All four double-stranded RNA binding domains of Staufen2 contribute to efficient mRNA recognition and transcript localization

of localization a of and Drosophila oocytes and mouse They consist of three to four double-stranded RNA binding domains (dsRBDs) and a C-terminal dsRBD-like domain. Mouse Staufen2 (mStau2) like Drosophila Stau (dmStau) contains four dsRBDs. Existing data suggest that dsRBDs 3-4 are necessary and sufficient for binding to Stau-target mRNAs. In contrast, dsRBDs 1-2 have been suggested to be non-canonical dsRBDs, fulfilling other functions such as protein-protein interaction. Here, we show that dsRBDs 1 and 2 of mStau2 bind RNA with similar affinities and kinetics as dsRBDs 3 and 4. While RNA binding by these tandem domains is transient and characterized by high on-and off-rates, an mStau2 fragment with all four dsRBDs recognizes its target RNAs with high stability. Rescue experiments with dsRBD1-2 mutant versions of mStau2 in otherwise staufen -deficient Drosophila oocytes confirmed the physiological relevance of these findings. Full-length mStau2 partially rescued Stau-dependent mRNA localization, whereas a rescue with dsRBD1-2 mutant mStau2 failed. In summary, our data show that the dsRBDs 1-2 play essential roles in the mRNA recognition and function of Stau-family proteins of different species. The combinatorial binding of two tandem-dsRBDs allows for the cooperative recognition of more complex secondary structures and thus contributes to our understanding of how Stau proteins achieve selectivity in RNA binding. 80 mg/mL and 0.5 mg/mL T7 RNA polymerase in 1x TRX (40 Tris/HCl RT spermidine, 0.1 Triton X-100, 5 mM The reaction was incubated for 2 h at 37°C. The reaction was stopped by removal of precipitants by centrifugation at 48,384 g in 5 min and subsequent RNA precipitation with 0.1 V 3 M NaOAc and 3 V absolute ethanol at -20 °C o/n. CcpNmr Analysis RNA assignment of imino groups was based on 1 H, 1 H-NOESY spectra; an initial protein backbone assignment was made with HNCACB spectra. Titration experiments with the single Stau2 dsRBDs 1 and 2 as well as the tandem domain dsRBD1-2 were performed at 50 µM protein concentration. After snap-cooling, the RNA ligand was added in molar ratios of 0.5, 1.0, 2.0 and 3.0 to the protein.


Summary
Proteins of the Staufen (Stau) family are core factors of mRNA localization particles in a number of metazoan organisms and cell types, including Drosophila oocytes and mouse neurons. They consist of three to four doublestranded RNA binding domains (dsRBDs) and a C-terminal dsRBD-like domain. Mouse Staufen2 (mStau2) like Drosophila Stau (dmStau) contains four dsRBDs. Existing data suggest that dsRBDs 3-4 are necessary and sufficient for binding to Stau-target mRNAs. In contrast, dsRBDs 1-2 have been suggested to be non-canonical dsRBDs, fulfilling other functions such as protein-protein interaction. Here, we show that dsRBDs 1 and 2 of mStau2 bind RNA with similar affinities and kinetics as dsRBDs 3 and 4. While RNA binding by these tandem domains is transient and characterized by high onand off-rates, an mStau2 fragment with all four dsRBDs recognizes its target RNAs with high stability. Rescue experiments with dsRBD1-2 mutant versions of mStau2 in otherwise staufen-deficient Drosophila oocytes confirmed the physiological relevance of these findings. Full-length mStau2 partially rescued Stau-dependent mRNA localization, whereas a rescue with dsRBD1-2 mutant mStau2 failed. In summary, our data show that the dsRBDs 1-2 play essential roles in the mRNA recognition and function of Stau-family proteins of different species. The combinatorial binding of two tandem-dsRBDs allows for the cooperative recognition of more complex secondary structures and thus contributes to our understanding of how Stau proteins achieve selectivity in RNA binding. Introduction mRNA localization is an essential mechanism for a range of cellular processes, including embryonic development, cell differentiation and migration, as well as neuronal plasticity (Buxbaum et al., 2015). For active transport of mRNAs along the cellular cytoskeleton, ribonucleoprotein particles (RNPs) are formed. Such mRNA-containing RNPs (mRNPs) contain motor proteins, RNA binding proteins, helicases, and translational regulators (Blower, 2013).
In the mature nervous system, neuronal mRNA localization to pre-and postsynaptic areas followed by local translation has been implicated in memory and learning (Holt and Schuman, 2013;Hutten et al., 2014). For instance, dendritically localized RNAs produce proteins with synaptic functions such as Ca 2+ /calmodulin kinase II (CaMKII), the cytoskeletal protein Arc or microtubule-associated protein 2 (MAP2), and AMPA or NMDA receptors.
The RNA binding protein Staufen (Stau) was originally identified in Drosophila as an mRNA transport factor required to establish the anterior-posterior axis of the embryo (St Johnston et al., 1991;St Johnston et al., 1992). Together with proteins of the exon-junction complex (EJC) and the translational repressor Bruno it binds to the oskar mRNA, which is transported from the nurse cells to the oocyte and then localized to its posterior pole (Lehmann, 2016). During neurogenesis, the asymmetric segregation of prospero mRNA into the ganglion mother cell requires Stau function as well (Li et al., 1997).
In mice, the two Staufen-paralogs Stau1 and 2 share about 50 % proteinsequence identity and have both been implicated in mRNA localization and RNA-dependent control of gene expression (Buchner et al., 1999;Furic et al., 2008;Wickham et al., 1999). Whereas Stau1 is ubiquitously expressed and required for Staufen-mediated decay (SMD) of its target mRNAs via UPF1 interaction, Stau2 expression is enriched in heart and brain (Duchaine et al., 2002;Gong and Maquat, 2011;Kiebler et al., 1999;Kim et al., 2005;Monshausen et al., 2001). The two paralogs Stau1 and Stau2 were reported function during early Drosophila development. The requirement of a combination of two dsRBD-tandem domains and thus binding to two stemloops allows recognition of combinations of secondary structure and thus a much more complex readout for specific binding. This observation might help to explain how Stau proteins can bind selectively to their RNA targets in vivo.

