Search

Despite the popularity of computer-aided study and design of RNA molecules, little is known about the accuracy of commonly used structure modeling packages in tasks sensitive to ensemble properties of RNA. Here, we demonstrate that the EternaBench dataset, a set of more than 20,000 synthetic RNA constructs designed on the RNA design platform Eterna, provides incisive discriminative power in evaluating current packages in ensemble-oriented structure prediction tasks. We find that CONTRAfold and RNAsoft, packages with parameters derived through statistical learning, achieve consistently higher accuracy than more widely used packages in their standard settings, which derive parameters primarily from thermodynamic experiments. We hypothesized that training a multitask model with the varied data types in EternaBench might improve inference on ensemble-based prediction tasks. Indeed, the resulting model, named EternaFold, demonstrated improved performance that generalizes to diverse external datasets including complete messenger RNAs, viral genomes probed in human cells and synthetic designs modeling mRNA vaccines.

The rapid spread of COVID-19 is motivating development of antivirals targeting conserved SARS-CoV-2 molecular machinery. The SARS-CoV-2 genome includes conserved RNA elements that offer potential small-molecule drug targets, but most of their 3D structures have not been experimentally characterized. Here, we provide a compilation of chemical mapping data from our and other labs, secondary structure models, and 3D model ensembles based on Rosetta's FARFAR2 algorithm for SARS-CoV-2 RNA regions including the individual stems SL1-8 in the extended 5′ UTR; the reverse complement of the 5′ UTR SL1-4; the frameshift stimulating element (FSE); and the extended pseudoknot, hypervariable region, and s2m of the 3′ UTR. For eleven of these elements (the stems in SL1–8, reverse complement of SL1–4, FSE, s2m and 3′ UTR pseudoknot), modeling convergence supports the accuracy of predicted low energy states; subsequent cryo-EM characterization of the FSE confirms modeling accuracy. To aid efforts to discover small molecule RNA binders guided by computational models, we provide a second set of similarly prepared models for RNA riboswitches that bind small molecules. Both datasets (‘FARFAR2-SARS-CoV-2’, https://github.com/DasLab/FARFAR2-SARS-CoV-2; and ‘FARFAR2-Apo-Riboswitch’, at https://github.com/DasLab/FARFAR2-Apo-Riboswitch’) include up to 400 models for each RNA element, which may facilitate drug discovery approaches targeting dynamic ensembles of RNA molecules., Graphical Abstract Graphical Abstract De novo 3D models of SARS-CoV-2 RNA elements and small-molecule-binding RNAs to aid drug discovery.

The discovery and design of biologically important RNA molecules is outpacing three-dimensional structural characterization. Here, we demonstrate that cryo-electron microscopy can routinely resolve maps of RNA-only systems and that these maps enable subnanometer-resolution coordinate estimation when complemented with multidimensional chemical mapping and Rosetta DRRAFTER computational modeling. This hybrid 'Ribosolve' pipeline detects and falsifies homologies and conformational rearrangements in 11 previously unknown 119- to 338-nucleotide protein-free RNA structures: full-length Tetrahymena ribozyme, hc16 ligase with and without substrate, full-length Vibrio cholerae and Fusobacterium nucleatum glycine riboswitch aptamers with and without glycine, Mycobacterium SAM-IV riboswitch with and without S-adenosylmethionine, and the computer-designed ATP-TTR-3 aptamer with and without AMP. Simulation benchmarks, blind challenges, compensatory mutagenesis, cross-RNA homologies and internal controls demonstrate that Ribosolve can accurately resolve the global architectures of RNA molecules but does not resolve atomic details. These tests offer guidelines for making inferences in future RNA structural studies with similarly accelerated throughput.

Thermostable reverse transcriptases are workhorse enzymes underlying nearly all modern techniques for RNA structure mapping and for the transcriptome-wide discovery of RNA chemical modifications. Despite their wide use, these enzymes' behaviors at chemical modified nucleotides remain poorly understood. Wellington-Oguri et al. recently reported an apparent loss of chemical modification within putatively unstructured polyadenosine stretches modified by dimethyl sulfate or 2' hydroxyl acylation, as probed by reverse transcription. Here, reanalysis of these and other publicly available data, capillary electrophoresis experiments on chemically modified RNAs, and nuclear magnetic resonance spectroscopy on (A)12 and variants show that this effect is unlikely to arise from an unusual structure of polyadenosine. Instead, tests of different reverse transcriptases on chemically modified RNAs and molecules synthesized with single 1-methyladenosines implicate a previously uncharacterized reverse transcriptase behavior: near-quantitative bypass through chemical modifications within polyadenosine stretches. All tested natural and engineered reverse transcriptases (MMLV; SuperScript II, III, and IV; TGIRT-III; and MarathonRT) exhibit this anomalous bypass behavior. Accurate DMS-guided structure modeling of the polyadenylated HIV-1 3' untranslated region requires taking into account this anomaly. Our results suggest that poly(rA-dT) hybrid duplexes can trigger an unexpectedly effective reverse transcriptase bypass and that chemical modifications in mRNA poly(A) tails may be generally undercounted.

Internet-based scientific communities promise a means to apply distributed, diverse human intelligence toward previously intractable scientific problems. However, current implementations have not allowed communities to propose experiments to test all emerging hypotheses at scale or to modify hypotheses in response to experiments. We report high-throughput methods for molecular characterization of nucleic acids that enable the large-scale video game–based crowdsourcing of RNA sensor design, followed by high-throughput functional characterization. Iterative design testing of thousands of crowdsourced RNA sensor designs produced near–thermodynamically optimal and reversible RNA switches that act as self-contained molecular sensors and couple five distinct small molecule inputs to three distinct protein binding and fluorogenic outputs. This work suggests a paradigm for widely distributed experimental bioscience.

