An in vitro evolved glmS ribozyme has the wildtype fold but loses coenzyme dependence

Abstract

Uniquely among known ribozymes, the glmS ribozyme-riboswitch requires a small-molecule coenzyme, glucosamine-6-phosphate (GlcN6P). Although consistent with its gene-regulatory function, use of GlcN6P is unexpected because all other characterized self-cleaving ribozymes employ RNA functional groups or divalent cations for catalysis. To determine what active site features make this ribozyme reliant on GlcN6P, and to evaluate whether it might have evolved from a coenzyme-independent ancestor, we isolated a GlcN6P-independent variant through in vitro selection. Three active site mutations suffice to generate a highly reactive RNA that adopts the wildtype fold but employs divalent cations for catalysis and is insensitive to GlcN6P. Biochemical and crystallographic comparisons of wildtype and mutant ribozymes show that a handful of functional groups fine-tune the RNA to be either coenzyme- or cation-dependent. These results indicate that a few mutations can confer novel biochemical activities on structured RNAs. Thus, families of structurally related ribozymes with divergent function may exist.

INTRODUCTION

The glmS ribozyme-riboswitch^1,2 controls expression of the mRNA encoding glucosamine-6-phosphate (GlcN6P) synthetase through negative feedback³ by site-specifically cleaving it in cis in response to elevated intracellular GlcN6P. GlcN6P accelerates the rate of ribozyme cleavage by ~10⁶, and functions as a coenzyme^4–6. Although consistent with its riboswitch⁷ function, use of GlcN6P by this ribozyme is unexpected because it is the only known natural catalytic RNA that depends on an exogenous small molecule coenzyme. Ribozymes that catalyze sequence-specific RNA cleavage through internal transesterification are widespread. In addition to the glmS ribozyme-riboswitch, which occurs throughout Gram-positive bacteria⁸, hammerhead and HDV-like ribozymes have been found in most animal phyla, plants, fungi, bacteria, and even phage^9,10. Such ribozymes have also been repeatedly isolated by in vitro selection, indicating that they are common in RNA sequence space^11–13. Except for the glmS ribozyme, however, none of the known nucleolytic ribozymes depend on a coenzyme. All others employ RNA functional groups, divalent cations, or both, for catalysis^14–16.

What is the structural basis of the unique coenzyme dependence of the glmS ribozyme? In physiologic Mg²⁺ concentrations, the ribozyme adopts its active conformation independent of GlcN6P (ref. 4,17), and undergoes no conformational changes as it traverses its reaction coordinate. The RNA is prefolded and poised to react upon binding GlcN6P^4–6. Four strategies are available to catalyze transesterification (reviewed in ref. 18). These are in-line positioning of the 2′-oxygen nucleophile, phosphorus electrophile, and the 5′-oxo leaving group as required for the S_N2 reaction; deprotonation of the nucleophile; protonation of the leaving group; and electrostatic stabilization of the transition state (Fig. 1a). Coenzyme-independent RNAs employ multiple strategies. The hairpin ribozyme (Fig. 1b) aligns the reactive groups in the ground state and it preferentially hydrogen bonds to the transition state^19,20. This ribozyme exhibits a bell-shaped pH-rate profile consistent with two rate-limiting ionizations²¹, and the two nucleobases that flank the scissile phosphate likely carry out general acid-base catalysis^19,21–23. The hammerhead ribozyme (Fig. 1c) also aligns its substrate and appears to employ the base of G12 and the 2′-OH of G8 as general base and acid catalysts, respectively²⁴. The GlcN6P-activated glmS ribozyme (Fig. 1d) exhibits¹ a log-linear pH-rate dependence (consistent with a single rate-limiting ionization) with a pK_a of ~7.0 that matches closely the microscopic pK_a of the amine^25–27 of the bound GlcN6P. Although an invariant nucleotide is positioned to hydrogen bond to the nucleophilic 2′-oxygen, analog studies²⁸ indicate that this G40 has an unperturbed pK_a of ~9.5. Thus, the glmS ribozyme appears to rely primarily on its coenzyme to achieve catalysis.

Figure 1. Comparison of the pre-cleavage active site structures and catalytic strategies of three natural nucleolytic ribozymes.

(a) Catalytic mechanisms¹⁸ for RNA transesterification: green, in-line arrangement of the functional groups; blue, deprotonation of the 2′-hydroxyl nucleophile; magenta and red, protonation or neutralization of negative charge on the NBPO or the 5′-oxygen, respectively. (b) Active site of the hairpin ribozyme^19,20 in complex with a non-cleavable substrate strand containing a single 2′-OMe substitution (oxygen, red; methyl group, magenta) on A(−1). Asterisk marks the scissile phosphate (CPK colors). Dashed lines depict hydrogen bonds. Potential general base and acid moieties are blue and red, respectively. Note near-in-line arrangement of the 2′ oxygen of A(−1), phosphorus, and 5′ oxo leaving group. (c) Active site of the hammerhead ribozyme in complex with a 2′-OMe containing substrate²⁴, colored as in (b). (d) Active site of the glmS ribozyme^4,41 in complex with a 2′-OMe containing substrate, colored as in (b). It is noteworthy that of all nucleolytic ribozymes of known structure, the glmS ribozyme is the only one that hydrogen bonds to both NBPOs of the scissile phosphate (employing G39 and G65) in the ground state.

The Bacillus cereus glmS ribozyme was previously subjected to in vitro selection for sequences that self-cleaved in the presence of mixtures of GlcN6P and several analogs²⁹. The resulting ribozymes were found to be maximally active with GlcN6P, but no faster than wildtype, suggesting that the structural framework of this RNA is optimized for GlcN6P utilization, and that the glmS ribozyme cannot readily evolve to employ other cofactors.

