Absence of interdomain contacts in the crystal structure of the RNA recognition motifs of Sex-lethal

Abstract

By binding specific RNA transcripts, the Sex-lethal protein (SXL) governs sexual differentiation and dosage compensation in Drosophila melanogaster. To investigate the basis for RNA binding specificity, we determined the crystal structure of the tandem RNA recognition motifs (RRMs) of SXL. Both RRMs adopt the canonical RRM fold, and the 10-residue, interdomain linker shows significant disorder. In contrast to the previously determined structure of the two-RRM fragment of heterogeneous nuclear ribonucleoprotein Al, SXL displays no interdomain contacts between RRMs. These results suggest that the SXL RRMs are flexibly tethered in solution, and RNA binding restricts the orientation of RRMs. Therefore, the observed specificity for single-stranded, U-rich sequences does not arise from a predefined, rigid architecture of the isolated SXL RRMs.

Proteins containing one or more copies of the 80- to 90-aa RNA recognition motif (RRM) often mediate the specific recognition of RNA required for numerous physiological processes, including splicing, nuclear transport, and translational regulation (1–3). Although an individual RRM can be sufficient for RNA binding (4), RRMs also commonly occur in repeats of 2–4 linked domains that act synergistically to increase affinity and specificity for target RNA sequences (5, 6). Proteins with functionally coupled, multiple RRMs are numerous. For example, the splicing factor hU2AF⁶⁵ requires all three of its RRMs for high-affinity RNA binding in vitro (7). Likewise, polyadenylate binding factor requires two of four RRMs (8), and hnRNP A1 and Sex Lethal (SXL) require two RRMs (6, 9). Interprotein interactions, as observed in the U2B′′/U2A′/RNA complex (10), also can modulate the activities of RRM proteins. Relatively little is known, however, about the structural basis for the RNA binding specificity of individual RRMs or the mechanisms of functional coupling between multiple RRM domains.

One structure containing two linked RRMs has been reported (11, 12), that of the heterogeneous nuclear ribonucleoprotein (hnRNP) component, hnRNP Al. The hnRNP Al protein binds single-stranded RNAs of varying sequences with a broad range of binding affinities in vitro (13, 14). In the absence of RNA, the two RRM domains of hnRNP Al form an interdomain interface that orients the domains in an antiparallel fashion. If this arrangement is preserved in the hnRNP A1/RNA complex, RNA binding would be favored by the interdomain contacts. The small size (≈850 Ų) of the interdomain interface and the apparent flexibility of the interdomain linker, however, raise further questions about the role of contacts between RRMs.

Do such contacts preorganize the RNA binding surface or do they form upon binding of specific RNAs? Are interdomain contacts conserved among tandem-RRM proteins? To identify general aspects of proteins with dual RRMs and to explore the roles of interdomain interactions, we have determined the structure of the tandem RRM motifs (amino acids 112–294 of SXL, RRM1+2) of the Drosophila melanogaster SXL protein.

SXL plays central roles in the alternative splicing and translational regulation of mRNAs, functioning as a master switch in sex determination and X chromosome dosage compensation in D. melanogaster (reviewed in ref. 15). By distinct mechanisms, SXL regulates expression of Sxl, transformer (tra), and male-specific-lethal-2 (msl-2) mRNAs, thereby ensuring female sex determination. By default, spliceosome components bind to tra and Sxl pre-mRNAs to produce non-sex-specific and male-specific mRNAs, respectively. SXL promotes female-specific splicing by blocking access to male-specific splice sites in its own message, and by binding at a distance from the non-sex-specific 3′ splice site in the tra pre-mRNA. SXL also inhibits translation of msl-2 mRNA in a manner that depends on the 5′ and 3′ untranslated regions (3, 16, 17). SXL contains several distinct domains, including an N-terminal Gly/Asn-rich domain that appears to be a protein–protein interaction domain (18), and tandem RRMs that are required for activity in vivo (19, 20) and in vitro (6). The natural SXL binding sites and a consensus site identified by in vitro selection demonstrate that the SXL RRMs have a strong sequence preference for single-stranded, U-rich sequences (21, 22).

MATERIALS AND METHODS

Sample Preparation.