Medicines based on messenger RNA (mRNA) hold immense potential, as evidenced by their rapid deployment as COVID-19 vaccines. However, worldwide distribution of mRNA molecules has been limited by their thermostability, which is fundamentally limited by the intrinsic instability of RNA molecules to a chemical degradation reaction called in-line hydrolysis. Predicting the degradation of an RNA molecule is a key task in designing more stable RNA-based therapeutics. Here, we describe a crowdsourced machine learning competition (‘Stanford OpenVaccine’) on Kaggle, involving single-nucleotide resolution measurements on 6,043 diverse 102–130-nucleotide RNA constructs that were themselves solicited through crowdsourcing on the RNA design platform Eterna. The entire experiment was completed in less than 6 months, and 41% of nucleotide-level predictions from the winning model were within experimental error of the ground truth measurement. Furthermore, these models generalized to blindly predicting orthogonal degradation data on much longer mRNA molecules (504–1,588 nucleotides) with improved accuracy compared with previously published models. These results indicate that such models can represent in-line hydrolysis with excellent accuracy, supporting their use for designing stabilized messenger RNAs. The integration of two crowdsourcing platforms, one for dataset creation and another for machine learning, may be fruitful for other urgent problems that demand scientific discovery on rapid timescales.

Currently approved anti-influenza drugs target viral proteins, are subtype limited, and are challenged by rising antiviral resistance. To overcome these limitations, we sought to identify a conserved essential RNA secondary structure within the genomic RNA predicted to have greater constraints on mutation in response to therapeutics targeting this structure. Here, we identified and genetically validated an RNA stemloop structure we termed PSL2, which serves as a packaging signal for genome segment PB2 and is highly conserved across influenza A virus (IAV) isolates. RNA structural modeling rationalized known packaging-defective mutations and allowed for predictive mutagenesis tests. Disrupting and compensating mutations of PSL2’s structure give striking attenuation and restoration, respectively, of in vitro virus packaging and mortality in mice. Antisense Locked Nucleic Acid oligonucleotides (LNAs) designed against PSL2 dramatically inhibit IAV in vitro against viruses of different strains and subtypes, possess a high barrier to the development of antiviral resistance, and are equally effective against oseltamivir carboxylate-resistant virus. A single dose of LNA administered 3 days after, or 14 days before, a lethal IAV inoculum provides 100% survival. Moreover, such treatment led to the development of strong immunity to rechallenge with a ten-fold lethal inoculum. Together, these results have exciting implications for the development of a versatile novel class of antiviral therapeutics capable of prophylaxis, post-exposure treatment, and “just-in-time” universal vaccination against all IAV strains, including drug-resistant pandemics.One Sentence SummaryTargeting a newly identified conserved RNA structure in the packaging signal region of influenza segment PB2 abrogates virus production in vitro and dramatically attenuates disease in vivo.

RNA nanotechnology seeks to create nanoscale machines by repurposing natural RNA modules. The field is slowed by the current need for human intuition during three-dimensional structural design. Here, we demonstrate that three distinct problems in RNA nanotechnology can be reduced to a pathfinding problem and automatically solved through an algorithm called RNAMake. First, RNAMake discovers highly stable single-chain solutions to the classic problem of aligning a tetraloop and its sequence-distal receptor, with experimental validation from chemical mapping, gel electrophoresis, solution X-ray scattering and crystallography with 2.55 A resolution. Second, RNAMake automatically generates structured tethers that integrate 16S and 23S ribosomal RNAs into single-chain ribosomal RNAs that remain uncleaved by ribonucleases and assemble onto messenger RNA. Third, RNAMake enables the automated stabilization of small-molecule binding RNAs, with designed tertiary contacts that improve the binding affinity of the ATP aptamer and improve the fluorescence and stability of the Spinach RNA in cell extracts and in living Escherichia coli cells. Automated 3D design produces rapid and near-atomically accurate predictions of RNA tertiary structure as well as the ability to generate complex RNA machines such as functional single-stranded tethered ribosomes, and enhancement of the binding properties of small-molecule RNA aptamers.

Riboswitches that couple binding of ligands to conformational changes offer sensors and control elements for RNA synthetic biology and medical biotechnology. However, design of these riboswitches has required expert intuition or software specialized to transcription or translation outputs; design has been particularly challenging for applications in which the riboswitch output cannot be amplified by other molecular machinery. We present a fully automated design method called RiboLogic for such “stand-alone” riboswitches and test it via high-throughput experiments on 2875 molecules using RNA-MaP (RNA on a massively parallel array) technology. These molecules consistently modulate their affinity to the MS2 bacteriophage coat protein upon binding of flavin mononucleotide, tryptophan, theophylline, and microRNA miR-208a, achieving activation ratios of up to 20 and significantly better performance than control designs. By encompassing a wide diversity of stand-alone switches and highly quantitative data, the resulting ribologic-solves experimental data set provides a rich resource for further improvement of riboswitch models and design methods.

Therapeutic mRNAs and vaccines are being developed for a broad range of human diseases, including COVID-19. However, their optimization is hindered by mRNA instability and inefficient protein expression. Here, we describe design principles that overcome these barriers. We develop an RNA sequencing-based platform called PERSIST-seq to systematically delineate in-cell mRNA stability, ribosome load, as well as in-solution stability of a library of diverse mRNAs. We find that, surprisingly, in-cell stability is a greater driver of protein output than high ribosome load. We further introduce a method called In-line-seq, applied to thousands of diverse RNAs, that reveals sequence and structure-based rules for mitigating hydrolytic degradation. Our findings show that highly structured “superfolder” mRNAs can be designed to improve both stability and expression with further enhancement through pseudouridine nucleoside modification. Together, our study demonstrates simultaneous improvement of mRNA stability and protein expression and provides a computational-experimental platform for the enhancement of mRNA medicines.