Discovery of the glmS ribozyme-riboswitch demonstrated how Gram-positive bacteria regulate the essential metabolite GlcN6P, and defined a new RNA target for antibiotic development^1,30,31. The mechanism of this RNA also suggested that ribozymes in the RNA world could have expanded their chemical versatility by employing coenzymes³². If they are common in sequence space, coenzyme-dependent RNAs other than the glmS ribozyme may await discovery. On the other hand, if the ligand-dependence of the glmS ribozyme results from an unusual architecture, coenzyme utilization by catalytic RNAs may be rare. To explore the evolutionary origin of the glmS ribozyme and its GlcN6P dependence, we subjected the RNA to in vitro selection for activity in the absence of GlcN6P, discovering a variant with just three mutations that is active in the presence of divalent cations alone. The mutant adopts the same fold as the wildtype and is a true metalloenzyme. Importantly, we find that the identities of a handful of active-site nucleotides are responsible for either enabling coenzyme utilization or allowing the ribozyme to function in a GlcN6P-independent manner. In addition to suggesting that the glmS ribozyme might have evolved from a coenzyme-independent ancestor, our results hint that acquisition of new functions by stable RNA folds is facile, and families of RNAs that share the same fold but have different functions probably exist in nature.

RESULTS

A GlcN6P-independent glmS ribozyme variant

We mutagenized nucleotides near the active site of wildtype glmS ribozyme from Thermoanaerobacter tengcongensis (hereafter glmS^WT) by 30% per position (Supplementary Results, Supplementary Fig. 1a) generating a starting pool of ~7.5 × 10¹⁴ variants. Unlike previous work in which the entire ribozyme was mutagenized at 9% per position²⁹, we did not alter nucleotides that define the triply pseudoknotted⁴ glmS ribozyme fold. This allowed mutagenesis of active site-residues at a higher rate. RNAs were 5′ end-labeled with 6-thioguanosine (6SG), and species capable of self-cleavage (thereby releasing the 6SG) in the absence of GlcN6P were separated from inactive sequences employing electrophoresis through mercury-containing polyacrylamide gels³³ (Online Methods and Supplementary Fig. 1b). After six rounds of selection, GlcN6P-independent, site-specific cleavage was detectable (Supplementary Fig. 2). Sequencing after ten rounds revealed three mutations (U49A, U51A and G65A) in all sequences resembling glmS^WT (Supplementary Fig. 3). A ribozyme containing just these three mutations (hereafter glmS^AAA) is active in Mg²⁺. Importantly, activity of glmS^AAA is not enhanced by high concentrations of GlcN6P that would saturate²⁶ glmS^WT (Fig. 2a and Supplementary Fig. 4).

Figure 2. glmS^AAA is a stably-folded metal-ion dependent ribozyme.

(a) Metal ion dependence of glmS^WT and glmS^AAA activity. Denaturing PAGE analysis of the cleavage of a 5′ end-labeled, 10 nt substrate at a 10 minute time-point. Rates below the autoradiogram are from triplicate full reaction profiles (Online Methods). Reactions contained 100 mM metal ions, except (*) cobalt hexammine (CoH), which was at 5 mM, or monovalent metal ions, which were at 3 M. Where present, GlcN6P was at 5 mM. (†)ND, not determined. The higher activity of glmS^AAA in Ca²⁺ over Mg²⁺ or Mn²⁺ does not indicate a preference for larger cations (ionic radii are 1.12, 0.72, and 0.67 Å, for Ca²⁺, Mg²⁺ and Mn²⁺, respectively, ref. 35), since the mutant ribozyme is only modestly active in Sr²⁺ (ionic radius 1.18 Å) and inactive in Ba²⁺ (ionic radius 1.35 Å). (b) Activity of glmS^WT (in 10 mM GlcN6P) in Mg²⁺ (K_1/2 = 24 mM;k_max = 62 min⁻¹) and Ca²⁺ (K_1/2 = 31 mM;k_max = 49 min⁻¹). (c) Activity of glmS^AAA (without GlcN6P) in Mg²⁺ (K_1/2 = 80 mM;k_max = 0.072 min⁻¹) and Ca²⁺ (K_1/2 = 93 mM;k_max = 3.4 min⁻¹). In all figures, means±standard error from at least 3 independent experiments are shown. (d) Three orthogonal views of a superposition (r.m.s.d. = 0.6 Å) of the crystal structures of glmS^WT (ref. 42) and glmS^AAA (in 50 mM MgCl₂). Throughout, glmS^WT and glmS^AAA are colored gray and cyan, respectively.

Like, glmS^WT, glmS^AAA generates products with 2′,3′-cyclic phosphate and 5′-OH termini (Supplementary Fig. 5). The metal ion requirements of the two ribozymes differ considerably (Fig. 2a). When employing GlcN6P, glmS^WT is basically indifferent to the nature of divalent cations, and is active in cobalt hexammine^34,35. Cobalt hexammine is a kinetically inert isoster of hexahydrated Mg²⁺ (ref. 36). Activity in this complex ion indicates that glmS^WT does not require inner-sphere cation coordination for function with GlcN6P. Although glmS^AAA was selected in Mg²⁺, it is equally active in Mn²⁺, and ~40 fold more active in Ca²⁺ (Fig. 2a). Its cleavage rate in Ca²⁺ (1.7 min⁻¹) approaches those of many natural self-cleaving ribozymes, and is only ~30-fold slower than the GlcN6P dependent activity of glmS^WT. Unlike glmS^WT, glmS^AAA is inactive in even molar concentrations of most monovalent metal ions, with the exception of Li⁺, which supports residual activity (Fig. 2a and Supplementary Fig. 6). Notably, glmS^AAA is inactive in cobalt hexammine, indicating a requirement for inner-sphere cation coordination.

