Structure-informed alterations in chemical structure impact binding to aptamer. Previous work demonstrated that a synthetic dibenzofuran ligand has high affinity and selectivity to both Bsu-PreQ1 and Tte-PreQ1 riboswitches.36 We used ICM MolSoft software38 to dock various chemical scaffolds related to the initial dibenzofuran hit compound that had been reported previously. We conducted structural modifications and synthesized various xanthone derivatives to assess their biologically relevant interactions with the PreQ1 riboswitch. By altering the side chains and incorporating other modifications, as detailed in Fig. 1, we synthesized several analogs and examined their recognition ability and activity. A goal of this exercise was to identify new, synthetically accessible chemical scaffolds potentially capable of improved affinity or activity by making more contacts to the RNA. Using this approach, we designed and synthesized nine small molecule ligands representing new heterocyclic cores or sidechains that could plausibly bind to the PreQ1 aptamer (Fig. 1).
To evaluate the binding of each compound to the RNA, we employed fluorescence titrations or microscale thermophoresis (MST).39 Compounds 1, 2, 4, 5, and 8 showed changes in ligand fluorescence with increasing concentrations of RNA. For these compounds, ligand fluorescence was plotted as a function of RNA concentration. Data were fitted using a one-site total binding model to measure an equilibrium dissociation constant (KD) for each of these compounds. Next, compounds that did not show any fluctuation of ligand fluorescence, (3, 6, 7, and 9) were evaluated using MST using a Cy5-labeled Bsu PreQ1 aptamer. By fitting the curves as a function of ligand concentration and using a one-site total binding model, an equilibrium dissociation constant (KD) was measured (Fig. 1A, Supplementary Fig. 1). In general, most of the ligands bound to the RNA with low micromolar affinity. However, compound 9 showed no binding up to a concentration of 500 µM. Of the remaining compounds, 4 showed micromolar binding (KD= 16 ± 21 µM), and therefore binding was also evaluated with another dye, AlexaFluor 647, to rule out potential effects due to the fluorophore used in binding analysis. Using labeled aptamers from Staphylococcus saprophyticus (Ssa)-PreQ1 and Tte-PreQ1 that have a conserved binding domain, compound 4 demonstrated apparent KD values of 21.9 ± 2.25 µM and 29.0 ± 2.4 µM, respectively, in MST (Supplementary Fig. 2), confirming direct binding to the RNA. Next, compounds were used in in vitro assays for functional evaluation.
Synthetic ligands are active in single round transcriptional termination assays. In vitro single round transcription termination assays were performed to biochemically analyze the activity of analogs in functional assays. The Ssa-PreQ1 riboswitch was subjected to transcription in the presence of increasing concentrations of ligands, which could either result into a full transcript read through (RT) or a transcription termination due to the formation of terminator hairpin (T). T and RT products were visualized on a denaturing PAGE and termination efficiency (T50) value was calculated by dividing the terminated band intensity by the total RNA intensity. These data show enhanced biochemical activity of 4 (T50 = 11.1 ± 0.10 µM) and 8 (T50 = 6.8 ± 0.45 µM) in-vitro relative to the initial dibenzofuran. (Fig. 1B, C, Supplementary Fig. 3). For the remaining compounds, saturation was not observed at the limit of solubility, and therefore accurate T50 values could not be measured. Since compounds 4 and 8 showed activity in functional assays, they were studied further.