Drug discovery campaigns against Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) are beginning to target the viral RNA genome (1, 2) . The frameshift stimulation element (FSE) of the SARS-CoV-2 genome is required for balanced expression of essential viral proteins and is highly conserved, making it a potential candidate for antiviral targeting by small molecules and oligonucleotides (3–6) . To aid global efforts focusing on SARS-CoV-2 frameshifting, we report exploratory results from frameshifting and cellular replication experiments with locked nucleic acid (LNA) antisense oligonucleotides (ASOs), which support the FSE as a therapeutic target but highlight difficulties in achieving strong inactivation. To understand current limitations, we applied cryogenic electron microscopy (cryo-EM) and the Ribosolve (7) pipeline to determine a three-dimensional structure of the SARS-CoV-2 FSE, validated through an RNA nanostructure tagging method. This is the smallest macromolecule (88 nt; 28 kDa) resolved by single-particle cryo-EM at subnanometer resolution to date. The tertiary structure model, defined to an estimated accuracy of 5.9 Å, presents a topologically complex fold in which the 5′ end threads through a ring formed inside a three-stem pseudoknot. Our results suggest an updated model for SARS-CoV-2 frameshifting as well as binding sites that may be targeted by next generation ASOs and small molecules.

Drug discovery campaigns against Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) are beginning to target the viral RNA genome1, 2. The frameshift stimulation element (FSE) of the SARS-CoV-2 genome is required for balanced expression of essential viral proteins and is highly conserved, making it a potential candidate for antiviral targeting by small molecules and oligonucleotides3–6. To aid global efforts focusing on SARS-CoV-2 frameshifting, we report exploratory results from frameshifting and cellular replication experiments with locked nucleic acid (LNA) antisense oligonucleotides (ASOs), which support the FSE as a therapeutic target but highlight difficulties in achieving strong inactivation. To understand current limitations, we applied cryogenic electron microscopy (cryo-EM) and the Ribosolve7 pipeline to determine a three-dimensional structure of the SARS-CoV-2 FSE, validated through an RNA nanostructure tagging method. This is the smallest macromolecule (88 nt; 28 kDa) resolved by single-particle cryo-EM at subnanometer resolution to date. The tertiary structure model, defined to an estimated accuracy of 5.9 Å, presents a topologically complex fold in which the 5′ end threads through a ring formed inside a three-stem pseudoknot. Our results suggest an updated model for SARS-CoV-2 frameshifting as well as binding sites that may be targeted by next generation ASOs and small molecules.

The computer-aided study and design of RNA molecules is increasingly prevalent across a range of disciplines, yet little is known about the accuracy of commonly used structure modeling packages in tasks sensitive to ensemble properties of RNA. Here, we demonstrate that the EternaBench dataset, a set of over 20,000 synthetic RNA constructs designed in iterative cycles on the RNA design platform Eterna, provides incisive discriminative power in evaluating current packages in ensemble-oriented structure prediction tasks. We find that CONTRAfold and RNAsoft, packages with parameters derived through statistical learning, achieve consistently higher accuracy than more widely used packages in their standard settings, which derive parameters primarily from thermodynamic experiments. Motivated by these results, we develop a multitask-learning-based model, EternaFold, which demonstrates improved performance that generalizes to diverse external datasets, including complete mRNAs and viral genomes probed in human cells and synthetic designs modeling mRNA vaccines.

The rapid spread of COVID-19 is motivating development of antivirals targeting conserved SARS-CoV-2 molecular machinery. The SARS-CoV-2 genome includes conserved RNA elements that offer potential small-molecule drug targets, but most of their 3D structures have not been experimentally characterized. Here, we provide a compilation of chemical mapping data from our and other labs, secondary structure models, and 3D model ensembles based on Rosetta’s FARFAR2 algorithm for SARS-CoV-2 RNA regions including the individual stems SL1-8 in the extended 5’ UTR; the reverse complement of the 5’ UTR SL1-4; the frameshift stimulating element (FSE); and the extended pseudoknot, hypervariable region, and s2m of the 3’ UTR. For eleven of these elements (the stems in SL1-8, reverse complement of SL1-4, FSE, s2m, and 3’ UTR pseudoknot), modeling convergence supports the accuracy of predicted low energy states; subsequent cryo-EM characterization of the FSE confirms modeling accuracy. To aid efforts to discover small molecule RNA binders guided by computational models, we provide a second set of similarly prepared models for RNA riboswitches that bind small molecules. Both datasets (‘FARFAR2-SARS-CoV-2’, https://github.com/DasLab/FARFAR2-SARS-CoV-2; and ‘FARFAR2-Apo-Riboswitch’, at https://github.com/DasLab/FARFAR2-Apo-Riboswitch’) include up to 400 models for each RNA element, which may facilitate drug discovery approaches targeting dynamic ensembles of RNA molecules.

Despite the critical roles RNA structures play in regulating gene expression, sequencing-based methods for experimentally determining RNA base pairs have remained inaccurate. Here, we describe a multidimensional chemical mapping method called M2-seq (mutate-and-map read out through next-generation sequencing) that takes advantage of sparsely mutated nucleotides to induce structural perturbations at partner nucleotides and then detects these events through dimethyl sulfate (DMS) probing and mutational profiling. In special cases, fortuitous errors introduced during DNA template preparation and RNA transcription are sufficient to give M2-seq helix signatures; these signals were previously overlooked or mistaken for correlated double DMS events. When mutations are enhanced through error-prone PCR,in vitroM2-seq experimentally resolves 33 of 68 helices in diverse structured RNAs including ribozyme domains, riboswitch aptamers, and viral RNA domains with a single false positive. These inferences do not require energy minimization algorithms and can be made by either direct visual inspection or by a new neural-net-inspired algorithm called M2-net. Measurements on the P4-P6 domain of theTetrahymenagroup I ribozyme embedded inXenopusegg extract demonstrate the ability of M2-seq to detect RNA helices in a complex biological environment.SIGNIFICANCE STATEMENTThe intricate structures of RNA molecules are crucial to their biological functions but have been difficult to accurately characterize. Multidimensional chemical mapping methods improve accuracy but have so far involved painstaking experiments and reliance on secondary structure prediction software. A methodology called M2-seq now lifts these limitations. Mechanistic studies clarify the origin of serendipitous M2-seq-like signals that were recently discovered but not correctly explained and also provide mutational strategies that enable robust M2-seq for new RNA transcripts. The method detects dozens of Watson-Crick helices across diverse RNA foldsin vitroand within frog egg extract, with low false positive rate (< 5%). M2-seq opens a route to unbiased discovery of RNA structuresin vitroand beyond.