The mutant ribozyme adopts the wildtype fold

Compared to the GlcN6P-catalyzed reaction of glmS^WT, the metal ion-dependent activity of glmS^AAA requires higher concentrations of divalent cations (Fig. 2b,c). Since divalent cations stabilize the active folds of RNAs³⁷, the elevated cation requirement of glmS^AAA could simply reflect global structural destabilization. We compared glmS^WT and glmS^AAA in solution by small-angle X-ray scattering (SAXS). Even at 2 mM Mg²⁺, well below K_1/2 for either RNA but approximately physiologic³⁸, the two RNAs exhibit near-identical radii of gyration (29.3 and 29.4 Å, respectively), and Kratky plots indicative of folded RNAs (Supplementary Fig. 7a). The similarity of their P(r) plots (Supplementary Fig. 7b) suggests that glmS^WT and glmS^AAA have similar structures. We determined crystal structures of glmS^AAA in the presence of Mg²⁺ or Ca²⁺ at 3.1 Å resolution (Online Methods). Consistent with SAXS, the structures of the cores of glmS^AAA and glmS^WT superimpose closely (Fig. 2d), with a root-mean-square difference of 0.65 Å, comparable to the precision of the atomic coordinates (Supplementary Table 1). There are modest conformational differences in the RNA periphery, but these are likely due to subtle crystal packing differences. Our SAXS and crystallographic results suggest that the elevated divalent cation requirement for glmS^AAA activity reflects metal ion participation in chemistry, not folding. If this were the case, it is likely that site-specifically bound metal ions are important for glmS^AAA catalysis, in addition to folding.

A candidate catalytic cation in the mutant ribozyme

Phosphorothioate interference allows location of functionally important metal ions in RNA. Replacement of a non-bridging phosphate oxygen (NBPO) with sulfur reduces the affinity of the phosphate for hard cations (e.g. Mg²⁺ and Ca²⁺), resulting in decreased activity. Such interferences can often be suppressed by thiophilic cations (e.g. Mn²⁺). To locate catalytic cations in glmS^AAA, we performed interference mapping³⁹. Transcripts with random sulfur substitution of pro-R_P NBPOs were separated based on their catalytic activity, and sites of deleterious substitutions identified. As noted previously in studies of the B. cereus ribozyme⁴⁰, the wildtype T. tengcongensis glmS ribozyme, in the presence of GlcN6P, exhibits strong interference at residue 2 (Supplementary Fig. 8a). In addition, glmS^AAA, in the absence of GlcN6P, exhibits a strong interference at residue 38 (Fig. 3a and Supplementary Fig. 8b,c). Although an interference at position 38 was previously reported for the B. cereus ribozyme, the effect for that wildtype RNA was very mild (interference value of 2.1 vs. > 6 at residue 2). In our analysis of glmS^WT, we did not detect significant interference at position 38 (interference value of 0.85). glmS^AAA also exhibits an interference at C60. A Ca²⁺ is near its pro-R_P NBPO. However, this ion is ~12 Å from the scissile phosphate; therefore, it likely plays a structural role.

Figure 3. A cation binding site in glmS^AAA important for catalysis.

(a) Strong phosphorothioate interference is present at position 38 (**) of glmS^AAA but not glmS^WT. Autoradiograms are of the self-cleaved pools of the RNAs. (b) Active sites of glmS^WT (ref. 45) and (c) glmS^AAA (Ca²⁺ crystal structure). Black dashes depict hydrogen bonds (distances ≤ 3.6 Å) and red dashes depict metal ion coordination. Asterisk denotes scissile phosphate. Metal ions M_A, M_B and M_C are in magenta. The glmS^WT structure bears a 2′-deoxy substitution at A(−1). The position of the 2′-OH of that residue (green) is conjectural. Residue A(−1) is disordered in glmS^AAA structures (drawn with thin sticks), and has been modeled based on the glmS^WT structure. (d) Histograms of activities of ribozymes with no phosphorothioates (light brown), or substitutions at the indicated positions of either pro-R_p or pro-S_p NBPOs (yellow and blue, respectively). Activity of a ribozymes with a diastereomeric mixture of phosphorothioates at the indicated position is denoted in green bars. Two and three asterisks denote statistically significant differences (0.001 < p < 0.01 and p < 0.001, respectively; two-tailed t-test).

The strong interferences can be explained by the glmS^WT and glmS^AAA crystal structures. All structures^4,41 of wildtype glmS ribozymes from T. tengcongensis and B. cereus share a single, highly ordered Mg²⁺. This cation (M_A, Fig. 3b) is coordinated by the pro-R_P NBPOs of residues 2, 36 and 37. Strong interference at residue 2 supports the importance of this inner-sphere coordination for wildtype stability and activity. Since M_A is distant from the scissile phosphate, and since wildtype glmS ribozyme catalysis is indifferent to the nature of the divalent cation, M_A is likely to play a structural role. The structure of glmS^AAA reveals a second cation in proximity to the pro-S_P and the pro-R_P NBPOs of residues 2 and 38, respectively (M_B, Fig. 3c). M_B is closer to the scissile phosphate than M_A (6.2 and 9.4Å from the phosphorus, respectively), and therefore might be involved in glmS^AAA catalysis. Although the wildtype B. cereus structures do show occupancy of the M_B site by a magnesium ion, the mild interference at position 38 in glmS ribozymes from that species argue against a prominent role of this metal ion in GlcN6P-stimulated catalysis.

We confirmed the interference mapping results by characterizing synthetic ribozymes with full sulfur replacement of NBPOs at single sites. As expected from the location of M_A, and consistent with the interference mapping result, sulfur substitution of the pro-R_P NBPO of residue 2 impairs both glmS^AAA and glmS^WT (Fig. 3d). Both can be rescued by Mn²⁺ (Supplementary Table 2). Substitution of the pro-S_P NBPO of residue 2 impairs only glmS^AAA. This can also be rescued by Mn²⁺. Because of the internal location of A38, we could only examine the effect of phosphorothioate substitution at this position using a diastereomeric mixture. We find that glmS^AAA with phosphorothioates at this site is markedly inhibited in Ca²⁺ (p=0.0004, two-tailed t-test), and can be rescued by Mn²⁺ (Fig. 3d and Supplementary Table 2). The corresponding NBPO substitution in glmS^WT had no significant effect (p=0.08). Previously, it was found that replacement of the NBPOs of the scissile phosphate of glmS^WT with sulfur only mildly affected ribozyme activity³⁴. Because a Ca²⁺ ion (M_C) coordinates the scissile phosphate in the glmS^AAA crystal structure (3.8Å from the phosphorus, Fig. 3c), we also examined the effect of replacing the scissile phosphate with the two diastereomeric phosphorothioates separately, and found no notable impairment of activity of the mutant ribozyme (Fig. 3d). Together, the results of SAXS, crystallographic and phosphorothioate interference studies suggest that an important part of the metal ion requirement of glmS^AAA arises from the use of M_B for catalysis.

