Targeting riboswitches with beta-axial substituted cobalamins

Abstract

RNA-targeting small molecule therapeutics is an emerging field hindered by an incomplete understanding of the basic principles governing RNA-ligand interactions. One way to advance our knowledge in this area is to study model systems where these interactions are better understood, such as riboswitches. Riboswitches bind a wide array of small molecules with high affinity and selectivity, providing a wealth of information on how RNA recognizes ligands through diverse structures. The cobalamin-sensing riboswitch is a particularly useful model system as similar sequences show highly specialized binding preferences for different biological forms of cobalamin. This riboswitch is also widely dispersed across bacteria and therefore holds strong potential as an antibiotic target. Many synthetic cobalamin forms have been developed for various purposes including therapeutics, but their interaction with cobalamin riboswitches is yet to be explored. In this study, we characterize the interactions of eleven cobalamin derivatives with three representative cobalamin riboswitches using in vitro binding experiments (both chemical footprinting and a fluorescence-based assay) and a cell-based reporter assay. The derivatives show productive interactions with two of the three riboswitches, demonstrating simultaneous plasticity and selectivity within these RNAs. The observed plasticity is partially achieved through a novel structural rearrangement within the ligand binding pocket, providing insight into how similar RNA structures can be targeted. As the derivatives also show in vivo functionality, they serve as several potential lead compounds for further drug development.

Introduction

RNA presents an underutilized target for therapeutics with the potential for wide-reaching effects.^1–3 While drugs to combat disease have historically targeted proteins, the revelation that only about 1.5% of the human genome is translated into protein⁴ while at least 75% is transcribed into RNA⁵ suggests that an RNA-focused approach may provide a wealth of targeting opportunities.^6–7 However, beyond antibacterial and antiviral drugs^8–9, small molecule-based RNA targeting therapeutics are mostly in the early discovery stages of development.¹ One of the main challenges is that our understanding of the basic principles of RNA-small molecule interactions is lacking as compared to that of proteins.^{2–3, 10} While many nonspecific RNA binding small molecules, such as general intercalators, are known, effective therapeutics need to be specific for a target RNA sequence or structure. Further exploration of how RNA specifically recognizes and binds various ligands will facilitate rational design of small molecules to target and drug RNA.

Perhaps no system better demonstrates specific and high affinity RNA-small molecule interactions that drive a biological outcome than riboswitches. Riboswitches are non-coding RNA elements typically found in the 5’-leader of messenger RNAs (mRNAs), mostly in bacteria.¹¹ Small molecule binding to the riboswitch regulates expression of the mRNA by directing formation of alternative RNA structures that inform the expression machinery. To date, over 55 classes of riboswitches have been discovered, binding a wide range of metabolites.¹¹ These ligands can be related to RNA structurally or metabolically, such as nucleotide derivatives, but can also be non-self-compounds, such as elemental ions and amino acids like lysine. Riboswitches can adopt a wide spectrum of RNA structures to create binding pockets with high affinity and specificity for these target ligands. These structures range from simple sites comprising formation of base triples with the ligand^12–13 to complex multi-junction pockets.^14–15 Riboswitch binding pockets often contain common motifs such as pseudoknots, suggesting that similar motifs in other RNAs can also be targeted by drug-like small molecules. Thus, the diversity of small molecules and RNA structures encompassed by riboswitches presents a rich palette of robust model systems for the study of RNA-small molecule interactions, and several studies have shown the strength of using riboswitches in this manner.^16–19

The cobalamin (Cbl, B₁₂) riboswitch controls the biosynthesis, salvage, and transport of Cbl, an essential metabolite for many forms of life.²⁰ The genes which maintain Cbl cellular levels can be essential for survival and virulence of bacterial pathogens, making the Cbl riboswitch a promising antibiotic target.²¹ Corresponding to the importance of this metabolite, the Cbl riboswitch is the second most widely distributed, found in almost every major clade of bacteria.²² The Cbl riboswitch is a particularly useful model system for the study of RNA-small molecule interactions, as different classes show highly specialized binding to the biological forms of Cbl despite similar sequences. Currently two classes, I and II, are recognized within the Cbl riboswitch family. Class-I specifically binds 5’-deoxyadenosyl-Cbl (AdoCbl) and class-II exhibits a spectrum of selectivity for AdoCbl and methylcobalamin (MeCbl).²³ All forms of Cbl contain a corrin ring that coordinates a cobalt atom and is flanked by methyl, acetamide, and propionamide groups. The sixth coordination site of the cobalt ion is the beta-axial position (Figure 1A), coordinating either the 5’-deoxyadenosyl (Ado) or methyl group. The interaction of the beta-axial group with a universally conserved four-way junction determines the selectivity of the riboswitch.²³ Understanding how similar RNA structures can bind distinct compounds and how that can be exploited using analogs can advance the optimization of lead compounds in drug discovery.

Figure 1.

