Structure–function studies of STAR family Quaking proteins bound to their in vivo RNA target sites

Mammalian Quaking (QKI) and its C. elegans homolog, GLD-1, are conserved STAR family RNA-binding proteins that post-transcriptionally regulate target genes essential for development and myelination. Here, Teplova et al. solve the crystal structure of QKI and GLD-1 STAR domains in complex with in vivo RNA targets. This study defines the principles underlying RNA target selection by STAR homodimers and provides key insights into the post-transcriptional regulatory function of mammalian QKI proteins.

Abstract

Mammalian Quaking (QKI) and its Caenorhabditis elegans homolog, GLD-1 (defective in germ line development), are evolutionarily conserved RNA-binding proteins, which post-transcriptionally regulate target genes essential for developmental processes and myelination. We present X-ray structures of the STAR (signal transduction and activation of RNA) domain, composed of Qua1, K homology (KH), and Qua2 motifs of QKI and GLD-1 bound to high-affinity in vivo RNA targets containing YUAAY RNA recognition elements (RREs). The KH and Qua2 motifs of the STAR domain synergize to specifically interact with bases and sugar-phosphate backbones of the bound RRE. Qua1-mediated homodimerization generates a scaffold that enables concurrent recognition of two RREs, thereby plausibly targeting tandem RREs present in many QKI-targeted transcripts. Structure-guided mutations reduced QKI RNA-binding affinity in vitro and in vivo, and expression of QKI mutants in human embryonic kidney cells (HEK293) significantly decreased the abundance of QKI target mRNAs. Overall, our studies define principles underlying RNA target selection by STAR homodimers and provide insights into the post-transcriptional regulatory function of mammalian QKI proteins.

Mammalian Quaking (QKI) and the QKI-related (QR) proteins GLD-1 (defective in germ line development) in Caenorhabditis elegans and How (held out wing) in Drosophila melanogaster are conserved RNA-binding proteins (RBPs) of the STAR (signal transduction and activation of RNA) family (Vernet and Artzt 1997). QR proteins post-transcriptionally regulate target gene expression and play essential roles in developmental processes in mice (Sidman et al. 1964; Ebersole et al. 1996), worms (Francis et al. 1995; Jones et al. 1996), and flies (Nabel-Rosen et al. 1999; Volk et al. 2008) and also function as tumor suppressors (Biedermann et al. 2010).

In mice, there are three main Qki splice isoforms (QKI-5, QKI-6, and QKI-7), and reduced expression of QKI-6 and QKI-7 isoforms leads to defective maturation of myelinating cells of the CNS (Zearfoss et al. 2008), while Qki-null mutations cause neural tube and vascular defects and are embryonic-lethal (Li et al. 2003). Low expression levels of human QKI in the CNS have been linked to the development of diseases such as ataxia and schizophrenia as well as the formation of glioblastomas through only partially defined mechanisms (Chenard and Richard 2008). The C. elegans and D. melanogaster homologs of QKI, GLD-1, and How also play a key developmental role and are required for multiple aspects of germline development in worms (Lee and Schedl 2010) and control muscle and tendon differentiation as well as glial cell maturation in flies (Volk 2010).

QKI proteins regulate the stability, export, and alternative splicing of multiple mRNAs associated with the formation of myelin (Larocque et al. 2002, 2005; Wu et al. 2002). Thus, mRNAs of the myelin basic protein (MBP) (Larocque et al. 2002), the oligodendrocyte cell differentiation factor CDKN1B/p27(Kip1) (Larocque et al. 2005), and the alternative splicing regulator HNRNPA1 (Zearfoss et al. 2011) in humans as well as the muscle-specific transcription factor Gli2a in zebrafish (Lobbardi et al. 2011) are stabilized as a result of specific interactions of QKI with specific sequence elements in their 3′ untranslated regions (UTRs). While QR proteins show high evolutionary conservation on the sequence level, their regulatory mechanisms are very different, as, in the case of GLD-1, binding of target mRNAs leads to translational repression in vitro and in vivo rather than mRNA stabilization (Lee and Schedl 2010).

Based on in vitro characterization of QKI interactions with synthetic RNAs derived from the 3′ UTR of MBP mRNA as well as in vitro evolution approaches (SELEX), the QKI RNA recognition element (RRE) was first defined as ACUAAY (where Y is pyrimidine) (Ryder and Williamson 2004; Galarneau and Richard 2005). A similar 6-nucleotide (nt) consensus sequence, UACU(C/A)A, was identified by a comprehensive mutational analysis of GLD-1-binding sites in the 3′ UTR of the tra-2 mRNA (Ryder et al. 2004). In a transcriptome-wide analysis of GLD-1 interactions by RNA immunoprecipitation (RIP) followed by microarray analysis (RIP-chip), 20 of the 35 most enriched 7-nt sequences contained a YUAAY core motif (Wright et al. 2011). While a single RRE was sufficient for GLD-1-dependent regulation, multiple RREs within the 3′ UTR increased the effect. Using PAR-CLIP (photoactivatable ribonucleoside-enhanced cross-linking and immunoprecipitation) in human embryonic kidney cells (HEK293) and C. elegans, our group and others expanded the QKI and GLD-1 target list and found similar RREs (Hafner et al. 2010; Jungkamp et al. 2011). In HEK293 cells, we found that >90% of the 2500 cross-linked binding sites present in 1500 transcripts contained at least one 5-nt YUAAY RRE. Interestingly, ∼50% of the, on average, 30-nt-long sites contained two or more RREs separated by a linker sequence of variable length.

All QR proteins bind RNA via their STAR domain, composed of a single heterogeneous nuclear ribonucleoprotein K homology (KH) domain flanked by two conserved Qua1 and Qua2 domains (Figs. 1, 2A,D; Vernet and Artzt 1997). Qua1 has been shown to be critical for homodimerization (Chen et al. 1997; Chen and Richard 1998), while KH–Qua2 is critical for RNA binding (Ryder et al. 2004). Even though the regulatory consequences of target RNA binding for worm GLD-1 and mammalian QKI are very different, they bind identical RREs due to their overall high protein sequence similarity and 57% identity of their STAR domains (Fig. 1).

Figure 1.