GlcN6P-dependent vs. cation-dependent catalysis

To examine the effects of each of the three mutations carried by glmS^AAA on GlcN6P-dependent and independent catalysis, we characterized ribozymes bearing either the wild-type or the mutant nucleotide, singly or pairwise (Fig. 4a,b and Supplementary Table 3). The three mutations contribute unequally to the response to GlcN6P. In Mg²⁺, G65A reduces activity in the presence of GlcN6P relative to glmS^WT by 422-fold, while U49A and U51A only reduce activation by 2.6 and 12-fold, respectively (in Ca²⁺, the reductions are 148, 2.5 and 5.9-fold, respectively). Analysis of mutants that revert glmS^AAA toward wild-type confirms the importance of residue 65 for GlcN6P-induced activity. A65G produces a 898-fold increase in GlcN6P-dependent activity (over that in Mg²⁺ alone), while A49U and A51U only increase GlcN6P response by 143 and 259-fold, respectively (the corresponding increases in Ca²⁺ are 117, 18 and 9.3-fold, respectively). Overall, nucleotides 49, 51, and 65 all contribute to the ability of glmS^WT to employ GlcN6P, but residue 65 is the most important.

Figure 4. Effect of active site residues in GlcN6P-dependent and metal ion-dependent catalysis.

(a) Cleavage rates of mutant ribozymes containing glmS^WT and glmS^AAA residues at positions 49, 51 and 61 in 100 mM Mg²⁺ or 100 mM Mg²⁺ with 5 mM GlcN6P. The ratio of GlcN6P-dependent to GlcN6P-independent rates for each species is indicated on the right. (b) Cleavage rates of same mutants as in (a) but in 100 mM Ca²⁺ or 100 mM Ca²⁺ with 5 mM GlcN6P. Reaction conditions near-optimal for glmS^WT and glmS^AAA were chosen to facilitate rate measurement for the slower mutants. Means±standard error from at least 3 independent experiments are shown. (c) Superposition of glmS^WT and glmS^AAA crystal structures showing details of glmS^WT GlcN6P binding pocket. Positions of the scissile phosphate (*), sugar and base of A49 and A51, and the nucleobase of A65 in the glmS^AAA Ca²⁺ crystal structure are shown in cyan. M_B from the glmS^AAA Ca²⁺ crystal structure is in magenta.

The increased activation of glmS^WT over glmS^AAA by GlcN6P is due not only to the ability of the former to employ the coenzyme, but also reflects the inability of glmS^WT to use divalent cations (Fig. 4a,b and Supplementary Table 3). In Ca²⁺, glmS^WT is 2788 times less active in GlcN6P-independent catalysis than glmS^AAA (243-fold decrease in Mg²⁺, where glmS^AAA is less active). Analysis of point revertants of glmS^AAA shows that the mutations A49U, A51U and A65G impair metalloribozyme activity by 165, 79, and 85-fold, respectively (decrease in 43, 86 and 71-fold in Mg²⁺, respectively). Nucleotides in the three positions appear to function synergistically in conferring maximal metalloribozyme activity to glmS^AAA, since all three possible double revertants toward wildtype reduced GlcN6P-independent activity to the background (glmS^WT) level.

Structural basis of GlcN6P utilization

Correlation of mutagenic analysis with crystal structures sheds light on the roles of nucleotides 49, 51 and 65. Of the three, U49 is the least conserved, and its role in GlcN6P utilization uncertain. In the T. tengcongensis crystal structures, its phosphate is solvent exposed, and its nucleobase and sugar are disordered. In B. cereus, U49 is deleted making J2/2.1 one nucleotide shorter. U51 and G65 are phylogenetically conserved, and were present in all active variants isolated by a previous in vitro selection experiment²⁹. In glmS^WT, U51 Watson-Crick pairs with A38 (Fig. 4c). In addition, the O4 of U51 hydrogen bonds to the amine of GlcN6P, the proposed general acid-base catalyst of the coenzyme^25–27. In glmS^WT, G65 makes three hydrogen-bonds simultaneously^4,41,42 (Fig. 3b and Fig. 4c). First, it donates a hydrogen bond from N1 to the anomeric oxygen of GlcN6P. Analog studies show that this oxygen and the amine of GlcN6P are the two functional groups of the coenzyme critical for ribozyme binding and activation⁴³. The anomeric hydroxyl (which itself hydrogen bonds to the pro-R_P NBPO of the scissile phosphate) is likely important for binding and precise positioning of the coenzyme. Second, the N2 of G65 donates a hydrogen bond to the pro-R_P NBPO of the scissile phosphate. Third, the same N2 donates a hydrogen bond to N3 of A(−1). The glmS ribozyme has a preference for an adenine at the (−1) position¹ consistent with this interaction.

