Molecular principles underlying dual RNA specificity in the Drosophila SNF protein

Abstract

The first RNA recognition motif of the Drosophila SNF protein is an example of an RNA binding protein with multi-specificity. It binds different RNA hairpin loops in spliceosomal U1 or U2 small nuclear RNAs, and only in the latter case requires the auxiliary U2A′ protein. Here we investigate its functions by crystal structures of SNF alone and bound to U1 stem-loop II, U2A′ or U2 stem-loop IV and U2A′, SNF dynamics from NMR spectroscopy, and structure-guided mutagenesis in binding studies. We find that different loop-closing base pairs and a nucleotide exchange at the tips of the loops contribute to differential SNF affinity for the RNAs. U2A′ immobilizes SNF and RNA residues to restore U2 stem-loop IV binding affinity, while U1 stem-loop II binding does not require such adjustments. Our findings show how U2A′ can modulate RNA specificity of SNF without changing SNF conformation or relying on direct RNA contacts.

Introduction

RNA-binding proteins (RBPs) often employ folded domains to specifically recognize their target RNAs^{1, 2}. Transcriptome-wide RNA binding studies^3,4,5 have revealed hundreds of RBPs bound to diverse RNA molecules. Many RBPs can bind to multiple RNA targets that lack obvious or strict consensus sequences⁶, suggesting that their RNA affinities and specificities can be modulated in the molecular context in which the recognition events take place. However, only few such examples have been analyzed in detail, leaving the molecular mechanisms that regulate RNA affinities and specificities of such RBPs in many cases unclear.

The U1A/U2B″/SNF family of RBPs is found in the U1 and/or U2 small nuclear ribonucleoprotein (snRNP) components of the spliceosome^7,8,9. Jawed vertebrates use U1A to bind stem-loop II of U1 snRNA (U1^SLII) and U2B″ in conjunction with the leucine-rich repeat (LRR) protein U2A′ to bind SLIV of U2 snRNA (U2^SLIV)¹⁰. Both U1A and U2B″ employ an N-terminal RNA recognition motif (RRM) to specifically recognize their RNA targets. RRMs are the most widespread type of RNA-binding domain in eukaryotes². While different RRMs can interact with target RNAs in diverse ways¹¹, crystal structures of human (h) U1A RRM1 (hU1A^RRM1) alone¹² and in complex with human U1 snRNA stem-loop II (hU1^SLII)¹³ have revealed the canonical mode of how RRMs bind single-stranded RNA sequences via the conserved RNP1 and 2 motifs on the exposed surface of a β-sheet. Moreover, structural analysis of human U2B″ RRM1 (hU2B″^RRM1) bound to human U2 snRNA stem-loop IV (hU2^SLIV) in context of the human U2A′ (hU2A′) protein¹⁴ have shown how the two proteins together provide a binding surface for the stem of hU2^SLIV, and how exchange of key residues leads to altered RNA loop specificities in hU1A^RRM1 and hU2B″^RRM1.

While segregation of the U1A^RRM1 and U2B″^RRM1 RNA-binding specificities and selective binding of U2B″^RRM1 to U2A′ can be understood based on their divergent sequences in jawed vertebrates¹⁵, other metazoan species use SNF to bind both U1^SLII and U2^SLIV10, only in the latter case in conjunction with U2A′. For example, Drosophila melanogaster (d) SNF RRM1 (dSNF^RRM1) binds Drosophila U1^SLII (dU1^SLII) with high affinity in vitro, but its affinity for Drosophila U2^SLIV (dU2^SLIV) is up to 100-fold weaker¹⁶, depending on the conditions. Analysis of binding thermodynamics revealed that Drosophila U2A′ (dU2A′) cooperatively restores high-affinity binding of dSNF^RRM1 to dU2^SLIV, under conditions where it does not influence dSNF^RRM1 binding to dU1^SLII17, although it can also associate with the latter complex¹⁶. Consistently, the ternary complex is found only in the U2 snRNP in vivo¹⁸. Until now the mechanistic principles that allow dSNF^RRM1 to bind dU1^SLII directly, yet be modulated by dU2A′ to selectively enhance its affinity for dU2^SLIV, remain unclear.

Here, we report crystal structures of four states of dSNF^RRM1 at 2 Å resolution or better: the free dSNF^RRM1 (residues 1–96; dSNF^1–96), dSNF^1–96–dU1^SLII, dU2A′–dSNF^1–96, and dU2A′–dSNF^1–96–dU2^SLIV. Altogether with the characterization of dSNF^RRM1 dynamics alone and in complexes by NMR spectroscopy and targeted mutagenesis combined with binding studies, we delineated how the intrinsic dSNF^RRM1 RNA-binding capacity and specificity are fine-tuned by networks of intra-molecular interactions that modulate dSNF^RRM1 dynamics, and how the auxiliary dU2A′ protein can capitalize on intrinsic dSNF^RRM1 flexibility to gear dSNF^RRM1 binding specifically to dU2^SLIV.

Results

dSNF^RRM1 exhibits high intrinsic structural flexibility

The sequence of dSNF^RRM1 is 84% identical to hU1A^RRM1 and hU2B″^RRM1 and includes the amino acid triad Tyr-Gln-Phe that defines this family (Fig. 1a). We determined a crystal structure of dSNF^RRM1 based on a construct containing residues 1–96 of dSNF (dSNF^1–96) at 1.49 Å resolution (Supplementary Table 1), showing that the protein adopts a classic RRM fold, with Tyr10-Gln51-Phe53 displayed on the surface of its four-stranded anti-parallel β-sheet (Fig. 1b), in agreement with the previous NMR structure of dSNF^{RRM1 19}. Comparison of the six crystallographically independent dSNF^1–96 molecules indicated intrinsic conformational flexibility in the RNA-binding surface of dSNF^1–96, predominantly affecting loop L3 (residues 42–51; Cα root-mean-square deviation (rmsd) 0.16–1.10 Å with a maximum displacement of 1.47 Å) and to a lesser extent L1 (residues 13–19; rmsd 0.10–0.14 Å with a maximum displacement of 0.21 Å), as well as side chain conformations in the Tyr-Gln-Phe motif (Fig. 1b, c). Furthermore, the conformations and positions of the N-terminal extensions (residues 1–7), which in one case forms an additional α-helix, and the C-terminal loops L6 (residues 84–87) and α3 helices (residues 88–94) differ among the six dSNF^1–96 copies, in agreement with solution NMR data showing that the latter element undergoes extensive ps-ns motions²⁰.

Fig. 1

dSNF^1-96 structure and conformational flexibility. a Sequence alignment of dSNF^RRM1, hU1A^RRM1 and hU2B″R^RM1. Identical residues—red; conserved residues —blue. The Tyr10-Gln51-Phe53 triad is highlighted in bold. Secondary structure elements as observed in a dSNF^1–96 crystal structure are shown above the alignment (a very short β-sheet within L5 has been omitted for clarity). b Overlay of the six dSNF^1–96 molecules (A-F) in an asymmetric unit of a dSNF^1-96 crystal (different shades of gray). Conserved RNA-binding residues (Tyr10-Gln51-Phe53) are shown as sticks. c Cα rmsd’s of crystallographically independent dSNF^1–96 copies relative to molecule A. The N- and C-terminal extensions and the L6 and α3 regions adopt very different conformations in the different dSNF^1–96 copies and have been omitted from the comparison

Crystal packing will limit the authentic conformational ensemble that the protein exhibits in solution. We therefore carried out NMR experiments to assess dSNF^RRM1 backbone amide dynamics in solution. For these experiments we used two dSNF^RRM1 constructs, dSNF^1–96 and dSNF^1-101. Both short and long versions of dSNF^RRM1 showed intermediate dynamics in most elements of the free protein (Fig. 2a, d). ¹⁵N/¹H ΔR₂ experiments (ΔR₂—enhancement of the intrinsic R₂ due to conformational exchange; here we measure ΔR_2,eff = R_2,app(ν_CPMG50 Hz)—R_2,app(ν_CPMG1000 Hz)) at 23 °C revealed motions on the µs-ms timescale affecting dSNF^RRM1 residues along its entire length. Helix α1 (residues 20–34) is particularly mobile, indicated by large ΔR_2,eff components (ΔR_2,eff = 20 s⁻¹ [50 ms]; Fig. 2a). Some L5 residues (70–76) also exhibit dynamics >10 s⁻¹ (100 ms). Truncating α3 at residue 96 causes the C-terminal region of the corresponding construct to become more flexible on the ns-ps and ms timescales as the helical structure is disrupted (Fig. 2a and see below), consistent with three copies of dSNF^1–96 in the crystal structure entirely lacking electron density for the α3 region. We did not obtain ΔR_2,eff values for several L3 residues because of line broadening or absence of a resonance (orange labels in Fig. 2a–c), indicating intermediate conformational exchange on this timescale or exchange with solvent protons. Only α2 residues exhibited no measurable exchange. Altogether these analyses document significant conformational flexibility throughout virtually the entire dSNF^RRM1, peaking at the termini, α1, L3, and L5.