Sequence alignment of the STAR domain family members. Sequence alignment between STAR domains of C. elegans GLD-1, Homo sapiens QKI, D. melanogaster How, and H. sapiens SAM68 and SF1 proteins. Secondary structure alignments of GLD-1 STAR and the SF-1 KH–Qua1 region are shown above (in color) and below (in gray) the sequences, respectively. Secondary structure elements of the GLD-1 Qua1, KH, and Qua2 domains are colored cyan, gold, and blue, respectively. Numbering above the sequences corresponds to C. elegans GLD-1. Black asterisks and triangles denote residues that form hydrogen bonds with RNA bases or sugar-phosphate backbone, respectively. Reverse triangles color-coded for the KH and Qua2 domains indicate residues involved in base stacking and hydrophobic interactions with RNA nucleotides. Dimer interface residues of the Qua1 domain that form intermolecular hydrogen bonds (cyan asterisks) and the hydrophobic patch (cyan triangles) are marked. The positions of the QKI and the GLD-1 spontaneous and N-ethyl-N-nitrosourea (ENU)-induced point mutations are indicated with arrowheads.

Figure 2.

Crystal structures of the QKI STAR–RNA and GLD-1 STAR–RNA complexes. (A) Domain architecture of the QKI STAR construct (seven to 214) consisting of the Qua1 domain (in cyan), KH domain (in gold), and Qua2 domain (in blue) used for crystallization of the complex. The RNA target is 5′-UUCACUAACAA-3′. (B) Crystal structure of the QKI STAR homodimer bound to two RNA molecules. The KH (gold) and Qua2 (blue) domains interact with the RNA (green), while the Qua1 domain (cyan) interacts with the Qua1 domain of the second STAR molecule in the crystallographic asymmetric unit. The linkers between the Qua1 and KH and between the KH and Qua2 domains are colored in red. (C) One monomer of the QKI STAR–RNA complex in B, with labeled RNA bases and STAR α helices and β strands. The loops in the KH domain that contact the RNA are colored in pink. (D) Domain architecture of the GLD-1 STAR construct (143–337) used for crystallization of the complex. The RNA target is 5′-p-CUAACAA-3′. (E) Crystal structure of the GLD-1 STAR homodimer bound to two RNA molecules. The color-coding is similar to B. (F) One monomer of the GLD-1 STAR–RNA complex from E, with labeled RNA nucleotides and STAR α helices and β strands.

To date, the structure of the entire STAR domain with or without bound RNA has not been solved, although several partial structures provide insights into STAR protein dimerization and RNA recognition. The NMR solution structure of Xenopus laevis KH–Qua2 without RNA ligands identified a well-defined KH domain structure in an undefined relative orientation to a highly dynamic Qua2 α-helical region connected with a flexible linker (Maguire et al. 2005). In contrast, the nuclear magnetic resonance (NMR) solution structure of the KH–Qua2 domain of splicing factor 1 (SF1) protein in complex with ssRNA (Liu et al. 2001) established that both KH and Qua2 domains contribute to sequence-specific RNA recognition and form an extended interface with bound RNA. The structures of Qua1 modules of GLD-1 (Beuck et al. 2010), Sam68 (Meyer et al. 2010), and QKI (Beuck et al. 2012) showed similar helix–turn–helix folds, which are aligned roughly perpendicular to each other to form a symmetric dimer, stabilized mainly by hydrophobic interactions.

Since many of the QKI and GLD-1 RNA targets contain two RREs in close proximity, it was hypothesized (Beuck et al. 2010) that these RBPs bind their targets as homodimers, with the two Qua1 modules mediating the homodimerization, and the KH–Qua2 interacting with the RNA targets. Therefore, molecular insights into the complete assembly of the Qua1, KH, and Qua2 modules are necessary to understand QKI recognition and regulation of RNA targets. To gain further insights into the molecular mechanism of RNA recognition by QR proteins, we solved the first crystal structures of intact GLD-1 and QKI STAR domains in complex with in vivo RNA targets. The structures of the complexes reveal an unanticipated relative arrangement of the Qua1, KH, and Qua2 domains within each subunit of a symmetric STAR homodimer, in which the canonical KH–Qua2 RNA-binding surfaces are positioned at opposite ends of the elongated complex, and dimerization is mediated exclusively by Qua1 domain. Such a symmetrical homodimer is capable of recognizing two RREs present on either separate RNA molecules or a single RNA chain and separated by a linker >10 nt. Expression of double mutants of QKI with reduced RNA-binding affinity in HEK293 cells followed by RNA sequencing demonstrated a stabilizing effect of QKI on RNA targets identified by PAR-CLIP, with the magnitude of stabilization depending on the number of binding sites.

Results

Crystal structures of QKI and GLD-1 STAR–RNA complexes

We first solved the 3.5 Å crystal structure of the QKI STAR domain (residues 7–214) bound to 5′-UUCACUAACAA-3′ (Fig. 2A), derived from the HOXD13 transcript characterized previously (Hafner et al. 2010) and containing one QKI RRE (underlined). The crystals belong to the P6₁22 space group and contain one STAR domain and one RNA molecule in the asymmetric unit. The structure of the complex was determined by single-wavelength anomalous dispersion (SAD) phasing on Se atoms using selenomethionine (SeMet)-labeled protein (Supplemental Table S1). The final model of the QKI STAR–RNA complex comprises Qua1 (residues 12–60, excluding disordered segment 32–39 between the two α helices), KH and Qua2 domains (residues 69–204), and 8 nt of the RNA (A4–A11) (Fig. 2B,C). The structure of the complex revealed that the RNA is contacted solely by the KH–Qua2 dual domain, while the Qua1 dimerization domain engages in protein–protein interactions with Qua2 and the symmetry-related Qua1, forming a symmetric dimer in the complex (Fig. 2B,C; Supplemental Fig. S1A).

We also solved the 2.85 Å crystal structure (P3₁21 space group containing two STAR domains and two RNAs in the asymmetric unit) of GLD-1 STAR (143–337) bound to a 7-nt 5′-p-CUAACAA-3′ RNA sequence (Fig. 2D) using molecular replacement (Supplemental Table S1). The entire STAR domain and the RNA could be traced with a high degree of confidence in the structure of the GLD-1 complex in the final refined structure, which showed an overall architecture of the STAR domain dimer and an RNA-binding mode similar to those in the QKI–RNA complex (Fig. 2E,F; Supplemental Figs. S1B, S2). The fully defined linker between Qua1 and KH of GLD-1 adopts an extended conformation alongside the α-helical surface of KH, allowing direct contact of the Qua1 and Qua2 motifs (Fig. 2F).

