Keywords

1 Introduction

RNA is central to life and has proven to be an exceptionally diverse molecule in terms of its functional repertoire (Serganov and Patel 2007; Clancy 2008). One activity of recent note is that certain structural elements, called riboswitches , can regulate gene expression by controlling mRNA transcription and translation (Winkler and Breaker 2003; Bastet et al. 2011) and less frequently by influencing intron excision (Kubodera et al. 2003; Lee et al. 2010; Chen et al. 2011), and mRNA stability via backbone self scission (Winkler et al. 2004); see Serganov and Nudler (2013) for a review. In the most common scenario, a riboswitch detects a small molecule that binds to its ‘aptamer domain .’ Such binding causes the riboswitch to undergo a conformational change that sequesters or unfurls mRNA sequences in a downstream expression platform , leading to gene “on” or “off” states (Mironov et al. 2002; Winkler et al. 2002; Nahvi et al. 2002; Sudarsan et al. 2003).

In this chapter, we review techniques used to characterize two phylogenetically unrelated riboswitches that bind the same secondary metabolite, preQ1 —an intermediate in the queuosine (Q) biosynthetic pathway (Fig. 1a). These riboswitches are designated class 1 (preQ1-I) and class 2 (preQ1-II) and were identified in the genomes of numerous bacteria within the Firmicutes phylum. PreQ1-I riboswitch es, such as that from B. subtilis, reside in the 5′-leader sequence of the queC gene located in the queCDEF operon, which encodes enzymes essential for synthesis of preQ1. PreQ1-I functions as a “switch” because high cellular levels of preQ1 favor metabolite binding to its aptamer domain, thus promoting formation of a rho-independent transcription terminator within its downstream expression platform. When preQ1 levels are low, the expression platform forms an anti-terminator stemloop that favors gene transcription (Roth et al. 2007). Other preQ1-I riboswitches appear to regulate translation , such as that from T. tengcongensis (Tte) (Fig. 1b). The Tte preQ1-I riboswitch is located within the 5′-leader sequence of COG4708 (DUF988) mRNAs that encode a putative preQ1 transporter. In the preQ1-bound state, the aptamer adopts a conformation that buries the first two bases of the ribosome-binding site (RBS), thus blocking ribosome access to the expression platform leading to attenuated gene expression. When preQ1 levels are low, the block is relieved and translation ensues (Roth et al. 2007; Spitale et al. 2009; Jenkins et al. 2011). Similar feedback regulation has been ascribed to preQ1-II riboswitch es, such as that from L. rhamnosus (Fig. 1c). Such riboswitches also govern COG4708 genes (Weinberg et al. 2007) although they bury the entire RBS when preQ1 binds to the aptamer (Liberman et al. 2013; Meyer et al. 2008). Because preQ1 is unique to the bacterial metabolome (McCarty and Bandarian 2012), and class 1 and 2 preQ1 riboswitches are present in a variety of pathogenic bacteria (Meyer et al. 2008; Roth et al. 2007; Weinberg et al. 2007), these riboswitches are of interest as antibacterial targets.

Fig. 1
figure 1

Queuosine biosynthesis in bacteria and schematic structure of preQ1-I and preQ1-II riboswitches reviewed herein. (a) The bacterial biosynthetic pathway from GTP to queuosine (Q) involves several enzymes (arrows) that lead to the soluble intermediate preQ1, which is incorporated into tRNA at that wobble position to ultimately give Q as a means to confer translational fidelity (reviewed in McCarty and Bandarian 2012). (b) Representative class 1 preQ1 riboswitch from Thermoanaerobacter tengcongensis (Tte) depicting the H-type pseudoknot with tertiary hydrogen-bond interactions (gray) from crystal structures (Jenkins et al. 2011; Spitale et al. 2009). Watson–Crick pairs are depicted as short, solid lines; arrows indicate connectivity. The ribosome-binding site (RBS) is highlighted in yellow. (c) Representative class 2 preQ1 riboswitch from Lactobacillus rhamnosus (Lra) depicting the HLout pseudoknot. Interactions and sequence features are as described in (b) based on the structure (Liberman et al. 2013). Boxed regions (gray) show wild-type sequences that were modified to produce constructs #1 and #2, respectively

Full size image

1.1 Learning Objectives of This Chapter

Here we describe approaches used to define the metal dependence of small-molecule binding to preQ1 riboswitches by isothermal titration calorimetry (ITC), as well as methods employed to introduce osmium(III) pentaammine and cesium(I) heavy-atom derivatives into the class 1 and 2 preQ1 riboswitches for multi-wavelength anomalous diffraction (MAD) or single-wavelength anomalous diffraction (SAD) phasing, respectively (Spitale et al. 2009; Jenkins et al. 2011; Liberman et al. 2013). The methods are written for graduate students or post-doctoral trainees with preliminary to intermediate knowledge of these topics; the information is meant as a practical guide rather than a pedagogical one. A key aspect of our methods is that they are broadly applicable to other functional RNAs and should stimulate the development of derivative approaches in the reader’s laboratory.