Materials & Methods
Molecular cloning DNA sequences of interest were amplified by polymerase chain reaction (PCR) from template plasmids. Cloning was performed with the In-Fusion HD Cloning kit (Clontech) according to the manufacturer's protocol. Point mutations or deletions were introduced by 3-point PCR with overlap-extension (Ho et al., 1989) or with the QuikChange II Site-directed mutagenesis kit (Agilent technologies), according to the manufacturer's instructions.
For in vivo experiments in D. melanogaster, rsEGFP2 was fused with mStau2 or with dmStau, via cloning into pBlueScript-KS. First, the primers pBSKS-rsEGFP2 FW and rsEGFP+3C RV were used to create an rsEGFP2 sequence with a pBlueScript-KS 5' overhang for In-Fusion cloning and a PreScission protease cleavage site as 3' overhang, and the primers 3C+Stau2 FW and pBSKS-mStau2 RV or 3C+dmStau FW and pBSKS-dmStaufen RV to create a mStau2 or dmStau sequence, respectively, with a PreScission cleavage site 5' overhang and a pBlueScript-KS 3' overhang for In-Fusion cloning. The PCR products from these reactions served as templates for a third PCR with the primers pBSKS-rsEGFP2 FW and pBSKS-mStau2 RV or pBSKS-dmStau RV to create rsEGFP2-mStau2/dmStau sequences with 5' and 3' overhangs for InFusion (Clontech) cloning into BamHI/XbaI-linearized pBlueScript-KS, according to the manufacturer's protocol. The resulting plasmids served as templates for PCR with primers pUASp-rsEGFP2 FW and pUASp-mStau2 RV or pUASp-dmStau RV to amplify rsEGFP2-mStau2/dmStau sequences with 5' and 3' overhangs for InFusion cloning into the BamHI/XbaI-linearized pUASp-attB plasmid.

Protein expression
Full-length mStau2 protein mStau2 FL was expressed as a HisSUMO-tagged fusion protein in High Five insect cells. After cloning in pFastBacDual, recombinant baculovirus was produced with the Bac-to-Bac Expression System (Invitrogen) in Sf21 insect cells as described by the manufacturer's protocol.
Truncated mStau2 dsRBD protein constructs mStau2 proteins were expressed after cloning into the expression vector pOPINS3C as fusion proteins with HisSUMO-tag in E. coli Rosetta cells using autoinduction ZY-medium (Studier, 2005).

Radioactive labeling of RNA
RNAs for electrophoretic mobility shift assays (EMSA) were labeled radioactively for sensitive detection. In vitro transcribed RNA was 5' dephosphorylated in 20 µL reactions containing 10 pmol RNA, 1x Tango Buffer with BSA (Thermo Fisher), 2 U FastAP thermosensitive alkaline phosphatase (Thermo Fisher) and 20 U of the RNase inhibitor SUPERaseIn (Thermo Fisher). After incubation at 37 °C for 30 min, the dephosphorylated RNA was phenol/chloroform extracted and precipitated with 0.1 V 3 M NaOAc, 3 V absolute ethanol and subsequent chilling at -20 °C for ≥ 15 min.
For radioactive labeling, 10 pmol dephosphorylated RNA or chemically synthesized RNA were 5'-phosphorylated with 32 P from γ-32 P ATP (Hartmann Analytic) in a 20 µL reaction with T4 polynucleotide kinase (New England Biolabs) in 1x buffer A. The labeling reaction was incubated at 37 °C for 30 min and subsequently stopped at 72 °C for 10 min.
Remaining free nucleotides were removed by purification on a NucAway™ Spin column (Ambion) according to the manufacturer's instructions. Eluted radiolabeled RNA was diluted to a final concentration of 100 nM in RNase-free H2O and stored at -20 °C.
. CC-BY-NC-ND 4.0 International license not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (which was this version posted August 21, 2018. . https://doi.org/10.1101/396994 doi: bioRxiv preprint RNA for annealing assays was labeled accordingly but purified by urea-PAGE. RNA was resolved on an 8 % urea-Midi Gel, detected with radiographic films, excised from the gel and eluted in 300 µL Crush'n'Soak buffer o/n. Subsequently, RNA was precipitated with 3 V 100 % EtOH, redissolved in 100 µL RNase-free H2O to a final concentration of 100 nM and stored at -20 °C.