In the glmS^AAA crystal structures, nucleotide 51 is flipped out of the GlcN6P pocket, thus relieving the steric clash with A38 resulting from the U51A mutation, which would juxtapose the Watson-Crick faces of A38 and A51 (Fig. 4c). When flipped out, residue 51 can no longer contribute to GlcN6P binding. The G65A mutation is highly disruptive because of the loss of the exocyclic N2 of guanine. In both of our glmS^AAA crystal structures, electron density is lacking for the sugar and nucleobase of A(−1), which, in contrast, is well ordered in all wildtype pre-cleavage structures. Since analysis of dissolved glmS^AAA crystals shows that A(−1) is intact (not shown), the lack of electron density indicates that the nucleotide adopts multiple conformations. This confirms that the interaction between G65 and A(−1) is critical in fixing the latter in a specific, reactive conformation. In addition, superposition of the glmS^WT and glmS^AAA structures (Fig. 4c) shows that the nucleobase at position 65 is displaced away from the scissile phosphate in the mutant, emphasizing the importance of the hydrogen bond made between the N2 of the guanine residue of the wildtype and the pro-R_P NBPO of the scissile phosphate. This displacement creates a gap between the Watson-Crick face of residue 65 and the scissile phosphate too wide for the anomeric hydroxyl of GlcN6P to bridge. Overall, the crystal structures indicate how the U51A and G65A mutations present in glmS^AAA lead to abrogation of coenzyme activity, and highlight the importance of the wild-type nucleotides for GlcN6P binding.

Mutant ribozyme activity with divalent cations

The inability of the mutant ribozyme to function in cobalt hexammine (Fig. 2a) indicates a requirement for inner-sphere metal ion coordination. Therefore, it might employ a metal ion-bound hydroxide as a specific base catalyst. To examine this, we determined its pH-rate profile in the presence of Ca²⁺, Mg²⁺, or Mn²⁺ (Fig. 5a). Regardless of the cation, the reaction pK_a is ~5.5, which is considerably lower than the pK_a of the metal ion-bound hydroxides (12.7, 11.4 and 10.6, respectively³⁵). This, and the fact that the pK_a of the glmS^AAA reaction does not parallel the trend in metal ion hydroxide pK_a's, argues against such a catalytic mechanism. Consistent with the lack of GlcN6P utilization by glmS^AAA, its pH-rate profile differs from that of glmS^WT (Fig. 5b) whose reaction pK_a is ~7.0, similar to the microscopic pK_a of the amine of ribozyme-bound GlcN6P^25,26. Our modification interference data demonstrate that a specifically bound divalent cation, M_B is critical for glmS^AAA activity (Fig. 3). Comparison of the structures of the wildtype and mutant ribozymes suggests that M_B may function analogously to GlcN6P in glmS^WT. In glmS^WT (Fig. 3b), the amine of GlcN6P is positioned to protonate the leaving group of the reaction⁴. M_B, while not a general acid, could function in glmS^AAA by neutralizing the charge of the 5′-oxo leaving group (Fig. 3c).

Figure 5. Chemical and structural analyses correlate metal ion utilization by glmS^AAA and glmS^WT.

(a) GlcN6P-independent cleavage rate of glmS^AAA as a function of pH in Mg²⁺, Ca²⁺ and Mn²⁺. (b) GlcN6P-dependent cleavage rate of glmS^WT as a function of pH in Mg²⁺ and Ca²⁺. In (a) and (b), error bars depicting standard errors are shown, but are smaller than the data symbols. Means± standard error from at least 3 independent experiments are shown. (c) Cartoon depiction of the active site of glmS^AAA seen from the direction of P2. J2/2.1 is on the right, colored according to B-factors of the Ca²⁺ structure (lowest values, blue; highest, red). Spheres represent nucleotides that are disordered in this structure. Scissile phosphate (*) is in CPK colors. (d) Cartoon depiction of the active site of glmS^WT showing the same region of the wildtype structure⁴². Coloring is analogous to (c), by B-factors of the glmS^WT structure.

Comparison of the glmS^WT and glmS^AAA structures suggest how the mutations enhance metalloribozyme activity of the latter. Important differences are that residue A(−1) is disordered and the scissile phosphate is displaced away from residue 65 in glmS^AAA towards M_B (Fig. 4c). The mobility of A(−1) brings the 5′-oxo leaving group of the reaction closer to M_B (5.6 and 6.8Å in glmS^AAA and glmS^WT, respectively), possibly facilitating catalysis by metallation. The presence of adenines at positions 49 and 51 may also contribute to the increased cation- dependent activity of glmS^AAA through flexibility, as the B-factors of nucleotides in J2/2.1, where these residues lie, are proportionately higher in the mutant structures (Fig. 5c). In glmS^WT, A50 stacks snugly between G66 and G68, helping to dock J2/2.1 with P2 (Fig. 5d). P2 stability is likely important because G65 lies at the top of this helix. In glmS^AAA, the place of A50 is taken by A49, which is geometrically constrained to make a less optimal stacking interaction between G66 and G68, and also fails to make the Hoogsteen pair with U48 made by A50 in the wildtype (Fig. 5c). These suboptimal interactions may result in increased flexibility in this region of the mutant. Analogously, extrusion of residue 51, when mutated to adenine, also may increase J2/2.1 disorder, and this may propagate to the P2 helical stack. Overall, correlation of mutagenesis results with crystal structures suggests that the tightly folded glmS^WT core (Fig. 5d) constrains the ribozyme to employ the GlcN6P-mediated pathway by precisely positioning the reactive functional groups in close proximity to GlcN6P, and away from M_B. Flexibility has been proposed to play a role in other ribozymes⁴⁴.