Structures of Cbl forms and riboswitches used in this study. (A) Top left: Chemical structure from which the Cbl forms are derived. Positions of interest are labeled as R₁ (beta-axial), R₂ (meso), c, and e. Top right: R₁ and R₂ groups for the standard Cbl forms AdoCbl, MeCbl, and CNCbl. Bottom: Modified groups of the Cbl derivatives, identified by numbers. (B) Secondary structures of the Cbl riboswitches. The preferred binding forms of Cbl are written in the common central four-way junction of each structure, colored in blue. Colored and labeled in green are additional areas relevant to this paper. Created using Biorender.com.

Synthetic forms of Cbl have been developed for various purposes, including therapeutics.²⁴ Cyanocobalamin (CNCbl) has a light-stable cyano group in the beta-axial position (Figure 1A) and is used as a diet supplement as it can be converted to the natural forms within the body. Cbl derivatives that cannot be transformed into the biologically active forms are called antivitamins.²⁴ Like CNCbl, B₁₂ antivitamins can have modifications at the beta- axial position. Additional modifications include formation of a lactone group at the corrin ring c position.²⁵ B₁₂ antivitamins are often used to study B₁₂ deficiency and have potential as novel antibiotics or cancer drugs.^26–27 Meso-modified Cbl derivatives have also been developed as delivery systems for drugs, imaging agents, oligonucleotides, and other agents.^28–30 These derivatives have modified chemical properties, caused by addition of electron-donating or -withdrawing groups around the corrin ring to affect the redox potential of the central cobalt ion.³¹ The interaction of these synthetic Cbl forms with Cbl binding proteins has been studied.^32–33 However, the potential use of structurally modified cobalamins as antibiotics through interaction with Cbl riboswitches has only been hypothesized^{24, 26} as their ability to bind to Cbl riboswitches has not been studied.

Here we directly address this question by examining the interactions of a series of Cbl derivatives with representative Cbl riboswitches. Three Cbl riboswitches were chosen to represent the structural and functional diversity across this family of RNAs:³⁴ the class-I E. coli btuB riboswitch that specifically binds AdoCbl, a class-II riboswitch that has moderate selectivity for large beta-axial form Cbl but also binds small beta-axial cobalamins (env50), and a class-II riboswitch that has high selectivity for small beta-axial forms of Cbl (env8) (Figure 1B, Table S1). Eleven Cbl derivatives (Figure 1A) synthesized with many of the modifications discussed above were compared to AdoCbl and CNCbl (used as a non-photolyzable proxy for MeCbl) in binding assays. These modifications include the presence of lactone groups at the e and c corrin ring positions (derivatives 1 and 2), various modifications to the meso position (derivatives 3-7), and substitutions at the beta-axial position (derivatives 8-11). Each ligand’s binding properties were evaluated using a fluorescent displacement assay and complementary chemical probing. The results reveal surprising plasticity of the class-II riboswitches, including a novel binding mode to accommodate bulky beta-axial moieties. While none of the tested Cbl derivatives bind the riboswitches more tightly than the native ligands, some of the interactions are still shown to be biologically active using an in vivo cell reporter assay, establishing those derivatives as promising lead compounds for drug development.

Results and Discussion

Validation of a fluorescence displacement assay to assess binding

To quantify binding of Cbl derivatives to the riboswitches, we used a fluorescent displacement assay. This approach is common for investigating RNA-small molecule interactions and has been used to investigate other riboswitches binding to alternative ligands.^{18, 35} The fluorescent probe used in this assay is a previously characterized fluorophore conjugated form of CNCbl, CNCbl-5xPEG-ATTO590 (Figure S1A).³⁶ Conjugation of an ATTO 590 fluorophore to the ribose 5’-hydroxyl group of CNCbl via a 5x polyethylene glycol (PEG) linker results in 90% quenching of the fluorophore. However, the conjugated probe binding to env8 results in 4.9-fold fluorescence induction. Titration of env8 into CNCbl-5xPEG-ATTO590 gives an apparent equilibrium dissociation constant (K_D) of 28 ± 7 nM as calculated from the fluorescence read-out (Figure 2A), comparable to measurements determined by isothermal the titration calorimetry (ITC) in the same conditions (34 ± 9 nM).³⁶

Figure 2.

Validation of fluorescence-based affinity measurements. A) Determination of K_D by fluorescence induction upon CNCbl-5xPEG-ATTO590 probe binding to env8 (n=9). B) Determination of K_D by fluorescence displacement assay, where CNCbl and CNCbl-5xPEG-ATTO590 compete for binding to env8 (n=4). Points represent the average of all replicates and error is represented as SEM. K_Ds determined by ITC are included for comparison.

To quantitatively assess derivative binding, we titrate the Cbl derivative into env8 saturated with CNCbl-5xPEG-ATTO590, displacing the fluorescent probe. Resultant fluorescence data can be used to determine the concentration of unlabeled competitor that reduces the binding of the probe by half (IC₅₀), which can be used to determine K_D. To validate this method, we titrated a ligand with a known binding affinity for env8 (CNCbl) into the env8-probe complex. The resultant K_D of 16 ± 5 nM (Figure 2B) is comparable to that determined by ITC (37 ± 1 nM)³⁶ and similar to the K_D of hydroxocobalamin, another Cbl form with a small beta-axial moiety, as measured by smFRET (5 ± 3 nM).³⁷ These data indicate the assay is a reliable method for quantitative assessment of Cbl derivative binding.