X-ray co-crystal structure establishes ligand binding mode. To further understand the binding mode of 4 and 8, we performed X-ray crystallography on the ligand aptamer complex. The co-crystal structures of the abasic mutant at positions 13, 14, and 15 in Tte-PreQ1 riboswitch aptamer (ab13_14_15) with 4 and 8 were determined at 2.15 Å and 2.25 Å resolution, respectively, by molecular replacement method (Fig. 2A, 2B and Supplementary table 4). Compound 4 binds at the PreQ1 binding site, where the xanthone core is sandwiched by one face with G11 and the other with G5 and C16, residues that are strictly conserved in the class I PreQ1 riboswitches. When the current co-crystal structure is superimposed onto the PreQ1-bound form, the planar rings of their ligands are well overlapped (Fig. 2C). However, because 4 is bulkier than PreQ1 and its heteroatom content is less than that of PreQ1, the binding pose of 4 slightly diverges from that of PreQ1. In the co-crystal structure with PreQ1, one side of the base containing the N2, N3, and N9 atoms of PreQ1 is recognized by strict hydrogen bonds with the N1 and N6 atoms of A29 and the O4 atom of U6 of the Tte-PreQ1 riboswitch, respectively (Fig. 2D). In contrast, the corresponding side of the xanthone moiety of 4 is further from these crucial atoms, resulting in a tilted binding axis of the heterocyclic core of 4 compared to PreQ1 of approximately 15 degrees (Fig. 2C). Consequently, the oxygen atom of the central ring of 4 is situated at 3.6 Å away from the N6 atom of the phylogenetically conserved A29. This finding suggests a weak hydrogen bonding interaction between the riboswitch and compound 4, unlike the strong interaction observed in the PreQ1-bound form where the distance between the N6 atom of A29 and the N3 atom of PreQ1, a counterpart of the oxygen atom of the central pyranoid ring of 4, is 3.1 Å. Superimposition of these two structures indicates that 4 collides with the base of C15 of the PreQ1-bound structure, due to the size of 4 being larger than that of PreQ1. Consequently, the conformation of the sugar and phosphate backbone at position 15 of the current structure is relocated to fit 4 into the ligand binding site, when compared to the PreQ1-bound form. It is important to note that C15 of the riboswitch is critical for recognizing PreQ1 via the canonical Watson-Crick base pairing. Therefore, the binding of 4 to the PreQ1 binding site would affect the conformations of L2 and S3 that are important for regulating the riboswitch function. Consistent with this, the co-crystal structure with 4 exhibits conformational differences of L2 and S3 when compared to those in the PreQ1-bound form. Since the abasic mutant at positions 13, 14, and 15 was used in this study, we cannot rule out the possibility that the conformational differences are due to the introduction of the abasic sites in the current construct. However, given the steric hindrance between 4 and C15, 4 probably has a major effect on the structure of these regions, which could be related to the differences in the results of biochemical analyses described below.
Like 4, compound 8 is situated in the PreQ1 binding site and is surrounded by the phylogenetically conserved nucleotides. While 4 forms a hydrogen bond with the riboswitch, 8 does not hydrogen bond with any nucleotides. Therefore, 8 is stabilized in the ligand binding site of the riboswitch by stacking and hydrophobic interactions. Compared to the binding site of 4, the binding site of 8 is shifted about 1–2 Å in the opposite direction of the L2 loop. This is likely because 8 has no hydrogen bond with the riboswitch, which anchors the compound in the ligand binding site (Fig. 2F). In our previous report36, we analyzed the effects of dibenzofuran and carbazole derivatives on PreQ1 riboswitch function and showed that the binding poses of these compounds differ due to changes in the acceptor/donor pair of hydrogen-bond between these compounds and the riboswitch. Ligand 8 is a derivative of harmol and has a nitrogen atom in the central ring like the previous carbazole derivative. However, the binding pose of 8 is quite different when compared other nitrogenous heterocycle ligands (such as PDB ID: 6E1V), and the nitrogen atom of the central ring of 8 faces in the opposite direction. Therefore, the conserved nucleotides, U6 and A29, that are crucial for recognizing PreQ1 by hydrogen-bonds only contact with the heterocycle of 8 via van der Waals interactions. Together, these structures provide a rationale for both how diverse ligands recognize the aptamer binding site and why they are active in functional assays.