Electrophoretic mobility shift assay (EMSA)
For EMSAs with short RNAs (<100 nt), protein at the indicated final concentration was mixed with 5 nM radiolabeled RNA in RNase-free protein buffer supplemented with 4 % glycerol and 30 µg/mL yeast tRNA as competitor in a final volume of 20 µL. In order to allow protein-RNA complexes to form, the mixtures were incubated for > 20 min at RT.
Separation of protein-RNA complexes was performed by native PAGE on 6 % polyacrylamide 1x TBE gels in 40 min at constant 110 V in 1x TBE running buffer. Subsequently, the gels were fixed in 30 % (v/v) methanol, 10 % (v/v) acetic acid for 10 min before drying in a vacuum gel drier (BioRad).
Visualization of radioactivity occurred after exposure of radiograph films (Kodak) in a Protec Optimax developer (Hohmann) or by PhosphorImaging with a Fujifilm FLA-3000.
In the case of long, unlabeled RNA (>100 nt), 10-100 nM RNA were used and separation of protein-RNA complexes was performed by 1-1.5 % agarose gel electrophoresis. Visualization of RNA was achieved by GelRed (Biotium) staining. Fluorescence was visualized with a Fusion SL imaging system (Vilber Lourmat) by UV at 254 nm.

Biotinylation of RNA
Ligand RNA for binding studies by Surface Plasmon Resonance was biotinylated to allow immobilization on a streptavidin-coated surface. For biotinylation, the Pierce TM RNA 3' End Biotinylation Kit (Thermo Fisher Scientific) was used according to the manufacturer's protocol. 50 pmol RNA were used per 30 µL reaction. After extraction and precipitation, the RNA was redissolved in 100 µL RNase-free water. Spectra were analyzed with SpectraManager (Jasco).

Structure determination of SRS2 RNA
Crystallization experiments were performed at the X-ray Crystallography Platform at Helmholtz Zentrum München. Initial crystals of Rgs4 SRS2 were obtained in 80 mM NaCl or 80 mM KCl, 20 mM BaCl2, 40 mM Na cacodylate pH 6.5, 40 % MPD, 12 mM spermine after 4 days at RT using the sitting-drop vapor-diffusion method. Crystals could be reproduced in at 34-46 % MPD.
Crystals were harvested with cryogenic loops and flash-frozen in liquid nitrogen. Native datasets were collected at the Swiss Light Source (SLS) synchrotron, beamline PXIII. Anomalous data for phasing were collected at the European Synchrotron radiation Facility (ESRF) at beam line ID 23-2.
Data were indexed and integrated using XDS and scaled via XSCALE.
Structure factor amplitudes were obtained with Truncate (CCP4 package, (Winn et al., 2011)). The structure was solved by MAD phasing with Barium from the mother liquor, using the AutoRickshaw web server (Panjikar et al., 2005). The structure was completed by iterative manual building in COOT and refinement with RefMac5 (CCP4 package). All crystallographic software was used from the SBGRID software bundle5. Images of the crystal structure were prepared with PyMol (Version 1.7; Schrodinger; http://www.pymol.org/).

Nuclear Magnetic Resonance (NMR)
RNA and RNA-protein complexes were dialyzed to 150 mM NaCl, 20 mM sodium phosphate buffer pH 7.0, 5 mM DTT prior to analysis and 5-10 % D2O was added for locking. Measurements were performed at 298 K on Bruker AVIII 600, AVIII 800 or AVIII 900 NMR spectrometers with cryogenic (TCI) triple resonance gradient probes. Data were processed with Topspin 3.0 or Topspin 3.5 and analyzed with Sparky 3 (Kneller, 2000) and CcpNmr Analysis (Vranken et al., 2005). RNA assignment of imino groups was based on 1 H, 1 H-NOESY spectra; an initial protein backbone assignment was made with HNCACB spectra. Titration experiments with the single Stau2 dsRBDs 1 and 2 as well as the tandem domain dsRBD1-2 were performed at 50 µM protein concentration. After snap-cooling, the RNA ligand was added in molar ratios of 0.5, 1.0, 2.0 and 3.0 to the protein.  Table) probe-sets for smFISH.
Drosophila ovaries were dissected and processed for smFISH analysis as described in (Gaspar and Ephrussi, 2017;Gaspar et al., 2017b).