Many of the Cbl derivatives bind to class-II Cbl riboswitches

Using the displacement assay, we surveyed the panel of Cbl derivatives for env8 binding (Figure 3, Table 1). Every derivative showed some degree of binding, and several were highly competitive. The derivatives that bound most tightly to env8 relative to CNCbl are beta-axial derivatives 8 and 9 which have K_rel values of less than 10. In general, the lactone and meso derivatives bound env8 with lower affinity than most of the beta-axial derivatives, having K_rel values of 30 and above. Together, these observations indicate that modifications to the beta-axial position of Cbl are more well tolerated by env8 than modifications to the periphery of the corrin ring, although all tested modifications still allow for some level of binding. Further exploration of the chemical space around these derivatives may be a productive route to compounds that rival CNCbl in their affinity for class-II Cbl riboswitches.

Figure 3.

Cbl derivatives binding to env8, analyzed by fluorophore displacement assay. A) Lactone and meso derivatives. B) Beta-axial derivatives. (n=2–4) Points represent the average of all replicates and error is represented as SEM.

Table 1.

Binding affinities of Cbl derivatives to env8 as determined by fluorescence displacement assay.

Ligand	Derivative Class	KD (μM)a	Krelb
CN-Cbl	N/A	0.016 ± 0.005	1.0
1. epi-e-lactone	corrin ring	0.49 ± 0.11	30
2. c-lactone	corrin ring	0.72 ± 0.42	44
3. meso-NH2	corrin ring	0.60 ± 0.35	37
4. meso-imine	corrin ring	0.68 ± 0.23	42
5. meso-2°amine	corrin ring	0.71 ± 0.04	43
6. meso-NO2	corrin ring	2.3 ± 0.8	142
7. meso-amide	corrin ring	21 ± 14	1288
8. phenyl-NO2	beta axial	0.091 ± 0.038	5.6
9. β-Si	beta axial	0.12 ± 0.06	7.2
10. phenyl-OMe	beta axial	0.39 ± 0.15	24
11. alkyne-OH	beta axial	1.5 ± 1.0	92

^an=3–4; error reported as SEM

^bK_rel = K_D,deriv/K_D,CNCbl

For fluorescence displacement assays with env50 and btuB, we synthesized an additional Cbl-fluorophore conjugate. Derivative 8 was modified in the same manner as CNCbl-5xPEG-ATTO590, resulting in a PhNO₂-Cbl-5xPEG-ATTO590 probe (Figure S1B). This allowed for direct analysis of binding to this Cbl derivative. Measurement of fluorescence induction upon binding showed env50 has slightly higher affinity for the PhNO₂ probe than the CN probe (Figure 4A). The observed K_D for the CN probe is consistent with a previously reported K_D for MeCbl as measured by ITC.²³ However, using either probe with env50 to quantitatively determine derivative K_D values requires saturation of the RNA. Given the high nanomolar affinity of these probes for env50, large amounts of each derivative would be necessary to fully define a binding curve. Additionally, we observed that high concentrations of derivative decreased the background fluorescence either from quenching of the probe or the inner filter effect, complicating quantification of the data.

Figure 4.

Cbl derivatives binding to env50, analyzed by fluorescence assays. A) Determination of K_D by fluorescence induction upon CNCbl-5xPEG-ATTO590 and PhNO₂-Cbl-5xPEG-ATTO590 probes binding to env50 riboswitch (n=3). Points represent the average of all replicates and error is represented as SEM. B) Binding of Cbl derivatives (1 μM) to env50 by fluorescence displacement assay (n=3). Values are normalized so “on” = probe fluorescence bound to env50 without competing ligand and “off” = probe fluorescence unbound to RNA. C) The same as B but competing ligands are at 100 nM. Error is represented as SEM.

Given the above complication, we changed the assay conditions to qualitatively assess env50 binding to the Cbl derivatives. Binding was evaluated at a 1:1 probe:derivative ratio for weakly binding compounds and a 10:1 probe:derivative ratio for high affinity derivatives (Figure 4B). The Cbl derivatives displayed a range of probe displacement. The observed trend in probe displacement is similar to that for env8. The beta-axial derivatives were the most effective competitors while the meso and lactone derivatives displayed poor to moderate competition. CNCbl was the second most effective competitor, although derivatives 9 and 11 performed nearly as well. Derivative 8 outcompeted CNCbl, consistent with the direct binding data. However, derivative 8 was not as effective a competitor as AdoCbl, reinforcing env50’s preference for this Cbl form. Using the beta-axial derivatives to compete at lower concentrations than the probe confirmed the tighter binding of AdoCbl (Figure 4C).