2 Isothermal Titration Calorimetry Methods for Riboswitches

2.1 Quantitative Procedures to Test Metal Ion Requirements for Ligand Binding

Ions play a significant role in RNA folding and function. In this section, we describe methods to probe riboswitch ion requirements by measuring ligand affinity in the absence or presence of various ions. K+ and Mg2+ are likely to be of greatest interest since they are prominent in the bacterial cell. K+ concentrations range from 0.1 to 0.6 M (McLaggan et al. 1994), whereas free Mg2+ is present at 1–2 mM (Alatossava et al. 1985). Mg2+ promotes RNA folding and tertiary interactions (Pyle 2002; Draper et al. 2005; Woodson 2005) but can also impart function by participating in the chemical reactions of ribozymes (Nakano et al. 2001; Chen et al. 2010; Toor et al. 2009; Marcia and Pyle 2012; Stahley and Strobel 2006). In the case of riboswitches, Mg2+ shifts the folding equilibrium of most aptamer domains into a “binding-competent” conformation that interacts with ligand (Montange and Batey 2008). Some riboswitches recognize their ligand in the absence of Mg2+, such as the guanine (xptG) riboswitch (Serganov et al. 2004), the SAH riboswitch (Edwards et al. 2010), as well as the preQ1-I and preQ1-II translational riboswitches (Sect. 2.5 and Liberman et al. 2013); however, Mg2+ still drives these RNAs into a binding-competent state. In addition to imparting structural stability, some riboswitches bind synergistically to Mg2+ and ligands, such as the glmS, TPP, FMN, and glycine riboswitches (Klein and Ferre-D’Amare 2006; Serganov et al. 2006, 2009; Lipfert et al. 2010); see Ferre-D’Amare and Winkler (2011) for a review. Such synergy has been noted for the FMN riboswitch interaction with K+ (Serganov et al. 2009). In addition to binding Mg2+, the Oceanobacillus iheyensis group IIc intron selectively binds K+ based on size, which forms part of its catalytic metal cluster (Marcia and Pyle 2012). Collectively these observations illustrate that complete characterization of a functional RNA requires analysis of the metal dependence of its activity, which should be the preface to any structural investigation.

2.2 Checking the Feasibility of Isothermal Titration Calorimetry Measurements

ITC is a quantitative method to characterize the binding thermodynamics for any two interacting molecules. This label-free approach offers a full characterization of the enthalpy (ΔH°), entropy (ΔS°), and free-energy (ΔG°) changes, as well as the equilibrium association constant, K a ; note K a equals 1/K d (the equilibrium dissociation constant). Although the approach can require large amounts of material, it is operative over a wide range of affinities between 1 mM and 1 nM. This range, however, is dictated by the respective ligand and receptor solubilities (i.e. low-affinity interactions), or what is detectable in terms of the measured heats of injection (i.e. very high-affinity interactions). Several key experimental design aspects should be considered as a preface to conducting ITC. The most important of these is the shape of the binding isotherm, which is determined by the unitless parameter c, where c = (([M] × n) ÷ K d). The c value is determined by K d, the concentration of the molecule [M] in the sample cell, and the stoichiometry (n) of binding . In practice, one should strive for 20 < c < 100 (Myszka et al. 2003) to generate measurements at the bottom, middle (inflection), and top of the titration curve; the overall shape for single-site binding should be sigmoidal. Populating the curve with data points is especially important near the inflection region since these observations are needed to accurately fit the K a. Because we typically have an estimate of the K d for the interaction in question, we employ the equation: [M] = (c × K d) ÷ n, which allows determination of the necessary molecular concentration in the sample cell. A good starting point for the titrant concentration in the syringe is 7–10 times the value of [M]. Upon collecting preliminary data, this relationship should be adjusted to reach saturation. Achieving a regular baseline after each injection is important. The heat of ligand dilution following saturation may be accounted for by subtracting the average of the last three points of the saturated thermogram from the total heat of each prior injection. This will provide greater accuracy in the integrated heats that are used in nonlinear least squares curve fitting to obtain the values of ΔH° , n and K a. Accurate knowledge of the concentrations of the molecules under investigation is imperative.