Fig. 2

dSNF^{1-101 15}N/¹H backbone dynamics change upon RNA binding. a–c Values of ΔR_2,eff, calculated from end points of a CPMG experiment; ΔR_2,eff = R_2,app(ν_CPMG50 Hz)–R_2,app(ν_CPMG1000 Hz). Positions of secondary structure elements are shown on the top. a dSNF^1-101, b dSNF^1–101 bound to dU1^SLII, c dSNF^1-101 bound to dU2^SLIV. Errors were determined from the propagation of base plane rms noise. In a, b and c, amides from residues L27 and I30 have no signal from exchange broadening, while in (a), amides from K44 and L46 are exchange-broadened and not quantifiable (orange labels). Prolines (orange labels) do not yield a signal. 50 mM KCl, 20 mM sodium cacodylate, pH 6.5, in 90% ¹H₂O, 10% ²H₂O; 23 °C; 700 MHz. d Orthogonal views of ΔR_2,eff values mapped onto the structure of dSNF^1–96 in isolation (top), bound to dU1^SLII (middle) and bound to dU2A′ and dU2^SLIV (bottom). A scale bar is shown on the top. The orientation of the left panels relative to Fig. 1b is shown at the bottom

Structure of a dSNF^1–96–dU1^SLII complex

We used a 22-residue, blunt-ended dU1^SLII construct to determine the crystal structure of a dSNF^1–96-dU1^SLII complex at 2.0 Å resolution (Supplementary Table 1). dU1^SLII contains a 10-nucleotide (nt) loop (A7-C16) closed by a C6:G17 Watson-Crick base pair (Fig. 3a; to simplify comparisons between the various RNAs, we numbered them identically according to the provided scheme). The center of the loop arches across the dSNF^1–96 β-sheet, L3 protrudes through the RNA loop, while L1, L6, and α3 border the outside of the loop (Fig. 3b). The overall structure closely resembles the previously analyzed hU1A^RRM1–hU1^SLII complex¹³ (Fig. 3b).

Fig. 3

dSNF^1–96–dU1^SLII structure compared to hU1A^RRM1–hU1^SLII. a Alignment of dU1^SLII (brown), hU1^SLII (light orange), dU2^SLIV (gold) and hU2^SLIV (light brown) RNAs (Drosophila RNAs as used in the present study; human RNAs as used in refs. ^{13, 14}). Loop residues are in bold. Sequence differences in the loops and the loop-closing base pairs of U2^SLIV compared to U1^SLII in the Drosophila and human systems are underlined. b Top: hU1A^RRM1–hU1^SLII complex (dark blue/light orange; PDB ID 1URN, chains B and Q¹³) superimposed on the crystal structure of a dSNF^1–96–dU1^SLII complex (gray/brown) according to the hU1A^RRM1/dSNF^1–96 subunits. Orientation of dSNF^1–96 as in Fig. 1b. Bottom: Scheme of selected protein-RNA interactions in the dSNF^1–96-dU1^SLII complex. c dSNF^1–96 contacts to the stem of dU1^SLII (top panel) and comparison to the human system (bottom panel). Lys20 Cδ, Cε and Nε coordinates are not contained in the human structure. d dSNF^1–96 contacts to the dU1^SLII loop and comparison to the human system. e C14 of dU1^SLII flipped towards dSNF^1–96 in one of the complexes in the crystal. In this and the following figures: Stick representations are colored by atom type; carbon—color of the respective molecule; nitrogen—blue, oxygen—red, phosphorus—orange, sulfur—yellow; water oxygens are shown as green spheres; dashed lines represent hydrogen bonds or salt bridges. Orientations relative to Fig. 1b are indicated by boxed rotation symbols

The similar positioning of the U1^SLII hairpins on dSNF^RRM1/hU1A^RRM1 is reflected in similar protein contacts to the 5′ branches of the RNA stems (Fig. 3c). However, molecular interactions in other parts of the complexes differ in detail. In the human system, the side chains of Ser46 and Ser48 together with backbone amides of Arg47 and Lys50 (L3) maintain an extensive, water-mediated hydrogen bonding network with nucleotides A12-G17 (Fig. 3d, bottom panel). In contrast in the Drosophila complex, the corresponding dSNF^RRM1 residues (Leu43, Thr45, Lys44, and Lys47, respectively) do not allow the formation of a similarly extensive water network. While Thr45, Lys44, and Lys47 engage in water-mediated contacts to the phosphate oxygens of C16 and G17, a more expanded network is prevented by Leu43, which replaces Ser46 of hU1A^RRM1 in dSNF^RRM1 (Fig. 3d, top and middle panels). As a consequence, C14 can adopt two alternative conformations. In one conformation, it is bulged out and does not contact dSNF^RRM1 (two of the three crystallographically independent complexes in the crystal; Fig. 3d, top panel); in the other, its base is flipped towards dSNF^RRM1 (Fig. 3d, middle panel), where it rests on a hydrophobic surface formed by Val41 (β2) and Leu43 (L3) and where it engages in hydrogen bonds to the side chain of Lys24 (α1) and the backbone carbonyl group of Ile40 (β2; one of the crystallographically independent complexes; Fig. 3e). The corresponding U14 in the human system is disordered in two of the observed complexes (PDB ID 1URN)¹³; in the third, it is stabilized above C15 by water-mediated interactions to the phosphate of C13 and a direct contact to the base of C16 (Fig. 3d, bottom panel). Lys47 (L3) contacts the phosphate on the 5′ side of A12 in the Drosophila complex (Fig. 3d, top and middle panels), while the equivalent hU1A^RRM1 Lys50 does not (Fig. 3d, bottom panel).

Several residues on the RRM β-strand surface and in the C-terminal extension of the RRM are unique to dSNF and lead to different contacts to the RNA loops in Drosophila and human (Fig. 3d). While Asn12 (Asn15 in hU1A^RRM1) hydrogen bonds to N7 of G10 in both organisms, it is stabilized by Gln80 (Arg83 in hU1A^RRM1) only in Drosophila. Gln85 of hU1A^RRM1 hydrogen bonds to the C11 4-amino group, while the corresponding Ala82 in Drosophila does not allow for a similar interaction. Conversely, Ser84 in dSNF^RRM1 engages in a hydrogen bond to the 6-amino group of A12, while the equivalent Ala88 of hU1A^RRM1 cannot. Instead, the 6-amino group of A12 and the 4-amino group of C13 are jointly bound by Thr89 in hU1A^RRM1, while the equivalent Ser86 does not engage in such interactions in Drosophila. Despite these differences in detail, our structural data are consistent with similar patterns of recognition by dSNF^RRM1 and hU1A^RRM1 for their respective U1^SLII21.

dU2^SLIV binds in a similar fashion to dSNF^1–101 as dU1^SLII

NMR chemical shift perturbations of dSNF^1–101 backbone amides upon dU1^SLII binding²² agree well with the crystal structure (Fig. 4a, d). L3 and the C-terminal region experience the largest changes between free and dU1^SLII-bound dSNF^1–101, consistent with L3 and L6 directly contacting the RNA loop. Additional perturbations are seen in β1, L1 and β4, in agreement with direct contacts of L1 to the RNA stem (Lys19-G2/C3; Figs. 3c) and β4 binding the RNA loop (Lys77-U9), as well as Gln80–Asn12 interactions that are stabilized when RNA is bound (Fig. 3d, top and middle panels). Little chemical shift perturbation was observed in α1, α2, and L5 that form the backside of dSNF^1–101 and do not directly engage the RNA ligand.

