydrogen bond network involving N76, N78, C85, and N86, which defines the Ura4 binding site. (Right) Summary of hydrogen bonds to Ura4.

Gua2, Gua3, and Ura4 make a number of hydrogen bonds to side chains and the backbone of the protein (Fig. 4 C and D). A striking feature is the bidentate interaction of both Gua2 and Gua3 with an arginine side chain. Gua2 forms 2 hydrogen bonds to R81 side chain protons: the carbonyl oxygen O6 with H^ε and 1 of the H^η protons with N7. O6 also forms a water-mediated hydrogen bond with the backbone amide proton of R81. The imino proton on the Watson–Crick face of Gua2 forms a second water-mediated hydrogen bond, whereby the water interacts with both the D68 carboxylate group and the A80 backbone amide.

Gua3 forms a bidentate interaction with R82, and again the O6 forms a second hydrogen bond to a backbone amide, in this case W79. The backbone carbonyl of V77 forms hydrogen bonds with both the imino proton and the amino group of Gua3, and a water-mediated hydrogen bond connects the 2′ hydroxyl group of Gua3 with the side chain of N86. Ura4 forms 3 side-chain-mediated hydrogen bonds to F2. The O4 carbonyl group recognizes side-chain amide protons in both N76 and N86, and the side-chain carbonyl group of N76 makes a hydrogen bond with the Ura4 imino proton.

These interactions explain the strong observed preference for a core GGU sequence. The pattern of hydrogen bond donors and acceptors observed for the interaction with each guanine (Fig. 4C) is not compatible with either adenine or cytosine, and although uracil has a similar polarity to guanine (O4 replaces O6 and N3 replaces N1 on the Watson–Crick face) it would be unable to form the additional hydrogen bond made to each guanine N7. Recognition of Ura4 relies on a hydrogen-bond network involving N76, N78, C85, and N86 (Fig. 4D), which orients N76 such that the polarity of its side chain is complementary to uracil.

Conformation of the Three Adenines.

Whereas electron density for the GGU was well-defined and unambiguous, refinement of the adenines was more difficult. Ade1 was modeled at 50% occupancy into a conformation in which the base extends the Gua2-W79-Gua3-Ura4 stack (Fig. 4 and Fig. S3C), but this nucleotide does not directly contact F2. Ade5 was modeled into 2 different conformations (Fig. 5 A and B), and for one of these, density could additionally be observed for Ade6 (Fig. 5A). In this latter conformer, the RNA backbone changes direction by ≈90° and folds back on itself. Ade5 makes contacts with the backbone and ribose ring of Ura4 and the Ade6 base is stacked coplanar with Ade5. In the alternate conformer, Ade5 is oriented away from the protein (Fig. 5B).

Fig. 5.

Conformation of Ade5 and Ade6. (A) Structure of 1 conformer, in which Ade5 and Ade6 were modeled with 50% occupancy. Residues 1–4 are shown in light brown and residues 5 and 6 are shown in light pink. (B) Structure of the second conformer, in which Ade5 points away from F2. No density for Ade6 was observed. (C) Model of the F2:RNA complex calculated by using the intermolecular NOEs between V77/M87 and Ade5 H2. The X-ray conformer from A is shown in green and the NOE-restrained model is in gray.

Our chemical-shift perturbation analysis revealed that M87 underwent one of the largest chemical-shift changes upon RNA binding and the alanine point mutation M87A reduced the RNA binding affinity of F2 (Fig. 1F). However, no contacts are observed between this residue and the RNA in the crystal structure. Further, the few intermolecular NOEs that we could assign with confidence were between the H2 proton of a single adenine and the side chains of M87 and V77 (Fig. S2C). Neither of the conformations in the crystal is consistent with these NOEs, suggesting that either the Ade5–Ade6 dinucleotide undergoes significant motion in solution or the conformation differs in this region in solution. Examination of packing in the crystal (Fig. S3D) shows that a tyrosine side chain from a symmetry-related molecule is next to M87, and it is possible that crystal-packing forces disturb Ade5 and Ade6 from their preferred solution positions.

