To study Sxl pre-mRNA splicing, we established an in vitro Sxl pre-mRNA splicing assay. We chose the region from exon 3 (including its 3? splicing site) to the end of exon 4 in Sxl pre-mRNA as the splicing substrate (referred to as E3-4), as both the 3? splicing site and the 5? splicing site of Sxl exon 3 are involved in splicing regulation even though the 5? splicing site is dominant in regulation ^[21], ^[22]. Under standard in vitro splicing conditions using Hela cell nuclear extract, the splicing of E3-4 RNA was very inefficient and there was almost no reaction products after two hours (data not shown). Similar attempt has been made previously in establishing in vitro splicing assay with Sxl transcript from exons 2 to 4, but no splicing could be detected ^[23]. Fortunately, we found that the in vitro splicing of E3-4 was dramatically accelerated by the addition of splicing-enhancer SR proteins (e.g. SC35), as evident by the amounts of splicing intermediates accumulated (Figure 1A, lanes 1?3). However, even with SC35 (lanes 4?5), the second step of Sxl splicing was still very slow and the final splicing products were hardly visible (Figure 1A, lanes 4 and 5). Nevertheless, by monitoring the amounts of splicing intermediates, we were able to observe directly the regulatory effect of SXL in Sxl splicing. When SXL was added, E3-4 splicing was inhibited in a dose-dependent manner (Figure 1A, lanes 6?8). Thus, the inhibition by SXL should occur at an early splicing step as the amounts of both splicing intermediates were decreased, but the ratio of the two bands remained the same. This result validates our splicing assay as a useful method for studying the regulation of Sxl splicing.

To our surprise, we found that the splicing of E3-4 could be inhibited directly by SNF using the established in vitro splicing assay, and this inhibition was also dose-dependent, similar to that of SXL (Figure 1B, lanes 2?4). However, the inhibitory effect of SNF was not as strong as that of SXL, and it required 15 times more SNF to achieve a comparable inhibition to that of SXL. As controls, we found that RNA splicing factors U2AF⁶⁵ and hnRNP L did not cause any observable changes in the in vitro splicing assay (Figure 1B, lane 8 and lane 10). PTB, which negatively regulates a number of splicing events in human ^[24], seemed to down-regulate E3-4 splicing slightly (Figure 1B, lane 9). We have also tested human U1A protein, an SNF homologue protein, and no effect of U1A on E3-4 splicing was observed (Figure 1B, lanes 5?7). On the other hand, we also added SNF to splicing reactions with a panel of commonly used in vitro splicing substrates, HIV tat for example, and did not observe any significant effect (data not shown). Therefore, Sxl exon 3 splicing suppression by SNF seems to be specific.

In Sxl pre-mRNA, there are multiple polyuridine (poly(U)) sequences scattered in the introns flanking the male-specific exon 3, and four of them contain 8 or more uridine residues (Figure 2A, labeled as RNA A, B, C and D). Previous studies have shown that these poly(U) sequences are critical for SXL to regulate its pre-mRNA splicing, in which SXL inhibits the splicing of exon 3 through a blockage mechanism by binding to poly(U) sequences ^[21], ^[25]?^[27]. Since SNF can directly regulate Sxl pre-RNA splicing in a similar manner to SXL, we decided to explore if SNF can also bind poly(U) sequences.

A band-shift assay was used to test the RNA binding ability of SNF. As expected, SNF could bind U1 snRNA, but did not bind U2 snRNA without U2A? (Figure 2B) ^[15], ^[17]. We then performed an RNA binding experiment in which SNF was mixed with different RNA fragments (RNA A, B, C and D) of the Sxl transcript (Figure 2A). Interestingly, we found that SNF binds weakly to RNA A which contains only a single U-run (?8 poly(U) sequence), and binds strongly to the double U-run-containing RNA C and RNA D, but does not bind RNA B without U-run (Figure 2B). To confirm that such binding is due to the recognition of U-runs by SNF, mutants of RNA D were constructed in which either or both U-runs were changed to UC-runs (Figure 2B). SNF bound mutant RNA D much more weakly when one U-run was changed to a UC-run, and did not bind mutant RNA D when both U-runs were changed to UC-runs (Figure 2B, lanes 1?12, right panel). These results show clearly that the presence of U-runs is necessary for SNF to bind Sxl pre-mRNA.