Three derivatives do not follow the same trend in affinities between env8 and env50. The two lactone derivatives, 1 and 2, which were similar to the meso derivatives in affinity for env8, appear to have higher affinity for env50. Additionally, derivative 11 shows tighter binding to env50 compared to env8; 11 ranked with the weakest of the binders for env8, while it ranks among the tightest for env50. Overall, results between env50 and env8 are similar in that none of the tested Cbl derivatives are tighter binders than the preferred Cbl form, although the beta-axial derivatives display impressive binding abilities.

To examine binding of our panel of Cbl derivatives to btuB, we first assessed direct binding of the CN and PhNO₂ probes. The highest tested concentration of RNA (10 μM) did not induce fluorescence in the probes (Figure S2). Due to the lack of binding detected with either probe, we could not use the displacement assay to determine binding affinities for the Cbl derivatives to btuB and an alternative method of chemical footprinting was explored for this and the other riboswitches.

Bulky beta-axial substitutions displace a purine residue in the binding pocket of a MeCbl-selective riboswitch

Given the high affinity of some of the Cbl derivatives to env8, we asked how the riboswitch structurally accommodates the Cbl modifications. The crystal structure of env8¹⁵ (Figure 5A) as well as other Cbl riboswitches^38–39 (Figure 5B) reveals the ligand is almost completely encapsulated by RNA and recognition is achieved by mutual shape complementarity. Docking analysis using the env8 crystal structure (PDB ID 4FRN)¹⁵ suggests that all tested modifications sterically clash with key RNA architectural elements (Figure 5A, S3). The beta-axial derivatives in particular present an obstacle with their bulky moieties due to clashes with nucleotides A20 (J3/4) and A68 (J6/3) in the most conserved region of the binding pocket. Derivative 1 appears to clash with G70 (J6/3), 2 with G19 (J3/4) and C71 (P3), and the meso derivatives with all three of these nucleotides. Thus, it is surprising that the Cbl derivatives productively bind env8, especially the beta-axial derivatives which are similar to the non-binder AdoCbl in having a sterically bulky beta-axial moiety. To enable productive binding, we hypothesized that the binding pocket must have previously unappreciated plasticity. This hypothesis is supported by studies with other riboswitches that have revealed unexpected modes of plasticity that enable binding of chemical analogs.^{14, 40–42}

Figure 5.

Molecular docking and structural probing by SHAPE of Cbl derivatives binding to env8. A) Crystal structure of env8 (PDB ID 4FRG) with CNCbl (left) or phenyl-NO₂ Cbl derivative (8, Cambridge Structural Database deposition number 961228, right), superimposed to show steric clash with position A20 (green) and A68 (orange). The beta-axial moieties are colored red. B) Crystal structure of Sth AdoCbl riboswitch (PDB ID 4GXY) with AdoCbl (left) or 8 superimposed (right). Equivalent nucleotides and beta-axial moieties are colored the same as in A. C) Lactone and meso derivatives, highlighting the L5 region. D) Same as A, but with the beta-axial derivatives. Nucleotide A20 is also included. Representative gels are shown and the results are quantified (n=6–10, error is reported as SEM).

To visualize binding of Cbl derivatives to env8, we used selective 2’-hydroxyl acylation analyzed by primer extension (SHAPE)⁴³, which has previously been used to characterize Cbl riboswitches.^{15, 44} Corroborating the fluorescence displacement assays, SHAPE showed that many of the tested Cbl derivatives bind to env8 (Figure 5C,D and S4). CNCbl was used as a standard for chemical reactivity protections and enhancements in the RNA in response to ligand binding as it has been used in this capacity previously and gives predictable reactivity patterns with this RNA.^{15, 44} The presence of CNCbl promoted the previously observed set of reactivity changes in L5, J6/3, and J1/13 that are indicative of binding.⁴⁴ The derivatives promoted the same protections and deprotections in the env8 riboswitch to varying extents. There is general agreement that the derivatives with lower K_Ds induce a higher degree of SHAPE protection. Notably, the reactivity pattern of the beta-axial derivatives was similar to that of CNCbl, despite the predicted steric clash. The lactone and meso derivatives displayed a larger range of induced reactivity changes, with some behaving similarly to CNCbl and several similar to the no ligand condition, reflecting their range of binding affinities. The lack of significant differences between the footprinting patterns for CNCbl and most of the derivatives, even at nucleotides with expected steric clash, suggests that all of the ligands interact with the RNA in a similar fashion, indicating that the RNA can accommodate many of these modifications without large structural rearrangements detectable by SHAPE. Alternative chemical probes such as DMS that interrogate other structural features of the RNA may reveal conformational changes likely required for derivative binding.

The only nucleotide with predicted steric clash to display a significant difference in chemical reactivity for several derivatives is A20. Whereas CNCbl binding protects this position, binding of beta-axial derivatives 8 and 9 show a strong enhancement of chemical reactivity. Notably, these two derivatives have the lowest K_rel values compared to CNCbl as observed in the fluorescence displacement assay. A20 is in a region with structural significance, being one of the four nucleotides from J3/4 and J6/3 (G19, A20, A68, and A67) that comprise a cross-strand purine stack^45–46 forming one half of the ligand binding pocket (Figure 5A). The beta-axial face of CNCbl forms direct van der Waals contacts with this region of the RNA. We propose that the nucleotide deprotection is the result of A20 being displaced from the purine stack by the derivative beta-axial group. This is similar to the observed displacement of a key nucleotide in the purine riboswitch by guanine derivatives containing modifications at the N2 position.⁴⁰ Removal of the nucleotide from the protective environment of the purine stack would enhance the reactivity of its 2’-hydroxyl group towards the probing reagent and enable the bulky beta-axial group to intercalate between residues G19 and A69.