Internet-based scientific communities promise a means to apply distributed, diverse human intelligence towards previously intractable scientific problems. However, current implementations have not allowed communities to propose experiments to test all emerging hypotheses at scale or to modify hypotheses in response to experiments. We report high-throughput methods for molecular characterization of nucleic acids that enable the large-scale videogame-based crowdsourcing of functional RNA sensor design, followed by high-throughput functional characterization. Iterative design testing of thousands of crowdsourced RNA sensor designs produced near-thermodynamically optimal and reversible RNA switches that act as self-contained molecular sensors and couple five distinct small molecule inputs to three distinct protein binding and fluorogenic outputs—results that surpass computational and expert-based design. This work represents a new paradigm for widely distributed experimental bioscience.One Sentence SummaryOnline community discovers standalone RNA sensors.

The discovery and design of biologically important RNA molecules is dramatically outpacing three-dimensional structural characterization. To address this challenge, we present Ribosolve, a hybrid method integrating moderate-resolution cryo-EM maps, chemical mapping, and Rosetta computational modeling, and demonstrate its application to thirteen previously unknown 119-to 338-nucleotide protein-free RNA-only structures: full-length Tetrahymena ribozyme, hc16 ligase with and without substrate, full-length V. cholerae and F. nucleatum glycine riboswitch aptamers with and without glycine, Mycobacterium SAM-IV riboswitch with and without S-adenosylmethionine, and computer-designed spinach-TTR-3, eterna3D-JR_1, and ATP-TTR-3 with and without AMP. Blind challenges, prospective compensatory mutagenesis, internal controls, and simulation benchmarks validate the Ribosolve models and establish that modeling convergence is quantitatively predictive of model accuracy. These results demonstrate that RNA-only 3D structure determination can be rapid and routine.

Riboswitches that couple binding of ligands to recruitment of molecular machines offer sensors and control elements for RNA synthetic biology and medical biotechnology. Current approaches to riboswitch design enable significant changes in output activity in the presence vs. absence of input ligands. However, design of these riboswitches has so far required expert intuition and explicit specification of complete target secondary structures, both of which limit the structure-toggling mechanisms that have been explored. We present a fully automated method called RiboLogic for these design tasks and high-throughput experimental tests of 2,875 molecules using RNA-MaP (RNA on a massively parallel array) technology. RiboLogic designs explore an unprecedented diversity of structure-toggling mechanisms validated through experimental tests. These synthetic molecules consistently modulate their affinity to the MS2 bacteriophage coat protein upon binding of flavin mononucleotide, tryptophan, theophylline, and microRNA miR-208a, achieving activation ratios of up to 20 and significantly better performance than control designs. The data enable dissection of features of structure-toggling mechanisms that correlate with higher performance. The diversity of RiboLogic designs and their quantitative experimental characterization provides a rich resource for further improvement of riboswitch models and design methods.

The predictive modeling and design of biologically active RNA molecules requires understanding the energetic balance among their basic components. Rapid developments in computer simulation promise increasingly accurate recovery of RNA's nearest-neighbor (NN) free-energy parameters, but these methods have not been tested in predictive trials or on nonstandard nucleotides. Here, we present, to our knowledge, the first such tests through a RECCES-Rosetta (reweighting of energy-function collection with conformational ensemble sampling in Rosetta) framework that rigorously models conformational entropy, predicts previously unmeasured NN parameters, and estimates these values' systematic uncertainties. RECCES-Rosetta recovers the 10 NN parameters for Watson-Crick stacked base pairs and 32 single-nucleotide dangling-end parameters with unprecedented accuracies: rmsd of 0.28 kcal/mol and 0.41 kcal/mol, respectively. For set-aside test sets, RECCES-Rosetta gives rmsd values of 0.32 kcal/mol on eight stacked pairs involving G-U wobble pairs and 0.99 kcal/mol on seven stacked pairs involving nonstandard isocytidine-isoguanosine pairs. To more rigorously assess RECCES-Rosetta, we carried out four blind predictions for stacked pairs involving 2,6-diaminopurine-U pairs, which achieved 0.64 kcal/mol rmsd accuracy when tested by subsequent experiments. Overall, these results establish that computational methods can now blindly predict energetics of basic RNA motifs, including chemically modified variants, with consistently better than 1 kcal/mol accuracy. Systematic tests indicate that resolving the remaining discrepancies will require energy function improvements beyond simply reweighting component terms, and we propose further blind trials to test such efforts.

We report a new algorithm and a battery of blind challenges for the prediction of complex RNA structures at atomic accuracy., Prediction of RNA structure from nucleotide sequence remains an unsolved grand challenge of biochemistry and requires distinct concepts from protein structure prediction. Despite extensive algorithmic development in recent years, modeling of noncanonical base pairs of new RNA structural motifs has not been achieved in blind challenges. We report a stepwise Monte Carlo (SWM) method with a unique add-and-delete move set that enables predictions of noncanonical base pairs of complex RNA structures. A benchmark of 82 diverse motifs establishes the method’s general ability to recover noncanonical pairs ab initio, including multistrand motifs that have been refractory to prior approaches. In a blind challenge, SWM models predicted nucleotide-resolution chemical mapping and compensatory mutagenesis experiments for three in vitro selected tetraloop/receptors with previously unsolved structures (C7.2, C7.10, and R1). As a final test, SWM blindly and correctly predicted all noncanonical pairs of a Zika virus double pseudoknot during a recent community-wide RNA-Puzzle. Stepwise structure formation, as encoded in the SWM method, enables modeling of noncanonical RNA structure in a variety of previously intractable problems.

