Crystal structure of human Acinus RNA recognition motif domain

Cloning, expression and purification of Acinus RRM

The protein fragment encoding the AcRRM domain was selected based on the secondary structure prediction of Acinus protein performed with the GeneSilico Metaserver (Kurowski & Bujnicki, 2003). The human Acinus cDNA (ID 9021673) was purchased from IMAGE Consortium Source BioScience. The DNA sequence encoding AcRRM (residues 1008–1100, Uniprot sequence ID Q9UKV3) was PCR amplified and subcloned into pGEX-4T1 (GE Healthcare Life Sciences, Little Chalfont, UK). E. coli BL21(DE3) strain (New England BioLabs, Ipswich, MA, USA) was used to overexpress the glutathione S-transferase (GST) tagged AcRRM. Expression was carried out in LB medium, induced with 1 mM isopropyl-D-1-thiogalactopyranoside solution (IPTG) at OD₆₀₀ of 0.6 and conducted at 37 °C with shaking at 200 rpm for 4 h. The cells were harvested by centrifugation at 4,000 × g, at 4 °C for 20 min. The pellet was resuspended in 1× PBS buffer, pH 7.9, supplemented with 1 mM phenylmethylsulfonyl fluoride (PMSF) (Sigma-Aldrich, St. Louis, Missouri, USA). The cells were subsequently lysed by the French press at 18,000 psi and the cell debris was removed by centrifugation at 20,000 × g for 30 min. The GST-AcRRM fusion protein was purified by GST affinity with Glutathione-Agarose beads (Sigma-Aldrich, St. Louis, Missouri, USA) according to the manufacturer protocol (Fig. S1A). The affinity-purified recombinant protein contained the thrombin recognition site and was subsequently cleaved with thrombin from bovine plasma (Sigma-Aldrich, St. Louis, Missouri, USA) to remove the GST tag (100 μg lyophilized thrombin was used to cleave 1 mg of GST-AcRRM fusion). The cleavage was performed for 2–4 h at room temperature in buffer containing 50 mM Tris–HCl pH 9, and 50 mM NaCl. After cleavage, AcRRM maintained two additional amino acid residues (GS) at its N-terminus. AcRRM was separated from GST tag by heparin affinity using a HiTrap Heparin HP column (GE Healthcare Life Sciences, Uppsala, Sweden) (Fig. S1B), pre-equilibrated in buffer containing 50 mM Hepes-NaOH pH 8 and 1 mM DTT. The bound protein was eluted with gradient of 0 to 1 M NaCl in the 50 mM Hepes-NaOH pH 8 and 1 mM DTT buffer.

Finally, AcRRM was purified by size exclusion chromatography on a Superdex 75 pg column (GE Healthcare Life Sciences, Little Chalfont, UK) pre-equilibrated in 50 mM Hepes-NaOH pH 8, 100 mM NaCl, 1 mM DTT buffer (Fig. S1C). The purified AcRRM protein was concentrated to 8 mg/ml using a 3,000 molecular-weight cut-off Amicon Ultra-4 Centrifugal Filter Units (Merck Millipore, Burlington, MA, USA).

Crystallization

Human AcRRM was crystallized at 18 °C using 0.2 + 0.2 μl sitting drop vapor diffusion method. The Phoenix nano-dispensing robot (Art Robbins instruments, Sunnyvale, CA, USA) was used to set the crystallization drops with the Morpheus Screen (Molecular Dimensions), Index and Crystal Screen (Hampton Research, Aliso Viejo, CA, USA) in 96 well crystallization plates (Hampton Research, Aliso Viejo, CA, USA). The first needle-like crystals appeared within two days. The crystals collected from Index Screen 81 (G9) crystallization reagent (0.2 M Ammonium acetate, 100 mM Tris–HCl pH 8.5, 25% w/v Polyethylene glycol 3,350) were cryo-protected for 10 s in reservoir solution supplemented with 25% w/v Polyethylene glycol 400, flash-frozen, stored in liquid nitrogen and used for X-ray data collection.

X-ray data collection and structure determination

Native X-ray diffraction data sets extending to 1.65 Å resolution were collected using the CCD detector at beamline 14.2 at the Helmholtz-Zentrum Berlin (BESSY II, Germany) (Gerlach, Mueller & Weiss, 2016). A data set consisting of 120 frames was collected with an oscillation width of 1° per frame and a crystal-to-detector distance of 150 mm. The data were indexed and scaled using the XDS software package (Kabsch, 2010).