2.3 General Notes About ITC Instrument Settings

The following settings apply to a MicroCal VP-ITC system (GE Healthcare). If it has not been filled recently, or if prior experiments were run at elevated temperatures, the reference cell should be emptied and refilled with degassed, distilled H2O. We begin investigating most binding interactions at a cell temperature of 30 °C; the temperature can be adjusted on a case-by-case basis to alter ΔH (and raw heats) to optimize the binding isotherm. A typical experiment begins with ~30 injections from a 250 μl syringe with an initial injection of 3 μL (discarded during fitting). The remaining injections are 10 μL with 20 s duration per injection. The reference power is 15 μcal s−1, stirring speed 307 rpm, and the feedback mode/gain is ‘high’. These settings are suitable to interrogate most macromolecular interactions (Suddala et al. 2013; Liberman et al. 2013; Salter et al. 2012). The time between injections should be sufficient for the interaction heat to return to baseline before the next injection begins; 240 s is a reasonable starting point but can be adjusted as needed. One should also strive for an average integrated heat of 3–5 μcal evolved (or absorbed) per injection for the first two-thirds of the injections. The sensitivity of the VP-ITC is about 0.1 μcal.

2.4 Preparation of Samples for ITC

The Tte preQ1-I riboswitch 33-mer (Fig. 1b) was produced by chemical synthesis and HPLC purified by C18 reverse-phase separation (Wedekind and McKay 2000; Spitale et al. 2009; Spitale and Wedekind 2009). The Lra preQ1-II riboswitch 77-mer (Fig. 1c) was generated by in vitro transcription and purified by denaturing PAGE (Lippa et al. 2012; Liberman et al. 2013). The preQ1 ligand used for ITC analysis was prepared by chemical synthesis (LeadGen Labs, LLC). Lyophilized RNA samples were stored at −20 °C until needed. Samples were dissolved in 0.050 M Na-HEPES pH 7.0 containing 0.10 M NaCl. For MgCl2-free conditions 0.5 mM EDTA at pH 7.0 was included in lieu of multivalent ions. The RNA solution was heated to 65 °C for 5 min, then MgCl2 was added slowly to 6.0 mM or Co(NH3)6Cl3 was added to 1.0 mM, followed by slow cooling to 24 °C. MgCl2, and Co(NH3)6Cl3 were omitted completely for folding conditions containing EDTA. Each riboswitch sample was dialyzed overnight at 4 °C against 4 L of 0.10 M NaCl with 6.0 mM MgCl2, 1 mM Co(NH3)6Cl3, or 0.5 mM EDTA, buffered at pH 7.0 by 0.050 M Na-HEPES. Following dialysis, samples were diluted with dialysis buffer to: 4.7 μM for the preQ1-I and 3.3 μM for preQ1-II wild-type riboswitches; 5.7 μM for the preQ1-I and 1.5–5.7 μM for wild-type sequences folded in 0.5 mM EDTA; and 3.6 μM for wild-type preQ1-II with Co(NH3)6Cl3. PreQ1 was dissolved in dialysis buffer to a concentration tenfold higher than the RNA.

ITC measurements were conducted by syringe injection of preQ1 into a sample cell containing the riboswitch (cell volume ~1.7 mL) using 28 or 29 injections of 10 μL each—except for the first 3 μL injection. An interval of 120 or 260 s was used between injections. Thermograms were analyzed with Origin 7.0 (MicroCal) using a 1:1 binding model. Experiments for publication should be performed at least twice. Representative titrations and curve fits are in Fig. 2.

Fig. 2
figure 2

Representative ITC titrations and curve fits for ligand binding to preQ1-I and preQ1-II riboswitches in the presence of EDTA or various Mg2+ or Co(NH3) 3 +6 levels. (a) PreQ1 interaction with the Tte 33-mer (class 1) wild-type sequence in 0.10 M NaCl, 0.006 M MgCl2, and 0.050 M HEPES pH 7.0 at 25 °C; the c value was 522. (b) PreQ1 binding to the Tte 33-mer under conditions in (a) except that MgCl2 was replaced with 0.5 mM EDTA; the c value was 228. (c) PreQ1 binding to the wild-type Lra 77-mer under conditions in (a); the c value was 184. (d) PreQ1 interaction with the wild-type Lra 77-mer under conditions described in (b); the c value was 63. (e) PreQ1 binding to the wild-type Lra riboswitch under conditions in (a) except that MgCl2 was replaced with 0.001 M Co(NH3)6Cl3; the c value was 101. In cases where c > 100, such values are a compromise because reduced RNA concentrations gave low heats of injection. The resulting K d values should be treated circumspectly. The average values for two or more measurements are reported in the main text

