Apo raver1 structure reveals distinct RRM domain orientations

The crystal structure of unbound raver1 RRM1–3 domains

To further define the function of raver1, we solved the crystal structure of apo raver1 RRM1–3 domains at 2 Å resolution [Fig. 1(A)]. As seen in our Vt-bound complex structure,²³ the three RRM domains exhibit the canonical RRM fold, with a four-stranded β-sheet and two α-helices. An additional β-hairpin is seen between α-helix α2 and β-strand β4 in the RRM1 and RRM2 domains. Further, α-helices that are amino-terminal to the RRM1 domain (residues 39–58) and those that are carboxy-terminal to the RRM3 domain (residues 304–321) were essential to obtain crystals (J.H.L., personal communication), suggesting roles in restraining interdomain movements.

We identified two electron-dense features within the RRM1 domain of the unbound RRM1–3 domains structure that were modeled as sulfate ions, given the high concentration of lithium sulfate in our crystallization conditions. Among the two sulfate ions, the binding interaction for one was near the α1–β2 loop [Fig. 1(A)] and involves contributions from the side-chain amino groups of Arg121 located on β-strand β4 and the main-chain nitrogen of Gly68 located on the α1–β2 loop. Notably, Arg121 contributes to electrostatic interactions with vinculin residues Glu932 and Asp953 in the vinculin-bound state.²³ Thus, this sulfate-binding site coincides with the vinculin binding site. Notably, a sulfate ion was also found in the RRM domain of human nuclear cyclophilin 33 (PDB entry 3lpy) at an equivalent position.

A second sulfate anion located near the RRM1–RRM2 interface is directly coordinated by RRM1 residues Arg59 and Lys60 located on the linker region that connects the N-terminal α-helix and β-strand β1 and by Lys85 on β-strand β2, by water-mediated interactions, via RRM1 residues Tyr86 on β-strand β2 and Thr99 on β-strand β3, and by RRM2 residue Glu159 on β-strand β2 [Fig. 1(B)]. By contrast, raver1 Arg59 in the Vt:RRM1–3 structure engages in electrostatic interactions with Glu159, which are disrupted by the sulfate ion.

Superposition of the raver1 RRM1 domain onto the second RRM domain of the sxl-lethal protein:RNA complex (PDB entry 1b7f) places our second sulfate ion within ∼5 Å, of the nearest phosphate of the RNA that is bound to sxl-lethal [Fig. 1(C)]. This suggests that this sulfate ion occupies a position corresponding to the RNA binding site of RRM1. Indeed, inspection of the electrostatic surface potential shows a positive surface area surrounding the sulfate ion (Supporting Information Fig. S1). Moreover, in addition to the linker region (residues 208–219) that connects the RRM2 and RRM3 domains, the RRM2 and RRM3 β-sheet surfaces usually involved in RNA binding are occupied by interactions with the N- and C-terminal α-helices^23,24 and, accordingly, sulfate ions are not bound to the RRM2 or RRM3 domains [Fig. 1(A)].

A C-terminal hydrophobic α-helix interacts with the RRM2 and RRM3 domains

The individual RRM domains of the Vt-bound and apo raver1 structures are very similar, with r.m.s.d. of 0.67 Å (for 497 RRM1 atoms), about 0.5 Å (for 540 RRM2 atoms), or 0.6 Å (for ∼500 RRM3 atoms), respectively. However, when superimposing tandem RRM domains, the two RRM2–RRM3 domains in the Vt:raver1 asymmetric unit superimpose better onto each other (r.m.s.d. of 0.34 Å for 1301 atoms) [Fig. 1(D)] versus their superposition onto the apo RRM2–RRM3 structure (r.m.s.d. of 0.6 Å–0.7 Å for 1223 atoms). Indeed, residues 210–212 that connect RRM2 with RRM3 form a well-defined 3₁₀-helix in the apo structure while this region has a random coil structure in the Vt:RRM1–3 complex [Supporting Information Fig. S2(A)].²³ Further, the C-terminal raver1 α-helix is about half a turn shorter in the apo structure, where it ends at Gln314 versus at Ala317 in its vinculin-bound state.

As seen in the Vt:RRM1–3 structure,²³ the C-terminal hydrophobic α-helix (residues 304–316) is wedged into a hydrophobic cleft between the RRM2 and RRM3 domains [Fig. 1(A)]. The interdomain hydrophobic residues of the RRM2 and RRM3 domains and the C-terminal α-helix are strictly conserved among the raver1 protein family (human, mouse, rat, and bovine) and deletion of this α-helix rendered the protein insoluble (J.H.L., personal communication), suggesting that this α-helix is important for proper stabilization and/or folding of the RRM2 and RRM3 domains. Indeed, the C-terminal α-helix buries over 1170 Ų of its solvent-accessible surface area on its interaction with the RRM2 and RRM3 domains.