DISCUSSION

Our characterization of a GlcN6P-independent mutant demonstrates that rather than being constrained to rely on an exogenous small molecule coenzyme, the glmS ribozyme fold can support efficient catalysis employing divalent cations alone. This could indicate that glmS^WT harbors an intrinsic metalloribozyme activity which precedes activation by GlcN6P, and the mutations present in glmS^AAA simply enhance this activity. Alternatively, glmS^WT has suppressed metalloribozyme activity, which only exists in ribozymes resembling glmS^AAA. Two lines of evidence support the latter hypothesis. First, the difference between the coenzyme-independent cleavage rate of glmS^WT (~10⁻⁴ min⁻¹) and the uncatalyzed rate of cleavage¹⁸ for a generic RNA (estimated at 10⁻⁸ min⁻¹) can be explained without invoking basal metalloribozyme activity. This could be achieved by in-line arrangement of reactive groups by the stably pre-folded ribozyme (10 to 100-fold enhancement¹⁸), and a weak general base function of G40, ~0.1 % of which exists in the deprotonated state at pH 7 by virtue of having an unperturbed pK_a of ~10, thereby achieving 1/1000 of the maximum possible rate enhancement¹⁸ of a fully ionized general base (or acid¹⁸) catalyst of 10⁶. [This weak general base function of G40 is unlikely to be its only role, since the G40A mutation dramatically suppresses GlcN6P-activated cleavage^28,41,45, and also suppresses glmS^AAA activity (data not shown)]. Second, while the M_B site is structurally present in glmS^WT, the results of phosphorothioate mapping shows that it exhibits no or little interference in the T. tengcongensis or B. cereus⁴⁰ ribozymes, respectively. We further examined the role of M_B by determining the effects on GlcN6P-independent activity of phosphorothioate replacement at position 38 in glmS^WT and two mutants, glmS^AAG and glmS^UAG (Supplementary Table 2). For glmS^WT and glmS^UAG, both of which exhibit rates of 10⁻⁴ min⁻¹ in the absence of GlcN6P, and are both activated by the coenzyme, no difference in reaction rate was observed upon sulfur modification. In contrast, glmS^AAG, which cleaves at ~0.02 min⁻¹ in the absence of GlcN6P, and is not activated by the coenzyme, shows a rate decrease of ~35 fold. Thus, the ability of our glmS ribozyme variants to employ a metalloribozyme mechanism correlates with the functional importance of a pre-existing cation bound at the M_B site, as well as their inability to use GlcN6P.

Although it is the result of an in vitro experiment, glmS^AAA is a plausible evolutionary ancestor to glmS^WT. Its cleavage rate of >1.7 min⁻¹ is comparable to those of many nucleolytic ribozymes in contemporary biology (e.g. ref. 9,15), and a point mutation at position 65 endows this RNA with nearly 900-fold GlcN6P-enhanced activity over its metabolite-free cleavage rate (Fig. 4a,b). If acquisition of GlcN6P-dependence required many mutations, the probability of such an activity arising spontaneously from glmS^AAA would be very small. Nonetheless, because there are no known natural glmS ribozyme homologs that carry the G65A mutation^8,46 or another subset of the three mutations, this evolutionary link remains speculative. On the other hand, our analysis has uncovered the critical role that the active site-nucleotide G65 plays enabling glmS^WT to employ GlcN6P. While the importance of the highly conserved G65 has been documented by site directed mutagenesis⁴⁷ and implied by the dense network of interactions apparent from crystal structures^4,41, its key role enabling the ribozyme to employ the coenzyme-dependent mechanism would not have been apparent without comparison with the metal-ion dependent glmS^AAA. This study also highlights the importance of nucleotides in J2/2.1 for GlcN6P-dependent function.

Our work shows how an RNA can acquire a new biochemical activity through a few mutations that do not alter its fold. Previously, the relationship between the sequence spaces of two RNAs with different biochemical activities by constructing an RNA capable of folding into either an RNA ligase or the self-cleaving HDV ribozymes was explored. This RNA exhibited modest ligase or self-cleaving activity depending on which of the two mutually exclusive folds it adopted, and differed from either the parental ligase and HDV RNAs by 40 substitutions⁴⁸. That work, therefore, indicated that while a continuous mutational path existed between RNAs with two different activities, several dozen mutations may be needed for evolving a fully active form of one into another. Our delineation of the short evolutionary path separating glmS^AAA from glmS^WT suggests that RNAs that have attained stable folds can readily evolve by acquiring new biochemical functions. As a corollary, families of RNAs of related fold but different activities (analogous, for instance, to the many different activities of proteins that share the TIM barrel⁴⁹ fold) may exist in nature. In addition, our finding supports de novo evolution of new ribozyme activities as an alternative to gradual adaptive improvement of pre-existing functions, something that may be difficult in light of the rugged fitness landscapes occupied by some catalytic RNAs⁵⁰.

ONLINE METHODS

RNA constructs

Seven types of constructs were employed. ^TypeA and ^TypeB comprise 40 and 60 nt leaders, respectively, followed by a 145 nt ribozyme (Supplementary Fig. 1b). ^TypeC consists of ^Oligo1 annealed to a 139 nt trans-ribozyme starting at position 7 (GGA CUU … CAG GAA) (Supplementary Table 4). ^TypeD consists of ^Oligo2, annealed to the trans-ribozyme starting at position 23. ^TypeE consists of ^Oligo3 annealed to the trans-ribozyme construct as described⁴. ^TypeF consists of ^Oligo4 annealed to the same 139 nt trans-ribozyme as ^TypeC. ^TypeG is the same as ^TypeF, except with a biotin tethered to the 3′ end of ^Oligo4.

RNA pool synthesis

Single-stranded 205 nt DNAs (^Oligo5, Supplementary Table 4) comprised of the T7 promoter, a 40 nt leader and the 145 nt mutagenized ribozyme pool sequences (Supplementary Fig. 1b) were synthesized at 1 µmol scale (W.M. Keck Foundation Biotechnology Resource Laboratory, Yale University). Double-stranded DNAs were generated by primer extension using ^Oligo6 at 72°C for 5 minutes, followed by five cycles of large-scale PCR (96°C for 4 minutes; 50°C for 5 minutes; 72°C for 8 minutes) with ^Oligo7, using Taq DNA polymerase (Invitrogen). Approximately 20 nmol of DNA (~7.5 × 10¹⁴ unique sequences) were used as template for transcription and the resulting ³²P body-labeled RNA was gel purified as described⁴, except 30 mM Tris-HCl pH 8.1 was replaced with 50 mM HEPES-KOH, pH 7.5 and 200 µM of self-cleavage inhibitor ^Oligo8 was added.