Microscopy and image analysis
Drosophila egg-chambers mounted onto slides in Vectashield were imaged on a Leica TCS SP8 confocal laser scanning microscope using a 20x dry (NA=0.75) objective for imaging the RNA distribution and a 63x oil immersion (NA=1.4) objective to obtain high resolution images for co-localization analysis of oskar or bicoid mRNA and GFP-Stau. Analyses of RNA distribution within stage 9 and stage 10-11 oocytes were carried out as described in (Gaspar et al., 2014). Briefly, the outlines of the oocytes and the anteroposterior (AP) axis . CC-BY-NC-ND 4.0 International license not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (which was this version posted August 21, 2018. . https://doi.org/10.1101/396994 doi: bioRxiv preprint were manually specified and the smFISH signal was redistributed into a 100x100 matrix. Each column of this matrix represents the relative amount of signal found under 1% of the AP axis length with anterior on the left (column 1) and posterior on the right (column 100). Such matrices are then averaged to obtain a mean and the variability of the RNA localization during a certain stage of oogenesis. Moreover, descriptors such as the center of mass (relative to the geometric center of the ooctye) and the amount of RNA localizing to the posterior domain (defined on the minimum two-fold enrichment of the signal over what is expected) were extracted and compared statistically using a Kruskal-Wallis test followed by pair-wise Mann Whitney U test against the stau null condition.
Co-localization between the mRNAs and GFP-Stau was assayed as described in (Gaspar et al., 2017a). Briefly, images deconvoluted using Huygens Essential were segmented, and nearest neighbor pairs between oskar or bicoid mRNPs and GFP-Stau particles were established. To determine the number of mRNA molecules in an mRNP and to normalize GFP-Stau signal intensity, we fitted multiple Gaussian functions to the corresponding signal intensity distributions taken from the nurse cells (as in (Little et al., 2015) and (Gaspar et al., 2017b)) using the mixtools package in R (https://cran.r-project.org/web/packages/mixtools/index.html). The µ value of Gaussian fit that described the largest portion of the distribution (for oskar mRNPs ~ 60%, for bicoid mRNPs ~80%, GFP-Stau >85%) was taken as the signal intensity of a unit (for mRNPs the intensity of a signal mRNA molecule). The copyright holder for this preprint (which was this version posted August 21, 2018. . https://doi.org/10.1101/396994 doi: bioRxiv preprint least-squares means analysis (Lenth, 2016). All statistical analyses were carried out in R (R Development Core Team (2008). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, URL http://www.R-project.org. (Team, 2008)) using RStudio (www.rstudio.com). All graphs were plotted by the ggplot2 library in R (Wickham, 2016).
. CC-BY-NC-ND 4.0 International license not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (which was this version posted August 21, 2018. . https://doi.org/10.1101/396994 doi: bioRxiv preprint

Results mStau2 binds directly to SRS motifs in the Rgs4 3'UTR
To probe a potential direct interaction between mStau2 and the Rgs4 mRNA, we performed in vitro binding experiments with mStau2 and the two previously predicted SRS motifs of the Rgs4 3'UTR (Figure S1 A). EMSAs with fulllength mStau2 showed binding with apparent equilibrium dissociation constants (Kd) in the low µM range for Rgs4 SRS1 as well as for SRS2 (Figure 1 A). The entire 3'UTR of Rgs4 mRNA was bound by full-length Stau2 with higher affinity (Figure S1 B). Deletion of SRS1 and SRS2 did not reduce RNA binding, indicating that regions other than the SRS motifs contribute to Stau2 binding. Since EMSAs with very long RNAs do not yield very precise results, we also performed experiments with a 3'UTR fragment consisting of 634 bases of the Rgs4 3'UTR (Rgs4-mini) that contains both predicted SRSs.
EMSAs with the previously reported RNA binding dsRBDs 3-4 using either wild-type Rgs4-mini RNA or a mutant version, in which SRS1 and SRS2 were deleted, showed similar affinities (Figure S1 C).
These observations indicated that other cryptic SRSs might be present in the Rgs4 3'UTR. In the Rgs4-mini RNA another region was predicted to fold into a stable imperfect stem-loop with 26 paired bases interrupted by two bulges.
Although this stem-loop is longer and predicted to be more stable, it could still serve as cryptic SRS. Surprisingly mStau2 bound to this stem loop with a KD in the hundred nanomolar range (Figure 1 A). This stem loop is termed SRS* (Figure S1 A).

The Stau2 tandem domain dsRBD1-2 binds dsRNA
Next, we tested whether binding is indeed only mediated by dsRBD3-4, as indicated by previous studies (Duchaine et al., 2002;Gleghorn and Maquat, 2014). Surprisingly, in EMSA experiments mStau2 dsRBD1-2 bound the SRS2 RNA with an affinity comparable to that of dsRBD3-4 (Figure 1 B). This finding shows that dsRBDs 1 and 2 are not inactive pseudo-RBDs as previously suggested but contribute to RNA recognition of Stau2. Binding of mStau2 dsRBD1-2 to SRS2 RNA was further confirmed by NMR titration experiments ( Figure S2). Upon addition of the stem-loop RNA to the tandem domain dsRBD1-2, chemical shift changes and differential line-broadening of To unambiguously confirm the dsRNA fold of Rgs4 SRS2 we solved its crystal structure at 1.73 Å resolution (Figure 1 D). The RNA adopts a typical A-form double-stranded helix, characterized by a wide and shallow minor groove and a deep and narrow major groove. Whereas electron density in the stem region of both molecules is very well defined, the density map in the loop region is poor ( Figure S3), indicating flexibility of the RNA in this region. Imino signals of base pairs are observed in imino 1 H, 1 H-NOESY spectra and are consistent with the base pairing observed in the crystal structure ( Figure S4). Our NMR data and the crystal structure thus confirm that SRS2 folds into a canonical stem-loop structure.