Intercalation of a functional group into the purine spine of Cbl riboswitches is observed for AdoCbl binding (Figure 5B).^{15, 38–39} In all AdoCbl-riboswitch structures, nucleotides from J3/4 and J6/3 form a continuous base-stack as one half of the binding pocket, and an opening within the stack allows for insertion of the Ado moiety. The opening is created by J6/3 being pulled back and away from the binding pocket, supported through interactions with another region of the RNA. In the class-I structures, the adenosine equivalent to A68 in J6/3 of env8 (A157 in Figure 5B) forms a trans A·A Watson-Crick-Hoogsteen pairing with the nucleobase of the Ado moiety.^{15, 39} This nucleotide is not reactive in SHAPE analysis due to its interactions with the Ado.¹⁵ Due to its selectivity against AdoCbl, it was unexpected that env8 could act similarly to the AdoCbl riboswitches by binding bulky beta-axial groups that displace a nucleotide from the purine stack. This is also the first evidence of displacement of a nucleotide in J3/4 as opposed to J6/3. Thus, both halves of the riboswitch central spine display plasticity that can be exploited by Cbl derivatives with bulky beta-axial substitutions.

The SHAPE results for the beta-axial derivatives can be further interpreted to infer how to successfully target env8 and potentially other MeCbl specific Cbl riboswitches. The exception to the tighter binding of the beta-axial derivatives is 11 which was the third weakest binder with a K_rel of 92. This is also the only beta-axial derivative with a methylene bridge proximal to the alkyne in the beta-axial moiety, making it more structurally similar to AdoCbl (Figure 1A). This sp³-hybridization likely projects the aromatic ring towards J6/3, consistent with the lack of increased SHAPE reactivity of A20 upon 11 binding. If 11 projects into the same segment of the purine stack as AdoCbl, this also likely causes its reduced binding affinity. This structural similarity to AdoCbl may also explain 11’s tighter affinity to env50, which does bind AdoCbl. The other beta-axial derivative that does not displace A20 is derivative 10, although it is most structurally similar to 8. We hypothesize that this is due to the difference between the OMe versus NO₂ terminal groups. The nitro group in derivative 8 withdraws electrons from the phenyl ring, which enhances π-stacking interactions.⁴⁷ Thus, 8 may help stabilize the purine stack after A20 displacement. In contrast, the methoxy group in 10 is an electron donating group which increases electronic repulsion during base stacking,⁴⁷ disfavoring the replacement of A20 with 10 in the purine stack. This hypothesis is supported by the reduced affinity of 10, which is about four times weaker than 8. The most surprising result of the beta-axial derivatives is 9, which has productive binding through displacement of A20 despite not possessing an aromatic ring to intercalate into the purine stack, indicating that π-stacking is not essential for facilitating displacement of A20.

Binding by Cbl derivatives does not yield SHAPE signatures in AdoCbl riboswitches

Given the valuable insights from the env8 SHAPE results, we applied the same approach to assess derivative binding to env50 and btuB. For these riboswitches, the preferred ligand AdoCbl was used as the standard for determining ligand binding. Env50 showed protections in J6/3 and deprotections in J1/13 in the presence of AdoCbl, indicative of binding (Figure 6A and Figure S5). Unexpectedly, CNCbl did not induce detectable pattern changes in the RNA despite the known binding of MeCbl to this riboswitch and the binding results with the CNCbl probe. While CNCbl may have a slightly different binding affinity to env50 than MeCbl, it was still expected to exhibit a SHAPE signature, especially at high ligand concentration (30 μM). Additionally, none of the tested Cbl derivatives showed indications of binding, either. This is also in contrast to the fluorescence displacement assay results which showed binding of many of the Cbl derivatives. Unfortunately, without a ligand-dependent SHAPE signature, we were unable to determine whether any novel binding mode exists between env50 and the Cbl derivatives as we were able to with env8.

Figure 6.

Structural probing by SHAPE of beta-axial Cbl derivatives binding to AdoCbl-selective riboswitches. Representative gels are shown, and the results are quantified (n=3, error is represented as SEM). A) Env50, J6/3 and J1/13 regions. B) btuB, L5 and J11/10 regions.