In vitro selection

Five nmol of the recovered 185 nt RNAs were incubated in the initial round in incubation buffer containing 50 mM HEPES, 200 mM KCl, 25 mM MgCl₂, and 40 µM of ^Oligo9 (designed to hybridize to the leader sequence and prevents its use for catalysis) for 1 hr at 23°C, and stopped by addition of one volume of loading buffer. Active RNAs were isolated using 8% denaturing PAGE (145 nt glmS^WT loaded in an adjacent lane as a marker), eluted, and treated with DNase I (400 mM Tris-HCl, 100 mM NaCl, 60 mM MgCl₂, 10 mM CaCl₂, pH 7.9, Roche) for 30 minutes at 37°C. DNase I was inactivated by heating at 75°C for 10 minutes, followed by extraction with phenol and chloroform. RNAs were recovered by ethanol precipitation, and reverse transcribed (50 mM Tris-HCl, 75 mM KCl, 3 mM MgCl₂, 10 mM dithiothreitol (DTT), 560 ⌠M of each dNTP, pH 7.5, 5 ⌠M ^Oligo6, 10 U/⌠L Superscript II, Invitrogen) at 48°C for 1 hr. RNAs were hydrolyzed with 100 mM KOH at 90°C for 10 minutes, and neutralized with HCl before being subjected to 8 cycles of PCR using ^Oligo10 and ^Oligo6. Amplified 145nt DNAs were purified (1.5% agarose electrophoresis), and recovered DNAs were subjected to a second PCR amplification and purification using ^Oligo7 and ^Oligo6, before the next round of selection (Supplementary Fig. 1b). Starting from round 3, 6-thioguanosine 5′-monophosphate (^6SGMP) was added in the transcription (8 mM of ^6SGMP and 2 mM of each NTP) and 5′ ^6SGMP-labeled RNAs were purified using 8% PAGE containing 3.75 ⌠M of [(N-acryloylamino)phenyl]mercuric chloride (APM, ref. 33). The recovered RNAs were incubated as described above, and active species (unshifted, 145 nt) were separated from inactive RNAs (supershifted) using a second 8%, 3.75 ⌠M APM, PAGE. From rounds 4 to 6, ^Oligo11 was used instead of ^Oligo7 during the second PCR step to give a different leader sequence, to prevent selection of ribozymes recognizing and cleaving specific leader sequences. ^Oligo8 and ^Oligo9 were also changed in these rounds to DNA ^Oligo12 and ^Oligo13. In rounds 7 and 8, a longer leader (61nt) sequence was introduced during PCR by switching ^Oligo11 to ^Oligo14. Starting from round 6, selection pressure was increased by shortening the incubation time to enhance isolation of the fastest ribozymes.

Characterization of glmS^AAA cleavage products

For characterization of the 5′ cleavage product, 5′ ³²P-labeled ^Oligo4 was cleaved with ^TypeF glmS^WT (50 mM MgCl₂, 2 mM GlcN6P), or glmS^AAA (50 mM CaCl₂) for 20 minutes. Half of each sample was subsequently removed and quenched with gel loading buffer, while 100 mM HCl was added to the remaining half and incubated for an additional 2 hrs before neutralizing with 100 mM NaOH. Samples in lanes 3, 5, 9 and 11 were acid treated after the cleavage reaction. For characterization of the 3′ cleavage product, ^TypeG RNAs were cleaved with glmS^WT (50 mM MgCl₂, 2 mM GlcN6P), glmS^AAA (50 mM CaCl₂), or no ribozyme, followed by incubation with 0.5 mg of streptavidin containing magnetic beads (Dynabeads, M-270 Streptavidin, Invitrogen) for 30 minutes. Immobilized RNAs containing 3′ biotin were washed 3 times with 300 µL of washing buffer (5 mM HEPES, 0.5 mM EDTA, 1 M NaCl, pH 7.5) as recommended by the manufacturer, released from streptavidin by heating at 95 C for 5 minutes, and 5′ ³²P radiolabeled.

Kinetics

Single-turnover kinetics were performed throughout using ^TypeC RNA constructs, except kinetics of phosphorothioate-substituted RNAs, where ^TypeE constructs (with adenosine at position −1) were also employed (Supplementary Table 2). ^TypeF RNA constructs were used in the cleavage assays shown in Fig. 2a, but the rates reported below the autoradiogram were obtained using ^TypeC constructs., Trans-ribozyme (glmS^WT or glmS^AAA) were heated at 80°C for 2 minutes, cooled to 21°C for 5 minutes and pre-incubated at 5 µM at the specified metal ion and GlcN6P concentrations for 6 minutes in 50 mM HEPES pH 7.5. Reactions were initiated by addition of ^Oligo1, which was 5′ radiolabeled with γ-[³²P] ATP using T4 Polynucleotide Kinase (NEB) and purified by 20% PAGE, and time points were quenched with 5 volumes of loading buffer (90% formamide, 50 mM EDTA). Percentage cleaved substrate was determined by 20% denaturing PAGE or thin-layer chromatography as described⁵¹, except 0.2 M KH₂PO₄ was used as the mobile phase. Reaction rates were determined by simultaneously fitting the fraction reacted for at least three independent time courses (quantified using a Storm 865 phosphorimager, GE) to the equation F = β(1 - e^–k_obst) using KaleidaGraph (Synergy), F being the fraction reacted at time t, k_obs the first-order apparent rate constant, and β the total fraction able to react. The resulting rates were fit to the Hill equation k_obs = k_max × [metal ion]ⁿ / ([K_1/2]ⁿ + [metal ion]ⁿ), where n is the Hill coefficient. Under our experimental conditions, the slowest rate we can determine is ~5 × 10⁻⁵ min⁻¹. For pH rate-profiles, cleavage reactions were performed as described in kinetics buffer (100 mM MgCl₂ or 100 mM CaCl₂, with either 25 mM potassium acetate or 25 mM cacodylate-KOH or 25 mM HEPES-KOH or 25 mM TAPSO-KOH) adjusted to the desired pH (ref. 52). First order reaction rates were determined using the equation: k_obs = k_max × 10^-pKa / (10^-pH + 10^-pKa). For glmS^WT GlcN6P-dependent time courses, a three-syringe Kintek rapid-quench mixer was used, where equal volumes of solution A (glmS^WT, buffer, metal ion, and GlcN6P) were mixed with solution B (5′ radiolabeled ^Oligo1) to initiate the reaction, and quenched by mixing with 5 volumes of solution C (gel loading buffer) after the indicated reaction time. Throughout, error bars are standard errors of the mean calculated from three or more independent time courses.