Homodimerization interface in the GLD-1 STAR–RNA complex

The homodimerization interface in both the GLD-1 and QKI STAR–RNA complexes is mediated solely by the Qua1 domain (Fig. 2B,E), which adopts a similar overall helix–turn–helix fold (core root-mean-square deviation [RMSD] is 0.9 for 38 matching Cα atoms in α1 and α2) in the two complexes (Supplemental Fig. S3A,B). The Qua1 fold is stabilized by conserved hydrophobic “zipper” residues supplemented by a hydrogen bond between a conserved Tyr and a Glu side chain, previously shown to be essential for homodimerization in vivo (Fig. 3A, shown for the GLD-1 complex; Ebersole et al. 1996; Chen and Richard 1998). The Qua1–Qua1 interface buries 1510 Ų of protein surface in the GLD-1 complex and is dominated by van der Waals contacts between conserved hydrophobic and aromatic residues of both Qua1 molecules (Fig. 3B; Supplemental Fig. S4A). The dimer interface is further stabilized by a network of intermolecular hydrogen-bonding interactions formed by conserved residues of each monomer (Fig. 3B; Supplemental Fig. S4B). A detailed description of the homodimerization interface stabilizing the GLD-1 complex is outlined in the Supplemental Material.

Figure 3.

Protein–protein interactions in the GLD-1 STAR–RNA complex. (A) Structure of the Qua1 domain of molecule A in the complex is shown in ribbon representation, with dimer interface residues shown as pink sticks and the monomer “zipper” shown as green sticks. (B) Structure of the Qua1 domains of molecules A and B in the complex are shown in ribbon representation, with dimer interface residues of molecule A shown as pink and molecule B shown as cyan in a stick representation. (C) An overview of the interface between Qua1 and Qua2 in the complex, with side chains of contact residues shown in a stick representation. α Helices of Qua1 and Qua2 are labeled as secondary structure elements of the STAR domain. (D) An overview of the interface between the KH and Qua2 domains in the complex with highlighted side chains of contact residues. Hydrogen bonds are indicated by black dashed lines. (E) Conformation of the Qua1–KH interdomain linker (red) in the complex. The contact residues of the KH domain and the Qua1–KH interdomain linker are highlighted. Hydrogen bonds are indicated by black dashed lines. The colors of the domains are the same as in Figure 2. (F) Variable relative alignment of the Qua1 and KH–Qua2 domains in the QKI STAR–RNA and GLD-1 STAR–RNA complexes. Superposition of the STAR domain components in RNA-bound complexes of GLD-1 (chains A and B in the asymmetric unit) and QKI showing differences in the positioning of the Qua1 domain relative to the KH–Qua2 segment. Orientations of α2 helices of Qua1 in each STAR domain structure are shown by red–green arrows.

Alignment of Qua1 and KH–Qua2 in the GLD-1 STAR–RNA complex

The association of the Qua1 and Qua2 motifs defines the orientation of the KH–Qua2 RNA-binding domain relative to the Qua1 dimerization module in the QKI (Fig. 2B,C) and GLD-1 (Fig. 2E,F) complexes. As viewed in Figure 3C, α7 of Qua2 packs against the α1 and α2 of Qua1 to form a three-helix bundle that is maintained by hydrophobic interactions involving highly conserved residues, a hydrogen bond, and a salt bridge. The Qua1–Qua2 interface-buried surface area is estimated at 970 Ų for molecule A and 1140 Ų for molecule B in the dimeric GLD-1 complex.

The relative alignment of the Qua2 and KH motifs in the complex appears to be driven by RNA binding, given that these two domains were shown not to interact with each other in the absence of RNA (Maguire et al. 2005). The contacts between Qua1 and Qua2 also stabilize the Qua2 orientation relative to the KH domain (Figs. 2C,F, 3D). The KH–Qua2 interactions are mediated by hydrophobic contacts involving residues of the α2-GPRG-α3 segment (219–232) of KH and aliphatic chains of Qua2, with hydrogen bonding interactions further strengthening the interaction (Fig. 3D).

The interdomain linkers facilitate the relative arrangement of the Qua1, KH, and Qua2 domains within each STAR monomer in the complex. A 15-amino-acid-long linker between Qua1 and KH of GLD-1 adopts an extended conformation and contacts α helices 4 and 6 of the KH domain via hydrogen bonding and van der Waals contacts (Figs. 2F, 3E). Residues at the C terminus of the Qua1–KH interdomain linker are more conserved in GLD-1 and QKI, consistent with similar structures for this part of the linker in the two complexes (Fig. 3F), whereas the N-terminal half of the 21-amino-acid QKI linker is disordered in the complex, indicative of a higher conformational flexibility, probably allowing for a larger degree of relative motion of Qua1 and KH in the absence of RNA.

Variable relative alignment of Qua1 and KH–Qua2 in STAR–RNA complexes

Comparison of the three available complexes—namely, the QKI–RNA complex and the two molecules (molA and molB) of the GLD-1–RNA homodimer—revealed a variable relative position of the Qua1 and KH–Qua2 domains within STAR domains in the RNA-bound state. The motion of Qua1 relative to the KH–Qua2 domain that occurs in the three complexes is visualized in Figure 3F, and a pairwise comparison of the three STAR domains in terms of this motion is outlined in Supplemental Figure S5A. Thus, two conformations were observed for the GLD-1 STAR domain that both differ from the conformation observed for QKI (Fig. 3F; Supplemental Figs. S5B–D). A detailed description of the variable relative alignment of Qua1 and KH–Qua2 in GLD-1 and QKI STAR–RNA complexes is outlined in the Supplemental Material.

RNA recognition by KH and Qua2 in the GLD-1 STAR–RNA complex

The RNA-binding KH–Qua2 modules of QKI and GLD-1 are highly similar, with an RMSD value of 0.92 Å for 131 matching Cα atoms of QKI (amino acids 72–202) and GLD-1 (amino acids 195–325) (Fig. 3F). Furthermore, amino acids directly contacting RNA are 100% identical (Fig. 1), clearly explaining their identical binding specificities.

The relative arrangement of the KH and Qua2 modules of the GLD-1 complex creates a hydrophobic surface that accommodates the RNA bases, whereas the sugar-phosphate backbone appears largely exposed to solvent and surrounded by positively charged patches on the surfaces of both KH and Qua2 (Supplemental Fig. S1B). The 5′-p-C1-U2-A3-A4-C5-A6-A7-3′ RNA is bound onto a canonical RNA-binding surface of the KH domain composed of residues from helices α3 and α4, strand β2, the conserved GPRG, and variable loops and extended by helix α7 of Qua2 (Figs. 2F, 4A, stereo view of one monomer in the dimeric complex). The Qua2 domain contacts the 5′-p-C1-U2 fragment (A4 to U6 in the structure of QKI complex), while the KH binds U2 to A7 (U6 to A11 in the structure of QKI complex).

Figure 4.