Role of the length of dsRNA for Stau2 tandem domain binding
To test the effect of stem-loop length on RNA binding, the SRS2 stem was extended by five base-pairs (SRS2+5, see Figure S1 A). EMSAs with dsRBD1-2 or with dsRBD3-4 showed significantly improved binding (Figure 1 E), indicating that the length of the stem has great influence on the affinity. In  (Koh et al., 2013;Tants et al., 2017;Wang et al., 2015).

Kinetics of Stau2 RNA binding
In order to understand the kinetics of mStau2 binding, surface plasmon resonance (SPR) experiments with biotin-labeled SRS* RNA or SRS2+5 RNA coupled to a streptavidin sensor chip surface were performed. For the tandem domains dsRBD1-2 and dsRBD3-4, rapid binding and dissociation kinetics were observed for both RNAs already at the lowest tested concentration of 10 nM (Figure 2 A and B). Because of the fast kinetics, the on-and off-rates could not be properly quantified. However, the steady-state binding is best described by a two-site binding fit for dsRBD1-2 with a KD1 of 130 nM for the SRS2+5 RNA and of 25 nM for the SRS* RNA (Figure 2 A,Tables 1 and 2).
KD2 could not be determined because binding was not saturated at the highest measured concentration of 1 µM. Because mStau2 tends to oligomerize at low micro-molecular concentrations, even higher concentration ranges could not be tested.
Also, dsRBD3-4 bound with similar properties, yielding a KD1 of 18 nM for SRS2+5 RNA and of 9 nM for the SRS* RNA (Figure 2 B; Tables 1 and 2). As with dsRBD1-2, KD2 were in the µM range and could not be determined.
Together these findings confirm the RNA binding activities of dsRBD1-2 and  Table 3).

The individual dsRBDs 1 and 2 bind RNA dynamically
In order to obtain structural insights into RNA binding preferences of dsRBDs 1 and 2, 1 H, 15 N-HSQC NMR spectra of the individual dsRBDs and of the tandem dsRBD1-2 were measured. The spectra of both individual domains showed that they are well-folded ( Figure S6). The two spectra also nicely Tables 1, 2). This binding was observed with fast on-and off-rates, similar to the tandem domain (Figure 2 A, B). Furthermore, dsRBD2 shows no sign of cooperativity, as indicated by Hill coefficients close to 1 (Figure 4 C, D). The lack of detectable RNA binding by SPR experiments with dsRBD1 compared to the detected interaction in NMR titrations can be explained by the much higher RNA concentrations used in the NMR experiments (50 µM).
We confirmed these findings by EMSA experiments, in which dsRBD1 did not bind to SRS2+5 RNA (Figure S7 A). Also, dsRBD2 bound SRS2+5 RNA much more weakly than the tandem domain dsRBD1-2, as binding was observed only at concentrations >10 µM (Figure S7 A). The two domains are connected by a linker region of 19 amino acids. Thus, an alternative explanation to the cooperative binding of the dsRBDs could be that the linker region together with one dsRBD provides cooperative RNA binding. We tested this possibility by performing EMSAs either with a dsRBD1-linker fragment or with a fragment consisting of linker-dsRBD2. In neither of these cases did we observe any improved binding to SRS2+5 RNA (Figure S7 B). Also, mixing the two individual dsRBDs with linker did not improve RNA binding activity of the tandem domain. Together these results indicate that the two domains bind dsRNA with some cooperativity that is mediated by the presence of the linker, which itself does not contribute to the RNA recognition.