RNA is a functionally versatile molecule that plays key roles in genetic regulation and in emerging technologies to control biological processes. Computational models of RNA secondary structure are well-developed but often fall short in making quantitative predictions of the behavior of multi-RNA complexes. Recently, large datasets characterizing hundreds of thousands of individual RNA complexes have emerged as rich sources of information about RNA energetics. Meanwhile, advances in machine learning have enabled the training of complex neural networks from large datasets. Here, we assess whether a recurrent neural network model, Ribonet, can learn from high-throughput binding data, using simulation and experimental studies to test model accuracy but also determine if they learned meaningful information about the biophysics of RNA folding. We began by evaluating the model on energetic values predicted by the Turner model to assess whether the neural network could learn a representation that recovered known biophysical principles. First, we trained Ribonet to predict the simulated free energy of an RNA in complex with multiple input RNAs. Our model accurately predicts free energies of new sequences but also shows evidence of having learned base pairing information, as assessed by in silico double mutant analysis. Next, we extended this model to predict the simulated affinity between an arbitrary RNA sequence and a reporter RNA. While these more indirect measurements precluded the learning of basic principles of RNA biophysics, the resulting model achieved sub-kcal/mol accuracy and enabled design of simple RNA input responsive riboswitches with high activation ratios predicted by the Turner model from which the training data were generated. Finally, we compiled and trained on an experimental dataset comprising over 600,000 experimental affinity measurements published on the Eterna open laboratory. Though our tests revealed that the model likely did not learn a physically realistic representation of RNA interactions, it nevertheless achieved good performance of 0.76 kcal/mol on test sets with the application of transfer learning and novel sequence-specific data augmentation strategies. These results suggest that recurrent neural network architectures, despite being naïve to the physics of RNA folding, have the potential to capture complex biophysical information. However, more diverse datasets, ideally involving more direct free energy measurements, may be necessary to train de novo predictive models that are consistent with the fundamentals of RNA biophysics.Author SummaryThe precise design of RNA interactions is essential to gaining greater control over RNA-based biotechnology tools, including designer riboswitches and CRISPR-Cas9 gene editing. However, the classic model for energetics governing these interactions fails to quantitatively predict the behavior of RNA molecules. We developed a recurrent neural network model, Ribonet, to quantitatively predict these values from sequence alone. Using simulated data, we show that this model is able to learn simple base pairing rules, despite having no a priori knowledge about RNA folding encoded in the network architecture. This model also enables design of new switching RNAs that are predicted to be effective by the “ground truth” simulated model. We applied transfer learning to retrain Ribonet using hundreds of thousands of RNA-RNA affinity measurements and demonstrate simple data augmentation techniques that improve model performance. At the same time, data diversity currently available set limits on Ribonet’s accuracy. Recurrent neural networks are a promising tool for modeling nucleic acid biophysics and may enable design of complex RNAs for novel applications.

The emerging field of RNA nanotechnology seeks to create nanoscale 3D machines by repurposing natural RNA modules, but successes have been limited to symmetric assemblies of single repeating motifs. We present RNAMake, a suite that automates design of RNA molecules with complex 3D folds. We first challenged RNAMake with the paradigmatic problem of aligning a tetraloop and sequence-distal receptor, previously only solved via symmetry. Single-nucleotide-resolution chemical mapping, native gel electrophoresis, and solution x-ray scattering confirmed that 11 of the 16 ‘miniTTR’ designs successfully achieved clothespin-like folds. A 2.55 Å diffraction-resolution crystal structure of one design verified formation of the target asymmetric nanostructure, with large sections achieving near-atomic accuracy (< 2.0 Å). Finally, RNAMake designed asymmetric segments to tether the 16S and 23S rRNAs together into a synthetic singlestranded ribosome that remains uncleaved by ribonucleases and supports life in Escherichia coli, a challenge previously requiring several rounds of trial-and-error.

The structural interconversions that mediate the gene regulatory functions of RNA molecules may be different from classic models of allostery, but the relevant structural correlations have remained elusive in even intensively studied systems. Here, we present a four-dimensional expansion of chemical mapping called lock-mutate-map-rescue (LM2R), which integrates multiple layers of mutation with nucleotide-resolution chemical mapping. This technique resolves the core mechanism of the adenine-responsive V. vulnificus add riboswitch, a paradigmatic system for which both Monod-Wyman-Changeux (MWC) conformational selection models and non-MWC alternatives have been proposed. To discriminate amongst these models, we locked each functionally important helix through designed mutations and assessed formation or depletion of other helices via compensatory rescue evaluated by chemical mapping. These LM2R measurements give strong support to the pre-existing correlations predicted by MWC models, disfavor alternative models, and suggest additional structural heterogeneities that may be general across ligand-free riboswitches.

RNA-Puzzles is a collective experiment in blind 3D RNA structure prediction. We report here a third round of RNA-Puzzles. Five puzzles, 4, 8, 12, 13, 14, all structures of riboswitch aptamers and puzzle 7, a ribozyme structure, are included in this round of the experiment. The riboswitch structures include biological binding sites for small molecules (S-adenosyl methionine, cyclic diadenosine monophosphate, 5-amino 4-imidazole carboxamide riboside 5′-triphosphate, glutamine) and proteins (YbxF), and one set describes large conformational changes between ligand-free and ligand-bound states. The Varkud satellite ribozyme is the most recently solved structure of a known large ribozyme. All puzzles have established biological functions and require structural understanding to appreciate their molecular mechanisms. Through the use of fast-track experimental data, including multidimensional chemical mapping, and accurate prediction of RNA secondary structure, a large portion of the contacts in 3D have been predicted correctly leading to similar topologies for the top ranking predictions. Template-based and homology-derived predictions could predict structures to particularly high accuracies. However, achieving biological insights from de novo prediction of RNA 3D structures still depends on the size and complexity of the RNA. Blind computational predictions of RNA structures already appear to provide useful structural information in many cases. Similar to the previous RNA-Puzzles Round II experiment, the prediction of non-Watson–Crick interactions and the observed high atomic clash scores reveal a notable need for an algorithm of improvement. All prediction models and assessment results are available at http://ahsoka.u-strasbg.fr/rnapuzzles/.