Protein–RNA intermolecular contacts in the GLD-1 STAR–RNA complex. (A) Stereo view highlighting intermolecular contacts in one monomer of the GLD-1 STAR–RNA complex. The color-coding is as follows: Qua1 in cyan, Qua2 in blue, KH in gold, RNA-interacting loops on KH in red, and RNA in green. Hydrogen bonds are shown as dotted lines. Protein (in ribbon representation) and RNA (in stick representation) are color-coded as in Figure 2, C and F, with the RNA backbone phosphorous and oxygen atoms in yellow and red, respectively. Secondary structure elements of the STAR domain are labeled. The conserved Gly–Pro–Arg–Gly and the variable loops of the KH domain are colored pink. (B) Intermolecular contacts between the 5′-p-C1-U2-A3 RNA segment and the Qua2 and KH domains of STAR in the complex. This view highlights recognition of the C1 and U2 bases by Qua2 residues and the stacking interactions of U2 with the protein backbone and Pro of the GPRG loop as well as hydrogen-bonding contacts involving the U2-A3 sugar-phosphate backbone. (C) Intermolecular contacts between the A3-A4-C5-A6 RNA segment and the KH domain of STAR in the complex. This view highlights the stacking interaction of A3 with the protein backbone and Arg and Val side chains, recognition of the A4 and C5 bases by the main chain and Arg side chain of KH, and A4-C5 backbone recognition by the Lys residues of the KH domain. Protein and RNA contact residues in A–C are shown as sticks colored by atom type (phosphorous in yellow, nitrogen in dark-blue, and oxygen in red), and hydrogen bonds are indicated by black dashed lines. (D) Schematic representation of protein–RNA interactions in the complex generated using the NUCPLOT software. Hydrogen-bonding and hydrophobic/stacking interactions between RNA bases and the sugar-phosphate backbone with amino acid residues of KH (in tan) and Qua2 (in blue) are shown by black and red dashed lines, respectively. Asterisks denote interactions involving protein main chain atoms.

The residues involved in specific protein–RNA contacts in the GLD-1 STAR–RNA (5′-p-CUAACAA-3′) complex that form the molecular basis for specific recognition of the YUAAY RRE are shown in Figure 4, B and C, and schematically in Figure 4D. C1, U2, A4, and C5 are involved in base-specific recognition, residue A3 is positioned within a pocket, and there is continuous stacking within the A4-C5-A6-A7 segment (Fig. 4B–D). Hydrogen-bonding interactions of cytosine bases at the first and the fifth positions with lysine and arginine side chains indicate that uracil substitutions can be tolerated at these positions, consistent with the YUAAY consensus element. A detailed description of the intermolecular contacts stabilizing the GLD-1 complex is outlined in the Supplemental Material.

Impact of QKI mutants on in vitro RNA-binding affinity

We used isothermal titration calorimetry (ITC) to measure RNA-binding affinities of QKI, with mutations of key amino acids within the Qua2 and KH modules contacting specific bases of the QKI RRE (Fig. 5A). Wild-type QKI bound the 11-nt RNA target containing one RRE with a K_d of 0.07 μM (Fig. 5B). Mutations of both Lys190 and Gln193 of QKI (equivalent to GLD-1 Lys313 and Gln316 involved in C1-U2 recognition) (Fig. 4B) to Ala resulted in a drop in binding affinity of two orders of magnitude (K_d = 8.9 μM) (Fig. 5C). Mutations of both Asn97 and Arg130 of QKI (equivalent to GLD-1 Asn220 and Arg253 involved in A3 recognition) (Fig. 5C) to Ala resulted in a 10-fold reduction in binding affinity (K_d = 1.0 μM) (Fig. 5D). Mutations of both Lys120 and Arg124 of QKI (equivalent to GLD-1 Lys243 and Arg247 involved in C5 base and sugar recognition) (Fig. 4C) to Ala resulted in a 20-fold reduction in binding affinity (K_d = 1.9 μM) (Fig. 5E).

Figure 5.

ITC-binding curves of complex formation between the QKI STAR protein and its RNA targets. (A) Sequence of 11-mer RNA (5′-UUCACUAACAA-3′) used for ITC titration studies in B–E. (B–E) ITC-binding curve complex formation of wild-type QKI STAR (B), the QKI STAR K190A/Q193A double mutant, (C) the QKI STAR N97A/R130A double mutant (D), and the QKI STAR K120A/R124A double mutant (E) with the 11-mer RNA target. Solid lines indicate nonlinear least-squares fit to the titration curve, with ΔH (binding enthalpy kcal mol⁻¹), K_a (association constant), and N (number of binding sites per monomer) as variable parameters. Calculated values for K_d (dissociation constant) and N are indicated.

Impact of QKI mutants on in vivo RNA binding

To investigate the impact of the double-mutant QKI proteins characterized above on in vivo binding to RNA, we generated stable HEK293 cell lines inducibly expressing Flag/HA-tagged QKI K120A/R124A, QKI K190A/Q193A, and QKI N97A/R130A as well as wild-type QKI isoform 5. We performed PAR-CLIP after culturing cells in medium containing 4-thiouridine and inducing expression of the tagged protein. Cross-linked RNA–protein complexes were recovered by immunoprecipitation, and the covalently bound RNA was partially digested by RNase T1, radiolabeled, and quantified by autoradiography (Fig. 6A). The signal for K120A/R124A was reduced 20-fold, the signal for K190A/Q193A was reduced fivefold, and the signal for N97A/R130A was reduced 10-fold compared with wild-type protein. These in vivo results are consistent with the reduced binding affinity determined by ITC.

Figure 6.

Impact of expression of the double-mutant QKI proteins on RNA binding in HEK293 cells. (A) Phosphorimage of an SDS-PAGE fractionating PAR-CLIP immunoprecipitate from stable HEK293 cells inducibly overexpressing Flag/HA-tagged QKI proteins. The intensity of radioactivity was quantified for the bands corresponding to RNA–QKI protein complex relative to the wild-type QKI RNP complex. Anti-HA immunoblot control for expression and loading is shown in the bottom panel. (B–D) Correlation of gene expression values (fragments per kilobase per million [FPKM], cutoff value of 5) for QKI targets (red triangles) (see Supplemental Table S2 for PAR-CLIP target list) and QKI nontargets (blue triangles) determined by RNA sequencing experiments of HEK293 cells expressing Flag/HA-tagged wild-type QKI and QKI K120A/R124A (B), QKI K190A/Q193A (C), and QKI N97A/R130A (D).

Impact of QKI mutants on QKI target mRNA abundance

We rederived QKI target RNAs from our previously reported PAR-CLIP data sets (Hafner et al. 2010) using PARalyzer (Corcoran et al. 2011). Groups of overlapping PAR-CLIP sequence reads were considered binding sites if they (1) passed thresholds of ≥0.25 for T-to-C conversion frequency, (2) contained more than five reads with T-to-C conversion, and (3) showed at least two independent T-to-C conversions. We obtained 2534 binding sites of 30-nt average length, of which 2090 were present in 1500 mRNAs (Supplemental Table S2). Seventy-eight percent of mRNA-binding sites were found in intronic regions, and the rest were found mainly in the 3′ UTR, reflecting the predominantly nuclear localization of QKI.