One potential explanation for the discrepancy between the SHAPE and fluorescence displacement assay results for env50 is a reaction between the SHAPE chemical probing agent and Cbl that inactivates binding. However, incubation of CNCbl with the probing agent did not alter its binding to env50 (Figure S6). Another potential explanation is a lack of conformational change upon ligand binding. If so, conducting chemical probing at higher temperatures could reveal additional flexibility and ligand-dependent changes not appreciable in the current probing conditions. However, the strong signature upon AdoCbl binding suggests this is not the issue. The negative SHAPE results may also be due to other reaction conditions, indicative of a high k_off rate of the derivatives, or the higher K_Ds may present a limitation for the methodology. Evidence suggests that SHAPE-based screening may be biased towards high affinity ligands, although K_Ds up to hundreds of micromolar have been detected.⁴⁸ Disagreement between the results of SHAPE and another method has been seen previously; the Tte-PreQ₁ riboswitch did not show an altered SHAPE signature for synthetic ligand binding despite a change in RNA flexibility suggested by molecular dynamics simulations.⁴⁹ The inconsistency between our results suggests that SHAPE may not be ideal for screening weakly binding compounds during initial phases of discovery of an RNA targeting compound. These observations highlight the utility of parallel approaches to assess binding in novel systems.

SHAPE analysis did not indicate any further Cbl derivative interactions with btuB, either. We observed a set of protections in btuB’s L5 and J11/10 in the presence of AdoCbl, consistent with previous studies (Figure 6B and Figure S7).¹⁵ However, similar to env50, chemical probing in the presence of CNCbl or any of the panel of Cbl derivatives did not yield reactivity pattern differences in the RNA. Unlike for env50, these results do align with those of the fluorescence assays, which did not indicate binding between btuB and either the CNCbl or PhNO₂ probe. This was expected for CNCbl and the lactone and meso derivatives because btuB does not bind small beta-axial Cbl forms. But these results were unexpected for the beta-axial derivatives, again because btuB does bind large beta-axial form AdoCbl. Together, the btuB fluorescence and SHAPE data suggest that btuB binding is highly specific for AdoCbl. In light of these and the fluorescence assay results, we do not expect the derivatives to productively bind class-I Cbl riboswitches. Targeting AdoCbl-specific riboswitches likely requires development of stable beta-axial moieties that better mimic the Ado moiety.

Env8’s binding pocket is flexible but largely unalterable for beta-axial derivative binding

To determine whether observed plasticity of the env8 J3/4-J6/3 purine spine is specific to adenosine, we examined the role of A20 in Cbl binding. This position was mutated to each of the other nucleotides, and the mutant sequences’ ability to bind CNCbl, 8, and 9 was examined using SHAPE (Figure 7A). Despite this position being a purine across the Cbl riboswitch family, all mutants exhibit the same patterns of SHAPE reactivity in the presence of the tested Cbls, including the signature for displacement of nucleotide 20. Thus, the conformational flexibility of this nucleotide is independent of its identity.

Figure 7.

Env8 binding pocket mutants binding to Cbl derivatives. A) A20 mutants were analyzed for binding to CNCbl, 8, and 9 by SHAPE. Nucleotide A20 and region L5 are shown in a representative gel and are quantified (n=2). B) Determination of K_D by fluorescence induction upon CNCbl-5xPEG-ATTO590 (top) and PhNO₂-Cbl-5xPEG-ATTO590 (bottom) binding to env8 A20 mutants. C) Same as B, but mutations to regions J6/3 (A68G) and J1/3 (ΔA9, G10U, G12A), alone and in combination with A20G. (n=3–4) Error is represented as SEM.

Since pyrimidines also support Cbl binding, we hypothesized that they may be easier to displace and thus promote PhNO₂ binding. We quantified the binding of the CN and PhNO₂ probes to the above mutants along with mutations in the flanking positions (G19U and A21U). Mutation of positions flanking nucleotide 20 do not support binding, indicating the importance of these residues (Figure 7B, Table 2). The position 20 mutants are less detrimental to the PhNO₂ probe binding than the CN probe. Notably, each of the A20 mutants prefers binding the PhNO₂ probe over the CN probe. This is significant especially for the A20G mutation, as guanosine at this position is observed in phylogeny. In fact, env50 has a guanosine at the equivalent position and has tighter binding to 8 than CNCbl. Together, this indicates that for some Cbl riboswitches, the beta-axial PhNO₂ Cbl derivative may have higher binding affinity than the natural ligand.

Table 2.

Binding affinities of CN-Cbl-5xPEG-ATTO590 and PhNO₂-Cbl-5xPEG-ATTO590 probes to env8 variants.

env8 variant	CN probe KD (nM)a	PhNO2 probe KD (nM)a	Krelb
Wild type	28 ± 7b	73 ± 8	0.4
A20U	270 ± 20	90 ± 2	3.0
A20C	290 ± 20	130 ± 10	2.2
A20G	450 ± 70	230 ± 50	2.0
G19U	NDc	ND
A21U	ND	ND
J6/3	90 ± 6	ND
J6/3 + A20G	540 ± 10	ND
J1/3	820 ± 30	ND
J1/3 + J6/3 + A20G	ND	ND

^an=3–4; error reported as SEM

^bK_rel = K_D,CN/K_D,PhNO2

^cND: binding not detectable.