Full size image

2.5 Outcomes for Metal Dependence of Ligand Binding Based on ITC

Previously we demonstrated that the wild-type Tte preQ1-I riboswitch binds preQ1 with an affinity of 2.1 ± 0.3 nM using surface plasmon resonance (SPR). This method required 5′-biotinylation of the RNA aptamer for immobilization to an SPR chip (Jenkins et al. 2011). SPR was conducted in 0.010 M Na-cacodylate pH 7.0 containing 3 mM MgCl2 at 25 °C. A comparable K d of 7.4 ± 2.3 nM was measured at pH 7.0 in 6 mM MgCl2 with an average n value of 0.98 ± 0.07 (Fig. 2a and Suddala et al. 2013). To test the metal-ion dependence of ligand binding , the Tte RNA was folded in 0.5 mM EDTA in place of metals. The resulting K d of 25.0 nM (Fig. 2b) suggested that MgCl2 was not required for ligand binding but the relatively low n value of 0.7 suggested that a significant portion of the riboswitch does not adopt a binding-competent conformation. A similar trend was observed for the Lra preQ1-II riboswitch in which average K d values for preQ1 binding were 17.9 ± 0.6 nM and 72.5 ± 6.3 nM in the presence and absence of Mg2+, respectively (Fig. 2c, d and Liberman et al. 2013). Despite a modest factor-of-four decrease in affinity, the average n value in the presence of EDTA was 0.57 ± 0.05, suggesting that slightly more than half of the riboswitch adopts a preQ1 binding-competent state. In this manner, the ion dependence of ligand binding can be explored, as well as the need for inner-sphere or outer-sphere interactions (reviewed in Wedekind 2011). For example, the crystal structure showed Mg2+ binding by the preQ1-II riboswitch via inner- and outer-sphere contacts (Liberman et al. 2013). However, ITC measurements indicated Co(NH3) 3 +6 can substitute for Mg2+ in ligand binding (Fig. 2e), based on a minimal increase in K d. This result suggests inner-sphere ion coordination is not a strict requirement for preQ1 binding (Liberman et al. 2013).

3 General Approaches to Introduce Heavy Atoms into Riboswitches

3.1 Mother Liquor Manipulations to Promote Heavy Atom Binding

For novel RNA folds, it is often necessary to determine phases for the measured structure factor amplitudes (F obs) by use of experimental approaches. This requires an empirical search for conditions that lead to the identification of one or more “heavy atom ” binding sites that can be used to establish phases for the non-hydrogen RNA atoms (Wedekind and McKay 2000). For the preQ1-I and preQ1-II riboswitches, determination of the respective phases presented unique challenges that required manipulation of the mother liquor to promote heavy atom binding (Spitale and Wedekind 2009; Liberman et al. 2013). PreQ1-I riboswitch crystals grew from high-salt mother liquors containing the negatively charged SO 2 −4 ion , which competes with the RNA for binding Os(NH3) 3 +5 and variants thereof, thus requiring sulfate removal from crystals. By contrast, the poor reproducibility of preQ1-II crystals—approximately one useful crystal for every seven screened for X-ray diffraction—required co-crystallization of Cs+ with the riboswitch in order to determine the structure by use of single-crystal approaches. Here we review techniques that assisted heavy-atom binding and ultimately led to high-resolution structure determinations of these, heretofore unknown, RNA folds.

3.2 Eliminating SO 2 −4 to Promote Os(NH3) 3 +5 Binding

High-salt mother liquors are useful to grow RNA crystals but can predispose subsequent heavy atom-derivative searches to grave failure. Although strategic placement of halogenated pyrimidines can assist with heavy-atom phasing (Golden 2000; Pley et al. 1994; Correll et al. 1997b), there are examples in which high-salt mother liquors have been supplanted by low ionic strength conditions that favor site-specific ion binding due to reduced electrostatic shielding. Such changes can be made at the level of RNA crystal growth or by in situ methods that entail slow replacement of salt within the crystal. In the case of the minimal hairpin ribozyme, crystals could be grown from solutions of 1.8 M (NH4)2SO4 as well as lower ionic strength 0.25 M Li2SO4 supplemented with 22–24 % (w/v) poly(ethylene) glycol 2000 monomethyl ether (Alam et al. 2005).

