Automated RNA structure prediction uncovers a missing link in double glycine riboswitches

The tertiary structures of functional RNA molecules remain difficult to decipher. A new generation of automated RNA structure prediction methods may help address these challenges but have not yet been experimentally validated. Here we apply four prediction tools to a remarkable class of double glycine riboswitches that exhibit ligand-binding cooperativity. A novel method (BPPalign), RMdetect, JAR3D, and Rosetta 3D modeling give consistent predictions for a new stem P0 and kink-turn motif. These elements structure the linker between the RNAs’ double aptamers. Chemical mapping on the F. nucleatum riboswitch with SHAPE, DMS, and CMCT probing, mutate-and-map studies, and mutation/rescue experiments all provide strong evidence for the structured linker. Under solution conditions that separate two glycine binding transitions, disrupting this helix-junction-helix structure gives 120-fold and 6to 30-fold poorer association constants for the two transitions, corresponding to an overall energetic impact of 4.3 ± 0.5 kcal/mol. Prior biochemical and crystallography studies from several labs did not include this critical element due to over-truncation of the RNA. We argue that several further undiscovered elements are likely to exist in the flanking regions of this and other RNA switches, and automated prediction tools can now play a powerful role in their detection and dissection. Non-coding RNA sequences play critical roles in cellular biochemistry and genetic regulation, and their number is growing rapidly. Many of these RNAs’ behaviors are intimately tied to their three-dimensional conformations, but determining these structures has challenged both experimentalists and modelers, especially with regards to tertiary interactions mediated by non-Watson-Crick base pairs. Recent years have seen the development of automated algorithms for detecting and modeling RNA tertiary structure, especially modular recurrent motifs. However, the predictive power of these methods has yet to be demonstrated through rigorous experiments. Here we report the application and chemical validation of RNA structure prediction to discover a previously missed tertiary element in a paradigmatic class of natural RNA riboswitches. The cooperative binding of small molecules is a fundamental feature of functional biopolymers that has been studied in numerous model systems, most of which were protein-based until 2004. In that year, Breaker and co-workers reported ligand-binding cooperativity in an RNA riboswitch that binds two glycine molecules. This discovery has inspired many biophysical studies, culminating in the recent publication of crystallographic models of double aptamers, but a predictive model for ligand-binding cooperativity has remained elusive. Prior investigations have mostly focused on constructs FIGURE 1. A new stem P0 and kink-turn tertiary element in double glycine riboswitches on a sequence alignment (A), on the secondary structure of the F. nucleatum riboswitch (B), and built into this RNA’s 3D crystallographic model (C). Coloring in (B) shows normalized chemical mapping data: SHAPE (on letters), DMS (rectangles at A/C), and CMCT (rectangles at G/U). 2 pared down to minimal sequences required for glycine binding. We therefore sought to explore outer flanking sequences that have so far remained uncharacterized. We started by applying four recently developed RNA structure modeling tools and found remarkably consistent predictions for a new motif. As input into modeling, we extracted aligned sequences of 360 double glycine riboswitches, collating single-aptamer entries in RFAM and extending these alignments into 5 ́ and 3 ́ flanking regions by 100 nucleotides. Inspired by consensus approaches for protein fold recognition, our new BPPalign tool (SI Methods) searches for novel stems by averaging basepair probability calculations for Watson-Crick secondary structure across homologs. In addition to new candidate elements in the 3 ́ region (SI Fig. S1) containing potential ‘expression platforms’, we found signals involving nine nucleotides preceding the conventional start of the riboswitch. For 160 sequences, including the widely studied riboswitches from Vibrio cholerae and Fusobacterium nucleatum (FN), a three base-pair interaction between the riboswitch inter-aptamer linker and nucleotides in the 5 ́ flanking sequence could occur (SI Fig S1 and Fig. 1). We call this putative stem P0. In addition, the nearby P1 stem of the first aptamer was found to potentially form an extension of three purine-purine pairs (129 sequences; Fig. 1) or, in some cases, three Watson-Crick pairs (41 sequences). Finally, while P0 and this extended P1 are contiguous on the 3 ́ strand, there is an intervening three-residue bulge in the 5 ́ strand. We noted that the lengths and sequences of these features matched the published consensus for the kink-turn motif. This motif, originally described based on high-resolution ribosome crystallography, has been annotated in numerous functional RNAs but never previously reported in glycine riboswitches. Three independent lines of bioinformatic/computational evidence supported the presence of the kink-turn motif. First, the automated RMdetect software gave a strong-confidence detection of a kink-turn (101 sequences, SI Fig. S2) when given our extended multiple sequence alignment. Second, the JAR3D server returned a kink-turn motif when given the putative P0/P1 sequences (SI Fig. S3). Third, we applied Rosetta/FARFAR 3D modeling to replace the linker in the FN riboswitch crystallographic model with a kink-turn motif forming a continuous helical interface with P1. The resulting model (Fig. 1C) demonstrated that a kink-turn can bridge the riboswitch aptamers with reasonable geometry and no clashes. Furthermore, we carried out de novo modeling of the entire linker and added 5 ́ strand; encouragingly, the lowest-energy models exhibited kink-turn conformations (SI Fig. S4). Finally, an unexpected concordance was observed: the linker backbone within the modeled kink-turn approximately followed the linker path in the deposited crystallographic electron density and coordinates (4.4 Å C4 ́ RMSD), despite the absence of any pairing partners in the crystallized molecule (SI. Fig. S5). The computational approaches applied above gave consistent predictions, but these tools are largely untested, can give false positives (SI Figs. 1 and 2), and do not provide information on the energetic significance of the putative element. Therefore, we carried out experiments to confirm and further characterize the kink-turn motif using mutate-and-map and mutate/rescue trials, with quantitative readouts from nucleotide-resolution chemical mapping. As a model RNA, we chose the smallest known glycine riboswitch, the FN system, which has been extensively studied by mutation, chemical mapping, and crystallography, albeit in truncated form. We focused on a sequence with a 9-nucleotide natural extension restored to the 5 ́ end compared to the prior construct (Fig. 1B; SI Methods). We called this sequence FN-KTtest. High throughput measurements of dimethyl sulfate (DMS) and carbodiimide (CMCT) modification reported on the chemical accessibilities of Watson-Crick nucleobase edges for A/C and G/U, respectively. Nucleotides outside the linker region served as controls in this analysis; we confirmed that their reactivities in 10 mM MgCl2, 50 mM Na-HEPES, pH 8.0, and 10 mM glycine correlated with the burial of bases in the glycine-bound FN crystallographic model (Fig. 1B). Within the linker region and within the added natural 5 ́ flanking sequence, the DMS and CMCT reactivities were consistent with the 3D model of the kink-turn (protections of nts –2, –4 to –6, and 72 to 77; and exposure of 0, –1, and –3; Figs. 1B and C). Further analysis, based on 2 ́OH acylation with N-methylisatoic acid (SHAPE chemistry), gave protections of linker nucleotides 72-79 (Fig. 1B), as would be expected for a structured element; these nucleotides gave high SHAPE reactivities in previous studies without the 5 ́ flanking sequence. More stringent tests of the predicted kink-turn structure were achieved with mutate-and-map experiments, shown FIGURE 2. Evidence for the predicted kink-turn tertiary element and P0 stem from disruption of chemical reactivities by mutation (MutA and MutB) and restoration (MutAB) of P0 Watson-Crick pairs. Fluorescence from aligned capillary electrophoretic traces from chemical mapping are shown. Arrows mark SHAPE effects discussed in text.