A novel RRM1 domain orientation

The most pronounced difference between the RRM1–3 domains alone or in complex with Vt is the relative orientations of their respective RRM1 domains [Fig. 1(D)]. Indeed, analysis of the apo versus Vt-bound raver1 RRM1–3 structures by the DynDom server²⁵ showed a 35° rotation of the RRM1 domain (residues 59–128) relative to the RRM2–RRM3 domain [residues 129–314; Fig. 1(D)]. The hinge region involves the Arg58-Arg59 and Leu127-Gln128-Pro129 psi and phi dihedral angles. In particular, the phi dihedral angle for the Arg58-Arg59 peptide bond differed by about 33° with a psi dihedral change of about 25° for Arg59 and corresponding consequences on RRM1–RRM2 interdomain interactions [Fig. 1(E)]. In the Vt:RRM1–3 complex, RRM2 residue Glu159 likely hinders the relative movement of RRM1 through its electrostatic interaction with Arg59 [Fig. 1(B)]. However, in the unbound RRM1–3 structure, the sulfate ion disrupts this interaction by directly binding to Arg59 and to Glu159 by water-mediated interactions. Thus, the phosphate group of RNA when bound to raver1 might cause similar RRM1–RRM2 interdomain movements. In contrast, as the vinculin binding interface resides over 20 Å distant from this hinge, vinculin binding does not appear to affect the orientation of the RRM1 domain [Fig. 1(D,E)].

The change in RRM1 orientation is accompanied by other local side-chain movements of vinculin–raver1 interface residues as seen for Tyr92 that binds to Vt residues Phe885, Pro886, Met899, and Arg935, where this residue is rotated by about 128° in the unbound raver1 structure [Fig. 1(D,E)]. Similarly, residues Tyr86 and Phe88 that are situated on β-strand β2 of the RRM1 domain are rotated by about 80° and 121°, respectively. Finally, the N-terminal α-helix engages in extensive interactions with the RRM2 domain [Fig. 1(A)] and might restrict the RRM1 domain flexibility. In support of this notion, recombinant RRM1–3 proteins lacking this amino-terminal α-helix would not crystallize (J.H.L., personal communication).

Comparison with the Prp24 three-tandem RRM domain structure

The only other three-tandem RRM structure reported is that of the splicing factor Prp24 at 2.7 Å resolution.²⁶ The individual RRM domains of raver1 and Prp24 readily superimpose [Supporting Information Fig. S2(B,C)] even though superimposition of the RRM domains in our 2 Å raver1 structure is superior, with r.m.s.d. ranging from 1.2 Å to 2.3 Å for 113–381 atoms for the superposition of RRM1 onto RRM2 domains, of RRM1 onto RRM3, or of RRM2 onto RRM3 domains, respectively. However, the individual orientations of the three RRM domains of raver1 and Prp24 are distinct [Fig. 1(F)]. While the RRM2 domain in Prp24 forms extensive interdomain contacts with RRM1 and RRM3, they only bury a small solvent-accessible surface area (∼550 Ų), suggesting their intramolecular interactions might be dynamic.²⁶ In contrast, the interdomain contacts of the raver1 RRM2 domain (residues 133–206) differ and the buried surface areas in the RRM1 (residues 60–127) versus RRM3 (residues 222–296) interfaces are about 400 Ų and almost 1700 Ų, respectively.

RRM1–3 crystallization and structure determination

The RRM1-3 domains (residues 39–321) were purified as described.²³ The crystals of the RRM1–3 were grown at 20 °C by sitting-drop vapor diffusion from protein concentrated to 18.2 mg/mL against solution of 100 mM Tris–HCl buffer (pH 8.5), 24% polyethylene glycol 3350, and 200 mM lithium sulfate. Flash freezing was achieved by transferring the crystals directly to a reservoir solution containing 15% glycerol, and they were allowed to equilibrate for 2 min. Native X-ray diffraction data were collected at beamline 22ID of the Advanced Photon Source (SER-CAT) to 2 Å Bragg spacings and integrated and scaled using autoProc²⁷ that uses XDS²⁸ as the data-processing engine. The hexagonal raver1 RRM1–3 domain crystals belong to space group P6₅ with unit cell dimensions a = b = 104 Å, c = 50.4 Å, α = β = 90°, and γ = 120°. There is one RRM1–3 molecule in the asymmetric unit resulting in a solvent content of 51.1% (V_M = 2.51 ų/Da; Table I).

The RRM1–3 structure was determined by molecular replacement using the program MOLREP.²⁹ Attempts to determine the structure using the Vt-bound RRM1–3 structure as a search model were unsuccessful. The refined RRM2–RRM3 domains of our Vt:RRM1–3 complex structure²³ (PDB entry 3h2u) were used as the search model. Subsequently, more than 65% of the model was built automatically with ARP/wARP³⁰ and REFMAC.³¹ The remaining model was then manually built into F_obs − F_calc electron density maps using Coot³² in several iterative rounds. The RRM2–RRM3 model was fitted, which allowed the determination of the RRM1 structure by molecular replacement using the RRM1 model of the Vt:RRM1–3 structure²³ (PDB entry 3h2u). Crystallographic refinement was performed using autoBUSTER³³ applying translation, libration, and screw-motion to account for the anisotropic nature of the RRM3 domain and resulted in a final crystallographic R value of 0.19 (free R value = 0.22). The final refined model has an overall all-atom clash score of 3.93 as determined by MolProbity³⁴ with a score of 1.63 and all residues were in favorable regions of the Ramachandran Plot. In the RRM1-3 structure, Cys239 and Cys251 are involved in disulfide bond formation and regions corresponding to amino acids residues 256–258 and 288–290 showed weak main chain electron density. The final crystallographic refinement statistics are shown in Table I.