By contrast, useful preQ1-I riboswitch crystals grew only from solutions of 1.8 M Li2SO4 (Spitale et al. 2009). Although a well diffracting, low-ionic-strength crystal form grew from 2-methyl-2,4-pentanediol (Lippa et al. 2012), merohedral twinning thwarted phasing attempts (Torelli 2008). Efforts to prepare heavy-atom derivatives of the Li2SO4 crystal form using halogenated pyrimidines and heavy-atom soaks proved unsuccessful. To overcome the limited solubility and poor reactivity of heavy-atom compounds in the high-salt mother liquor, crystals were first placed into a “stabilization” solution comprising 2.0 M Li2SO4, 0.10 M Na-cacodylate pH 6.0, 0.02 M MgSO4, 5 % (v/v) 1,3-propanediol, 2 mM spermine, and saturating amounts of ligand. Crystals were then serially transfer red into a “derivative” solution of 4.0 M LiOAc , 0.02 M Mg(OAc)2, 0.10 M Na-cacodylate pH 6.5, 5 % (v/v) 1,3-propanediol, 2 mM spermine, saturating ligand, and 0.10 M pentaammine-(trifluoro-methane-sulfonato) Os(III) triflate (Sigma-Aldrich). Our rationale was that we previously grew hammerhead ribozyme crystals from 1.9 to 2.2 M NH4SO4 as well as 3.2 M KOAc (Wedekind and McKay 2000), suggesting that both anions were compatible with crystals. Additionally, sulfate adopts multiple modes of metal coordination via uni- or bidentate interactions (Cotton and Wilkinson 1988) and is known to produce insoluble group II and transition-metal complexes. By contrast, acetate coordinates unidentately making it a relatively weak metal binder.

During the exchange of SO 2 −4 with OAc−, we maintained constant ionic strength by assuming 4.0 M LiOAc was equivalent to 2.0 M Li2(SO4). The exchange solutions were mixed in the following derivative-to-stabilizing solution ratios: 1:3, 1:1, and 3:1; each transfer lasted 15 min and crystals showed no signs of osmotic shock. After the final crystal transfer into the derivative solution, crystals were incubated for 3 h. High acetate concentrations also serve as a cryoprotectant (Wedekind and McKay 2000), which allowed samples to be flash frozen directly by plunging into N2(l). MAD phasing of preQ1-I crystals has been described elsewhere (Spitale et al. 2009).

3.3 The Mode of Os(NH3) 3 +5 Binding to the PreQ1-I Riboswitch

Inspection of the preQ1-I riboswitch structure revealed the mode of Os(NH3) 3 +5 binding to RNA (Spitale et al. 2009). Anomalous difference Fourier maps revealed two binding sites: a minor site at the sugar edge of Uri21 and a major site on the twofold axis between preQ1-I riboswitch molecules (Fig. 3). The latter doubly weighted peak had a 30 σ height and the ion received coordination from six oxygens. Due to the isotropy of pentaammine binding , it was not possible to discern the identity of the sixth, inner-sphere ligand in the osmium coordination sphere; such spatial averaging is not always the case (Cate and Doudna 1996). Notably, our observations differ from osmium(III) hexammine binding to the P4–P6 domain of the group I intron, which involved binding at tandem G•U wobble base pairs (Cate and Doudna 1996). Similar G•U binding modes were reported for Ir(NH3) 3 +6 in a comprehensive analysis of major-groove ion binding motifs (Keel et al. 2007); at present, it is unclear whether Os(NH3) 3 +5 can bind G•U pairs in the manner documented for hexammine ions. Nonetheless, we favor pentaammine-(trifluoro-methane-sulfonato) Os(III) triflate as a first choice RNA derivative because it is site selective, commercially available, does not require G•U pairs in the sequence, and circumvents the need for in-house synthesis of Os(NH3) 3 +6 or Ir(NH3) 3 +6 (reviewed in Wedekind 2011).

Fig. 3
figure 3

Site-bound Os(III) at the dyad interface of the Tte preQ1-I riboswitch. The outer-sphere Os(NH3) 3 +5 amines bind to the riboswitch via interactions with the O2-keto and O2′ sugar of Cyt1, as well as O4 keto contacts from Uri21 (red model). Because the Os(NH3) 3 +5 resides on a proper twofold axis, these interaction are duplicated by a symmetry-related molecule (blue model) denoted by a prime (′). The anomalous difference electron density map (orange mesh) is contoured at 10σ. (Inset) The average octahedral geometry for Os(NH3) 3 +6 (reviewed in Wedekind 2011)