SXL RRM1+2 (containing residues 112–294) was expressed from a pET-3b vector in BL-21(DE3) pLysS Escherichia coli cells (23). Selenomethionine (SeMet) was incorporated to >98% by growth in M9 minimal medium supplemented with Leu, Ile, Val, Phe, Lys, and Thr and 75 mg⋅l⁻¹ SeMet at mid-log phase immediately before induction with isopropyl β-d-thiogalactoside (24). SeMet incorporation was measured by electrospray mass spectrometry. RRM1+2 was purified as described (6).

Crystals were grown initially from hanging drops containing a 1:1 mixture of 18 mg⋅ml⁻¹ RRM1+2, 10% glycerol, 5 mM Hepes (pH 7.5), and mother liquor [20% monomethyl ether (MME) PEG 2000, 0.1 M Hepes buffer (pH 6.0), and 0.2 M ammonium sulfate] equilibrated with the mother liquor by vapor diffusion. Because of the rarity of de novo crystallization, initial crystals were microseeded into subsequent crystallizations. Optimized crystals grew at pH 7.6 by using 28% MME-PEG 2000 as the precipitant; all other conditions remained constant. Crystals were transferred to a cryoprotectant [30% MME-PEG 2000, 0.1 M Hepes (pH 7.6), 0.2 M lithium sulfate] and flash-frozen in liquid nitrogen. The crystals have the symmetry of space group P6₁ at room temperature (a = b = 52.20 Å, c = 124.26 Å), but freezing reduced the symmetry to that of space group to P2₁ (a = 50.91 Å, b = 124.47 Å, c = 50.93 Å, β = 120°), with three SXL monomers in the asymmetric unit.

Phase Determination and Structure Refinement.

Experimental phases were obtained by multiwavelength anomalous dispersion methods (Table 1). Data were collected at the Stanford Synchrotron Radiation Laboratory beamline 1–5, and an additional native data set was collected at the Advanced Light Source beamline 5.0.2 at the Lawrence Berkeley National Laboratory. Diffraction data were processed by using mosflm (25) and scaled with the CCP4 suite of programs (26). A modified Matthews difference Patterson map (27) was produced by combining the anomalous differences from the peak wavelength (f′′) and dispersive differences between the remote (“f_low”) and inflection (f′) wavelengths. Six of 15 Se atoms were located on the basis of the Patterson map, including three sites evenly spaced along the y axis. The other SeMets are disordered or have alternate conformations. Phases were calculated by using the program sharp (28). The electron density map calculated at 2.7-Å resolution after solvent flattening with dm (29) was traced by using the program o (30) using the hnRNP A1 and U1A RRM structures as guides.

Table 1.

Data collection and refinement statistics

Wavelength, Å	1.06879 (λ1)	0.98001 (λ2)	0.97957 (λ3)	1.2000 (Native)
Rsym*	0.062	0.068	0.072	0.069
Completeness, %	84.1	85.0	85.0	99.8
Multiplicity	5.1	5.7	5.9	6.7
I/SD†	20.5  (6.2)	19.6  (6.2)	18.6  (5.6)	13.4  (5.7)
Phasing power‡, anom/disp	2.99/0.75	—/1.43	0.88/2.14	—
Resolution: 20.0–2.7 Å
Mean figure of merit (20.0–2.8 Å resolution)§: 0.70 (0.74 after solvent flattening)
Crystallographic Rcryst/Rfree (20.0–2.7 Å)¶: 0.218/0.262
rms Δbonds, rms Δangles‖: 0.015 Å, 1.990°

^*R_sym = Σ|I − 〉I〈|/Σ I; I, intensity. 

^†I, intensity. Numbers in parentheses denote I/SD for highest resolution bin. 

^‡Phasing power = [Σ_n|F_H|²/Σ_n|E|²]^1/2; F_H, calculated heavy atom scattering factor; E, lack of closure error. 

^§Mean figure of merit = 〈|Σ_αP(α)e^iα/Σ_αP(α) |〉; α, phase; P(α), phase probability distribution. 

^¶R_cryst = Σ |F_o − F_calc|/Σ F_o; F_o, observed structure-factor amplitude; F_calc, calculated structure-factor amplitude. 

^‖rms deviations from ideal values. 