Fig. 4

dSNF^1–101-RNA NMR chemical shift perturbations. a Backbone amide chemical shift differences between dSNF^1–101 and dSNF^1–101–dU1^SLII complex. Δδ = [(ΔδHN)² + ((ΔδN)²*0.154)]^1/2. dSNF^1–101 at 300 µM, dU1^SLII at 700 µM. b dSNF^1–101 backbone amide chemical shift perturbations (Δδ) upon dU2^SLIV binding. dSNF^1–101 at 300 µM, dU2^SLIV at 1.1 mM. c Difference plot of chemical shift perturbations (Δ(Δδ) = [Δδ(dSNF^1–101–dU2^SLIV) – Δδ(dSNF^1–101–dU1^SLII)]). Positions of secondary structure elements are shown on the top. 50 mM KCl, 20 mM sodium cacodylate, pH 6.5, in 90% ¹H₂O, 10% ²H₂O; 23 °C; 700 MHz. d Orthogonal views of Δδ (top and middle) or Δ(Δδ) values (bottom) mapped onto the structure of dSNF^1–96. A scale bar is shown on the top. The orientation of the left panels relative to Fig. 1b is shown at the bottom

dSNF^RRM1 binds dU2^SLIV with 10 to 100-fold weaker affinity compared to dU1^SLII, depending on the salt concentration, thus hampering crystallization. However, super-stoichiometric amounts of dU2^SLIV ([dSNF^1-101]:[dU2^SLIV] 1:4) allowed analysis of the dSNF^1–101–dU2^SLIV complex by solution NMR. The pattern of chemical shift perturbations following dU2^SLIV binding to dSNF^1-101 resembles that seen for dSNF^1–101–dU1^SLII (Fig. 4b–d), indicating a similar overall binding mode, but exhibits notable differences in detail. Chemical shift differences are most extensive in the β2-L3-β3 and C-terminal regions (Fig. 4c, d). The most prominent differences are seen for Leu46 (L3), whose backbone amide hydrogen bonds to the phosphate of G17 and whose side chain stacks on the G17 base of the loop-closing base pair in the dSNF^1–96–dU1^SLII complex, and Asp89 (α3) that caps C13 in the dSNF^1–96–dU1^SLII complex. Notably, in dU2^SLIV a U6:G17 wobble pair replaces the Watson-Crick C6:G17 loop-closing base pair of dU1^SLII, and C13 in the loop of dU1^SLII is replaced by G13 in dU2^SLIV (Fig. 3a). While chemical shift differences in Asp89 may thus be explained by similar interactions with different types of residues in dU1^SLII and dU2^SLIV, chemical shift differences in Leu46 suggest that dSNF^RRM1 differentially recognizes the different configurations of the loop-closing base pairs in the two RNAs.

dSNF^1–101 backbone dynamics upon binding dU1^SLII or dU2^SLIV

Fast timescale (ps-ns) backbone dynamics of dSNF^1-101 are quite similar in the absence or presence of RNA (dU1^SLII or dU2^SLIV; Supplementary Fig. 1). C-terminal regions of the protein are an exception; in the free protein, residues 89–101 (α3 and C-terminus) are mostly disordered on this timescale, as shown by hetNOE experiments. When the hairpins are bound, residues in α3 become less dynamic (Supplementary Fig. 1). Those residues do not make contact with the RNAs, but residues Tyr83-Ser84-Lys85-Ser86-Asp87-Ser88-Asp89 are critical for interactions with nucleotides at the top of the loop, using their backbone amides and carbonyl oxygens to engage in hydrogen bonds to nucleobases and riboses, as also previously described in the hU1A^RRM1–hU1^SLII crystal structure¹³. We suspect that these contacts restrict the motions of α3, anchoring it to the body of the complex.

In contrast, the intermediate timescale (µs-ms) dynamics of dSNF^1–101 bound to dU1^SLII or dU2^SLIV are strikingly different compared to isolated dSNF^1-101 (Fig. 2a–d). In the complexes, backbone amides on the RNA-binding surface of dSNF^1–101 are no longer dynamic on the µs-ms timescale, and only in α1 and L5 do they retain µs-ms motions (Fig. 2b–d). These findings suggest that dSNF^1–101 limits the entropic cost of complex formation by retaining high-frequency molecular motions throughout and, in addition, low-frequency motions in elements not directly involved in RNA binding. Notably, α1 and L5 are the binding sites for dU2A′, suggesting that they need to retain flexibility to adapt to that binding surface (see below).

dU2Aʹ binds flexible elements in dSNF^RRM1–RNA complexes

We next determined a crystal structure of a dU2A′–dSNF^1–96 complex at 1.42 Å resolution (Supplementary Table 1), which resembles the arrangement of the two proteins in the hU2A′–hU2B″^RRM1–hU2^SLIV complex¹⁴. In the dU2A′–dSNF^1–96 complex, dSNF^1–96 helix α1 rests on a shallow, concave surface formed by the parallel β strands of the dU2A′ LRR motif and is held laterally by protruding loops that act like pincers (Fig. 5a), with electrostatic contacts in the periphery and hydrophobic interactions in the center of the interface. L5 of dSNF^1–96 additionally comes to lie on the C-terminal pincer loop of dU2A′ (Fig. 5a). Thus, dU2A′ directly binds dSNF^1–96 elements α1 and L5, which remained flexible when dSNF^1–101 alone binds dU1^SLII or dU2^SLIV (see above). Superposition of all copies of isolated dSNF^1–96 on the two crystallographically independent and virtually identical dU2A′–dSNF^1–96 complexes revealed that, with the exception of the α3 region, the overall conformation of dSNF^1–96 does not change upon dU2A′ binding (Fig. 5a). We presently do not know if conformational changes in dU2A′ occur upon complex formation.

Fig. 5

Structure and dynamics of the dU2A′–dSNF^1–96 complex. a Crystal structure of the dU2A′–dSNF^1–96 complex (dU2A′ - green; dSNF^1-96 - steel blue) in two orientations, superimposed on isolated dSNF^1–96 (gray). The orientation of dSNF^1–96 in the left panel is identical to the view in Fig. 1b, the orientation of the right view is indicated by the boxed rotation symbols. b dSNF^{1-96 15}N/¹H backbone amide dynamics measured with ΔR_2,eff NMR CPMG experiments. dSNF^1-96 is 300 µM in 50 mM KCl, 20 mM sodium cacodylate, pH 6.5, in 90% ¹H₂O, 10% ²H₂O; 23 °C; 700 MHz. c ¹⁵N-dSNF¹⁻⁹⁶ + dU2A′ (1:1) (220 µM/220 µM) in 200 mM KCl, 20 mM sodium cacodylate pH 6.5; 23 °C; 700 MHz. Amino acid residues indicated in orange have no detectable signal due to conformational exchange or exchange with proximal protons, hatched bars are line-broadened and not quantifiable. Errors in (b) and (c) were determined from  the propagation of base plane rms noise. d Orthogonal views of ΔR_2,eff values mapped onto the structure of dSNF^1–96 in isolation (top) and bound to dU2A′ (bottom). A scale bar is shown on the top. The orientation of the left panels relative to Fig. 1b is shown at the bottom. e Representative thermogram and binding isotherm of an ITC experiment assessing binding thermodynamics of the dU2A′-dSNF^1–101 interaction

We analyzed motions of dSNF^1–96 alone and in complex with dU2A′ by NMR. When dU2A′ is bound to dSNF^1-96, the overall pattern of ΔR_2,eff terms throughout the body of dSNF^1–96 is substantially altered (Fig. 5b–d). Only α3 retains its ΔR_2,eff contributions, indicating motions on the µs-ms timescale; many backbone amide resonances within α1 and L5 (labeled orange in Fig. 5c) are absent or too low in intensity to analyze, while amides from L3 do not show evidence of dynamics on this timescale. These samples were not deuterated, so exchange of dSNF^1-96 amide protons with proximal protons from dU2A′ is likely to be responsible for their disappearance in the ΔR_2,eff experiments. These data support the co-crystal structure, where dSNF^1–96 α1 is tightly packed against the surface of dU2A′, and dSNF^1–96 L5 is in contact with the LRR surface. In contrast, backbone amides from L3 and its flanking β2 and β3 do not undergo exchange on this timescale in the binary complex. Consistent with these observations, isothermal titration calorimetry (ITC) experiments measured dU2A′-dSNF^1–101 binding thermodynamics ΔH=−1008 kJ/mol, TΔS = −975 kJ/mol-K (ΔS=−3.3 kJ/mol-K), K_d = 4.2 nM (22 °C; 100 mM arginine, 50 mM KCl, 10 mM sodium cacodylate pH 7; Fig. 5e). The large favorable (negative) enthalpy is offset by the large negative TΔS term, which could reflect a loss of conformational flexibility in dSNF^RRM1 and/or dU2A′.