To assess the likely conformation of Ade5 and Ade6, we carried out a restrained molecular dynamics calculation by using HADDOCK (17). In this calculation, the positions of the AGGU sequence and the entire protein other than M87 and V77 were fixed. Ade5 and Ade6 were allowed to reorient freely under the influence of the force field and the NOEs to the protein. When the NOEs are directed to Ade5 H2, the calculations revealed a single conformer that is consistent with all of the NOEs (Fig. 5C). In this structure, Ade5 is in a position similar to that exhibited in the X-ray conformer shown in Fig. 5A (but translated ≈5 Å toward the protein), suggesting that Ade5 and Ade6 do spend at least a proportion of their time in such a conformation.

The Binding Preferences of ZRANB2 Corroborate Functional Splicing Data.

It has been demonstrated that ZRANB2 can alter the splicing of Tra2-β, GLUR-B, and SMN2 reporter genes in splicing assays (6, 7). The Tra2-β reporter gene contains 4 exons (Fig. 6A), and addition of ZRANB2 promotes the exclusion of exon 3. Given our SELEX data, it is notable that the sequence of the 5′ splice site of exon 3 is AG/GUAA, whereas those of exons 1 and 2 are GG/GUAC and AA/GUGA, respectively (where the slash represents the cleavage site in the splicing reaction). To test whether the tandem zinc finger domain of ZRANB2 will bind preferentially to the 5′ splice site of exon 3, we carried out surface plasmon resonance competition experiments. A 5′-biotinylated oligonucleotide containing the sequence AGGUAA was immobilized on a streptavidin-coated Biacore chip and a solution of ZRANB2-F12 containing unlabeled oligonucleotides corresponding to the 5′ splice sites of exons 1, 2, or 3 was injected. As shown in Fig. 6B, the oligonucleotide from the 5′ splice site of exon 3 caused the largest reduction in binding of ZRANB2 to the chip, indicating that the protein has a clear preference for this sequence above the others.

Fig. 6.

The RNA-binding preferences of ZRANB2 correlate with its observed splicing activity. (A) Schematic of the Tra2-β minigene. Normally, the dominant transcript contains exons 1, 3, and 4. After the addition of ZRANB2, a transcript containing only exons 1 and 4 is observed (6). The sequences of the 5′ splice sites of each exon are shown (the vertical line indicates the intron–exon boundary). (B) Surface plasmon resonance data. Responses are shown after the injection of F12, in the presence of a competitor RNA, on to a chip bearing the sequence of the 5′ splice site of exon 3. (C) Sequence alignment of human RanBP2 ZnFs showing that several contain the residues shown to meditate RNA recognition in ZRANB2. Color is ascribed by residue type and RNA binding residues in F2 are highlighted in gray boxes. Swissprot accession codes are given in SI Text.

Discussion

ZRANB2 ZnFs Are ssRNA-Binding Domains.

The RanBP2 ZnFs of ZRANB2 bind with high specificity to ssRNA containing a core GGU sequence, defining another class of sequence-specific ssRNA-binding domain. The interaction is mediated predominantly by hydrogen bonds between protein side chains and the bases, together with a tryptophan stacking motif. The bidentate arginine–guanine hydrogen bonds observed here have been observed previously, although only in complexes involving double-stranded nucleic acids (the interaction still permits normal Watson–Crick base pairing). For example, Ets domain transcription factors contain a conserved arginine in the recognition helix that recognizes a guanine (e.g., ref. 18), as do type II restriction enzymes such as BamH1 (19). Complexes formed between the arginine-rich peptides of retroviral proteins such as HIV Rev also exhibit this type of interaction (e.g., ref. 20); here the target is dsRNA. A related interaction, in which the 2 distal Nη groups form a bidentate hydrogen bond with guanine, is observed in a number of complexes, including classical ZnF–DNA complexes (21).