The structure was refined with the Crystallography and NMR Suite (CNS) (31) by using data > 2 σ from 20.0- to 2.7-Å resolution, excluding 10% of the data reserved to calculate the free R value. Procedures carried out with CNS included torsional angle dynamics, simulated annealing by using a maximum likelihood target function, restrained individual B-factor refinement, conjugate gradient minimization, and bulk solvent correction. The program o was used to adjust the model to fit sigma-A-weighted and phase-combined 2F_o − F_c electron density maps.

Structure Comparisons and Modeled Structures.

Structural superpositions were calculated by using the program lsqman (32). Variable loops, linkers, or the additional residues of U1A and U2B′′ α₁ and α₃ were not used in the superpositions. Linker residues for the hnRNP Al-like arrangement of RRMl+2 were produced by using the “lego_loop” command in o. The loop chosen for the model showed the best fit to the endpoints, and it was unique among the solutions for its lack of steric clashes with neighboring RRMs.

RESULTS AND DISCUSSION

Canonical RRMs Adopt a Novel Orientation.

We determined the structure of SXL RRM1+2 at 2.7-Å resolution by using multiwavelength anomalous diffraction methods (Fig. 1A, Table 1). The refined model contains 156 of 184 amino acids in each of the three SXL molecules in the asymmetric unit. Residues lacking electron density were not included in the model. Disordered segments include the 14 N-terminal residues, five C-terminal residues (290–294), six residues in the α₂-β₄ loop in RRM2 (276–282), and two residues (204–205) of the interdomain linker.

Figure 1.

Structure of the SXL RRM domains. (A) Sigma-A-weighted, 2F_o − F_c electron density map contoured at 1σ illustrates the quality of the map. Residues shown are Tyr-164–Phe-173. (B) Ribbon diagram of SXL RRM1+2. Residues left out of the model for lack of interpretable electron density include 111–123, 204–205 (interdomain linker), 276–282 (α2-β4 loop of RRM2), and 290–294 (C terminus). RNP-1 and RNP-2 sequences are colored in magenta. (C) Electrostatic surface potential of SXLl+2 calculated with grasp (blue = +5 kT > white = 0 kT > red = −5 kT). The absence of interdomain contacts in the SXL RRMl+2 crystal structure is evident. Based on the structures of the RNA complexes of UlA and U2B′′/U2A′, the most positive regions are likely RNA binding surfaces.

As expected from sequence homologies to RRM proteins and the NMR structures of the individual SXL RRMs (33, 34), the RNA binding domains adopt the conserved βαββαβ topology (Fig. 1B, Table 2). The central β-strands, β₁ and β₃, contain the conserved RNP motifs, RNP-2 and RNP-1, respectively. Three of four β₂-β₃ and α₂-β₄ loops, which are dynamically flexible in NMR-derived structures, are ordered in each SXL chain in the crystal structure. The ordered 1oops make intermolecular contacts in the crystals that presumably trap the observed conformations. The α₂-β₄ 1oop of RRM2 is the only loop lacking electron density and also the only α₂-β₄ loop that does not make intermolecular crystal contacts.

Table 2.

rms deviations (Å) of equivalent alpha carbons show close structural similarities among RRM domains

RRM	SXL RRM1	SXL RRM2
SXL RRM2	0.9	–
hnRNP A1 RRM1 (11)	1.4	1.5
hnRNP A1 RRM2 (11)	1.9	1.9
U1A (37)	1.1	1.1
U2B" (10)	1.1	1.1

Residues Asn-126–Arg-158, Val-171–Val-191, Lys-194–Ala-201, Asn-212–Arg-220, Thr-223–Asp-245, Arg-252–Ile-276, and Gln-282–Ala-289 of SXL RRM1 +2 were used for the comparisons. 

Unlike the structures of the individual RRM domains, the crystal structure of SXL RRM1+2 can reveal the conformation of the interdomain linker and any contacts between the domains. Characteristic of a poorly ordered region, the electron density for the 10-residue interdomain linker was weak and discontinuous in maps calculated with the multiwavelength anomalous diffraction data. In the later stages of refinement, electron density for part of the linker became apparent. After RRM1, Ala-201–Pro-203 form a random coil segment. No interpretable density was observed for Gly-204 and Gly-205. After the gap, Glu-206 begins a short helix that connects to β₁ of RRM2 at Asn-212. The refined model contains a 6.9-Å gap between Pro-203 and Glu-206. Distance constraints define unambiguously the intramolecular connection between RRM1 and RRM2. From the C terminus of RRM1, the second-closest RRM2 N terminus is 28.8 Å away, a distance that two residues could not possibly bridge. Weak electron density and the high B-factors of the interdomain linker imply significant flexibility in this region.