Structure probing reveals impacts of ligand binding on aptamer flexibility. PreQ1 riboswitches are among the most commonly evolved riboswitches, and as such, have been observed to have considerable diversity in terms of sequence, structure, and mechanisms. Given the diversity of RNA structures that recognize the PreQ1 metabolite, we asked whether evolutionarily diverse PreQ1 aptamers have differential effects on ligand-mediated recognition and flexibility. We utilized selective 2’-hydroxyl acylation analyzed by primer extension and mutational profiling (SHAPE-MaP) to assess the flexibility of bases at single nucleotide level in the presence and absence of both PreQ1 and 4.40 PreQ1 RNA aptamers from six different species were selected, representing all three classes of PreQ1 riboswitch. We studied aptamers from Tte19, 41, Bsu19, 41, Faecalibacterium prausnitzii (Fpr)24, Ssa36, Lactobacillus rhamnosus (Lrh)41–43, Streptococcus pneumoniae (Spn).44, 45
Each in-vitro synthesized RNA was folded and incubated with DMSO, PreQ1 or 4, followed by incubation with the SHAPE reagent 2A3.46 Modified and unmodified RNAs were reverse transcribed, and mutations were mapped by next generation sequencing. Data analysis using the Shapemapper pipeline47 revealed mutation rates and the reactivity profile for each nucleotide. Here, lower SHAPE reactivity depicts decreased flexibility (or stabilization) of each nucleoside in the presence of ligand. Next, SHAPE constraints for each nucleotide were utilized to predict secondary structure with RiboSketch software (Fig. 3A-F).48 Base-pairing probabilities using the SHAPE derived data are shown using the arc plots using Superfold49. Delta SHAPE analysis was then used to identify nucleotides specifically altered in flexibility upon binding to the ligand.50 After accessing significant changes in the SHAPE reactivity within different riboswitches, Bsu, Lrh, Spn, Fpr showed alteration of structure in presence of both ligands. Specifically, the Bsu riboswitch showed stabilization of structure at C20, A21, C22 belonging to aptamer domain, (consistent with crystallographic studies),18 as well as at the C53 and U60 bases within the terminator domain. However, A42, C43, G44 and terminator hairpin bases U55, U56, G57 display destabilization in presence of PreQ1 ligand (Supplementary Fig. 4A). With PreQ1 and the Spn aptamer, A52, G53, G54, A55, G56 (belonging to the loop J2-4) were stabilized and A41, U42, A43, A44, C45 (that makeup the P4 stem)44 are destabilized. This effect was strikingly similar in the presence of 4, as A41, U42, A43 were stabilized along with the destabilization of A41, U42, A43, A44 bases (Supplementary Fig. 4B,C). The Lrh aptamer in presence of PreQ1, A31, U32, U33, C36, U37, U38 (J2-3 loop), G57 (P4 region) were observed to have positive delta SHAPE inferring stabilization and bases U49, A50, U 51, U52, A53 (J2-4 loop), A59, A60 (P4 region) had negative delta SHAPE (informed from crystal structure).51 With 4, the effect was similar to PreQ1, where bases U30, A31, U32, U33, C36, U37 along with U49, A50 displayed stabilization in the aptamer. However, bases G40, A41, U42 (P3), U52, A53 (J2-4), A69, G70, G71, A72, incorporated in the ribosome binding site (RBS) showed significant destabilization (Supplementary Fig. 4D,E). In Fpr PreQ1-riboswitch, with ΔSHAPE, only destabilizing events were captured with both ligands, as bases G110, G111, A112, G113 (constituting ribosome binding site) had enhanced reactivities. With PreQ1, only A115 showed positive ΔSHAPE, meaning stabilization (Supplementary Fig. 4F,G). In contrast, the Tte PreQ1 riboswitch displayed only destabilization in the presence of PreQ1 (at A24, C25, A26, A27, A28, A29, which have no interaction with the PreQ1 ligand) and had no significant ΔSHAPE reactivity when 4 was bound (Supplementary Fig. 4H). In general, all structures displayed decreased reactivity in the presence of both ligands in comparison to the DMSO control, reflecting an overall stabilization of the structure. Both natural (PreQ1) and synthetic ligand (4) had strikingly similar effects on RNA conformation except for a few nucleotides. While these secondary structures are experimentally informed, they do not necessarily reflect three-dimensional aptamer structure with perfect accuracy. Still, this comparative analysis is a powerful demonstration that chemically distinct ligands can have similar effects on structurally diverse RNAs that recognize a common cognate ligand.