Next, we measured mRNA abundance in double-mutant QKI as well as wild-type QKI Flag/HA-tagged HEK293 cell lines using three conditions in duplicate experiments: (1) Transgene was not induced, and endogenous QKI was not depleted by siRNA transfection; (2) transgene was not induced, and cells were mock transfected; and (3) transgene expression was induced, and endogenous QKI was depleted by siRNA transfection. cDNA libraries were prepared from poly(A) RNA and Illumina-sequenced at a depth of 16–20 Mio reads per sample. The Cufflinks software package (Trapnell et al. 2012) was used to align sequence reads to the genome for determination of mRNA abundance (Supplemental Table S3). Subsequently, we computed the difference in mRNA abundance after depletion of endogenous QKI protein (Supplemental Fig. S6) in cells expressing QKI double mutants and cells expressing Flag/HA-tagged wild-type QKI. (Fig. 6B–D; Supplemental Table S4). Cumulative distribution analysis (Fig. 7A–C) showed a reduction of QKI PAR-CLIP target abundance of 11% (P = 3 × 10⁻¹⁷) for QKI K120A/R124A, 7% (P = 2.4 × 10⁻¹²) for QKI K190A/Q193A, and 4% (P = 0.002) for QKI N97A/R130A, respectively, consistent with the reduced RNA-binding affinity of the mutant proteins and a loss of function.

Figure 7.

Impact of expression of double-mutant QKI proteins on target mRNA abundance in HEK293 cells. (A–C) Changes in transcript stability from QKI double-mutant overexpression—QKI K120A/R124A (A), QKI K190A/Q193A (B), and QKI N97A/R130A (C)—relative to QKI wild-type overexpression were inferred from RNA sequencing analysis. (D–F) QKI PAR-CLIP targets were binned according to the number of target sites on the mRNA (nontargets [black line], one target site [brown line], two target sites [green line], or three or more target sites [orange line]) and changes in transcript abundance upon QKI double-mutant expression (QKI K120A/R124A [D], QKI K190A/Q193A [E], and QKI N97A/R130A [F]) quantified relative to QKI wild-type overexpression. (G–I) QKI PAR-CLIP target sites were binned according to the number of contained RREs (no target sites [black line], no RRE [brown line], one RRE [green line], two RREs [orange line], or three or more RREs [red line]).

We refined our analysis and asked whether the number of QKI PAR-CLIP sites on a transcript correlated with the extent of QKI-mediated stabilization and binned target mRNAs according to whether they contained one, two, or more PAR-CLIP sites (Fig. 7D–F). Irrespective of the number of sites identified, QKI binding resulted in significant stabilization of target transcripts. By analyzing the double-mutant K120A/R124A, which showed the strongest impact on target mRNA stability (Fig. 7A), we found that having two or more target sites reduced median target mRNA abundance by 13.5% compared with 9.5% for mRNAs with single sites (P = 0.065) (Fig. 7D), in line with structural and genetic data showing the importance of QKI homodimerization for regulatory function. QKI dimerization may be necessary to bring together distant RREs on a mRNA for regulation because the presence of two or more RREs in close proximity of up to 20 nt within PAR-CLIP sites of 30-nt average size did not result in increased target RNA stabilization by QKI (Fig. 7G–I).

Discussion

Domain architectures in STAR–RNA complexes

The helix–turn–helix motif of the Qua1 module in both the QKI and GLD-1 STAR–RNA complexes reported in this study (Fig. 2B,C,E,F) was similar to previously observed structures for isolated Qua1 domains of QKI (Beuck et al. 2012) and GLD-1 (Beuck et al. 2010). Furthermore, the relative orientation of the two Qua1 domains in the GLD-1 homodimer observed in the earlier structure (Beuck et al. 2010) is unchanged upon formation of the complex (Supplemental Fig. S3B), while that in the QKI homodimer is slightly different in the free (Beuck et al. 2012) and RNA-bound states, most likely due to crystal packing involving the Qua1 region in the structure of the QKI STAR–RNA complex (Supplemental Fig. S3A).

The KH domain adopts a β1α3α4β2α5β3α6 topology with three α helices (α3, α4, and α6) packed against a three-stranded β sheet, as observed in the RNA-free structure of the KH–Qua2 dual domain of X. laevis QKI determined by NMR in solution (pairwise Cα RMSD value between KH domains of 3.3 Å for 98 matching residues; Protein Data Bank [PDB] ID 2BL5) (Maguire et al. 2005). In contrast, α helix 7 of Qua2, whose orientation is undefined in the RNA-free state in solution, packs against the α3-GPRG-α4 motif of the KH domain in the complex, thereby extending the canonical RNA-binding surface of KH situated between conserved GXXG and the variable loops and allowing binding of the 8-nt RNA sequence ACUAACAA. A very similar RNA-binding mode was observed previously for the RNA-bound structure of the SF1 protein that contains the KH and Qua2 domains but lacks the Qua1 domain typical of STAR proteins (Liu et al. 2001).

Role of Qua1 in facilitating RNA recognition by KH–Qua2

The presence of Qua1 increases the RNA-binding affinity of the KH and Qua2 domains by an order of magnitude (Ryder et al. 2004) as a result of contacts between Qua1 and Qua2, which stabilize the Qua2 orientation relative to the KH domain helping to form an optimal RNA-binding interface within each STAR subunit. In addition, the Qua1 homodimer provides an important bridging contact to both KH–Qua2 units, leading to synergistic binding by the two STAR subunits to two separated conserved nucleotide sequences in the pre-mRNA. Furthermore, these RNA element sequences are positioned in an anti-parallel orientation within the protein–RNA complex, which would allow RNA looping of a target containing multiple RREs. The mutational analysis of the QKI homodimerization interface (Beuck et al. 2012) showed that hydrophobic interactions made by key conserved residues within the dimer interface or the monomer “zipper” of Qua1 are critical for maintaining the precise orientation of the two STAR subunits required for RNA recognition and that the stability of the dimer is important for the ability of the protein to bind RNA.