Primary molecular replacement searches were performed with the MrBUMP/Phaser pipeline (McCoy et al., 2007; Keegan & Winn, 2008) using RRM domain of Rna15 (PDB code 2X1F (Pancevac et al., 2010)) as a search model. It produced a marginal solution with two monomers in the asymmetric unit, a Z score of 11.5 and a final log-likelihood gain (LLG) of 104. The model was then automatically rebuilt with ARP/wARP (Langer et al., 2008) excluding some resolution shells that were marked as ice rings. Re-inspection of diffraction images and the Wilson plot revealed two problematic resolution shells (2.29–2.22 and 1.94–1.90 Å). They were then excluded from a second iteration of indexing and scaling. Statistics for the diffraction data collection and (re)processing are presented in Table 1. Iterations of manual rebuilding, using the Coot program (Emsley & Cowtan, 2004), and restrained refinement using CCP4 REFMAC5 (Murshudov et al., 2011) that included TLS optimization with one group per monomer, produced the AcRRM model characterized by crystallographic R factor of 18.5% and R_free of 22.4%. The data processing and refinement statistics are summarized in Table 1. Final model coordinates and the corresponding structure factors were deposited at Protein Data Bank with the 6G6S accession code.

Overall structure

Acinus RRM domain exhibits the α/β sandwich fold (Figs. 2A and 2B). The core βαββαβ is followed by an additional C-terminal α-helix not present in the canonical RRM domains. The two molecules in the asymmetric unit of the crystal structure are very similar (rmsd of 0.25 Å for 91 Cα atoms), despite the fact that no non-crystallographic symmetry (NCS) restraints were used during the refinement. The N- and C-termini display good electron density for all residues of molecule A, but are not fully ordered in molecule B. AcRRM has three prolines within its sequence and two of them, P1022 and P1075, are in the cis-conformation in both molecules. The two AcRRM monomers are covalently bound in the crystal structure via a disulfide bond (Fig. 2B), that generates an interface area of approximately 500 Å for each of them (calculated by the PISA server (Krissinel, 2015)). For the two linked AcRRM monomers present in the crystal, the Complexation Significance Score (CSS) calculated by PISA is 1. However, computational modeling of the disulfide bond removal by replacement of C1083 with alanine, results in the CSS of 0. This indicates that in the reducing conditions the dimerization does not take place and the dimeric assembly is likely to be a crystal-packing artifact. The monomeric state of AcRRM in reducing conditions was confirmed by gel filtration (Fig. S2). C1083 is conserved in AcRRMs of human, mouse and fly, but in other RRM domains this position is generally not conserved (Fig. S3).

RNA recognition motif domains typically bind RNA via the aromatic residues of the RNP1 and RNP2 motifs located in the two central β-strands. These residues are also present in AcRRM. RNP1 comprises residues 1049–1056 [KSHCFVTY], with the deviations found at second, third and fourth position of the consensus sequence (underlined). RNP2 comprises residues 1014–1019 [VHISNL] with deviations at the second position of the consensus (Fig. 2A). The aromatic residues of RNPs are exposed at the β-sheet surface. In their close vicinity, there are additional aromatic amino acids (Y1094 and H1095) of the C-terminal α-helix packed against the β-sheet on the opposite side to the α-helices 1 and 2. Furthermore, the C-terminal α-helix contains the DELD caspase-3 target site.

The AcRRM domain has a single conserved region on the surface, which correlates with a part of the protein that is predicted to be positively charged (Fig. 3, Fig. S4). The region covers the RNP1 and RNP2 strands of the β-sheet and extends towards its edge. It correlates well with the RNA binding sites of the most similar RRM domains. Therefore, it seems plausible that Acinus binds its RNA target via this region in a way similar to other RRMs.

Comparison with other RRMs

The AcRRM crystal structure was compared with other RRM structures deposited in the Protein Data Bank using the DALI server (Holm & Rosenstrom, 2010). AcRRM was found to be structurally most similar to RRM of Rna15, a subunit of the 3′-end processing factor from Saccharomyces cerevisiae (PDB code: 2X1B (Pancevac et al., 2010)). The βαββαβ cores of both structures align with an rmsd of 1.6 Å (sequence identity of 17%) (Fig. 4A). The RRM of U1A (U1 small nuclear ribonucleoprotein A) was found to be the second closest structural relative of AcRRM (PDB: 5FJ4 (Huang, Wang & Lilley, 2016)). It harbors the C-terminal α-helix and aligns with an rmsd of 2.1 Å (sequence identity of 17%) (Fig. 4B).

There are two sequence insertions in AcRRM compared to the most similar known RRM structures (Figs. 4 and 5A): the first insertion (residues 1040 and 1041) distorts the following β-strand making it shorter and increases the size of the loop-2 preceding β2. The second insertion (residues 1076 to 1078) is located in loop-5, making it larger than the corresponding loop in both other structures (RRMs of Rna15 and U1A) (Figs. 4 and 5A). In contrast, loop-3 is substantially shorter than in the similar structures and lacks the DXXT motif implied in the binding of uridine rich RNAs (Muto & Yokoyama, 2012). While loop-2 is located on the side of the RRM domain that usually does not participate in the RNA binding, the length of loops-3 and 5 might be functionally relevant.