As SNF bound much more strongly to double-U-run-containing RNA than single-U-run-containing RNA, we studied the binding of SNF to different double-U-run-containing RNAs in which the spaces between two U-runs are different. For a consecutive double-U-run-containing RNA, the size of the SNF-RNA complex formed remained unchanged with increasing SNF concentration (data not shown). This is different from SXL which forms a bigger size complex with the same RNA at high concentrations ^[25], and should indicate that the two consecutive U-runs are simultaneously recognized by one SNF molecule. When the two U-runs were separated by 13 bases (2U-13), the size of SNF-RNA complex formed was still independent of the SNF/RNA ratio (Figure 2C, lanes 4?6), while SXL still formed a bigger complex with the same RNA at high concentrations (Figure 2C, lanes 1?3). Moreover, when the sequence of one of the two U-runs was altered (1U-13), it could still form a complex with SNF and the size was the same as that for 2U-13 (Figure 2C, lanes 7?9). However, the complex of SNF and 1U-13 was less abundant, consistent with a weaker binding affinity. These results support the idea that a strong SNF binding site is composed of two U-run sequences, even if they are separated by more than 10 bases.

To further examine this idea, we separated two U-runs by 120 bases (2U-120) to see if it would still work as a strong binding site (Figure S1, left). Astonishingly, it appeared that a single SNF can still recognize the two U-runs across a significant distance because 2U-120 still formed a single defined complex with SNF, albeit weaker than 2U-13. Furthermore, we compared the binding of SNF to 2U-13-2U and to 2U-13-UC, and found that two complexes could form when there were two double-U-runs (Figure S1, right).

SNF¹⁶²¹ has a point mutation in RRM1 (R49H) and functions as a dominant-negative factor in female-specific Sxl splicing ^[19]. We used the same in vitro RNA-binding assay to examine the RNA-binding ability of SNF¹⁶²¹. With respect to U1 snRNA, SNF¹⁶²¹ behaved like the wide type SNF and bound U1 snRNA efficiently (Figure 2D, left, lanes 2?3). However, no complex was observed when SNF¹⁶²¹ was mixed with the double U-run-containing RNA D (Figure 2D, right, lanes 3). These results suggest that the poly(U) binding ability of SNF is important for this general splicing component to specifically participate in Sxl splicing. As a circumstantial support for this suggestion, human U1A which did not inhibit Sxl exon 3 splicing (Figure 1B), bound fly snRNA U1 but not poly(U) RNA (data not shown).

Even though SNF shows a very high amino acid sequence identity to U1A (Figure S4A), SNF is able to recognize both U1 snRNA and poly(U) RNA while U1A is not able to recognize poly(U) RNA. This novel dual RNA recognition ability of SNF led us to study the solution structure of SNF in order to reveal the structural basis for its unique functions.

Near-complete backbone ¹H/¹⁵N chemical shift assignments for full-length SNF were obtained (Figure S2). The ¹H-¹⁵N correlation peaks of full-length SNF were quite similar to the overlay of those from isolated RRM1 (residues 1?104) and RRM2 (residues 134?216) (data not shown), indicating that each RRM is relatively independent and there lacks inter-domain interaction. Both RRMs of SNF are quite rigid, whereas the linker loop (96?140) is flexible as indicated by low {¹H}-¹⁵N NOE values (Figure S3). Although RRM1 and RRM2 have similar molecular weights, the average ¹⁵N R₂/R₁ ratio for RRM1 (?17.4) and RRM2 (?9.5) are significantly different, also indicating that RRM1 and RRM2 tumble independently in solution. As the quality of NMR spectra of full-length SNF was very poor, the ¹H, ¹⁵N and ¹³C chemical shift assignments for SNF RRM1 and RRM2 were obtained, respectively (BioMagResBank database under accession numbers 6930 and 6844). The structures of SNF RRM1 and RRM2 were also solved separately (Protein Data Bank accession numbers 2K3K and 2AYM). Statistic data indicate that both structures are well defined (Table 1).