Contribution of protein dimerization to RNA target selection

Modular proteins have been shown to use pseudodimers formed by the same type of RNA-binding domains, thereby positioning their canonical RNA-binding surfaces at opposite sides of the molecule to enable RNA looping upon binding to separate recognition sites. The use of these preformed dimeric modules is especially suited for proteins that interact with short conserved RNA sequences and act on many targets, such as observed for PTB (Oberstrass et al. 2005), Nova (Teplova et al. 2011), and MBNL (Teplova and Patel 2008), involved in splicing regulation; ZBP1 (Patel et al. 2012), involved in pre-mRNA localization; and polyC-binding protein (Du et al. 2008), involved in multiple DNA and RNA regulation mechanisms. Although the arrangement of the STAR dimer is more complex, as it involves three domains connected by flexible linkers rather than one globular domain, the rationale for the use of dimerization is the same: targeting a pair of RNA sequence elements separated by a spacer of sufficient length and thereby modulating spatial arrangement of the RNA ligand.

Principles underlying specificity of RNA recognition

Our crystal structures of the GLD-1 and QKI STAR domains in complexes with RNA explained the preferences for the YUAAY PAR-CLIP-deduced consensus RNA element (Hafner et al. 2010). The RRE is recognized through shape-specific van der Waals interactions and hydrogen bonds with residues of both the canonical RNA-binding surface of the KH domain and the Qua2 α helix motif, and mutations of amino acids involved in the specific base recognition lead to loss of RNA affinity and concomitant loss of QKI target regulation in vivo. NMR studies showed that the atypical STAR domain family member SF1 has a mode of CUAAC segment recognition similar to that of GLD-1 and QKI (Liu et al. 2001), showing the independence of RNA recognition of KH–Qua2 from Qua1-mediated dimerization.

QKI binding is coupled to target mRNA abundance increase

The distribution of QKI cross-linking sites across pre-mRNAs and mature mRNAs is similar to those of other nuclear or shuttling RBPs, such as ELAVL1, FUS, TAF15, or EWSR1 (Hoell et al. 2011; Lebedeva et al. 2011; Mukherjee et al. 2011). QKI protein is localized predominantly in the nucleus and was reported to also influence alternative splicing of target transcripts (Wu et al. 2002); however, our RNA sequencing data analysis using the Cufflinks and Cuffdiff software (Trapnell et al. 2012) after overexpressing wild-type and mutant QKI proteins revealed no significant variation in QKI target mRNA splicing patterns. Considering that MAG, PLP, and MBP, the only genes for which splicing was altered in the brains of Qki-deficient mice, were not expressed in our HEK293 cells and that QKI predominantly regulated transcript abundance, it is conceivable that splice regulation may result from secondary consequences of direct down-regulation of other splicing factors. Given the observation of nuclear accumulation of Qki target transcripts in Qki mutant mice (Larocque et al. 2002), it is conceivable that QKI plays a more direct role during nucleocytoplasmic transport, possibly acting as an adapter protein for transport.

The presence of more than a single QKI RRE at a RNA target site slightly enhanced target RNA stabilization, indicating that multiple sites can work in concert. Synergy of multiple binding sites within 3′ UTRs was also observed for GLD-1-dependent target RNA repression in C. elegans (Wright et al. 2011). Recognition of two RREs by the QKI dimer could involve sequential steps, with the RNA-binding domain from one subunit initially binding the RNA target followed by a search for another RRE by the second RNA-binding domain, suggesting coordinated conformational changes.

Understanding relevant loss-of-function mutations

Our structures of STAR–RNA complexes provide a molecular framework for the explanation of the deleterious effects (Justice and Bode 1988) of several spontaneous and N-ethyl-N-nitrosourea (ENU)-induced point mutations in model organisms, which have linked RNA binding to defects in glial cell fate and myelination (for review, see Chenard and Richard 2008).

Two ENU-induced point mutations are located within the Qua1 and KH domains of QKI (Fig. 1). The qk^kt3/4 mutant was identified to alter Glu48 to Gly in the Qua1 domain and prevent dimerization (Chen and Richard 1998). This conserved glutamate (Glu177 in GLD-1) forms a hydrogen bond with conserved tyrosine (Tyr17 in QKI and Tyr149 in GLD-1) that stabilizes the helix–turn–helix fold of Qua1 (Fig. 3A), and its replacement by Gly eliminates a hydrogen bond and most likely destabilizes the α-helical structure of the Qua1 motif and its ability to dimerize, as the E48G point mutation was shown to strongly impair protein stability in contrast to an E48A mutant (Beuck et al. 2012). The other ENU mutant, qk^k2, was one of the first described mutations affecting RNA binding (Larocque et al. 2002) and changed Val157 on the central (β3) strand of the β sheet within the KH domain (Figs. 1, 3E). Substitution of the hydrophobic Val side chain pointing toward the core of the KH domain for a polar Glu is likely to abrogate hydrophobic contacts with Lys167 and Leu168 of α6 and C119 at the beginning of the β2 strand and thereby destabilize the overall fold of the KH domain. Recently, another mutation in the hydrophobic β3/α6 interface in zebrafish quaking protein has been found to lead to severe developmental defects (Lobbardi et al. 2011) by changing a conserved Ile (Ile155 in QKI) (Fig. 3E) to Asn with destabilizing effects possibly similar to those of the mouse V157E mutation.

Eight missense mutations exist in the KH domain or flanking loop regions (Fig. 1) that abolish essential GLD-1 function in the germline (Jones and Schedl 1995). Interestingly, five affected residues are highly conserved glycines found in various loop regions of the STAR domain. Gly227 is part of the KH hallmark GXXG loop (Fig. 4B), Gly248 and Gly250 are located in the variable loop of KH (Fig. 4C), and Gly200 (Fig. 3E) and Gly308 (Fig. 3D) are found in KH motif boundaries. The presence of glycine residues tends to interrupt the secondary structure, and indeed, all five mutated glycines are involved in turns and were replaced in each case by long side chain amino acids, leading to unfavorable steric constraints. A structure-based rationalization of the impact of these glycine mutants is outlined in the Supplemental Material.

We now understand that the relative positioning of the Qua2 and KH domain may be compromised by another set of mutations. E310N probably disrupts hydrogen bonding with the KH domain residues Tyr219 and Arg224 as well as Qua2 backbone amides (Fig. 3D), thereby destabilizing the relative positioning of the Qua2 and KH domains. Last, the P217L mutation in the β1–α3 loop (Fig. 1) and the A294T mutation in the hydrophobic core of the KH domain (Fig. 3E) are likely to destabilize the fold of the RNA-binding domain.

In sum, combining the information from our structural studies of QKI and GLD-1 with the known point mutations disrupting their function demonstrates that the structural integrity of both RNA-binding and dimerization domains as well as their relative orientation are critical for the high-affinity RNA binding implicated in multiple developmental functions of QR proteins.