Full size image

3.4 Co-crystallization with Cs+ for Single-Wavelength Anomalous Diffraction Phasing

To identify crystals of the preQ1-II riboswitch, we screened sequences from several species including: Streptococcus gordonii, S. pneumoniae, Lactobacillus casei, and L. rhamnosus. The best X-ray-diffraction-quality crystals grew from a modified variant of the L. rhamnosus sequence (Lippa et al. 2012; Liberman et al. 2013) in which nonconserved stem loops in P1 and P4 were replaced with stable GNRA and UNCG tetraloops , respectively (Fig. 1c, construct #1) (Liberman et al. 2013). In another variation, P1 was modified to include a G•U wobble (Fig. 1c, construct #2) that was shown to function as a phasing module due to its avidity for Os(NH3) 3 +6 , Ir(NH3) 3 +6 or Cs+ (Keel et al. 2007). Constructs #1 and #2 yielded crystals by the vapor-diffusion method from solutions of: 14.4–14.8 % (w/v) poly(ethylene) glycol 6000 (PEG6K), 0.14–0.16 M MgOAc2, 0.05 M Na-cacodylate pH 6.0, 1–2 mM spermine, and 0.20 M NaCl. Crystals appeared at 20 °C and grew as rectangular plates to 0.20 mm × 0.05 mm × 0.02 mm within 2–3 weeks. Cryo-protection was achieved by a 30 s transfer into synthetic mother liquor comprising: 17.2–17.8 % (w/v) PEG6K, 168-192 mM MgOAc2, 0.24 M NaCl, 0.06 M Na-cacodylate pH 6.0, and 1.2 mM spermine supplemented with 20 % (v/v) 2-methyl-2,4-pentanediol and 20 % (v/v) ethanol. The best crystals diffracted in house to 2.8 Å resolution and were stored in N2(l) dewars prior to X-ray diffraction data collection. Diffraction analysis revealed that construct #1 belonged to space group C2221 with one molecule per asymmetric unit (a = 58.0 Å, b = 86.0 Å, c = 98.1 Å and α = β = γ = 90°) (Liberman et al. 2013). Construct #2 crystallized in space group P21 with four molecules per asymmetric unit (a = 83.6 Å, b = 60.7 Å, c = 101.5 Å and α = γ = 90° with β = 106.8°) (Liberman, Salim and Wedekind, unpublished).

Although substantial effort was expended to prepare heavy-atom derivatives, both crystal forms were recalcitrant to derivatization. One obstacle was that only one in seven crystals from the same drop exhibited useful X-ray diffraction, which hindered efforts to generate a starting supply of crystals that could be subjected to heavy-atom screening. Another shortcoming was a paucity of Ir(NH3) 3 +6 binding based on the absence of a measurable anomalous signal. The latter phenomenon was most likely the result of elevated Mg2+ concentrations in mother liquors that were necessary to obtain high-quality X-ray diffraction. Even though construct #2 possessed a G•U wobble, it had no appreciable Ir(NH3) 3 +6 binding.

To surmount difficulties, we undertook co-crystallization in the presence of CsCl in lieu of NaCl. Substitution of Na+ with Cs+ caused no apparent changes in space group despite notable differences between these ions. Cs+ is electron rich (54 e−) relative to Na+ (10 e−) and forms as many as 11 inner-sphere contacts. A search of the Cambridge Structural Database (Allen 2002) version 5.34 using ConQuest v1.15 (Bruno et al. 2002) revealed an average Cs+ to O–C distance of 3.23 ± 0.16 Å, an average Cs+ to N–C distance of 3.31 ± 0.15 Å, (where the O–C and N–C bond was of any type), and an average Cs+ to water distance of 3.27 ± 0.16 Å. Such distances are significantly longer than those for Na+ to O–C, as well as K+ to O–C, which exhibited values of 2.53 ± 0.17 and 2.83 ± 0.08 Å, respectively.

The presence of Cs+ improved the resolution of X-ray diffraction for preQ1-II crystals, as well as the reproducibility of well-diffracting crystals. Of particular significance is the fact that Cs+ exhibits a +9 electron anomalous-diffraction signal (f″) at 1.70 Å resolution, which enabled use of the site-bound Cs+ ions of construct #1 as a means to generate high-quality SAD phases for a complete structure determination (Liberman et al. 2013). By comparison, the low-symmetry space group of construct #2 crystals hampered collection of the anomalous signal, preventing an independent structure determination. Nonetheless, the latter structure was solved by molecular replacement, which yielded the locations of a number of Cs+ ions whose identities were corroborated by anomalous difference Fourier maps (Liberman, Salim and Wedekind, unpublished). We also conducted co-crystallization of the Lra preQ1-II riboswitch from Br− and I− salts in lieu of Cl−. The latter ions produced well diffracting crystals using Na+ as a counter ion, but no significant anomalous diffraction signal was detected. Our results suggest that halide salts are not useful for RNA phasing , in contrast to proteins (Dauter and Dauter 2001).