Structure of a dU2A′–dSNF^1-96–dU2^SLIV complex

As in the dSNF^1–96–dU1^SLII complex, the overall structure of dSNF^1–96 is not significantly altered in a crystal structure we determined of a ternary dU2A′–dSNF^1–96–dU2^SLIV RNP (Supplementary Table 1; Fig. 6a, b). Significant differences in the shear of the loop-closing U6:G17 wobble pair in U2^SLIV compared to the C6:G17 Watson-Crick pair in dU1^SLII lead to a different orientation of the two stem regions on dSNF^1-96 (Fig. 6a). The different positioning of the RNA hairpins is supported by their different interactions with Lys17 and Lys19 (L1; Fig. 6c). Lys17 and Lys19 do not face the backbone of the 5′ branch of dU2^SLIV (Fig. 6c, top panel) as they do in complex with dU1^SLII (Fig. 6c, bottom panel). Instead, Lys17 engages in contacts to the major groove side of U6:G17, as well as to G5 of the adjacent base pair of dU2^SLIV; note that an equivalent interaction of Lys17 to the C6:G17 pair in dU1^SLII is not equally possible. To enable the Lys17–U6:G17 interaction, the dU2^SLIV stem must be directed away from Lys19, which lacks dU2^SLIV contacts (Fig. 6c, top panel). Concomitantly, Arg49 (L3) is displaced from its position next to A7 in dU1^SLII and instead contacts the phosphate of G17 in dU2^SLIV. G17 is thereby pulled below the first loop nucleotide (A7; Fig. 6c, top panel), preventing dSNF^1–96 Leu46 from stacking on the loop-closing base pair as in the dU1^SLII complex (Fig. 6c, bottom panel). No other dSNF^RRM1 contacts to the stem of dU2^SLIV are seen in the structure.

Fig. 6

Structure comparison of ternary and binary complexes. a Ribbon diagram of the ternary dU2A′–dSNF^1–96–dU2^SLIV complex (green/steel blue/gold) superimposed on the binary dSNF^1–96–dU1^SLII complex (gray/brown) according to the dSNF^1–96 subunits. The orientation of dSNF^1–96 is as in Fig. 1b. b Scheme of selected protein-RNA interactions in the dU2A′–dSNF^1–96–dU2^SLIV complex. dSNF^1–101 and dU2^SLIV mutations and resulting relative binary affinities (assessed in the absence of dU2A′) are indicated by arrows and labels. Because binding data had to be acquired at different salt concentrations to measure affinities accurately, the ratios of the dissociation constants of wild type and mutant dSNF^1–101 or dU2^SLIV are used to report changes. nx = K_d(wt)/K_d(mut). nx < 1 indicates wt affinity is higher; nx > 1 indicates mutant binds tighter. c Comparison of dSNF^1–96 contacts to the stems and regions around the loop-closing base pairs of dU2^SLIV (top panel) and of dU1^SLII (bottom panel). d Comparison of dSNF^1–96 contacts to the loops of dU2^SLIV (top panel) and of dU1^SLII (middle and bottom panel, representing two different complexes in the crystal structure with C14 contacting or turned away from dSNF^1–96, respectively). Orientations relative to Fig. 1b are indicated by boxed rotation symbols

Notably, a U6C substitution introducing a C:G loop-closing base pair in U2^SLIV led to threefold tighter binding to dSNF^RRM1 (Fig. 6b), indicating that the interaction network ensuing around a stem oriented as in dU1^SLII is more stable and thus providing one reason why dU2^SLIV binding additionally requires U2A′. Furthermore, both types of Lys17-based interactions add similarly to the stability of the respective RNP, as a dSNF^1-101 Lys17Ala variant exhibited a tenfold weaker affinity for both U1^SLII and U2^SLIV (Fig. 6b). Thus, due to the different positioning of the stem, dU2^SLIV experiences essentially a net loss of the Lys19-based backbone interactions compared to dU1^SLII.

Five loop residues of dU2^SLIV, U8-U9-G10-C11-A12, are sequence-identical to dU1^SLII and bind dSNF^1–96 in almost the same fashion in the two RNAs. The only difference is that the 6-amino group of A12 is contacted by Ser84 in the dSNF^1–96–dU1^SLII complex and by Ser86 in the dU2A′-dSNF^1–96–dU2^SLIV complex. The importance of this region is illustrated by a G10A mutation in dU2^SLIV, which led to a 1000-fold reduction in dSNF^1–101 affinity (Fig. 6b).

G13 of dU2^SLIV adopts a syn conformation to occupy an equivalent position as C13 of dU1^SLII, sandwiched by A12 and Asp89 (α3) of dSNF^1-96 and hydrogen bonding to the backbone amide of Asp89 (Fig. 6d). Mutational analysis showed that C13 in dU1^SLII and G13 in dU2^SLIV are critical determinants of RNA affinity and specificity. A dU2^SLIV variant bearing a G13A replacement lacks the ability to support the syn conformation by an interaction with its own phosphate and might additionally lose a contact to the Asp89 backbone, resulting in 200-fold reduced affinity to dSNF^1–101 (Fig. 6b). Conversely, a dU1^SLII-like G13C substitution in dU2^SLIV led to 16-fold higher affinity for dSNF^1-101 (Fig. 6b), presumably due to the ability of cytidine at this position to additionally contact Asp87 (see above).

Significant differences compared to dU1^SLII are again seen in the following 3′ sequence of the dU2^SLIV loop. U14 adopts the same position in both copies of the ternary complex, resembling the minor conformation of the equivalent C14 of the dSNF^1–96–dU1^SLII complex, with the base occupying a binding pocket on dSNF^1–96 (Fig. 6d, top and bottom panels). U14 of dU2^SLIV can engage in more complementary interactions than a cytidine at the equivalent position, with its Watson-Crick hydrogen bonding potential fully saturated by contacts to Lys20, Lys24 (α1) and the backbone carbonyl of Ala42 (L3; Fig. 6d, top panel). Thus, there are more extensive contacts of U14 in dU2^SLIV to dSNF^1–96 helix α1 compared to C14 in dU1^SLII. Consistently, a U14C substitution reduced dSNF^1–101 affinity twofold (Fig. 6b). In addition, U14 is held in its position on dSNF^1–96 by Arg20 from the N-terminal pincer loop of dU2A′, acting like a lid (Fig. 6d, top panel).

The additional loop nucleotide A14a in U2^SLIV forms an extended base stack with the following residues, C15 and C16 (Fig. 6d, top panel). Whereas the equivalent U15 and C16 of dU1^SLII do not contact dSNF^1–96, the additional nucleotide in dU2^SLIV allows this RNA region to approach dSNF^1–96 more closely, so that Lys20 (α1) can directly contact the phosphates of C15 and C16 (Fig. 6d, top panel). Stabilization of dSNF^1–96 by dU2A′ (see above) might reinforce these interactions.

Comparison to the hU2A′–hU2B″^RRM1–hU2^SLIV complex

Globally, the hU2A′–hU2B″^RRM1–hU2^SLIV and the dU2A′–dSNF^1–96–dU2^SLIV ternary complex structures¹⁴ closely resemble each other. The protein moieties of the Drosophila and human complexes superimpose well (rmsd of 0.74 Å for 252 common Cα positions across both proteins; Fig. 7a). All loop nucleotides (A7-C16) of hU2^SLIV are identical to dU2^SLIV and the central and 3′ portions of the loops likewise superimpose well in the two structures. Like dSNF^1–96, hU2B″^RRM1 L3 harbors Leu46 and Thr48 (Leu43 and Thr45 in dSNF^1–96, respectively), which do not permit an extensive water-mediated hydrogen bonding network between the RRM and the RNA, as fostered by the two equivalent serines (Ser46, Ser48) in the hU2^SLII–hU1A^RRM1 structure¹³ (Fig. 7b). As a consequence, U14 of the respective U2^SLIV is accommodated by dSNF^1–96 and hU2B″^RRM1 in the same fashion (Fig. 7b).