Future opportunities

We plan in the future to undertake structure–function studies of STAR family QKI proteins bound to RNA sequences containing tandem repeats so as to generate a complex containing two STAR proteins bound per RNA target. From the distances between two STAR domains bound to two independent RREs in the QKI and GLD-1 homodimers reported in this study (Supplemental Fig. S5E), it appears that the spacer separating two RREs should be >10 nt, in line with the observation that multiple RREs within one PAR-CLIP-binding site of 30 nt do not increase QKI regulatory effects in vivo (Hafner et al. 2010). The proposed studies will further delineate the importance of contacts in the Qua1–Qua2 interface as well as of the spacer length between multiple RREs on QKI RNA affinity and in vivo function.

Once we solve the structure of STAR family QKI proteins bound to an RNA sequence containing appropriately spaced tandem repeats, we will be in a position to test the extent to which GLD-1 loss-of-function mutations of highly conserved glycines found in the KH invariant GXXG (G227) and variable (G248 and G250) loops as well as in loop regions flanking the KH domain (G200 and G308) of the STAR domain affect protein–RNA recognition.

Materials and methods

Protein and RNA preparation

The PCR-amplified cDNA fragments encoding the STAR domains of human QKI (7–214) and C. elegans GLD-1 (143–337) V143H mutant protein sequences were cloned into a modified pET28a (Novagen) vector between the BamHI and XhoI restriction sites. The plasmids containing the DNA inserts of interest were transformed into Escherichia coli strain BL21-CodonPlus(DE3)-RIL (Stratagene) grown in Luria-Bertrani (LB) medium supplemented with 30 μg/mL kanamycin. The SeMet-substituted QKI construct was expressed by growing cells in a M9 minimal medium using a standard protocol to saturate the biosynthetic pathway for methionine production (Doublié 1997). The recombinant protein expression was induced by 0.4 mM IPTG at 37°C, followed by 12 h of incubation at 18°C. The cell pellets were lysed using a French press and further clarified by centrifugation at 40,000 rpm. The proteins were then purified from the soluble fraction by a nickel-chelating affinity column, followed by cleavage of the N-terminal His₆-Smt3 tag with the Ulp1 protease and additional purification by sequential chromatography on monoQ, Heparin HP, and Superdex 75 columns (GE Healthcare). Protein purity was monitored on a polyacrylamide–SDS denatured gel (PAG).

RNA oligonucleotides were commercially synthesized (Dharmacon Research), deprotected as recommended by the manufacturer, and purified by anion exchange chromatography on a monoQ column (GE Healthcare) according to the manufacturer's instructions, followed by desalting.

ITC measurements

ITC measurements were performed at 25°C using an iTC200 (Microcal) calorimeter. Protein and RNA samples were dialyzed in 20 mM HEPES buffer (pH 7.5) containing 150 mM NaCl and 1 mM DTT. Protein concentrations in the cell of volume 200 μL were 0.012 mM for the wild type, 0.03 mM for the K190A/Q193A mutant, 0.021 mM for the K120A/R124A mutant, and 0.025 mM for the N97A/R130A mutant. RNA concentrations in the injection syringe of volume 60 μL were 0.18 mM for the wild type, 0.34 mM for the K190A/Q193A mutant, and 0.21 mM for the K120A/R124A and N97A/R130A mutants. The data were analyzed with the Microcal ORIGIN software using a single site-binding model.

Crystallization and data collection

Crystals of the complex of QKI STAR with 5′-UUCACUAACAA-3′ RNA were grown by hanging-drop vapor diffusion. Crystallization conditions were determined with sparse matrix screens (Hampton Research). The complex was crystallized by mixing 1 μL of 0.5 mM complex solution in 0.15 M NaCl, 25 mM Tris-HCl (pH 7.5), and 1 mM DTT with 1 μL of reservoir solution containing 0.2 M ammonium sulfate, 0.1 M Bis-Tris (pH 6.5), and 22% (w/v) PEG 3500. Droplets were equilibrated against 1.0-mL reservoirs at 20°C. Prior to data collection, crystals were stabilized by incubation in the above crystallization buffer containing 38% (w/v) PEG 3500 and 5% MPD for 2 d at 4°C. For data collection, crystals were flash-frozen (100 K) without additional cryoprotection. The SAD data set was collected on SeMet QKI–RNA complex crystals at a 0.97920 Å wavelength and 3.5 Å resolution on the 24-ID beamline at the Advanced Photon Source (APS), and the data were processed on site by using HKL2000 (Otwinowski and Minor 1997). Crystals belonged to space group P6₁22, with one molecule of protein–RNA complex per asymmetric unit.

The complex of the GLD-1 STAR V143H mutant with the 5′-p-CUAACAA-3′ RNA was crystallized at 20°C by mixing 1 μL of solution containing 0.5 mM protein and 0.6 mM RNA in 0.1 M NaCl, 25 mM Tris-HCl (pH 7.5), and 1 mM DTT with 1 μL of reservoir solution containing 0.1 M Na-cacodylate (pH 6.5) and 0.7 M Na-acetate. For data collection, crystals were flash-frozen (100 K) in the above reservoir solution supplemented with 1.4 M Na-acetate and 20% glycerol. Diffraction data at 2.85 Å resolution were collected on the X29 beamline at the National Synchrotron Light Source. Crystals belonged to space group P3₁21 with two protein–RNA complexes per asymmetric unit. Crystal and diffraction data characteristics for all QKI and GLD-1 complexes with bound RNA are summarized in Supplemental Table S1.

Structure determination and refinement

The structure of the complex of QKI STAR bound to RNA was determined by SAD phasing using the anomalous diffraction data collected at Se peak wavelength. A total of eight Se sites were located using SHELEXD (Schneider and Sheldrick 2002), and the AutoSol Wizard of PHENIX package (Adams et al. 2010) was used for phasing and density modification. The initial experimental map showed clear density for most regions of the KH and Qua2 domains, 7 nt of the 11-nt RNA, and partial helical regions of Qua1. Iterative manual model building and refinement with phenix.refine produced the current model of the one QKI STAR (183 amino acids) and one 8-nt RNA in the asymmetric unit. The five N-terminal and 10 C-terminal residues and segments 32–39 and 61–68 were disordered in the crystal.