3.5 The Binding Mode of Cs+ to a G•U Wobble Pair and Other Sites

We included a single-site G•U wobble pair in stem P1 of the preQ1-II riboswitch with the goal of using it as a phasing module (Fig. 1c, construct #2). This approach was described in a thorough analysis on the propensity of single and tandem G•U wobble sites to bind Os(NH3) 3 +6 or Ir(NH3) 3 +6 for heavy-atom phasing (Keel et al. 2007). We chose the sequence 5′-GGC-3′/3′-CUG-5′, which features a single-wobble pair, dubbed phasing module 14 (PM14) (Keel et al. 2007). PM14 was stated to have advantages over other G•U motifs including (1) metal binding with one of the highest anomalous signals and lowest B-factors relative to a reference hexammine site and (2) inclusion of a single wobble requires only three nucleotides to be altered in the target RNA sequence rather than four (Keel et al. 2007). It was also noted that the G•U wobble of PM14 binds Cs+, which proved useful in phasing the SAM-II riboswitch wherein NH4 + in the mother liquor was replaced with Cs+, introduced by crystal soaking (Keel et al. 2007; Gilbert et al. 2008). As noted above, the preQ1-II riboswitch was co-crystallized with 0.15 M Cs+ yielding 15 site-bound Cs+ ions in construct #1 (one molecule per asymmetric unit) (Liberman et al. 2013) and 35 Cs+ ions in construct #2 (four molecules per asymmetric unit) (Liberman, Salim and Wedekind, unpublished).

The strongest anomalous difference Fourier signal in electron density maps was observed for Cs+ bound at the G•U wobble of construct #2; the average peak height was 21 ± 2 σ δ for all four riboswitch copies in the asymmetric unit. Because construct #1 did not possess such a phasing module (Fig. 1c), Cs+ was not detected at a spatially equivalent location. Cs+ coordination at the G•U wobble site of construct #2 entailed contacts in the P1 major groove (Fig. 4a), which concur with the mode of Cs+ coordination at PM14 in the SAM-II riboswitch (Gilbert et al. 2008). A variety of other single G•U wobble pairs have been reported to bind Cs+ including: 5′-GGA-3′/3′-CUG-5′ in the lysine riboswitch (Serganov et al. 2008); 5′-GGA-3′/3′-CUU-5′ in the FMN riboswitch (Serganov et al. 2009); as well as 5′-GGU-3′/3′-CUA-5′ and 5′-CGU-3′/3′-GUA-5′ from the O. iheyensis group IIc intron (Marcia and Pyle 2012). Each of the latter, site-bound ions was introduced by soaking crystals in solutions of Cs+. (Wobble sites engaged in tertiary or crystal contacts were not considered in the sequences above).

Fig. 4
figure 4

Representative site-bound Cs+ ions in the Lra preQ1-II riboswitch G•U wobble, GNRA tetraloop and secondary structures with accompanying anomalous difference Fourier maps. (a) The mode of G•U wobble coordination to Cs+ in construct #2 entails inner-sphere contacts with the keto oxygens of Gua5, 6 and 14, and Uri15 in the P1 major groove. Here and elsewhere the secondary structure is shown; asterisks (*) indicate base keto-oxygen coordination to Cs+. (b) GNRA tetraloop coordination to Cs+ utilizes N7 of Ade10, O3′ of Gua9, and nonbridging phosphate oxygens from Ade10 and Uri55′; the latter interaction arises from a crystal contact (blue ball-and-stick-model). A nearby Mg2+ shows inner-sphere coordination to the nonbridging phosphate oxygen from Ade10. (c) Major groove Cs+ binding by tandem Gua bases. The O6 groups of Gua58 and Gua59 make inner-sphere Cs+ contacts whereas N7 of Gua58 and the O4 keto of Uri57 serve to position waters for ion coordination. (d) Minor groove Cs+ binding at the sugar-edge of Cyt5. Coordination entails the O2 keto and 2′-O groups of Cyt5 and the cross-strand O2 keto group of Uri17