In contrast to the RRM domains of hnRNP A1, the structure of SXL RRM1+2 reveals a complete absence of interdomain contacts. In the structure of hnRNP Al (11, 12), α₂ of RRM2 packs against the N- and C-terminal segments of RRMl, establishing a defined orientation of the domains (Fig. 2A). The RRM domains of SXL make only intermolecular contacts in the crystal, however, implying that the domains are flexible with respect to each other in solution. Thus, preorganization of the tandem binding surface cannot contribute to the specificity of SXL RRM1+2 for U-rich, single-stranded RNA sites. Instead, if interdomain contacts are formed, they must be coupled to RNA binding.

Figure 2.

Relationship of RRM domains in SXL and hnRNP Al. (A) Ribbon diagrams of Sex-lethal (green) and hnRNP A1 (red) superimposed by using the secondary structural elements of RRM1 of each protein. In contrast to hnRNP A1 RRM2, RRM2 of SXL makes no contacts with RRM1 and differs in orientation from RRM2 of hnRNP A1 by a 92.5° rotation. (B) NMR chemical shift changes for SXL RRM1+2 upon binding the sequence r(GU₈C) (35). (Left) SXL RRM1+2. (Right) RRM1+2 modeled with the RRM domains arranged as in an hnRNP Al. Chemical shift changes are colored as follows: blue (0–200 Hz), green (201–400 Hz), yellow (401–600 Hz), orange (601–800 Hz), and red (801–1000 Hz). Changes caused by RNA binding are concentrated along the interdomain linker and RNA binding surfaces.

The conclusion that the RRMs of the isolated SXL protein are flexibly tethered is consistent with solution studies. Residues in the linker show large changes in NMR chemical shifts upon binding of the oligonucleotide r(GU₈C) to RRMl+2, but only minimal chemical shift changes are seen in the linker residues upon binding of a “linker-RRM2” protein to the oligonucleotide d(T₄GT₂GT₈CTAG) (6, 35). These results argue against direct contacts between the linker and RNA, and they suggest that the linker undergoes a conformational change upon RNA binding. An extensive quantitative analysis of the effect of linker length on the overall RNA affinity of RRM proteins also suggests that the linkers in SXL and hnRNP Al function to increase the effective concentration of the RRMs (36). Interestingly, these binding studies in conjunction with the observation that proteolysis of the linker leads to dissociation of the RRM domains of both hnRNP Al (9) and SXL (34) suggest that in solution hnRNP Al, like SXL, may access an “open” conformation lacking interdomain contacts.

The crystal structure of hnRNP Al clearly reveals a “closed” conformation in which contacts between the RRM domains favor a particular overall architecture. It is of interest to consider whether or not the RRM domains of SXL could form a similar closed arrangement. Analysis of 54 protein sequences containing two RRMs reveals covariance of core residues, but coupled sequence changes in the putative interdomain interfaces are not apparent (S.M.C. and T.A., unpublished results). These results suggest that the interface between the RRMs of hnRNP A1 may not be conserved in all multiple RRM proteins.

To further explore the specific potential of SXL to form conserved interdomain contacts, we modeled the SXL RRMs in the arrangement observed in hnRNP Al (Fig. 2B). The interdomain interface in the resulting SXL model lacks steric clashes and shows general chemical complementarity, including an interdomain salt bridge analogous to Asp-266–Arg-184 found in hnRNP Al. With the SXL RRMs in the orientation seen in hnRNP A1 (Fig. 2B), however, a single, minimal, 10-base, specific binding site could not span the β-sheets of SXL RRMs in a manner analogous to that observed for single-stranded RNA segments bound to U1A and U2B′′/U2A′ (10, 37). These results argue that the closed conformation of hnRNP A1 is unlikely in the RNA complex of SXL, assuming that the SXL RRMs make canonical RNA contacts. Restriction of the orientation of the SXL RRMs upon RNA binding is consistent with previous NMR observations (35) that show chemical shift changes upon binding of r(GU₈C) concentrated on the RNA-binding platforms of the RRMs as well as the interdomain linker (Fig. 2B). Taken together, these results support the idea of a distinct closed architecture for the SXL/RNA complex.