Influenza A virus’s (IAV’s) frequent genetic changes challenge vaccine strategies and engender resistance to current drugs. We sought to identify conserved and essential RNA secondary structures within IAV’s genome that are predicted to have greater constraints on mutation in response to therapeutic targeting. We identified and genetically validated an RNA structure (packaging stem–loop 2 (PSL2)) that mediates in vitro packaging and in vivo disease and is conserved across all known IAV isolates. A PSL2-targeting locked nucleic acid (LNA), administered 3 d after, or 14 d before, a lethal IAV inoculum provided 100% survival in mice, led to the development of strong immunity to rechallenge with a tenfold lethal inoculum, evaded attempts to select for resistance and retained full potency against neuraminidase inhibitor-resistant virus. Use of an analogous approach to target SARS-CoV-2, prophylactic administration of LNAs specific for highly conserved RNA structures in the viral genome, protected hamsters from efficient transmission of the SARS-CoV-2 USA_WA1/2020 variant. These findings highlight the potential applicability of this approach to any virus of interest via a process we term ‘programmable antivirals’, with implications for antiviral prophylaxis and post-exposure therapy.

The three-dimensional conformations of noncoding RNAs underpin their biochemical functions but have largely eluded experimental characterization. Here, we report that integrating a classic mutation/rescue strategy with high-throughput chemical mapping enables rapid RNA structure inference with unusually strong validation. We revisit a 16S rRNA domain for which SHAPE (selective 2′-hydroxyl acylation with primer extension) and limited mutational analysis suggested a conformational change between apo- and holo-ribosome conformations. Computational support estimates, data from alternative chemical probes, and mutate-and-map (M2) experiments highlight issues of prior methodology and instead give a near-crystallographic secondary structure. Systematic interrogation of single base pairs via a high-throughput mutation/rescue approach then permits incisive validation and refinement of the M2-based secondary structure. The data further uncover the functional conformation as an excited state (20 ± 10% population) accessible via a single-nucleotide register shift. These results correct an erroneous SHAPE inference of a ribosomal conformational change, expose critical limitations of conventional structure mapping methods, and illustrate practical steps for more incisively dissecting RNA dynamic structure landscapes.

The predictive modeling and design of biologically active RNA molecules requires understanding the energetic balance amongst their basic components. Rapid developments in computer simulation promise increasingly accurate recovery of RNA’s nearest neighbor (NN) free energy parameters, but these methods have not been tested in predictive trials or on non-standard nucleotides. Here, we present the first such tests through a RECCES-Rosetta (Reweighting of Energy-function Collection with Conformational Ensemble Sampling in Rosetta) framework that rigorously models conformational entropy, predicts previously unmeasured NN parameters, and estimates these values’ systematic uncertainties. RECCES-Rosetta recovers the ten NN parameters for Watson-Crick stacked base pairs and thirty-two single-nucleotide dangling-end parameters with unprecedented accuracies – root-mean-square deviations (RMSD) of 0.28 kcal/mol and 0.41 kcal/mol, respectively. For set-aside test sets, RECCES-Rosetta gives an RMSD of 0.32 kcal/mol on eight stacked pairs involving G-U wobble pairs and an RMSD of 0.99 kcal/mol on seven stacked pairs involving non-standard isocytidine-isoguanosine pairs. To more rigorously assess RECCES-Rosetta, we carried out four blind predictions for stacked pairs involving 2,6-diaminopurine-U pairs, which achieved 0.64 kcal/mol RMSD accuracy when tested by subsequent experiments. Overall, these results establish that computational methods can now blindly predict energetics of basic RNA motifs, including chemically modified variants, with consistently better than 1 kcal/mol accuracy. Systematic tests indicate that resolving the remaining discrepancies will require energy function improvements beyond simply reweighting component terms, and we propose further blind trials to test such efforts.SignificanceUnderstanding RNA machines and how their behavior can be modulated by chemical modification is increasingly recognized as an important biological and bioengineering problem, with continuing discoveries of riboswitches, mRNA regulons, CRISPR-guided editing complexes, and RNA enzymes. Computational strategies to understanding RNA energetics are being proposed, but have not yet faced rigorous tests. We describe a modeling strategy called ‘RECCES-Rosetta’ that models the full ensemble of motions of RNA in single-stranded form and in helices, including non-standard nucleotides such as 2,6-diaminopurine, a variant of adenosine. When compared to experiments, including blind tests, the energetic accuracies of RECCES-Rosetta calculations are at levels close to experimental error, suggesting that computation can now be used to predict and design basic RNA energetics.

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 class of double glycine riboswitches that can bind two ligands cooperatively. A novel method (BPPalign), RMdetect, JAR3D, and Rosetta 3D modeling give consistent predictions for a new stem P0 and a kink-turn motif. These elements structure the linker between the RNAs' double aptamers. Chemical mapping on the Fusobacterium nucleatum riboswitch with N-methylisatoic anhydride, dimethyl sulfate and 1-cyclohexyl-3-(2-morpholinoethyl)carbodiimide metho-p-toluenesulfonate probing, mutate-and-map studies, and mutation/rescue experiments all provide strong evidence for the structured linker. Under solution conditions that permit rigorous thermodynamic analysis, disrupting this helix-junction-helix structure gives 120- and 6-30-fold poorer dissociation constants for the RNA's two glycine-binding transitions, corresponding to an overall energetic impact of 4.3 ± 0.5 kcal/mol. Prior biochemical and crystallography studies did not include this critical element due to over-truncation of the RNA. We speculate that several further undiscovered elements are likely to exist in the flanking regions of this and other functional RNAs, and automated prediction tools can play a useful role in their detection and dissection.