The structure of the complex of GLD-1 STAR bound to RNA was determined by molecular replacement with PHASER in CCP4 suite using the coordinates of the KH–Qua2 domain (71–182) and the bound RNA fragment C5-A11 from the structure of the QKI STAR–RNA complex and the structure of the GLD-1 Qua1 domain in the free state (PDB ID 3K6T; chain A 147–187) as independent search models. The final structure refined with phenix.refine comprises two copies of the entire STAR domain (143–332) and two bound 7-nt RNA molecules. All protein residues in both the QKI and GLD-1 STAR–RNA structures are in allowed regions of the Ramachandran plot as evaluated in phenix.refine. Refinement statistics are given in Supplemental Table S1, and simulated annealing omit maps of the bound RNA segments in the complexes are shown in Supplemental Figure S2. The atomic coordinates and structure factors for the QKI STAR–RNA and GLD-1 STAR–RNA complexes have been deposited in the Research Collaboratory for Structural Bioinformatics (RCSB) PDB with accession codes 4JVH and 4JVY, respectively.

Preparation of vectors for generation of stable cell lines

Double mutations of QKI K120A/R124A, QKI N97A/R130A, and QKI K190A/Q193A were introduced into the plasmid pFRT-TO-Flag/HA-QKI (plasmid available on http://www.addgene.org; Landthaler et al. 2008) in one or two rounds of mutagenesis using the QuikChange II XL kit (Agilent) according to the manufacturer's instructions with the following primers (mutagenized nucleotides are in lowercase letters): QKI K120A/R124—first mutagenesis (fw primer [GTAAAATCATGGTCgcaGGCAAAGGCTCAATG], rev primer [CATTGAGCCTTTGCCtgcGACCATGATTTTAC]) and second mutagenesis (fw primer [GCAGAAACCGGATGTgcaATCATGGTCgcaGGC], rev primer [GCCtgcGACCATGATtgcACATCCGGTTTCTGC]); QKI K190A/Q193A (fw primer [GAAGACAGCCTGgcgAAGATGgcgCTGATGGAGCTTGCG], rev primer [CGCAAGCTCCATCAGcgcCATCTTcgcCAGGCTGTCTTC]); and QKI N97A/R130A—first mutagenesis (fw primer [GAATACCCAGATTTTgctTTTGTTGGGAGAATCC], rev primer [GGATTCTCCCAACAAAagcAAAATCTGGGTATTC]) and second mutagenesis (fw primer [GGCAAAGGCTCAATGgcgGATAAAAAAAAGGAGG], rev primer [CCTCCTTTTTTTTATCcgcCATTGAGCCTTTGCC]).

Cell culture

Flp-In HEK293 cells expressing Flag/HA-tagged wild-type and double-mutant QKI were prepared as described (Landthaler et al. 2008). Cells were maintained in DMEM containing 10% FBS, 2 mM glutamine, 15 μg/mL blasticidin, and 100 μg/mL hygromycin. Flag/HA-tagged transgene expression was induced by addition of 1 μg/mL doxycycline to the cell medium for at least 16 h prior to harvesting.

PAR-CLIP for characterization of in vivo interaction of QKI double-mutant proteins

PAR-CLIP of QKI double mutants was performed according to the protocol published previously (Hafner et al. 2010). Briefly, 10 × 10⁶ cells were cultured in medium supplemented with 100 μM 4-thiouridine and 1 mM doxycycline for 16 h to induce Flag/HA-QKI or Flag/HA-QKI double-mutant expression. Next, cells were washed with PBS and irradiated with 0.15 mJ/cm² 365-nm UV light in a SpectraLinker XL-1500 to cross-link RNA to RBPs. Cells were harvested and lysed in the equivalent of three cell pellet volumes of NP40 lysis buffer (50 mM HEPES-KOH at pH 7.5, 150 mM KCl, 2 mM EDTA, 1 mM NaF, 0.5% [v/v] NP40, 0.5 mM DTT). The cleared cell lysates were treated with 1 U/μL RNase T1 (Fermentas). QKI proteins were immunoprecipitated with monoclonal anti-Flag antibodies (M2, Sigma) bound to Protein G Dynabeads (0.25 mg of antibody per milliliter of beads, 10 μL of bead suspension per milliliter of cell lysate; Invitrogen). The beads were resuspended in one original bead suspension volume of lysis buffer (10 μL of bead suspension per milliliter of lysate), and the immunoprecipitate was further treated with 100 U/mL RNase T1. Beads were washed in high-salt buffer (50 mM HEPES-KOH at pH 7.5, 500 mM KCl, 0.05% [v/v] NP40, 0.5 mM DTT) and resuspended in one bead volume of dephosphorylation buffer (50 mM Tris-HCl at pH 7.9, 100 mM NaCl, 10 mM MgCl₂, 1 mM DTT). RNA was dephosphorylated and radioactively labeled with γ-³²P-ATP using T4 polynucleotide kinase. The protein–RNA complexes were separated by SDS-PAGE, and RNA–protein complexes were visualized by autoradiography and quantified using ImageGauge (Fuji) software version 4.1.

siRNA knockdown of endogenous QKI

A cocktail of three siRNA duplexes targeting the 3′ UTR of endogenous QKI (3.3 nM final concentration for each duplex; siRNA_1 [UCAAGAAGUUGGUGAAAUUUdT, AAUUUCACCAACUUCUUGAUdT], siRNA_2 [CCCAAAGGAUUCUGAGAUAUdT, UAUCUCAGAAUCCUUUGGGUdT], and siRNA_3 [GGAGAAGGCGAAUGAAUUAUdT, UAAUUCAUUCGCCUUCUCCUdT]) was transfected in a 12-well format using Lipofectamine RNAiMAX (Invitrogen) as described by the manufacturer and incubated for 72 h. Knockdown efficiency was assessed by Western blotting using a Pan-QKI antibody (1:1000 dilution, University of California at Davis/National Institute of Neurological Disorders and Stroke (NINDS)/ National Institute of Mental Health (NIMH) NeuroMab facility/Antibodies, Inc., clone N147/6). Total RNA of transfected cells was extracted using TRIzol reagent according to the instructions of the manufacturer.

RNA sequencing experiments

One microgram of total RNA from knockdown and overexpression experiments performed in biological duplicates was used as input for poly(A) purification and cDNA library construction using the TruSeq version 1.5 kit (Illumina). cDNA was barcoded using the Illumina Multiplexing Sample Preparation Oligonucleotide kit and analyzed on two lanes of an Illumina HiSeq 2000 in a 100-base-pair (bp) single-end sequencing run. The resulting sequence reads were aligned to the reference genome (GRCh37/hg19) using TopHat version 2.0.5 (Trapnell et al. 2009, 2012), allowing for up to two alignment errors. Gene expression was estimated using the Ensembl GRCh37 gene model using the Cufflinks software version 2.0.2 (Trapnell et al. 2009, 2012). Gene expression changes and splicing differences between samples were calculated using Cuffdiff version 2.0.2 after update of the gene model based on all experimental evidence using Cuffmerge software version 2.0.2 (Trapnell et al. 2009, 2012).