Fig. 7

Comparison of ternary complexes from Drosophila and human. a Ribbon diagram of the hU2A′–hU2B″^RRM1–hU2^SLIV complex (fern green/cyan/light brown; PDB ID 1A9N¹⁴) superimposed on the dU2A′–dSNF^1–96–dU2^SLIV complex (green/steel blue/gold) according to the hU2B″R^RM1/dSNF^1-96 subunits. The orientation of dSNF^1–96 is as in Fig. 1b. b Contacts to the RNA loops in the Drosophila (top) and human (bottom) systems. c Interaction networks around the loop-closing base pairs in the Drosophila (left) and human (right) systems. d Interactions between the loop-closing base pairs and L3 in the Drosophila (left) and human (right) systems. e Contacts of the respective U2A′ protein to the RNA stems in the Drosophila (left) and human (right) systems. Orientations relative to Fig. 1b are indicated by boxed rotation symbols

A more in-depth comparison revealed subtle differences in U2^SLIV binding to dSNF^1–96 and hU2B″^RRM1, which arise from an interplay of the different loop closing base pairs (U6:G17 in dU2^SLIV; U6:U17 in hU2^SLIV) and key residue variations between the proteins (Supplementary Fig. 2). On the major groove side, the loop closing base pairs are recognized directly by Lys17/Lys20 (L1) in an equivalent fashion (Fig. 7c). The side chain of hU2B″^RRM1 Met49 (L3) is longer than that of the equivalent dSNF^1–96 Leu46 and abuts the U17 ribose of the loop-closing base pair (Fig. 7d). As a consequence, hU2B″^RRM1 Met49 seems to push U17 of hU1^SLIV underneath A7. A7, in turn, is pulled towards U17 by its N6 amino group hydrogen-bonding to Asp19 (L1) of dSNF^1–96, and so A7 stacks efficiently on the center of the U6:U17 loop-closing base pair (Fig. 7d, right panel). Moreover, Asp19, Arg52 (L3), and G10 interact with the Watson–Crick face of U8, which positions this nucleobase above A7, thereby extending the base stack. In the Drosophila system, the ideal stacking position for A7 on the larger U6:G17 loop-closing base pair is farther remote from the dSNF^1–96 surface (Fig. 7d, left panel). As a consequence, A7 does not directly interact with Glu16 (L1; the equivalent of Asp19 in hU2B^RRM1). Instead, Glu16 is positioned between U8 and G10, allowing U8 to be positioned more remote from the dSNF^1–96 surface and in ideal stacking position on A7. As a consequence, U8 loses its interaction with Arg49 (L3; the equivalent of Arg52 in hU2B″^RRM1).

Due to the slightly different geometry of the loop-closing base pairs, the RNA stems approach the C-terminal end of U2A′ marginally differently. In the Drosophila system, both Gln152 and Lys153 contact the RNA backbone between residues G2 and C4 (Fig. 7e, left panel), while in the human system, hU2A′ Lys149 contacts the U2 phosphate (Fig. 7e, right panel). The orientations of the U2^SLIV RNA stems also leads to a unique hydrogen bond between hU2B″^RRM1 Lys22 and C2, which is absent in the equivalent dSNF^1-96 Lys19. Notably, several of the interaction networks revealed here that control specificity and affinity of dSNF^RRM1, were proposed by Price et al.¹⁴ based on analysis of the hU2B″^RRM1–hU2A′–hU2^SLIV structure.

Differential energetic contributions to dSNF^RRM1-RNA binding

To compare the energetic driving forces for association of the RNAs to the proteins, we used ITC to directly measure the enthalpy and calculate the entropy and dissociation constants (K_d; Table 1; Fig. 8a–d). To avoid partial dissociation of the dU2A′–dSNF^1–96 complex during the titrations, we conducted the experiments at 10 °C. Under the chosen conditions (150 mM KCl, 1 mM MgCl₂, 20 mM HEPES-NaOH, pH 7.5, 5% (v/v) glycerol), binding to both RNAs is enthalpically driven, although entropically unfavorable for dU1^SLII and entropically favored for dU2^SLIV. The affinity of dSNF^1–96 to dU2^SLIV increased about sevenfold in the presence of dU2A′, while dU2A′ increased the affinity of dSNF^1–96 to dU1^SLII only twofold. Notably, in both cases increased binding in the presence of dU2A′ was due to a more favorable (U2^SLIV) or less unfavorable (U1^SLII) entropic contributions, while the enthalpic terms were reduced. These observations are consistent with reduced conformational flexibility of dSNF^1–96 in complex with dU2A′ resulting in a reduced entropic cost of RNA binding. Due to differential entropy–enthalpy compensation effects, the affinity for U2^SLIV is increased relative to U1^SLII (sevenfold difference in affinity), compared to the situation without dU2A′ (23-fold difference in affinity; Table 1).

Table 1 Thermodynamics of RNA binding

Fig. 8

ITC analysis of RNA binding by dSNF^1–96 or dU2A′–dSNF^1–96. a–e Representative thermograms and binding isotherms of ITC experiments assessing the thermodynamics of binding of dSNF^1–96 (a, c), dU2A′–dSNF^1–96 (b, d), dU2A′^Arg20Ala–dSNF^1–96 (e), and dU2A′^mutC–dSNF^1–96 (f) to the indicated RNA hairpins

We also used ITC to test the contributions of dU2A’ Arg20 (which contacts U14 in the loop of dU2^SLIV; Fig. 6d, top panel) and of dU2A′ C-terminal residues Gln152 and Lys153 (which contact the ribose backbone of the dU2^SLIV stem; Fig. 7e, left panel). We generated two variants of dU2A′, in which Arg20 was replaced by an alanine (dU2A’^Arg20Ala) or in which Gln152, Lys153, as well as the preceding Arg143, Lys149, and Lys151 were replaced by alanines (dU2A′^mutC). Complexes of dU2A′^Arg20Ala-dSNF^1-96 and dU2A′^mutC-dSNF^1-96 bound dU2^SLIV with only slightly reduced affinities compared to wild type dU2A′ (Table 1; Fig. 8e, f). While the dU2A′ variants might exhibit larger effects on RNA affinity at higher temperature, the results show that under certain conditions dU2A′ can enhance dU2^SLIV binding by dSNF^1–96 by modulation of dSNF^1–96 alone and without fostering additional RNA contacts.

Discussion

RBPs are often modular, with repeated structural motifs, such as RRMs, that provide opportunities to tune RNA binding affinity and specificity²³. Alternatively, RNA affinities and specificities of RBPs can be modulated by interacting proteins. For instance, an RRM of the U2AF35 protein can shift the structural equilibrium of the two RRMs of the U2AF65 protein towards a more open conformation to facilitate recognition of weak polypyrimidine tracts in human pre-mRNAs²⁴. As another example, the cold shock domain of Upstream-of-N-ras (Unr) and two RRMs of the sex lethal (SXL) protein form an intimately intertwined complex, in which the RNA affinities and specificities of the proteins are mutually reprogrammed to specifically recognize an RNA element in the 3′ untranslated region of msl2 mRNA during dosage compensation in Drosophila²⁵. Yet another case is afforded by the spliceosomal Snu13 and Prp31 proteins that recognize 5′ stem-loops in the major spliceosomal U4 and the minor spliceosomal U4atac snRNAs²⁶. Snu13 positions its αβα sandwich fold to bind kink-turns in the RNAs using the edge of a helix and sheet to contact residues that are identical in U4 and U4atac^{27, 28}. In contrast, Prp31 uses a NOP domain to interact with Snu13 and with different residues in the two RNAs^{26, 29}. Different local structures in the capping pentaloops of the RNAs allow them to adapt differently to Prp31, resulting in different complex stabilities²⁶. Moreover, Snu13 also binds kink-turns in box C/D snoRNAs³⁰. However, in the latter case it is aided by different NOP domain proteins, NOP56/58^{29, 31}. In all of the above cases, the cooperating proteins modulate their respective conformations and/or both proteins exhibit RNA contacts that are important for the stability of RNPs. Among the hundreds of RBPs^{2, 4, 32}, there are certain to be many others whose specificities are modulated through protein–protein interactions.

Here, we studied how the affinity of dSNF^RRM1 for related RNA hairpin loops is modulated by the LRR domain of the dU2A’ protein. We investigated structures and dynamics of dSNF^RRM1 alone and bound to all of its partners, providing a comprehensive catalogue of its free and bound states. Our results demonstrate that, except for its termini, dSNF^RRM1 retains the same overall conformation in isolation or when interacting with dU1^SLII, dU2A′ or both dU2A′ and dU2^SLIV (Cα rmsd 0.39–0.58 Å for dSNF residues 4–83). Moreover, one side of dSNF^RRM1 binds RNA, while the other side binds dU2A’; while dU2A′ fosters contacts to dU2^SLIV in the dU2A′–dSNF^1–96–dU2^SLIV complex, which are not possible with dU1^SLII, these interactions do not contribute to higher dU2^SLIV affinity under all conditions. Thus, in stark contrast to the cases described above, dU2A′ can tune RNA affinity and specificity of dSNF^RRM1 (i) without inducing a conformational change in dSNF^RRM1 and (ii) without relying on additional, own RNA interactions.