Mutations in dsRBD1 moderately impair RNA binding of dsRBD1-2
For further verification of the observed binding properties and to allow for functional in vivo studies of the RNA binding activity of Stau2 dsRBD1-2, RNA binding mutants of the dsRBDs 1 and 2 were designed. For dsRBD1, mutants were designed based on the NMR titration experiments and multiple sequence alignments with dmStau ( Figure S8 A). A partial assignment allowed for the identification of residues with chemical shift perturbations upon RNA titration, pointing at their location within or close to the binding interface.
These residues map to the predicted end of helix α1, loop 2 and the beginning of helix α2, which are the regions that mediate RNA binding in a canonical dsRBD ( Figure S 8). Conserved dsRBD residues close or within these regions were chosen for mutation. We mutated glutamate in helix α1 (E15), histidine in loop 2 (H36), lysines from the conserved KKxxK motif (K59 and K60) and phenylalanine in the beta strand β2 (F40). Mutation of these residues in dsRBD3 from D. melanogaster to alanines had been shown to abolish RNAbinding completely (Ramos et al., 2000).
The dsRBD1-2 tandem domain with a range of mutations in dsRBD1 were tested for binding to SRS2+5 ( Figure S10; Table 5). For the mutations E15A, H36A, F40A, K59A, K60A and K59A K60A binding kinetics were fast. Except for E15A the steady-state binding curves are best described by Hill-fits with Hill coefficients n≈1, indicating non-cooperative binding. Whereas the . CC-BY-NC-ND 4.0 International license not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (which was this version posted August 21, 2018. . https://doi.org/10.1101/396994 doi: bioRxiv preprint observed KDs of dsRBD1-2 H36A and K59A are similar to that of dsRBD2 alone, dsRBD1-2 mutations F40A, K60A and K59A K60A bind with even lower affinity than RBD2 alone. These results indicate that binding-activity of RBD1 was abolished by these mutations. The only exception was RBD1-2 E15A, where steady-state binding to SRS2+5 was fitted with a first order binding reaction and a KD of 132±31 nM, indicating that RNA binding activity of RBD1 might be compromised but not completely abolished. When the same RBD1 mutations in context of the dsRBD1-2 fragment were tested for binding to the SRS* RNA, the mutations H36A, F40A, K59A and K59A K60A behaved again very similarly to their binding to SRS2+5, confirming that RNA binding activity of RBD1 is abolished by these mutations ( Figure S11; Table 4). However, RBD1-2 E15A binds to SRS* RNA similar to the wild-type protein, showing two-site binding with KD1=15±11 nM and

Mutations in dsRBD2 impair RNA binding of dsRBD1-2 fragment
Due to the lack of NMR assignments for RBD2, to design mutations in this domain we had to rely on sequence homology. A sequence alignment of twelve species was used to identify conserved, positively charged or aromatic residues for mutation (Figure S8 B). These residues, E99A, K106A, F157A and H169A, were individually mutated in the context of the dsRBD1-2 tandem domain, and subsequently tested for RNA binding by SPR. All mutants showed strongly decreased binding to SRS2+5 (Figure S12, Table 5) and fitting of binding curves indicated that two-site binding was lost in all mutants.
All dsRBD2 mutations in the context of dsRBD1-2 were also tested for binding to SRS* RNA. Unlike SRS2+5, Stau2 dsRBD1-2 E99A and K106A bound SRS* with properties similar to the wild-type protein ( Figure S13, Table 4), indicating that the effects of these mutations are less dramatic and can be compensated for by the longer RNA. Binding of dsRBD1-2 H169A, however, . CC-BY-NC-ND 4.0 International license not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (which was this version posted August 21, 2018. . https://doi.org/10.1101/396994 doi: bioRxiv preprint was still strongly decreased, such that a KD could not be determined. Binding of dsRBD1-2 F157A was again strongly impaired and no KD could be determined, thus confirming the results obtained for SRS2+5.

Double mutations in RBD1 and RBD2 strongly impair RNA binding of the tandem RBD1-2 domain
Based on the SPR results for the single point mutations in RBD1 and RBD2, double-mutants were designed in the context of Stau2 dsRBD1-2. In RBD1, the mutation F40A was chosen because it had a strong effect on binding to both tested RNAs, its resonance shifted upon RNA titration in the 1 H, 15 N-HSQC spectra, and it is conserved in dmStau. In RBD2, the mutations F157A and H169A were chosen. F157A showed altered binding kinetics and H169A had the strongest effect on RNA binding of all tested RBD2 mutations. Both residues are conserved in dmStau. Correct folding of double-mutant proteins was verified by CD spectroscopy ( Figure S9). As expected, in the SPR experiments all double-mutant versions of Stau2 RBD1-2 lacked binding to SRS2-5 and to SRS* RNAs ( Figure S14, Tables 3-4).

Double mutations in dsRBD1 and dsRBD2 strongly impair RNA binding of Stau2 dsRBD1-4
To assess the contribution of dsRBD1-2 to RNA binding in the context of all four verified RNA binding dsRBDs, SPR experiments were performed with

Rescue experiments with Drosophila Stau -/embryos confirm the in vivo importance of RNA binding by RBD1-2
To assess the relevance of our findings for the in vivo function of Staufen proteins, we took advantage of the well-studied role of Stau in early In summary, these rescue experiments confirm the importance of RNA binding by dsRBD 1-2 for the in vivo function of Stau proteins.