Although several published structures [including the HIV nucleocapsid ZnF, which contacts a flipped out guanine at the end of an RNA stem loop (22, 23)] show Trp–purine stacking interactions, the ZRANB2:RNA structure appears to be an example of true intercalation of a tryptophan side chain between 2 bases. Stacking interactions involving tyrosine or phenylalanine are more common. The Tis11d ZnFs, which recognize AU-rich elements (24), and the pumilio repeat proteins, which bind elements in the 3′ untranslated regions of target mRNAs (25), both display such stacks.

Several water-mediated hydrogen bonds and hydrogen bonds to protein backbone atoms also contribute to binding specificity in ZRANB2, including a hydrogen bond to the 2′OH of Gua3 that provides support for the selectivity of ZRANB2 for ssRNA over DNA. No interactions with the phosphate backbone are observed, in contrast to several other common classes of ssRNA binding domains, such as RRMs (reviewed in ref. 26).

Despite these differences however, several similarities exist between the activities of the tandem ZRANB2 ZnFs and those of the Tis11d/TTP/Mex-5 family. In both cases, each ZnF recognizes ssRNA in a sequence-specific manner with micromolar affinity and there does not appear to be a requirement for the RNA to be presented in a specific conformation. In contrast, a number of other RNA-binding proteins, including Nova (27), RBMY (28), nucleolin (29), TFIIIA (30), and the nucleocapsid ZnFs (23), recognize RNA in the context of a specific secondary structure. The presence of 2 (or more) modular RNA-binding domains that recognize a single RNA sequence is also a common theme in nucleic acid recognition (31), and one that potentially allows double sites with variable spacing to be recognized.

Conservation of the ZRANB2–F12:RNA Interaction.

The ZnFs of ZRANB2 are highly conserved from Xenopus to humans (Fig. S4A), and a number of the residues that are important for RNA recognition are conserved in insects and nematodes. It is notable that the N-terminal finger of the Caenorhabditis elegans protein is missing one of the conserved zinc-binding cysteines and is disordered in solution (Fig. S4B). Yeast and rice also contain orthologues of ZRANB2 in which the RNA recognition surface of the ZnFs is partly conserved. Further, examination of the F2:RNA structure reveals that some of the observed substitutions could most likely retain specificity for GGU. For example, the asparagine that replaces R82 in the yeast protein could still hydrogen-bond, via its side-chain NH₂ moiety, to the O6 of Gua3. This high degree of conservation suggests that the RNA-binding activity and sequence specificity of ZRANB2 is part of an ancient and important function.

A Family of RanBP2 ZnF RNA-Binding Domains.

Alignment of protein sequences obtained from a BLAST search reveals that subsets of the residues in ZRANB2-F2 that directly contact the GGU motif are also present in several other human RanBP2 ZnFs (Fig. 6C). In fact, TLS has already been shown to bind GGUG RNA motifs (32, 33), consistent with our data. RBP56 shares all of these residues and so should exhibit the same sequence specificity and mode of recognition as ZRANB2. Both TLS and EWS carry a single change: R81 to W. Inspection of the F2:RNA structure reveals that the indole H^N of this tryptophan could still make a hydrogen bond with the O6 carbonyl of Gua2, thereby mimicking the base specificity imparted by the arginine. Tex13a, RBM5, and RBM10 all have changes to the 2 residues that specify uridine at position 4 (N76 to A/L/V and N86 to F). Notably, the N76L and N86F changes in RBM5 give rise to a surface that is similar in overall shape but lacks hydrogen-bonding capacity. It is therefore possible that RBM5 will accept either pyrimidine in this position. Further, many of these proteins have already been ascribed a function in splicing. For example, TLS, EWS, RBP56, RBM5, and RBM10 all have been shown to be associated with the early spliceosome (34–36). Most recently, a role for RBM5 in the splicing of apoptosis-related genes was demonstrated in 2 separate studies (37, 38).