One reason for the higher affinity of dSNF^RRM1 for dU1^SLII compared to dU2^SLIV is a more stable binding of the dU1^SLII stem and central loop region by dSNF^RRM1. Our structures and structure-guided mutational analyses indicate that the stem, the loop-closing C6:G17 base pair and C13 in the loop of dU1^SLII interact more intimately with dSNF^RRM1 than the corresponding stem, U6:G17 wobble pair and loop G13 in U2^SLIV. Conversely, the 3′ portion of the dU2^SLIV loop exhibits higher complementarity to dSNF^RRM1 than the corresponding region of dU1^SLII. In particular U14 in dU2^SLIV engages in more intimate contacts to dSNF^RRM1 helix α1 residues (Lys20 and Lys24), as opposed to the equivalent C14 in dU1^SLII. Furthermore, due to the additional loop nucleotide A14a in dU2^SLIV, dSNF^RRM1 α1 residue Lys20 can contact C15 and C16 phosphates in dU2^SLIV but not the equivalent U15 and C16 phosphates in dU1^SLII. While, therefore, the 3′ portion of the dU2^SLIV loop can in principle foster more stable interactions with dSNF^RRM1 than the corresponding region in U1^SLII, these interactions apparently require presence of dU2A’.

Several structural features of the dU2A′–dSNF^1–96–dU2^SLIV complex are fully consistent with NMR chemical shift perturbations seen in the binary dSNF^1–101–dU2^SLIV complex, such as G13 of dU2^SLIV contacting Asp89 of dSNF^1–96, de-stacking of dSNF^1–96 Leu46 and disruption of the hydrogen bond of its backbone amide to the RNA backbone, hydrogen bonding of Gln51 to RNA and packing of Ala42 against RNA. Therefore, dU2A′ apparently does not change the way dU2^SLIV is positioned on dSNF^1–96. Consistent with this notion, under certain conditions dU2A’ also enhances interaction of dSNF^1–96 with dU1^SLII, albeit to a much smaller extent than the dSNF^1–96–dU2^SLIV interaction.

Based on these observations, we suggest that the complementarity of the 3′ part of the dU2^SLIV loop to dSNF^RRM1 cannot be fully accessed in the binary dSNF^RRM1–dU2^SLIV interaction, most likely because interacting protein and RNA elements remain flexible. As shown by our NMR analyses, dSNF^RRM1 loses backbone flexibility in complex with dU2A′. In particular, dU2A′ rigidifies dSNF^RRM1 helix α1 and our crystal structure of the ternary complex shows that dU2A′ also stabilizes dU2^SLIV U14. Formation of a dU2A′–dSNF^RRM1 complex is driven by a large negative enthalpy and opposed by a large negative entropy, consistent with locking out conformational flexibility of both dU2A′ and dSNF^RRM1. Furthermore, subsequent RNA binding is characterized by a less favorable enthalpic contribution than in the binding of isolated dSNF^RRM1 to RNA, but also by more favorable/less unfavorable interaction entropy. Fully in line with interactions of the 3′ portion of the dU2^SLIV loop with dSNF^RRM1 depending on dU2A′-mediated stabilization, fluorescence measurements after replacement of A14a with 2-amino purine (2AP) showed that addition of dSNF^RRM1 leads to loss of 2AP stacking, which is recovered upon addition of dU2A′³³. Our thermodynamic analyses indicate that due to distinct enthalpy–entropy compensation effects, presence of dU2A′ enhances the dSNF^RRM1–dU2^SLIV interaction more strongly than the dSNF^RRM1–dU1^SLII interaction.

We presented a mechanism by which the RNA affinity and specificity of an RBP can be modulated via a protein interaction partner without the induction of significant conformational changes and without the second protein engaging in essential RNA contacts itself. In 2014, Gerstberger et al.² identified 1542 RBPs, comprising 7.5% of all protein-coding genes in humans. Many RBPs have multiple RNA targets in cells and the same RNA can be bound by several RBPs. Moreover, RBPs typically interact not only with RNAs but also with other proteins. We therefore expect that the principles uncovered here will also play a role in the assembly of many other RNPs. The present example of proteins modulating their RNA affinities and specificities is certain to be one of many in cells, with several other mechanisms already known³⁴. The challenge is to recognize when modulation of RBP specificity and affinity occurs, and to understand how these properties are tuned.

Methods

Preparation of complexes

For crystallographic investigations, DNA coding for residues 1–96 of Drosophila melanogaster SNF was cloned into pETM-11 using NcoI and XhoI restriction sites to yield a fusion protein with a tobacco etch virus (TEV)-cleavable N-terminal hexa-histidine tag. Production of dmU2A′ was based on a triple cysteine variant to prevent non-specific intermolecular disulfide formation, as described before¹⁶. For co-expression of dSNF^1–96 and dU2A′, a bi-cistronic expression construct of dSNF^1–96 and dU2A′ open reading frames was assembled by PCR and cloned into pETM-11 using NcoI and XhoI restriction sites. The expression construct comprised DNA coding for a hexa-histidine tagged, TEV-cleavable dSNF^1–96 followed by untagged dU2A′. Mutations encoding dU2A’ variants Arg20Ala and Arg143Ala/Lys149Ala/Lys151Ala/Gln152Ala/Lys153Ala (mutC) or SNF variants were introduced by QuikChange Mutagenesis (Agilent). All DNA oligonucleotides used in this study are listed in Supplementary Table 2.

For protein production, Escherichia coli BL21 T7 Express cells (NEB) were transformed with pETM-11:dSNF^1–96 or pETM-11:dSNF^1–96:dU2A′, grown in Terrific Broth to an OD₆₀₀ of 0.8 at 37 °C, cooled to 20 °C, induced with 0.5 mM IPTG and incubated at 20 °C overnight. Cells were harvested by centrifugation, resuspended in lysis buffer (400 mM NaCl, 1 mM DTT, 20 mM TRIS–HCl, pH 7.5) and stored at −20 °C. Cells were lysed using a homogenizer and cell debris were removed by centrifugation. All purification steps were carried out at 4 °C.

dSNF^1–96 was purified by affinity capture on 5 ml Ni-NTA equilibrated in 400 mM NaCl, 8 mM imidazole, 0.5 mM DTT, 20 mM TRIS−HCl, pH 7.5 (buffer A) and elution with a gradient (10 column volumes) to buffer A supplemented with 400 mM imidazole. Eluted fractions were incubated with His-tagged TEV protease during overnight dialysis into buffer A, followed by a recycling step over 1 ml Ni-NTA. The flow-through was diluted to a NaCl-concentration of 100 mM with 0.5 mM DTT, 20 mM TRIS–HCl, pH 7.5. Contaminating nucleic acids were removed by binding the sample to a 20 ml Heparin Sepharose column (GE Healthcare) equilibrated with 100 mM NaCl, 1 mM DTT, 20 mM Tris–HCl, pH 7.5, and eluting the protein with a gradient (5 column volumes) to 1 M NaCl, 1 mM DTT, 20 mM TRIS-HCl, pH 7.5. dSNF^1–96 was further purified by size exclusion chromatography using a Superdex 75 10/300 gel filtration column (GE Healthcare) in 200 mM NaCl, 1 mM DTT, 20 mM TRIS-HCl, pH 7.5 (gel filtration buffer) to yield pure protein as judged by SDS–PAGE. dSNF^1–96 was concentrated to 36 mg/ml, flash frozen in liquid nitrogen and stored at −80 °C.

In total 3.1 mg of a synthetic 22-mer dU1^SLII RNA (5′-GGCCGC[AUUGCACCUC]GCGGCC-3′; loop in brackets; Agilent Labs) were dissolved in 300 µl Milli-Q water, incubated at 65 °C for 5 min, cooled on ice and supplemented with 150 µl gel filtration buffer. For complex assembly, dSNF^1-96 was mixed with a twofold molar excess of dU1^SLII RNA and incubated for 30 min on ice. The protein-RNA complex was purified by size exclusion chromatography on a Superdex 75 10/300 gel filtration column in gel filtration buffer. dSNF^1-96–dU1^SLII was concentrated to 9 mg/ml, flash frozen in liquid nitrogen and stored at −80 °C.

The dU2A′–dSNF^1–96 complex (with dU2A′ wild type, dU2A′^Arg20Ala or dU2A′^mutC) was purified, as described for dSNF^1–96 alone, but all buffers were supplemented with 5% (v/v) glycerol and the minimal salt concentration during all steps of purification was 200 mM NaCl. Purification over heparin yielded two peaks, dU2A′-dSNF^1–96 and free dSNF^1–96. dU2A′-dSNF^1–96 fractions were pooled and further purified by size exclusion chromatography using a Superdex 75 10/300 gel filtration column. Purified dU2A′-dSNF^1–96 was concentrated to 10 mg/ml, flash frozen in liquid nitrogen and stored at −80 °C.