We next sought to assess whether nucleotides supporting the binding pocket influence the plasticity of nucleotide 20.²³ A series of mutations were made to nucleotides in regions J6/3 and J1/3, alone and in combination with the A20G mutation. Given that env50 shows reduced affinity for the beta-axial derivatives compared to env8, we mutated the env8 regions to match env50 to test our hypothesis. Within this set of mutations, binding to the CN probe is progressively diminished, recapitulating prior observations (Figure 7C, Table 2).²³ In stark contrast, mutations beyond A20G that support CN probe binding are severely deleterious to PhNO₂ probe binding. This reveals the flexibility of nucleotide 20 is dependent upon the surrounding RNA architecture which suggests that accommodation of beta-axial derivatives may be restricted to a subset of class-II riboswitches. In the development of an FMN riboswitch targeting compound, a similar phenomenon was observed in which the lead compound was highly specific for the Clostridium difficile FMN riboswitch.¹⁶ This property enabled the lead compound to target C. difficile in an animal model while leaving the remainder of the microbiome largely unaffected—a potentially important feature of a desired antimicrobial agent.^{16, 50}

The beta-axial Cbl derivatives can drive biological function

The above results demonstrate that a set of beta-axial Cbl derivatives productively bind the env8 riboswitch, but binding in vitro may not drive regulatory function in cells. Regulatory function may not occur for several reasons, including an inability to import, an inability to bind the riboswitch within a relevant timescale, or an inability to occlude the ribosome binding site to repress translation.⁴⁴ To determine whether the derivatives repress gene expression, we tested the beta-axial Cbl derivatives in a previously established cell-based reporter assay in E. coli.⁴⁴ In this assay, the env8 riboswitch is upstream of a fluorescent protein (FP) reporter whose expression is repressed by Cbl-dependent occlusion of the ribosome binding site. The beta-axial derivatives that productively bind env8 in vitro also repress FP expression (Figure 8). While the repressive activity of these ligands was less than that of CNCbl, they generally agree with their binding affinities. Derivative 7, the worst binder in vitro, does not show this repressive effect, indicating that biological activity of the other derivatives is due to riboswitch binding. Thus, these derivatives serve as promising lead candidates for further development as inhibitors of genes involved in Cbl metabolism that are regulated by class-II Cbl riboswitches.

Figure 8.

In vivo fluorescence reporter assay shows Cbl derivatives are capable of regulating translation of a protein (mNeonGreen) under control of the Cbl riboswitch (n=3). Protein fluorescence is measured on the right y-axis and shown with symbols. Fluorescence fold repression compared to the no ligand condition is measured on the left y-axis and shown with green box-and-whisker plots. 7, shown to have low affinity for env8 by displacement assay (Figure 3), was used as a negative control. K_rel values of the Cbl derivatives from Table 1 are included for reference.

Summary and Outlook

In this work, we have demonstrated that class-II Cbl riboswitches are able to bind photostable beta-axial cobalamin derivatives and regulate gene expression in E. coli. Chemical probing revealed that a critical aspect of recognition of the beta-axial derivatives is displacement of a nucleotide from a purine stack central to the binding pocket. This displacement mechanism is likely supported through replacement of the displaced nucleotide by intercalation of the derivative functional group as has been seen in other RNAs.⁴⁰ Although none of the tested ligand analogs bound to env8 tighter or had higher in vivo repressive activity than the natural ligand, such qualities are not necessarily required for antibacterial activity.^{18, 51–52} For instance, ribocil, an antimicrobial ligand analog of the FMN riboswitch, binds with 13 times weaker affinity than the natural ligand.¹⁸ However, if additional antibacterial characteristics such as higher affinity are desired, these leads may be further optimized.⁵² For example, given the importance of stacking, heterocycles or multiring systems in the beta-axial position may yield even higher affinity compounds. Even so, the results with the env8 A20G mutant and env50 suggest that Cbl riboswitches that preferentially bind derivative 8 over the natural ligand likely exist. Thus, this study adds to a growing body of evidence that exploiting RNA’s structural plasticity is a productive path in drug discovery efforts.

Complementary to the class-II Cbl riboswitch findings is the observation that an increase in AdoCbl selectivity, integral to class-I Cbl riboswitches, appears to be deleterious to plasticity. It has been hypothesized that class-I Cbl riboswitches are incapable of discriminating between AdoCbl and similar antivitamin B₁₂ forms and some evidence supports this.^{24, 26, 53} However, we show that the E. coli btuB Cbl riboswitch did not bind any of the Cbl derivatives tested through SHAPE and fluorescent methods. This suggests that displacement of the equivalent A20 position in this riboswitch is disallowed. Examination of AdoCbl riboswitch structures provides insights into this behavior. In complex with AdoCbl, a purine in J6/3 (A157 in Figure 5B) is displaced from the purine spine to accommodate the 5’-deoxyadenosyl moiety. If a beta-axial substituted Cbl displaces the nucleotide equivalent to A20 in J3/4 (G42 in Figure 5B) and A157 is positioned out of the purine stack, this leaves a large cavity in the core of the RNA adjacent to the substitution. This cavity would be expected to be significantly deleterious to binding of these derivatives. It is likely that compounds that direct their beta-axial group towards A157 or its equivalent would promote binding by preserving purine spine stacking interactions.