Ligand binding stabilizes RNA structure. Having confirmed that ligand binding leads to significant changes in the structure of PreQ1 RNA using bulk methods in solution, we next used the MAGNA magnetic force spectroscopy platform to evaluate the effects of ligand binding on the stability of the aptamer’s structure at the single molecule level. This platform allows precise tracking of molecular extension in response to an applied force across hundreds of single molecules in parallel to gain insights into molecular dynamics and interactions. To use MAGNA, first, a biotinylated Bsu-PreQ1 aptamer was bound to a streptavidin paramagnetic bead and tethered to a flow cell floor via hybridization to a surface-bound oligonucleotide. A precisely controllable magnetic force was then applied to the beads whilst their vertical or Z-positions were tracked in real time. When the RNA was subjected to low force, it folded freely. As the force was increased, structural disruption or unfolding occurred, resulting in a sudden change in vertical bead position. The force could then be reduced, allowing the structures to return to a folded conformation (Fig. 4A). This non-destructive process was repeated over multiple cycles of slowly increasing, then decreasing forces (referred to as force ramp experiments, Supplementary Fig. 5A), while the forces at which individual structures unfolded and refolded were measured. Addition of ligands to the flow cell allowed tracking of their impact on the stability of the RNA structures, through their effect on these unfolding and folding forces. Separately, stepped constant-force experiments were performed where RNA molecules were subjected to the same force for a fixed amount of time before increasing the force in a stepwise manner (Supplementary Fig. 5A). During each force step, the transition of the RNA between the unfolded and folded states was tracked, and the time spent in the unfolded state was observed to increase with force until the RNA structures remained constantly unfolded. The equilibrium force at which the RNA spent equal time in each state was also determined. Constant force experiments in which the RNA was subjected to the equilibrium force for an extended period could then be performed, to allow the impact of ligand binding on folding and unfolding dynamics to be explored through changes in the equilibrium force and/or the frequency of folding-unfolding events (Supplementary Fig. 5A).
We conducted ramp experiments to probe RNA structure unfolding under varied conditions: control (1% DMSO), 4, and PreQ1 and plotted the distribution of the normalized forces required to unfold and refold the RNA structures. In the control condition, the force distribution formed a single peak (Fig. 4B) which was attributed to unfolding of a “pre-pseudoknot” structure. Introduction of a saturating concentration (500 µM) of 4 subtly shifted the peak of the force distribution toward higher forces, implying minor structural influence that increases the force needed to unfold and refold the structure (Fig. 4B & Supplementary Fig. 5B). In contrast, saturating concentrations of PreQ1 (500nM) induced a second peak in higher forces, indicating that the molecules sometimes required a much higher force to unfold, which was attributed to the formation of stable pseudoknot structures. However, the position of the first peak did not shift, demonstrating that PreQ1 did not change the stability of the pre-pseudoknot structure (Fig. 4B and Supplementary Fig. 5B. The second high force peak was notably absent with 4, highlighting that the compound did not trigger the formation of persistent/stable pseudoknots like those induced by PreQ1 (Fig. 4B).
The increase caused by compound 4 on both the median normalized unfolding and refolding forces was shown to be concentration dependent (Fig. 4C and Supplementary Fig. 5C respectively) with EC50 values of 394 ± 157 µM and 456 ± 258 µM respectively, suggesting that 4 stabilizes the riboswitch pre-pseudoknot structure and that this interaction helps to refold the RNA. For the PreQ1 ligand, concentration effect was assessed differently, using the fraction of high force cycles, to account for cycles in which the RNA in pseudoknot conformation did not unfold at the maximal force applied. PreQ1 binding increased the fraction of the high-force cycles in a concentration dependent manner indicating an increase in pseudoknot formation whilst RNA refolding was not affected by the cognate ligand (Fig. 4C and Supplementary Fig. 5D).