The structures of SNF RRM1 and RRM2 are characteristic of typical RNP-type RBDs ^[28] comprising a four-stranded anti-parallel ?-sheet packed against two ??helices (Figure 3). Two additional ??helices (?? and ??) are inserted in the loop regions of RRM1. Helix ?? directly follows ?A with a kink at Ile³⁰, and helix ?? is inserted in loop3 (Figure 3). Residues 88?95 in RRM1 form an ?-helix (?C) which is flexible, as indicated by the fact that {¹H}-¹⁵N NOE values for residues in helix ?C are significantly lower than those in the core regions of RRM1 (Figure S3). The other secondary structure elements (1?83) are well defined with root mean square deviation (RMSD) values for backbone heavy atoms of 0.4 ?. In RRM1, the conserved RNP-1 and RNP-2 submotifs lying in the center of strands ?1 and ?3 contain two conserved aromatic residues (Tyr¹⁰ and Phe⁵³) ^[29]. Compared to RRM1, RRM2 lacks helices ??, ?? and ?C, whereas two short ??strands (?? and ??) inserted in the loop connecting ?B and ?4 form a small anti-parallel ?-sheet (Figure 3).

SNF RRM1 has a close resemblance to U1A RRM1 (Figure S4B), and their backbone C^? atoms align well with a 1.6 ? RMSD (residues 7?86 of SNF, and residues 10?90 of U1A). Calculation of electrostatic potential surfaces showed that RRM1s of both U1A and SNF are highly charged with large clusters of positive charges on the RNA binding surface (Figure S4B). In general, SNF possesses a similar charge distribution on the surface of RRM1 to that of U1A. A small difference is that the residue Arg⁸³ in U1A, which contacts U1 snRNA directly, is replaced with a Gln in SNF (Figure S4B). Another notable difference is that SNF ?C points away from the ?-sheet surface and does not make interactions with residues in the ?-sheet surface, while U1A ?C forms a small hydrophobic core with residues in the ?-sheet surface (close conformation) and it swings away from the ?-sheet surface by ?130? upon U1 snRNA binding ^[29].

The RMSD of C^? atoms between SNF RRM2 (residues 141?216) and U1A RRM2 (residues 207?282) is 1.5 ?. SNF RRM2 lacks the N-terminal capping box in helix ?A, which is an important structural motif in U1A RRM2 and other RRM proteins ^[30]?^[32]. On the exposed ?-sheet surface, U1A RRM2 has more negatively charged residues than SNF RRM2 does (Figure S4C).

The interactions of full-length SNF with U1 snRNA and poly(U) RNA were analyzed by NMR chemical shift perturbation experiments, in which 2D ¹H-¹⁵N HSQC spectra of H²/¹⁵N/¹³C-labeled SNF were recorded with stepwise titration of RNAs (Figure 4). With the addition of the U1 snRNA stem-loop II segment (U1hpII), we observed significant exchange broadening of NH signals without chemical shift change (Figure 4A). These residues are all located in SNF RRM1 and the linker loop (Figure S5A). Locations of these residues on the structure of SNF RRM1 define the binding surface of SNF to U1 snRNA, and it is clear that this binding surface is similar to that of U1A to U1 snRNA (Figure 4C) ^[29], ^[33]. In contrast, no NH signal of residues in SNF RRM2 displays significant change in chemical shift or peak intensity during U1hpII titration (Figure 4A). These results demonstrate that SNF RRM1 is necessary and sufficient for binding U1hpII.