Other proteins, including MDM2 (a regulator of p53), contain RanBP2 ZnFs that display 1 or 2 of the RNA-binding residues that we have identified. If such domains can bind RNA, their sequence specificity will likely be very different. In contrast, the RanBP2 ZnF in Npl4 contains none of the RNA-binding residues and instead harbors a conserved Thr–Phe dipeptide that mediates an interaction with ubiquitin (10). Similarly, the RanBP2 ZnFs of Nup153 recognize RanGTP/GDP by using a Leu-Val motif in the same position (11). The RanBP2 ZnF in the putative regulator of cytokine signaling TRABID/ZRANB1 (39) contains several of the RNA-binding residues and a Thr-Tyr motif, and it will be interesting to ascertain whether both of the biochemical functions indicated are active in this domain.

The Function of ZRANB2.

The functional data available for ZRANB2 point strongly toward a role in alternative splicing. The observation that ZRANB2 binds both U170K and U2AF³⁵ suggests that it acts early in the splicing reaction, consistent with a role in splice site choice. The RNA-binding properties of ZRANB2 draw parallels with canonical SR proteins such as ASF/SF2 and SC35, although unlike these latter proteins ZRANB2 does not localize to nuclear speckles. It is also notable that, although SR proteins bind RNA with high affinity, SELEX data have rarely revealed clear consensus sequences (3). For example, the structure of SRp20 bound to RNA reveals that only 1 of the 4 nt is recognized in a sequence-specific manner (40). This finding contrasts sharply with the well defined consensus sequence obtained here and hints at a role for ZRANB2 in regulating specific transcripts, rather than a global role in constitutive splicing.

The target sequence for a single ZRANB2 ZnF strongly resembles the 5′ splice site, which is conserved across all metazoans. It is therefore possible that ZRANB2 might act by binding directly to a subset of 5′ splice sites so as to prevent recognition of those sites by the spliceosome. Such a mode of action is supported by the affinities of ZRANB2 for the different splice sites in the Tra2-β minigene. Indeed in each of the exons excluded from the transcripts of the GluR-B, SMN2 and Tra2-β minigenes after the addition of ZRANB2, a single or double (A)GGUA(A) site is present at or around the 5′ splice site of the major excluded exon. Given that 3′ splice sites also display a GG dinucleotide, it is also possible that the 2 ZRANB2 ZnFs might simultaneously contact both splice donor and splice acceptor sites within the same transcript to influence splicing.

Alternatively, ZRANB2 might recognize cryptic splice sites containing 1 or 2 AGGUAA sequences, either activating or suppressing their use in a similar manner to that of the Drosophila protein PSI (41, 42) and the pseudo 5′splice site in the P-element transposase pre-mRNA. The fact that ZRANB2 can accommodate a range of spacings between 2 AGGUAA motifs suggests that the protein might recognize clusters of these motifs rather than a strict tandem site. A similar situation has been observed for the splicing factor Nova (43), which has 3 KH domains separated by a long and short linker.

Methods

Expression and Purification.

RanBP2-type ZnF domains from ZRANB2 and other human proteins were expressed as GST-fusion proteins and purified by glutathione affinity chromatography and either gel filtration or cation exchange chromatography. Additional details are provided in SI Text.

SELEX.

The ZRANB2-F12 SELEX protocol was based on that used by Sakashita and Sakamoto (44). A library of ssRNA sequences was incubated with GST-F12 on glutathione Sepharose beads, and after washing protein–RNA complexes were eluted with glutathione and the selected RNA was reverse-transcribed and amplified by PCR. Sequencing of selected sequences was carried out after 7, 9, and 13 rounds of selection. A more detailed description of the protocol is provided in SI Text.

Gel Shifts and Fluorescence Anisotropy Titrations.

These experiments were carried out by using standard protocols. Details can be found in SI Text.

NMR Spectroscopy.

The structure of F2, RNA titrations, and full assignments for F12 and a F2:RNA complex were determined by using standard solution NMR experiments. Details can be found in SI Text.

Surface Plasmon Resonance.

Competition binding experiments were carried out by flowing a solution of F12 over a streptavidin chip coated with a biotinylated ssRNA oligonucleotide (containing the sequence AGGUAA) in the presence of an unlabeled competitor oligonucleotide. Details can be found in SI Text.