Full size image

The low-symmetry P21 space group of our preQ1-II riboswitch (construct #2) contains four molecules per asymmetric unit, which impeded high-resolution SAD phasing. As such, we focused on the higher symmetry C2221 space group of construct #1, which proved successful despite the absence of a G•U pair for ion binding. Serendipitous Cs+ binding prompted us to ask (1) what sequences common to both preQ 1 -II constructs exhibited appreciable levels of Cs + binding based on the measurable anomalous signal? and (2) which of the latter Cs + sites are present in secondary structure that does not strictly depend upon tertiary or crystal contacts for ion binding? Of the 15 Cs+ sites identified and modeled in the structure of construct #1, the average anomalous peak height was 8.9 σ and the average refined B-factor was 43 Å2. Our results revealed the highest occupancy and largest anomalous signal for Cs+ at Ade10, which resides in apposition to the phosphodiester bond between Gua9 and Ade10 within the GNRA tetraloop that caps P1 (Fig. 4b). This site has an average peak height of 17.4 ± 3.5 σ based on all five molecules of constructs #1 and #2. The strong occupancy is likely derived from four inner-sphere contacts originating from three GNRA ligands, as well as a nonbridging phosphate oxygen donated by a crystal contact (Fig. 4b). A nearby Mg2+ coordinates to the backbone of Ade10, which simultaneously makes an inner-sphere contact to the GNRA-bound Cs+. Mg2+ aside, a similar constellation of Cs+ ligands was observed in the GNRA tetraloop of the O. iheyensis group IIc intron, including N7 of base Ade33 in the “N” position of the tetraloop (Marcia and Pyle 2012). Like the preQ1-II riboswitch, the coordination sphere of this ion is completed by inner-sphere coordination to a crystal contact, which draws into question the generality of the GAAA motif as a “sure-fire” Cs+-binding site.

Two other Cs+ sites were prominent in preQ1-II structures of constructs #1 and #2. Both sites are located in helical regions devoid of tertiary or crystal contacts. The first site was bound in the major groove of P4 at the sequence 5′-UGG-3′/3′-ACC-5′. The average height of the anomalous difference Fourier peak was 4.9 ± 2.8 σ. Here, tandem Gua bases appear important in Cs+ binding (Fig. 4c). A comparable mode of coordination is observed at the base of a GNRA tetraloop in the O. iheyensis group IIc intron in the sequence 5′-UGG-3′/3′-ACA-5′, wherein the G•A pair corresponds to the first and last residues of a GNRA tetraloop. Specifically, Gua275 at position i of the tetraloop forms a sheared pair with Ade278 at position i + 3. The keto oxygens of the sequence 5′-UGG-3′ each make inner-sphere contacts to Cs+ (Marcia and Pyle 2012).

The involvement of O6 groups from tandem guanines to coordinate Cs+ has also been observed at Gua29/30 of the O. iheyensis group IIc intron (Marcia and Pyle 2012). Other tandem Gua sequences are present but these make prominent use of tertiary contacts for ion binding. Tandem Gua nucleobases—specifically tandem G•U wobbles—have long been known as hotspots for hexammine binding (Cate and Doudna 1996; Keel et al. 2007). Moreover, tandem Gua bases in RNA duplexes—outside the context of wobbles—are known to be sites of Mg(H2O) 2 +6 coordination (Correll et al. 1997a; Jovine et al. 2000; Shi and Moore 2000; Wedekind and McKay 2003). Our work and that of others suggests that such sequences bind Cs+, although binding is not necessarily mutually exclusive with respect to Mg(H2O) 2 +6 based on respective O. iheyensis group IIc intron structures soaked in Li+/Mg2+ and Cs+/Mg2+ (Marcia and Pyle 2012).

A second sequence-specific Cs+ binding site was located in the minor groove of preQ1-II constructs #1 and #2 (Fig. 4d). This site entailed sugar-edge binding at Cyt5 and Cyt16, respectively. In construct #1, Cs+ bound to Cyt5 with an anomalous difference Fourier peak height of 9.9 σ. In construct #2, Cs+ bound to Cyt16 in all four copies in the asymmetric unit with an average peak height of 4.5 ± 1.8 σ. Differences in Cs+ localization were the result of sequence differences resulting from use of the PM14 G•U wobble (Fig. 1c). The site of Cs+ binding in construct #1 is 5′-AC-3′/3′-UG-5′, whereas that of construct #2 is 5′-AG-3′/3′-UC-5′. At present, it is unclear what features surrounding the respective Cyt nucleobases are important for Cs+ localization since not all accessible 5′-AC-3′ and 5′-CU-3′ sequences in preQ1-II riboswitch structures coordinate Cs+. Higher resolution diffraction is needed to clarify this issue.