Non-coding RNAs fold into precise base-pairing patterns to carry out critical roles in genetic regulation and protein synthesis, but determining RNA structure remains difficult. Here, we show that coupling systematic mutagenesis with high-throughput chemical mapping enables accurate base-pair inference of domains from ribosomal RNA, ribozymes and riboswitches. For a six-RNA benchmark that has challenged previous chemical/computational methods, this 'mutate-and-map' strategy gives secondary structures that are in agreement with crystallography (helix error rates, 2%), including a blind test on a double-glycine riboswitch. Through modelling of partially ordered states, the method enables the first test of an interdomain helix-swap hypothesis for ligand-binding cooperativity in a glycine riboswitch. Finally, the data report on tertiary contacts within non-coding RNAs, and coupling to the Rosetta/FARFAR algorithm gives nucleotide-resolution three-dimensional models (helix root-mean-squared deviation, 5.7 Å) of an adenine riboswitch. These results establish a promising two-dimensional chemical strategy for inferring the secondary and tertiary structures that underlie non-coding RNA behaviour.

Accelerating discoveries of non-coding RNA (ncRNA) in myriad biological processes pose major challenges to structural and functional analysis. Despite progress in secondary structure modeling, high-throughput methods have generally failed to determine ncRNA tertiary structures, even at the 1-nm resolution that enables visualization of how helices and functional motifs are positioned in three dimensions. We report that integrating a new method called MOHCA-seq (Multiplexed •OH Cleavage Analysis with paired-end sequencing) with mutate-and-map secondary structure inference guides Rosetta 3D modeling to consistent 1-nm accuracy for intricately folded ncRNAs with lengths up to 188 nucleotides, including a blind RNA-puzzle challenge, the lariat-capping ribozyme. This multidimensional chemical mapping (MCM) pipeline resolves unexpected tertiary proximities for cyclic-di-GMP, glycine, and adenosylcobalamin riboswitch aptamers without their ligands and a loose structure for the recently discovered human HoxA9D internal ribosome entry site regulon. MCM offers a sequencing-based route to uncovering ncRNA 3D structure, applicable to functionally important but potentially heterogeneous states. DOI: http://dx.doi.org/10.7554/eLife.07600.001

The role of phosphatidylcholine turnover during hypersaline stress is investigated in Saccharomyces cerevisiae. In the wild-type strain, 2180-1A hypersaline stress induced the rapid turnover of phosphatidylcholine, a major membrane lipid. Yeast cells were grown in the presence of [14C]-choline to label phosphatidylcholine. Upon shifting the cells to medium with 0.8 M NaCl, phosphatidylcholine levels were diminished by c. 30% within 20 min to yield glycerophosphocholine, a methylamine osmoprotectant that has been previously identified in renal cells. High-performance liquid chromatography studies showed that osmotically mediated glycerophosphocholine production was enhanced if 10 mM choline was added as a supplement to synthetic dextrose medium with 1.6 M NaCl, but glycine betaine was not detected. Enhanced glycerophosphocholine production also correlated with improved growth in media containing 1.6 M NaCl and choline. Enhanced growth is specific to methylamines: salt-stressed cells supplemented with 10 mM choline or glycine betaine showed enhanced growth relative to unsupplemented control cultures, but other additives had no effect on growth or adversely affected it. Nutritional effects are ruled out because yeast cannot use choline or glycine betaine as carbon or nitrogen sources in normal or high-salt medium. Finally, enhanced growth in hypersaline media with choline or glycine betaine is dependent on the choline permease Hnm1. These results in yeast highlight a similarity with mammalian renal cells, namely that phosphatidylcholine turnover contributes to osmotic adaptation via synthesis of the osmoprotectant glycerophosphocholine.

For decades, dimethyl sulfate (DMS) mapping has informed manual modeling of RNA structure in vitro and in vivo. Here, we incorporate DMS data into automated secondary structure inference using an energy minimization framework developed for 2'-OH acylation (SHAPE) mapping. On six noncoding RNAs with crystallographic models, DMS-guided modeling achieves overall false negative and false discovery rates of 9.5% and 11.6%, respectively, comparable to or better than those of SHAPE-guided modeling, and bootstrapping provides straightforward confidence estimates. Integrating DMS-SHAPE data and including 1-cyclohexyl(2-morpholinoethyl) carbodiimide metho-p-toluene sulfonate (CMCT) reactivities provide small additional improvements. These results establish DMS mapping, an already routine technique, as a quantitative tool for unbiased RNA secondary structure modeling.

This paper is a report of a second round of RNA-Puzzles, a collective and blind experiment in three-dimensional (3D) RNA structure prediction. Three puzzles, Puzzles 5, 6, and 10, represented sequences of three large RNA structures with limited or no homology with previously solved RNA molecules. A lariat-capping ribozyme, as well as riboswitches complexed to adenosylcobalamin and tRNA, were predicted by seven groups using RNAComposer, ModeRNA/SimRNA, Vfold, Rosetta, DMD, MC-Fold, 3dRNA, and AMBER refinement. Some groups derived models using data from state-of-the-art chemical-mapping methods (SHAPE, DMS, CMCT, and mutate-and-map). The comparisons between the predictions and the three subsequently released crystallographic structures, solved at diffraction resolutions of 2.5–3.2 Å, were carried out automatically using various sets of quality indicators. The comparisons clearly demonstrate the state of present-day de novo prediction abilities as well as the limitations of these state-of-the-art methods. All of the best prediction models have similar topologies to the native structures, which suggests that computational methods for RNA structure prediction can already provide useful structural information for biological problems. However, the prediction accuracy for non-Watson–Crick interactions, key to proper folding of RNAs, is low and some predicted models had high Clash Scores. These two difficulties point to some of the continuing bottlenecks in RNA structure prediction. All submitted models are available for download at http://ahsoka.u-strasbg.fr/rnapuzzles/.