Discussion
Previous reports had identified dsRBDs 3 and 4 in Staufen proteins as the RNA binding domains (Duchaine et al., 2002;Gleghorn et al., 2013;Wickham et al., 1999), suggesting that dsRBDs 1 and 2 fulfill other functions. In our present work we have demonstrated that the mStau2 dsRBDs 1 and 2 also possess RNA binding activity. Furthermore, we could show that the two dsRBDs 1 and 2 work together as a tandem domain to achieve their full functionality. Our data further confirm that also dsRBD3-4 also act as a tandem domain. Our comparison of RNA binding affinities of single domains and tandem domains bearing mutations suggest that the first binding event with moderate affinity is achieved by the second dsRBD in each tandem domain, namely dsRBD2 and dsRBD4.
Based on our results, we propose a model in which sequential binding events lead to stable RNA recognition by Staufen (Figure 7). In this model, binding of the first tandem domain occurs initially at a random position, with dsRBD2 achieving the first interaction (Figure 7 A, left side). Our NMR data suggest that dsRBD2 unspecifically slides along the stem, thereby positioning dsRBD1 close to the dsRNA. Subsequently, dsRBD1 also binds, thereby increasing the affinity of the tandem domain to dsRNA ( Figure 7A, right side). In the tandem domain dsRBD3-4 the second domain, dsRBD4, binds with higher affinity ( Figure S17) and thus likely undergoes the first priming contact in a fashion similar to dsRBD2 (Figure 7 B, left side). Then, the other, free dsRBD of the tandem domain also joins the RNA-bound complex ( Figure 7B, right side).
Only when the two tandem domains dsRBD1-2 and dsRBD3-4 act together does the protein form a stable RNA complex. This interpretation is consistent with the stronger and more stable RNA binding of dsRBD1-4 and full-length mStau2.
. CC-BY-NC-ND 4.0 International license not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (which was this version posted August 21, 2018. . https://doi.org/10.1101/396994 doi: bioRxiv preprint To confirm in vivo that RNA binding by the dsRBD1-2 is important for the function of the full-length protein, we utilized the Drosophila oocyte as model system. In the germline of otherwise stau null flies, different variants of Stau were expressed and the rescue of the mutant phenotype assessed by analyzing oskar mRNA localization to the posterior pole. Surprisingly, in mStau2-expressing oocytes a moderate rescue of oskar mRNA localization was observed. In contrast, mStau2-rescue constructs bearing RNA binding mutations in dsRBD1-2 failed to rescue oskar localization. This observation confirms the importance of dsRBD1-2 for RNA binding and RNA localization in vivo. This observation, together with the fact that the long isoform of mStau2 has the same number of dsRBDs, indicate that mStau2 might be the functional homolog of dmStau. The fact that mStau1 lacks the first dsRBD raises the question how this paralog achieves full binding. One option is that its mode of RNA binding is different enough from mStau2 to allow for strong and stable binding even with only two or three dsRBDs. An alternative could be that the reported dimerization of mStau1 (Gleghorn et al., 2013) allows for the joint action of dsRBDs in trans and thus full, stable RNA binding.
Sliding as initial binding mode also occurs in other RNA binding proteins such as the Drosophila protein Loqs-PD, a member of the siRNA silencing pathway. Loqs-PD contains two canonical dsRBDs that show highly dynamic binding that involves sliding along RNA stems to scan for suitable secondary structures (Tants et al., 2017). A similar activity was also reported for the human ortholog of Loqs-PD, TAR RNA binding protein (TRBP) (Koh et al., 2013). In contrast to these examples, however, Stau2 involves two tandem domains with four dsRBDs for its sequential RNA-target recognition.
The feature of stable RNA binding is likely to be of great importance for transport of transcripts over longer distances. It is therefore not surprising that our rescue experiments of Drosophila stau mutants with Stau constructs required all four dsRBDs to be functional. The presented model (Figure 7) offers a mechanistic view on how Stau2 may recognize biological targets with high affinity and stability. RNA binding proteins in higher eukaryotes very often contain multiple RNA binding domains (Lunde et al., 2007). It is thought that these act in a . CC-BY-NC-ND 4.0 International license not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (which was this version posted August 21, 2018. . https://doi.org/10.1101/396994 doi: bioRxiv preprint combinatorial fashion such as we have shown for the dsRBDs of mStau2.
However, for most of these multidomain proteins the manner in which they act cooperatively for function and specificity is not well understood. mStau2 contains two tandem domains, each of which can bind to a stem-loop. It is likely that the choice and combination of the stem-loops as well as their specific spacing determines the specificity of Stau binding for transport of selected mRNAs in vivo.

Acknowledgement
We thank Vera Roman for technical support and Michael A. Kiebler for fruitful discussions and support. We acknowledge the support by the Bavarian NMR center and the X-ray crystallography platform at the Helmholtz Zentrum München. This work was supported by the Deutsche Forschungsgemeinschaft (FOR2333 to S.H., I.G., A.E., and D.N.).