Structural analysis of the novel binding modes discovered here will reveal how the beta-axial derivatives specifically interact with the Cbl riboswitch binding core and yield potential insights into how differences between class-I and class-II result in binding specificities. This work provides new insights to target a broadly distributed RNA in bacteria that regulates metabolism of an essential cofactor for survival and virulence in a number of medically important pathogens.

Materials and Methods

RNA synthesis and preparation

All RNAs used in this study (sequences in Table S2) were made using dsDNA templates amplified by PCR,⁵⁴ transcribed with T7 RNA polymerase, and purified using denaturing PAGE.⁵⁴ Purified RNA was buffer exchanged and concentrated into Milli-Q H₂O using centrifugal concentrators (Amicon). Final RNA concentrations were calculated using A₂₆₀ and molar extinction coefficients determined from the summation of the individual bases.

Cobalamin derivative and probe synthesis

Derivatives 1-11 and the CN probe were synthesized according to previously described procedures.^{28, 31, 36, 55–56} See Supporting Information for further details and synthesis of the PhNO₂ probe.

Fluorescence induction direct binding assay

RNAs were titrated into 30 μL reactions in a Corning 384-well plate such that the final concentrations were 100 nM CNCbl-5xPEG-ATTO590 and 1x of RNA buffer (100 mM KCl, 10 mM NaCl, 1 mM MgCl₂, 50 mM HEPES, pH 8.0).³⁶ The concentration of probe could not be substantially lowered due to the sensitivity of the plate reader. Reactions were done in at least triplicate. Reactions were incubated for at least 30 minutes at room temperature in the dark before reading. Time to equilibration was experimentally validated (Figure S8). ATTO 590 fluorescence was monitored (594 nm excitation, 620–670 nm emission) using a BMG Labtech CLARIOstarPLUS microplate reader. Fluorescence values were background corrected by subtracting buffer fluorescence values, integrated over all wavelengths, and normalized to the average integrated fluorescence of the no RNA control reactions. The corrected normalized integrated fluorescence values were plotted versus log(nM RNA) in GraphPad Prism. Technical replicates were fit to the quadratic binding equation with one transition,

$$Y=m+(n-m)*\left(\right((c+x+K)-\mathit{\text{sqrt}}\left(sqr\right(c+x+K)-(4*c*x\left)\right))/(2*c\left)\right)$$ (1)

where Y is the corrected normalized fluorescence value, m is the lower baseline, n is the upper baseline, c is the probe concentration, x is RNA concentration, and K the K_D. The quadratic equation was used due to the K_D being lower than the concentration of the probe.⁵⁷ Calculated K_Ds are the average of multiple biological replicates, and SEM was calculated from those average K_Ds. Biological replicates were combined for the final graphs shown in the figures.

Fluorescence displacement assays

In experiments using env8, competing ligands were titrated into 30 μL reactions in a Corning 384-well plate such that the final concentrations were 1 μM CNCbl-5xPEG-ATTO590, 100 nM RNA, and 1x of RNA buffer. Reactions were done in technical duplicates or triplicates and in 2–6 biological replicates. Equilibration, plate reading, background subtraction, and fluorescence value integration were done as above. Values were plotted as μM competitor ligand versus integrated fluorescence in GraphPad Prism. Technical replicates were fit to a nonlinear regression curve with the HillSlope parameter set equal to 1.0 and normalized. Fitting gave an IC₅₀ which was used to determine K_D using the equation

$${IC}_{50}=K_{D\left(AX\right)}\left[1+\frac{\left[B\right]}{K_{D\left(AB\right)}}\right]$$ (2)

where A is the RNA, B is CNCbl-5xPEG-ATTO590, and X is the competing ligand.⁵⁸ A K_D of 34 nM between env8 and CNCbl-5xPEG-ATTO590 was used.³⁶ Biological replicates consisting of normalized technical replicates were combined for the final graphs shown in the figures. They were refit to a nonlinear regression curve with the HillSlope set equal to 1.0, the bottom set equal to 1.0, and the top set equal to 0. Calculated K_Ds are the average of multiple biological replicates, and SEM was calculated from those average K_Ds. For qualitative analysis, 30 μL reactions had final concentrations of 1 μM CNCbl-5xPEG-ATTO590, 100 nM RNA, 1x of RNA buffer, and either 1 μM or 100 nM of competing ligand. Corrected and integrated fluorescence values were normalized to those of wells with no competing ligand (“on” = 1) and with no RNA (“off” = 0) and were plotted in GraphPad Prism.

Selective 2’-Hydroxyl Acylation Analyzed by Primer Extension (SHAPE)

Chemical structure probing was performed using N-methylisatoic anhydride (NMIA) as previously described with modifications.⁴³ See Supporting Information for details.

Cell-based reporter assays

Conducted as previously reported²³ with the following changes: 0.5 μL of the overnight culture was added to 500 μL of medium and 150 μL were added to wells in a Costar 96-well plate.