Upon the addition of a double-U-run containing RNA (2U-run, sequence [gene: 5?UUUUUUUUAUUUUUUUU3?]), the NH chemical shift change pattern indicates that both RRMs of SNF are involved in poly(U) binding, unlike U1 snRNA binding (Figure 4B, 4D). In RRM1, many NH cross-peaks display significant intermediate exchange line-broadening (Figure 4B). These residues are mainly located in the ?-sheet surface, the edge of ??helices and loop regions. While in RRM2, a number of NH cross-peaks shifted gradually as the RNA concentration increased (Figure 4B). These residue are mainly located in the vicinity of ?1, ?3, and the C-terminus of RRM2, along with the linker loop between RRM1 and RRM2 (Figure 4B, 4D, S5B, S5C). The poly(U) RNA binding surfaces derived from NMR titration data for SNF RRM1 and RRM2 are roughly similar to those of SXL RRM1 and RRM2 revealed by X-ray crystallography (Figure 4D) ^[34]. In addition, these observations suggest that RRM1 and RRM2 should bind U-runs independently, and it is probable that each RRM binds a different U-run. We have also titrated RRM2 alone with 2U-run RNA and observed a similar chemical shift perturbation pattern to that of RRM2 in the full-length protein (data not shown). This further proves that the two RRMs of SNF bind poly(U) RNA independently.

The binding of SNF to U1hpII and poly(U) RNA RNA (2U-run) was also probed by the SPR measurements (Figure S6). The results clearly show that the interaction between SNF RRM1 and U1hpII involves a high association rate and low dissociation rate (Figure S6A). The apparent equilibrium dissociation constant K_d value extracted from the SPR data was ?4.4 nM, which is about 2 orders higher than the K_d for U1A RRM1 binding U1hpII ^[35], ^[36]. Meanwhile, the K_d for SNF binding 2U-run RNA is ?0.3 ?M, which is about 2 orders higher than the reported K_d for SXL binding 1U RNA ([gene: 5?-GUUUUUUUUC- 3?]) ^[37]. Scatchard-plot analysis using results from the sensorgrams also confirmed a 1?1 complex between SNF and 2U-run RNA. As expected, SNF RRM1 alone binds 2U-run RNA with a K_d of ?6 ?M, about 20-fold weaker in binding affinity. The binding affinity of SNF RRM2 to 2U-run RNA is estimated to be at low mM range from NMR titration data, which is much weaker than that of SNF RRM1.

The 2D ¹H?¹⁵N HSQC spectrum of SNF at 0.4 mM revealed that NH peak intensities of residues in the RRM1 were significantly lower than those of residues in RRM2 (Figure S2). When the concentration of SNF was raised from 0.4 mM to 2 mM, about half of the NH signals in the 2D ¹H-¹⁵N HSQC spectrum, mainly from residues in RRM1, disappeared (data not shown). These results suggest that SNF self-associates through RRM1, and this is consistent with the above mentioned observation that the average ¹⁵N R₂/R₁ ratio of RRM1 is much bigger than that of RRM2 even though the two RRMs are about the same size. In addition, analytical ultracentrifugation analysis indicates that SNF RRM1 could self-associate into dimeric and higher oligomeric species (Figure S7). The apparent equilibrium dissociation constant K_d for monomer-dimer equilibrium of SNF RRM1 is estimated to be a few hundred micromolar.

Overlay of 2D ¹H?¹⁵N HSQC spectra of SNF at different concentrations (0.035?0.39 mM) revealed that some NH cross-peaks (e.g. residues Lys²⁵, Ser²⁶, Leu²⁷, Tyr²⁸, Gln³³, Phe³⁴, Gly³⁵, Phe⁷⁴, Tyr⁷⁵, Asp⁷⁶ and Met⁷⁹) displayed significant concentration-dependent line broadening (Figure 5A). These residues could be mapped to one area of SNF RRM1, which composes the self-association surface. Interestingly, the self-association surface does not interfere with U1 snRNA and poly(U) RNA binding surfaces of SNF RRM1 (Figure 5B).