Tables
. CC-BY-NC-ND 4.0 International license not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (which was this version posted August 21, 2018. . https://doi.org/10.1101/396994 doi: bioRxiv preprint    Stau2 dsRBD3-4 binding to surface-coupled SRS2+5 and SRS*. RNAs. The tandem domains dsRBD1-2 and dsRBD3-4 bind transiently with fast kinetics. The steady-state binding curves do not saturate up to 1 µM protein concentration but can be described by a two-site binding fit with KD1 of 18 nM and 130 nM, respectively, for dsSRS2+5 and KD1 of 9 nM and 25 nM, respectively, for SRS*. C) Stau2 dsRBD1-4 binding to surface-coupled SRS2+5 RNA and SRS* is stable with slower kinetics. The steady-state binding curve saturates at approximately 1 µM and is described by a Hill fit with an apparent overall KD of 357 nM and a Hill coefficient n=1.7 for SRS2+5 and an apparent overall KD of 330 nM and a Hill coefficient n=1.8 for SRS*, indicating positive cooperative binding. D) SRS2+5 and SRS* bind to surface-coupled Stau2 FL stably and with high affinity. The steady state binding curves can be described by a two-site binding fit with KD1 of 1.3 nM and 10.6 nM and KD2 of 185 nM and 195 nM, respectively. Two-site binding fit:     Figure 4: SPR sensorgrams and steady-state binding curves for Stau2 dsRBD1 or Stau2 dsRBD2 binding to SRS2+5 or SRS*. At micro-molar concentrations, Stau2 dsRBD1 alone binds neither to (A) SRS2+5 nor to (B) SRS*. In contrast, dsRBD2 binds to (C) SRS2+5 and to (D) SRS* with fast kinetics in a non-cooperative fashion with KDs of 650 nM or 829 nM, respectively. Steady-state binding curves are described by a Hill fit.    . CC-BY-NC-ND 4.0 International license not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (which was this version posted August 21, 2018. . https://doi.org/10.1101/396994 doi: bioRxiv preprint in the germline of stau R9 /stau D3 mutant females that lack the dmStau protein. Transgenic GFP-Stau protein is shown in red, bicoid and oskar mRNA molecules are shown in green and cyan, respectively. In the dmStau-expressing oocytes oskar localizes almost exclusively to the posterior pole (right side of images) and bicoid to the anterior pole (left side of images) during stage 9 of oogenesis. In oocytes that lack Stau (stau null), oskar is found at both poles, enriching slightly more at the anterior, while bicoid localization is unaffected during stage 9. Insets show magnified regions of the upper anterior corner. Scale bar is 20 µm and 1 µm for the insets. (B) Typical localization of oskar mRNA in oocytes as a function of the expressed Stau protein. Using image transformation algorithms, the RNA signal within the oocytes was redistributed into a 100x100 square matrix. Such matrices of different oocytes could be statistically evaluated to obtain the average (green) and the variability (magenta) of the RNA distribution under a given condition. In wild-type oocytes (top left) most signal is found close to the posterior pole (on the right of the panels) by stage 9. In absence of Stau (stau null), there is a strong accumulation of . CC-BY-NC-ND 4.0 International license not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (which was this version posted August 21, 2018. . https://doi.org/10.1101/396994 doi: bioRxiv preprint oskar mRNA at the anterior pole as well. Scale bar represents 20% of the length of the anteroposterior axis. (C, D) Evaluation of oskar mRNA localization by comparing the center of mass (relative to the geometric center at 0, C) and the fraction of oskar mRNA that localizes at the posterior pole (D) during stage 9. P-values show the result of pair-wise Mann-Whitney U tests versus the stau null condition (Bonferroni corrected alpha value is 0.01). Numbers in C show the number of different oocytes per condition. (E-H) Interaction of GFP-tagged Stau molecules with oskar (E, F) and bicoid (G. H) mRNPs. mRNPs were sorted by their mRNA content using quantitative smFISH. The fraction of Stau positive mRNPs (E, G) and the normalized GFP-Stau signal intensity (F, H) were plotted as a function of mRNA content of the mRNPs. The normalized GFP-Stau signal intensities were fitted linear models, the slope of which is indicated. In case of oskar mRNPs (F), all slopes are significantly different in pairwise comparisons (p<0.0001) with the exception of GFP-mStau2[F40A H169A] vs GFP-mStau2[F40A F157A] (p=0.016, alphacorrected=0.01), indicating that wild-type mStau2 has higher affinity to oskar mRNA than its mutant counterparts. In the case of bicoid mRNPs (H), the slope of GFP-mStau2[F40A F157A] is significantly different from the other three (p<0.01), which do not differ significantly from each other (p>0.9).
Figure 7: Model of the molecular recognition of dsRNA by Stau2. A) dsRNA recognition by the Stau2 tandem domain dsRBD1-2. dsRBD2 binds dsRNA promiscuously with moderate affinity and slides along the stem. Through this sliding, dsRBD2 positions dsRBD1 close to the dsRNA. When a suitable dsRNA structure is reached, dsRBD1 also binds, thereby strongly increasing the affinity of the tandem domain to dsRNA. B) Next to the tandem domain dsRBD1-2, dsRBD3-4 acts in a similar way. Here, dsRBD4 does the first promiscuous binding with moderate affinity. When dsRBD2 and dsRBD4 position dsRBD1 and dsRBD3, respectively, close to a suitable dsRNA, the respective domains also bind the dsRNA, thereby increasing affinity. Only when suitable dsRNA binding sites for both tandem domains are in sufficient spatial proximity can all four dsRBDs be bound and form a stable complex with the RNA target. . CC-BY-NC-ND 4.0 International license not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (which was this version posted August 21, 2018. . https://doi.org/10.1101/396994 doi: bioRxiv preprint