Under constant-force experiments, the bead position tracking of individual molecules showed that the PreQ1 ligand prolonged folded state duration compared to the control and 4 (at the same applied forces), revealing ligand-induced pseudoknot formation (Fig. 4D, Supplementary Fig. 5E). Analysis of the lifetimes of the folded states under the control, PreQ1 (500 nM) and 4 (500 µM) conditions showed an exponential distribution of the observed events with 4 inducing a slightly increased lifetime compared to the control. In the presence of the cognate ligand, the folded states fitted a combination of two distinct lifetime distributions (confirmed using the Bayesian information criterion) (Supplementary Fig. 5F, representing one single molecule). Of these two lifetimes, the shorter of the two showed a lifetime similar to that of the control, most likely corresponding to the pre-pseudoknot state, and the second corresponded to the stable folded state attributed to the pseudoknot.
To evaluate concentration dependency of the effects, lifetime data from multiple molecules were aggregated by evaluating log (1/lifetime) and normalizing each condition to the control for the same molecule before combining data from multiple molecules. Compound 4 was confirmed to decrease the unfolding rate (i.e., to cause the RNA to stay longer in the folded pre-pseudoknot state) in a concentration dependent manner, but only at concentrations above 100 µM (Fig. 4E). In contrast, the cognate ligand did not affect the unfolding rate of either the short lifetime form (pre-pseudoknot) or long lifetime form (pseudoknot) (Supplementary Fig. 5G) and no change in the refolding rate (Supplementary Fig. 5H). Instead, the probability of the pseudoknot state occurring increased with PreQ1 ligand concentration (Fig. 4F), confirming that binding of the PreQ1 ligand induces pseudoknot formation in a concentration-dependent manner. Compound 4 was also demonstrated to increase the refolding rate in a concentration dependent manner above 100 µM, suggesting that the molecule alters the rate of refolding of the RNA, perhaps by binding to a less folded form (Supplementary Fig. 5I). However, while the PreQ1 ligand affects the rate of pseudoknot formation, it has no effect on the less stable pre-pseudoknot structure’s folding dynamics.
Ligands affect riboswitch activity in cells. Next, we evaluated the ability of 4 to modulate riboswitch activity in vivo. We employed an engineered green fluorescent protein (GFPuv) reporter assay, as has been used previously to demonstrate riboswitch activity.52,53 The Bsu-PreQ1 riboswitch aptamer was cloned into a plasmid bearing GFPuv expression in parallel with a second, empty vector that expresses GFP but lacks any riboswitch. Next, GFPuv positive constructs were transformed into the JW2765 strain of E.coli bearing a ΔqueF mutation 54, 55 to generate a stable cell line expressing the reporter construct. The ΔqueF mutation leads to impaired PreQ1 biosynthesis and provides an ideal system to study the effects of ligands on the PreQ1 riboswitch, as it lacks endogenous PreQ1. Next, cells were grown on specialized CSB media (to further hinder any endogenous PreQ1 biosynthesis) in the presence of compounds or the DMSO control. When visualized under UV, cells grown in the presence of DMSO exhibited high levels of fluorescence (Fig. 5A). In contrast, treatment with PreQ1 and 4 led to a complete loss of fluorescence levels (Fig. 5B, 5C). The cell line expressing an empty vector was not responsive to ligand (Fig. 5D). In addition, a compound structurally similar to 4 that did not bind to the riboswitch (compound 9) was also inactive. Finally, cells treated with 8 also did not respond, even though 8 both binds to and modulates the function of the riboswitch in vitro. Importantly, these results demonstrate that both PreQ1 and 4 clearly exhibit gene modulation activity by directly binding to RNA structures in cells, rather than nonspecific or other off-target mechanisms.