Small angle X-ray scattering

^TypeD RNA constructs were mixed at a 1.2 : 1 ratio (glmS^WT or glmS^AAA : ^Oligo2) in 50 mM HEPES-KOH pH 7.5, 100 mM KCl, and 2 mM MgCl₂, to give a blunt end at the terminus of the P1 helix. After 15 minutes at room temperature, the complex was purified by size-exclusion chromatography (Superdex 200, GE), and exchanged into the final buffer containing 50 mM HEPES pH 7.5 and 2 mM MgCl₂ by ultrafiltration. Prior to SAXS experiments, samples were diluted to 0.5 g/L in the final buffer. All data collection was performed at the BioCAT beamline at Advanced Photon Source (APS), Argonne National Laboratories, and analyzed as described⁵³.

Crystallization, structure determination and refinement

^TypeE glmS^AAA RNA for crystallization was transcribed from plasmid template generated by site-directed mutagenesis (QuikChange, Agilent) of glmS^WT construct⁴, except that for the last of 5 hours, MgCl₂ was increased to 100 mM. RNA was purified⁴⁴, and mixed at a final concentration of 175 µM with 210 µM of ^Oligo3 (with deoxyadenosine at A-1) in 10 mM HEPES-KOH pH 7.5, and 50 mM either MgCl₂ or CaCl₂, and incubated at 65°C for 2 minutes. Samples were held at 21°C for 15 minutes before spermine-HCl was added to 1 mM. Hanging drops prepared by mixing equal volumes of this solution and a reservoir 10% (w/v) PEG 8000, 1 M LiCl, 50 mM Mg(OAc)₂ and 50 mM Cacodylate-NaOH pH 6.5 were equilibrated by vapor diffusion at 21°C. Leaf-shaped crystals⁵⁴ grew to 0.5 × 0.2 × 0.05 mm³ over 2–3 weeks. Crystals were mounted on nylon loops and transferred to 50 mM Cacodylate-NaOH pH 6.5, 18% (w/v) PEG 8000, 1.8 M LiCl, 1 mM spermine, and either 170 mM MgCl₂ or 170 mM CaCl₂ for ~1 minute before flash-freezing by plunging into liquid nitrogen. Diffraction data were collected at 100 K (1 Å X-radiation) at beamlines 5.0.1 and 5.0.2 at the Advanced Light Source, Lawrence Berkeley National Laboratory (ALS), and reduced with HKL2000 (ref. 55). Both Mg²⁺ and Ca²⁺ crystals have the symmetry of space group P2₁2₁2 (Supplementary Table 1) and contain one ribozyme-inhibitor complex per asymmetric unit. The structures were solved by molecular replacement with PHASER⁵⁶ using PDB ID 2Z75 (after deleting water, ions, GlcN6P, and nucleotides 49, 51, 65) as the search model. Iterative rounds of simulated annealing, restrained individual B-factor and energy refinement⁵⁷ interspersed with manual model building⁵⁸ produced the current models (Supplementary Table 1). The superposition of glmS^WT and glmS^AAA (in Mg²⁺) in Fig. 2d was calculated with 87 C1′ atom pairs from the RNA cores (residues 1–9, 32–48, 52–73, and 97–135; r.m.s.d. = 0.6 Å). The Mg²⁺ and Ca²⁺ structures superimpose closely (r.m.s.d. = 0.8 Å for the same 87 atom pairs as in Fig. 2d); the conformation of J2/2.1 is better defined in the latter. Structure figures were prepared with PyMol (ref. 59).

Phosphorothioate interference mapping

^TypeA glmS^WT and glmS^AAA cis-constructs were transcribed with ATPαS, GTPαS, CTPαS or UTPαS (Glen Research) and purified as described⁶⁰. Eluted glmS^WT and glmS^AAA RNAs with phosphorothioate modifications were incubated in 50 mM HEPES pH 7.5, 5 mM MgCl₂, 200 µM GlcN6P, or 50 mM HEPES pH 7.5, 25 mM CaCl₂, respectively, for 1 minute at 21°C. The "uncleaved" and "cleaved" RNAs were purified and 5′ radiolabeled as described⁴⁰, and positions of phosphorothioate incorporation were determined by subsequent cleavage of the RNAs with 1 mM iodine in 5 volumes of gel loading buffer for 2 minutes at 90°C, followed by 15% denaturing PAGE analysis. To calculate normalized interferences, the band intensities were first corrected for loading differences and non-specific RNA degradation. Then, per-position interferences were determined by first dividing the sum of the band intensities for self-cleaved and inactive RNAs by the band intensities of self-cleaved RNAs (averaged from at least 3 separate experiments). These quantities were then normalized by subtracting the average of interferences observed at all positions, for all four nucleoside phosphorothioates. Finally, interferences were expressed as standard deviations above mean interference by dividing the normalized interference by the standard deviation of interferences at all observed positions, for all four nucleoside phosphorothioates.

Atomic coordinates and structure factors for glmS^AAA Mg²⁺ and glmS^AAA Ca²⁺ crystal structures have been deposited in the Protein Data Bank (accession codes 4MEG and 4MEH).