Plasmids for constructing GST-hnRNP L, GST-PTB, GST-U2AF65 and GST-U1A were gifts from the laboratories of G. Dreyfuss (University of Pennsylvania), M. Garcia-Blanco (Duke University Medical Center), M. Green (University of Massachusetts Medical Center) and S. Mount (University of Maryland), respectively. Generation of GST-SNF, GST-SXL, and SNF¹⁶²¹ constructs has been described previously ^[46]. Small Sxl RNAs, RNA A, B, C and D, (containing Sxl sequences from the restriction sites PvuII (9278) to AflII (9373), EcoRV (10355) to BspMII (10448), StyI (8255) to PstI (10139), and SpeI (9809) to AflII (9892), respectively) were cloned into plasmid pGEM2, and transcribed under the control of the SP6 promoter. Constructs encoding RNA D mutants, U-13-U, and U-120-U were created by ligating annealed DNA oligos into pGEM4 (Promega). To make the SNF, SNF RRM1, and SNF RRM2 constructs, the snf gene (Met¹-Lys²¹⁶) and snf RRM1 (Met¹-Lys¹⁰⁴) were cloned into a pET-21d(+) expression vector, and snf RRM2 (Ala¹³⁴-Lys²¹⁶) was cloned into a pET-28a(+) expression vector.

In vitro splicing reactions were performed as described by Valc?rcel et al.^[47], ^[48]. Purified RNA samples were resolved on PAGE containing 6 M urea as previously described ^[49].

RNAs were generated by in vitro transcription using T7 or Sp6 RNA polymerase (Promega) with P³²-ATP or P³²-GTP. Labeled RNAs were precipitated before mixing with the desired amount of protein in binding buffer and run on 4% native polyacrylamide gels as described previously ^[25]. Binding assays were performed using a high protein/RNA ratio, and protein concentrations were in the order of 1 ?M.

The NMR samples contained about 0.4 mM ¹⁵N- or ¹⁵N/¹³C- or ²H/¹⁵N/¹³C labeled SNF (RRM1 and RRM2). The buffer contained 1 mM EDTA, 0.01% NaN₃, 0.006% 2,2-dimethyl-2-silapentane-5-sulfonate (DSS), 50 mM PBS, 90% H₂O, and 10% D₂O, at pH 7.2. All NMR data were collected at 298 K on a Bruker AVANCE 600 spectrometer. Backbone sequential and side-chain assignments were obtained using standard 3D NMR experiments ^[50]. Proton chemical shifts were referenced using DSS, whereas ¹³C and ¹⁵N chemical shifts were referenced indirectly to DSS ^[51]. NMR spectra were processed using NMRPipe ^[52] and analyzed using NMRView ^[53].

NOE constraints were obtained from automated analysis of NOESY spectra using the computer program SANE ^[54]. Angle constraints (? and ?) of the secondary structure were derived using TALOS ^[55]. Hydrogen bonds were assigned based on analysis of NOEs and secondary structure predictions by CSI ^[56] and TALOS ^[55].

The initial structures were calculated with the CANDID ^[57] program by using only NOE distance constraints. Hydrogen bond constraints and dihedral angle constraints were then added for CYANA ^[58] calculations. After several rounds, 200 structures were calculated and 100 structures with lowest target function were selected for further refinement using the AMBER program (version 7.0) ^[59]. The 20 structures with the lowest AMBER energy were used for the final analysis. The final structures were analyzed by using MOLMOL ^[60] and assessed by using PROCHECK-NMR ^[61].

Two RNA sequences [gene: 5?CUUGGCCAUUGCACCUCGGCUGAGT3?] (U1hpII) and [gene: 5?UUUUUUUUAUUUUUUUUU3?] (2U-run) were synthesized and purified by Allele Biotechnology & Pharmaceuticals Inc. RNAsin was added to prevent RNA degradation. For U1hpII binding, a series of 2D ¹H-¹⁵N HSQC experiments were carried out by adding this RNA oligonucleotide to the SNF sample to reach a final protein:RNA ratio of 1?7. For 2U-run RNA binding, the final ratio was 1?15.

Experimetal methods about NMR relaxation measurements, Biosensor analysis, and analytical ultracentrifugation can be found in supplemental file S1.