In total 7.8 mg of chemically synthetized 25-mer dU2^SLIV RNA (5′-GCGGCCGU[AUUGCAGUACC]GCGGCC-3′; loop in brackets;) were dissolved in gel filtration buffer to yield a concentration of 0.7 mg/ml, incubated at 95 °C for 5 min and cooled on ice. For complex assembly, dU2A′–dSNF^1–96 was mixed with a twofold molar excess of dU2^SLIV RNA, incubated at 25 °C for 10 min and cooled on ice. The protein–RNA complex was purified by size exclusion chromatography on a Superdex 75 10/300 gel filtration column. dU2A′–dSNF^1–96–dU2^SLIV was concentrated to 13 mg/ml, flash frozen in liquid nitrogen and stored at −80 °C.

For NMR and ITC experiments, dSNF^1–101 was produced and purified, as described previously²¹. The Lys17Ala exchange was introduced via QuikChange Mutagenesis (Supplementary Table 2). Briefly, protein constructs were isolated from E. coli BL21(DE3) cells (Invitrogen) that had been transformed with a plasmid carrying the protein of interest under control of the TAC promoter and ampicillin resistance. Cells were grown in LB medium at 37 °C and induced at OD₆₀₀ = 0.6–0.8 with 1 mM IPTG, then grown for an additional 4 h at 30 °C. Proteins for NMR experiments were grown in minimal media supplemented with either ¹⁵NH₄Cl and/or ¹³C-glucose. Cells were pelleted and stored at −70 °C until lysis in 100 mM NaCl, 2 mM EDTA, 8.5% (w/v) sucrose, 50 mM sodium acetate, pH 5.3, supplemented with Protease Inhibitor Cocktail (Sigma), PMSF and DNase II. The cell suspension was sonicated then centrifuged. Lysate was passed over an SP-XL Sepharose FPLC column (GE Healthcare), pre-equilibrated in 20 mM sodium cacodylate, pH 7.0, washed with 0 and 100 mM NaCl and eluted over a 100–400 mM NaCl gradient. Fractions containing protein were collected, concentrated and run over a Superdex 75 10/300 gel filtration column in 50 mM KCl, 1 mM EDTA, 20 mM sodium cacodylate, pH 6.5. Desired fractions were concentrated with 10 kDa MWCO Vivaspin concentrators (GE Healthcare). For dSNF^1–101–RNA complexes, dSNF^1–101 protein solution was added slowly to purified lyophilized folded RNA to create RNA-bound complexes with either dU1^SLII or dU2^SLIV.

To purify dU2A′-dSNF^1–96 complex for NMR analyses, dSNF^1–96 with a hexa-histidine tag at the N-terminal end and dU2A′ were isolated from Escherichia coli BL21(DE3) cells that had been transformed with a plasmid carrying the protein of interest under control of the pET promoter and kanamycin resistance. Cells were grown in LB medium (dU2A′) or minimal media supplemented with either ¹⁵NH₄Cl and/or ¹³C-glucose (dSNF^1–96) at 37 °C and induced at OD₆₀₀ = 0.8 with 0.5 mM IPTG, then grown an additional 16 h at 25 °C. Cells were pelleted and stored at −70 °C until lysis. Pellets from ¹⁵N or ¹⁵N/¹³C-labeled dSNF^1–96 and unlabeled dU2A′ were combined in 400 mM NaCl, 5% (v/v) glycerol, 20 mM HEPES-NaOH, pH 7.5 and supplemented with Protease Inhibitor Cocktail (Sigma), PMSF and DNase II. Cell suspension was sonicated then ultra-centrifuged. Lysate was filtered and bound to Ni-NTA beads, washed and eluted with 400 mM imidazole in lysis buffer. The hexa-histidine tag was cleaved with His-TEV protease overnight and removed with His-TEV after running over Ni-NTA beads a second time. Flow-through containing the unlabeled dU2A′-¹⁵N or ¹⁵N/¹³C dSNF^1–96 complex was run over a Superdex 75 10/300 gel filtration column in 200 mM KCl, 20 mM sodium cacodylate, pH 6.5. Fractions containing the complex were concentrated with 10 kDa MWCO Vivaspin concentrators.

RNA preparation

RNAs for NMR experiments were prepared by in vitro transcription with T7 RNA polymerase (300 U) from double-stranded oligonucleotides (Supplementary Table 2)³⁵. Reactions contained 4 mM of each rNTP and 1 mM rGMP (from stock solutions adjusted to pH 7.0), 26 mM MgCl₂, 1 mM spermidine, 0.1% (v/v) Triton-X100, 10 mM DTT, 40 mM Tris–HCl, pH 8.4 in 5 ml at 37 °C for 4 h. When solutions became cloudy, 2 µl inorganic pyrophosphatase (Sigma) were added. After addition of EDTA to a final concentration of 30 mM, reactions were vigorously mixed with an equal volume of equilibrated phenol, centrifuged at low speed for 10 min, and the reaction volume was transferred to acid-washed Corex tubes. Sodium acetate was added to 0.3 M final concentration before mixing with 3 × volumes of 100% ethanol. Tubes were stored at −20 °C overnight. Following centrifugation at 12,000 × g and 4 °C for 30 min, the solution was decanted, and pellets were dried and resuspended in 100 µl MilliQ water. After addition of 100 µl formamide, samples were heated to 95 °C for 3 min, quenched on ice and loaded onto 8 M urea/20% polyacrylamide gels in TRIS-Borate-EDTA. Bands were visualized by UV shadowing, cut out and chopped into small chunks, which were immersed in 0.3 M sodium acetate solution. RNA was extracted by soaking overnight at 37 °C with gentle shaking. Solution was removed, transferred to Corex tubes and centrifuged to clear any acrylamide contaminants, then lyophilized. Product was recovered in a minimal volume of MilliQ water, quantified by UV absorbance and frozen at −20 °C. RNA was dialyzed against buffer before use.

For nitrocellulose filter binding experiments, RNAs were synthesized using T7 RNA polymerase from 200 nM DNA oligonucleotides in 25 µL reactions. RNAs were internally labeled with α-³²P-UTP and α-³²P-CTP. Transcription products were purified on denaturing polyacrylamide gels, and bands were cut out and soaked overnight in 0.3 M sodium acetate. Solutions were centrifuged to remove residual gel residue, then 10 µg glycogen (Roche) was added and RNA was precipitated by addition of 3 × volumes of 100% ethanol overnight at −20 °C. Recovered RNA was washed with cold 70% ethanol, dried, resuspended in MilliQ water and stored at −20 °C until needed.

Nitrocellulose filter binding experiments

Nitrocellulose filter binding experiments were used to determine the affinity of dSNF^1–101 for wild type and mutant dU2^SLIV RNAs³⁶. Nitrocellulose filters (BA-45; Whatman) were prepared by soaking in KCl solutions at concentrations used in binding experiments. Samples for binding experiments contained pM concentrations of RNA, titrated with increasing concentrations of dSNF^1–101. Binding affinity was calculated using Kaleidagraph to give dissociation constants (K_d) and Gibbs’ free energies³⁶.

Isothermal titration calorimetry

ITC experiments measuring binding thermodynamics of the dU2A′–dSNF^1–101 interaction were conducted on a Nano ITC LV (TA Instruments) at 22 °C in 50 mM KCl, 100 mM arginine, 10 mM sodium cacodylate, pH 6.5, using dU2A’ as the sample (2.5 µM) and dSNF1-101 as titrant (25 µM). Proteins were dialyzed into ITC buffer. Data were analyzed using Origin software.

ITC experiments measuring thermodynamics of dSNF^1–96 or dU2A′–dSNF^1–96 binding to RNA were conducted on a iTC200 (Malvern Panalytical) at 10 °C in 150 mM KCl, 1 mM MgCl₂, 20 mM HEPES-NaOH, pH 7.5, 5% (v/v) glycerol, using the RNAs as samples (8.3–25 µM) and the proteins as titrants (50–150 µM). 100 µM solutions of synthetic RNAs (IBA GmbH; U1^SLII: 5′-CCAGGACGC[AUUGCACCUC]GCGUCCUGG-3′; U2^SLIV: 5′-CCAGGACGU[AUUGCAGUACC]GCGUCCUGG-3′; loops in brackets) were incubated at 80 °C for 3 min, snap-cooled on ice and adjusted to ITC buffer. dSNF^1–96 alone or in complex with dU2A′ (wild type, dU2A′^Arg20Ala or dU2A′^mutC) was dialyzed into ITC buffer. Data were analyzed using the Microcal PEAQ software.