RNA Binding Surfaces.

Although the current SXL RRMl+2 structure contains no RNA, several studies define residues involved in RNA binding. In SXL RRM1+2, the conserved RNP sequences show significant changes in NMR chemical shifts upon binding the oligonucleotide r(GU₈C), and the Phe-256–Ala mutation in RNP1 of RRM2 reduces SXL RRM1+2 binding to r(GU₈C) by ≈200-fold (35). The Arg-252 main-chain amide, at the end of the β₂-β₃ loop in SXL RRM2, undergoes one of the largest changes in NMR chemical shift upon binding the r(GU₈C) sequence. Consistent with a central role for this residue in RNA recognition, the Arg-252–Ala mutation in SXL decreases RNA affinity more than 20,000-fold.

Available biochemical evidence indicates that analogous residues in SXL and U1A make RNA contacts (38, 39). The crystal structures of RNA complexes of UlA (37) and U2B′′/U2A′ (10) reveal similar modes of sequence-specific binding of RRM motifs to single-stranded segments of RNA. In the structures of UlA and U2B′′/U2A′ bound to RNA, residues in the conserved RNP-1 and RNP-2 sequences stack with RNA bases. The nonconserved β₂-β₃ loop, which makes extensive RNA contacts, has been implicated in the sequence specificity of RRMs. Swapping the β₂-β₃ loops between the conserved UlA and U2B′′ proteins, for example, switches their RNA binding specificities (40).

Because similar regions of SXL, UlA and U2B′′ make important contributions to RNA binding, the RRMs of UlA and U2B′′ were superimposed on the SXL RRM domains to define possible positions of single-stranded RNA (Fig. 3). As expected, many residues in the RNP-1 and RNP-2 motifs are positioned to make stacking or base-specific hydrogen bonding interactions with RNA. Assuming that each RRM interacts with the same bases as observed in the U1A and U2B′′/U2A′ RNA complexes, the minimal, 10-base RNA binding sequence could not be modeled to bridge the two SXL RRMs. This result, in agreement with solution studies, is consistent with a rearrangement of the SXL RRMs upon RNA binding.

Figure 3.

Model of RNA binding sites on SXL RRMl (Left) and RRM2 (Right). The RRMs (red) are displayed in the orientation observed in the crystal structure, and the RNA positions (blue) were modeled after the single-stranded RNA segment of the U1A/RNA complex superimposed onto each SXL RRM in turn. The six bases of single-stranded RNA in the UlA complex were changed to U to better match the SXL binding site, r(GU₈C), and minor adjustments were made to improve the fit of the RNA on the surfaces of the RRMs. Both SXL RRMs are thought to simultaneously make contacts with a single r(GU₈C) site, and this model demonstrates that a single decamer cannot span the conserved RNA binding surfaces of both RRMs without a large change in domain orientation.

In summary, SXL RRM1 and RRM2 form two discreet RNA binding surfaces that are too far apart in the crystal structure to interact simultaneously with the high-affinity, r(GU₈C) RNA. These results indicate that the interdomain linker acts as flexible tether, and orientation of the RRM domains is coupled to RNA binding. A specific, preorganized architecture of the RRMs does not contribute to the target site specificity of SXL in isolation. A similar coupling mechanism has been reported for a fragment of Xenopus laevis transcription factor IIIA (TFIIIA) containing three DNA-binding, zinc finger domains (41). In the absence of DNA, the zinc fingers are randomly oriented in solution and the interdomain linkers are flexible. Nucleic acid binding is coupled to ordering of the linkers and stabilization of interdomain contacts. Thus, orientation of flexibly tethered protein domains upon nucleic acid binding may be a general feature of DNA- and RNA-protein interactions.

ABBREVIATIONS

SXL

Sex-lethal

RRM

RNA recognition motif

hnRNP

heterogeneous nuclear ribonucleoprotein

RRM1+2

amino acids 112–294 of Sex-lethal

RNP

ribonucleoprotein

SeMet

selenomethionine