Comparison with other RRM structures in complex with RNA

Focusing on the top 100 structures defined by Dali server (Holm & Rosenstrom, 2010) as most similar to AcRRM (Fig. 5B), 31 were RRMs in complex with RNA. They can be divided into two major groups: RRMs that bind single-stranded RNA and RRMs that bind an RNA hairpin.

The first group includes, among others, RRMs of splicing factors: (i) Tra2-β in complex with the (GAA)₂ sequence (PDB code 2RRA, (Tsuda et al., 2011) Fig. 5C), where the AGAA tetra-nucleotide is recognized through the N- and C-terminal extensions, as well as stacking interactions on the β-sheet surface; (ii) Sex-lethal protein (Sxl) which binds to the 5′-UGUUUUUUU-3′ via a V-shaped, strongly electropositive cleft formed by β-sheets of two consecutive RRM domains (PDB code 1B7F, (Handa et al., 1999) Fig. 5D); (iii) HuD (ELAVL4) which binds (AU)-rich elements (AREs) in the 3′ untranslated regions of many short-lived mRNAs via RNPs and C-terminus (PDB codes 1G2E and 1FX1 (Wang & Tanaka Hall, 2001)). The nucleotide sequences of all these ligands align poorly in the analyzed structures, but in all of them the RNA is placed in proximity of RNP1 (contacts are made with U, G and A residues) and RNP2 (contacts are made with U, G and C residues).

The second group comprises multiple structures of U1A RRM in complex with the same RNA loop sequence (AUUGCACUCC) (Fig. 5E) and one structure of the Sxl RRM in complex with UUUGAGCACGUGA (PDB code 4QQB (Hennig et al., 2014)) (nucleotides that are in contact with RNPs are underlined).

Putative RNA binding mode of AcRRM

RNA recognition motif domains are known for having diverse RNA binding modes (Maris, Dominguez & Allain, 2005; Clery, Blatter & Allain, 2008; Muto & Yokoyama, 2012; Daubner, Clery & Allain, 2013) that make the precise prediction of the Acinus interaction with the ligand very difficult. However, there are a few features that are likely to be conserved. In many RRMs, including the ones most similar to AcRRM, the RNA is bound at the face of the β-sheet with its 5′-end on the side of the β4 and 3′-end on the side of the β1-strand (Fig. 5). It seems likely that Acinus binds the RNA in analogous way. The RRMs usually use aromatic residues in position 2 of RNP2 (Fig. 5E) and 5 of RNP1 (Figs. 5C–5E) for the stacking interactions with the RNA bases. H1015 and F1053 are present in these positions in AcRRM and are likely to fulfill the same roles (Fig. 5B). The aromatic residue in position 3 of RNP1 is often stacked against the ribose groups of the bound nucleic acid (Fig. 5D). This position is occupied by a H1051 in AcRRM that might have the same function (Fig. 5B). The first amino acid of RNP1 is often positively charged and sometimes interacts with the phosphate backbone. K1049 in AcRRM might play this role, but homologous residues of the similar RRMs are not involved in the interactions with the RNA backbone. While RNP1 and RNP2 mediated stacking interactions with the bases provide the RNA binding affinity, the sequence specificity determinants of RRMs are often located in surrounding secondary structure elements. The β4-strand and the flanking regions provide part of these interactions in RRMs similar to AcRRM (indicated with blue color in Fig. 5), but the protein-RNA hydrogen bonding patterns are too variable for precise predictions (Fig. S5).

Some RRM domains can accommodate a spectrum of RNA sequences, and the actual contribution of RNPs to ligand discrimination may not always be exclusive (Clery, Blatter & Allain, 2008; Daubner, Clery & Allain, 2013). The determinants of sequence specific RNA recognition may be located in other structural elements of RRMs. For instance, RRM2 of Sxl (Handa et al., 1999) and RRM1 of PABP (Deo et al., 1999) interact with RNA via loops, in addition to RNPs. In extreme cases, the RRM domains do not bind RNA through the β-sheet at all but they do so through other structural elements, for instance qRRMs use loops-1, 3 and 5 (Dominguez et al., 2010). The comparison of AcRRM with similar RRM structures suggests that analogous elements might play a role in RNA binding (Fig. 5). These include the above-mentioned β4 strand and its flanking regions (residues 1080–1090) that may provide sequence specific interactions. Alternatively, loop-1 (residues 1017–1025) and/or loop-3 (residues 1046–1051) that are immediately adjacent to the RNPs and characterized by high sequence conservation may be involved in RNA binding by AcRRM (Figs. 3B and 5B).