Crystallographic analyses

Proteins and complexes were crystallized by sitting drop vapor diffusion (1 µl protein plus 1 µl reservoir for dSNF^1-96 and dU2A′-dSNF^1–96–dU2^SLIV or 100 nl protein plus 100 nl reservoir for dSNF^1–96–dU1^SLII and dU2A′–dSNF^1–96) at 4 °C (dSNF^1–96–dU1^SLII, dU2A′–dSNF^1–96, dU2A′–dSNF^1–96–dU2^SLIV) or 20 °C (dSNF^1–96). dSNF^1–96 crystallized with a reservoir containing 200 mM sodium chloride, 1 M sodium citrate, 100 mM TRIS–HCl, pH 7.5. Crystals were cryo-protected by transfer into reservoir solution supplemented with 20% (v/v) glycerol. The dSNF^1–96–dU1^SLII complex crystallized with a reservoir containing 200 mM Li₃citrate, 20% (w/v) PEG 3350 and 15 mM NiCl₂ as an additive. Crystals were cryo-protected by transfer into reservoir solution supplemented with 10% (v/v) PEG 400. dU2A′–dSNF^1–96 crystallized with a reservoir containing 200 mM Li₂SO₄, 30% (w/v) PEG 400, 100 mM sodium cacodylate, pH 6.5. The dU2A′–dSNF^1–96–dU2^SLIV complex crystallized with a reservoir containing 200 mM Li₂SO₄, 20% (w/v) PEG 3350 and 15 mM sarcosine as an additive. Crystals were cryo-protected by transfer into reservoir solution supplemented with 10% (w/v) PEG 400. All crystals were incubated in the respective cryo-protecting solution for 10–30 s and then flash-cooled in liquid nitrogen. Diffraction data were collected at 100 K on beamline 14.2 of the BESSY II storage ring (Berlin, Germany). All diffraction data were processed with XDS³⁷.

The structure of dSNF^1–96 was solved by molecular replacement using the program PHASER³⁸ and a homology model generated by HHpred³⁹ based on the structure of hU1A^RRM1 (PDB ID 1URN¹³). The structure of dSNF^1–96–dU1^SLII was solved by molecular replacement with PHASER employing the structure coordinates of hU1A^RRM1 in complex with hU1^SLII (PDB ID 1URN¹³), in which the hU1A^RRM1 coordinates had been replaced by the dSNF^1–96 structure and the residues of hU1^SLII had been exchanged for those of dU1^SLII. The structure of dU2A′-dSNF^1–96 was solved by molecular replacement with PHASER employing the structure coordinates of the hU2A′–hU2B″^RRM1–hU2^SLIV complex (PDB ID 1A9N¹⁴), in which the coordinates of hU2B″^RRM1 had been replaced by the dSNF^1–96 structure, hU2A′ had been replaced by a homology model of dU2A′ and the RNA had been omitted. The structure of the dU2A′–dSNF^1–96–dU2^SLIV complex was solved by molecular replacement with PHASER using the structure coordinates of dU2A′–SNF^1–96, to which an RNA model had been appended based on the structure coordinates of the hU2A′-hU2B″^RRM1–hU2^SLIV complex (PDB ID 1A9N¹⁴). Structural models were completed through alternating rounds of automated refinement using PHENIX.REFINE⁴⁰ and manual model building using COOT⁴¹.

NMR spectroscopy

NMR data were acquired either on a 700 MHz (¹H) Varian Inova spectrometer with z-axis pulsed field gradient triple resonance Varian probe, or a 600 MHz (¹H) Bruker Avance III spectrometer with QCI cryoprobe. dSNF^1–101 and dSNF^1–101-RNA samples included 50 mM KCl, 2 mM EDTA, 20 mM sodium cacodylate, pH 6.5, 10% (v/v) ²H₂O at 23 °C. dSNF^1–96 and dU2A′–dSNF^1–96 samples included 200 mM KCl, 1 mM EDTA, 20 mM sodium cacodylate, pH 6.5, 10% (v/v) ²H₂O at 23 °C. Temperature was calibrated against a standard methanol NMR sample at 700 MHz, and NMR Thermometer methanol-d4 (Bruker) for 600 MHz cryoprobes. 2,2-dimethyl-2-silapentane-5-sulfonate was used for chemical shift reference. Unbound dSNF^1–96 and dSNF^1–101 were assigned using established 3D triple-resonance backbone NMR experiments (HNCACB and CBCA(CO)NH)⁴². dSNF^1–101 complexed with unlabeled RNA (dU1^SLII or dU2^SLIV) or unlabeled dU2A′ protein was assigned at 600 MHz using BEST TROSY 3D pulse sequences²². In order to collect data in a usable timeframe, 3D experiments were collected with non-uniform sampling techniques and data (FID) reconstructed with Wagner Lab hmsIST additions⁴³ to NMRPipe. Non-uniform sample schedules were 20–30%. NMR spectra were processed in NMRPipe⁴⁴ and all data were analyzed in NMRViewJ.

The weighted change in backbone amide chemical shifts between free dSNF^1–101 and RNA-bound dSNF^1–101 was calculated as:

$$\Delta \delta = \left[ {\left( {\Delta \delta H^N} \right)^2 + \left( {\left( {\Delta \delta N} \right)^2 \ast n} \right)} \right]^{1/2}$$

(1)

in which ΔδH^N and ΔδN are calculated as ΔδH^N = (δH^N(wt)−δH^N(mutant)) and similarly Δδ¹⁵N = (δ¹⁵N(wt)−δ¹⁵N(mutant)) in parts per million (ppm) of the magnetic field. n = 0.154 is a standard normalization factor that allows the combination of the ¹H and ¹⁵N chemical shift ranges.

Heteronuclear TROSY ¹⁵N-{¹H} NOE data were acquired in pairs of spectra, one with and one without saturation of the proton resonances. Experiments with proton saturation used a 3 s relaxation delay with 3 s of saturation, while spectra with no saturation used 6 second relaxation delays between scans. Spectra were collected in duplicate, and for each pair, peak intensity ratios (I/I₀) were used to calculate the steady-state NOE. Propagation of the base plane noise was used to give the error in the data.

Relaxation dispersion experiments (ΔR_2,eff) were acquired at 700 ¹H MHz and 23 °C using a relaxation-compensated Carr–Purcell–Meiboom–Gill TROSY pulse sequence. Data were collected at two endpoints: at CPMG ν_CPMG = 50 Hz and ν_CPMG= 1000 Hz. In addition, two experiments were acquired: at T_relax = 40 ms, and a reference with ν_CPMG = 0 Hz (i.e., no CPMG interval). For all experiments, the relaxation delay was 2.5 seconds. R_2,app values were determined as⁴⁵:

$$R_{2,{\mathrm{app}}}\left( {{\mathbf{\nu }}_{{\mathrm{CPMG}}}} \right) = - 1/{\mathrm{T}}_{{\mathrm{relax}}} \cdot {\mathbf{ln}}\left[ {{\mathrm{I}}\left( {{\mathbf{\nu }}_{{\mathrm{CPMG}}}} \right)/{\mathrm{I}}_0} \right]$$

(2)

where T_relax = 40 ms is the total time for the CPMG refocusing period, I(ν_CPMG) is the peak intensity with CPMG refocusing, and I₀ is the peak intensity with no refocusing. To measure the contribution to intrinsic R₂ from conformational exchange,

$$\Delta {\mathrm{R}}_{2,{\mathrm{eff}}} = {\mathrm{R}}_{2,{\mathrm{app}}}\left( {{\mathbf{\nu }}_{{\mathrm{CPMG}}\;{\mathrm{at}}\;{\mathrm{50Hz}}}} \right)-{\mathrm{R}}_{{\mathrm{2,app}}}\left( {{\mathbf{\nu }}_{{\mathrm{CPMG}}\;{\mathrm{at}}\;{\mathrm{1000Hz}}}} \right)$$

(3)

or, for simplification:

$$\Delta {\mathrm{R}}_{{\mathrm{2,eff}}} = - 1/{\mathrm{T}}_{{\mathrm{relax}}} \cdot {\mathbf{ln}}\left[ {{\mathrm{I}}\left( {{\mathbf{\nu }}_{{\mathrm{CPMG}}\;{\mathrm{at}}\;{\mathrm{1000Hz}}}} \right)/{\mathrm{I}}\left( {{\mathbf{\nu }}_{{\mathrm{CPMG}}\;{\mathrm{at}}\;{\mathrm{50Hz}}}} \right)} \right].$$

(4)

Data availability

The coordinates and structure factor data have been deposited in the Protein Data Bank (www.pdb.org) with the accession codes 6F4I (dSNF^1–96), 6F4J (dU2A′-dSNF^1–96), 6F4H (dSNF^1–96–dU1^SLII) and 6F4G (dU2A′–dSNF^1–96–dU2^SLIV). Other data are available from the corresponding authors upon reasonable request.