Large RNAs control myriad biological processes but challenge tertiary structure determination. We report that integrating Multiplexed •OH Cleavage Analysis with tabletop deep sequencing (MOHCA-seq) gives nucleotide-resolution proximity maps of RNA structure from single straightforward experiments. After achieving 1-nm resolution models for RNAs of known structure, MOHCA-seq reveals previously unattainable 3D information for ligand-induced conformational changes in a double glycine riboswitch and the sixth community-wide RNA puzzle, an adenosylcobalamin riboswitch.

Emerging evidence suggests that the ribosome has a regulatory function in directing how the genome is translated in time and space. However, how this regulation is encoded in the messenger RNA sequence remains largely unknown. Here we uncover unique RNA regulons embedded in homeobox (Hox) 5' untranslated regions (UTRs) that confer ribosome-mediated control of gene expression. These structured RNA elements, resembling viral internal ribosome entry sites (IRESs), are found in subsets of Hox mRNAs. They facilitate ribosome recruitment and require the ribosomal protein RPL38 for their activity. Despite numerous layers of Hox gene regulation, these IRES elements are essential for converting Hox transcripts into proteins to pattern the mammalian body plan. This specialized mode of IRES-dependent translation is enabled by an additional regulatory element that we term the translation inhibitory element (TIE), which blocks cap-dependent translation of transcripts. Together, these data uncover a new paradigm for ribosome-mediated control of gene expression and organismal development.

The three-dimensional conformations of non-coding RNAs underpin their biochemical functions but have largely eluded experimental characterization. Here, we report that integrating a classic mutation/rescue strategy with high-throughput chemical mapping enables rapid RNA structure inference with unusually strong validation. We revisit a paradigmatic 16S rRNA domain for which SHAPE (selective 2′-hydroxyl acylation with primer extension) suggested a conformational change between apo-and holo-ribosome conformations. Computational support estimates, data from alternative chemical probes, and mutate-and-map (M2) experiments expose limitations of prior methodology and instead give a near-crystallographic secondary structure. Systematic interrogation of single base pairs via a high-throughput mutation/rescue approach then permits incisive validation and refinement of the M2-based secondary structure and further uncovers the functional conformation as an excited state (25±5% population) accessible via a single-nucleotide register shift. These results correct an erroneous SHAPE inference of a ribosomal conformational change and suggest a general mutate-map-rescue approach for dissecting RNA dynamic structure landscapes.

Chemical mapping methods probe RNA structure by revealing and leveraging correlations of a nucleotide's structural accessibility or flexibility with its reactivity to various chemical probes. Pioneering work by Lucks and colleagues has expanded this method to probe hundreds of molecules at once on an Illumina sequencing platform, obviating the use of slab gels or capillary electrophoresis on one molecule at a time. Here, we describe optimizations to this method from our lab, resulting in the MAP-seq protocol (Multiplexed Accessibility Probing read out through sequencing), version 1.0. The protocol permits the quantitative probing of thousands of RNAs at once, by several chemical modification reagents, on the time scale of a day using a tabletop Illumina machine. This method and a software package MAPseeker ( http://simtk.org/home/map_seeker ) address several potential sources of bias, by eliminating PCR steps, improving ligation efficiencies of ssDNA adapters, and avoiding problematic heuristics in prior algorithms. We hope that the step-by-step description of MAP-seq 1.0 will help other RNA mapping laboratories to transition from electrophoretic to next-generation sequencing methods and to further reduce the turnaround time and any remaining biases of the protocol.

Chemical mapping is a widespread technique for structural analysis of nucleic acids in which a molecule’s reactivity to different probes is quantified at single nucleotide resolution and used to constrain structural modeling. This experimental framework has been extensively revisited in the past decade with new strategies for high-throughput readouts, chemical modification, and rapid data analysis. Recently, we have coupled the technique to high-throughput mutagenesis. Point mutations of a base paired nucleotide can lead to exposure of not only that nucleotide but also its interaction partner. Systematically carrying out the mutation and mapping for the entire system gives an experimental approximation of the molecule’s “contact map.” Here, we give our in-house protocol for this “mutate-and-map” (M2) strategy, based on 96-well capillary electrophoresis, and we provide practical tips on interpreting the data to infer nucleic acid structure.

Chemical mapping methods probe RNA structure by revealing and leveraging correlations of a nucleotide's structural accessibility or flexibility with its reactivity to various chemical probes. Pioneering work by Lucks and colleagues has expanded this method to probe hundreds of molecules at once on an Illumina sequencing platform, obviating the use of slab gels or capillary electrophoresis on one molecule at a time. Here, we describe optimizations to this method from our lab, resulting in the MAP-seq protocol (Multiplexed Accessibility Probing read out through sequencing), version 1.0. The protocol permits the quantitative probing of thousands of RNAs at once, by several chemical modification reagents, on the time scale of a day using a tabletop Illumina machine. This method and a software package MAPseeker ( http://simtk.org/home/map_seeker ) address several potential sources of bias, by eliminating PCR steps, improving ligation efficiencies of ssDNA adapters, and avoiding problematic heuristics in prior algorithms. We hope that the step-by-step description of MAP-seq 1.0 will help other RNA mapping laboratories to transition from electrophoretic to next-generation sequencing methods and to further reduce the turnaround time and any remaining biases of the protocol.

Chemical purity of RNA samples is important for high-precision studies of RNA folding and catalytic behavior, but photodamage accrued during ultraviolet (UV) shadowing steps of sample preparation can reduce this purity. Here, we report the quantitation of UV-induced damage by using reverse transcription and single-nucleotide-resolution capillary electrophoresis. We found photolesions in a dozen natural and artificial RNAs; across multiple sequence contexts, dominantly at but not limited to pyrimidine doublets; and from multiple lamps recommended for UV shadowing. Irradiation time-courses revealed detectable damage within a few seconds of exposure for 254 nm lamps held at a distance of 5 to 10 cm from 0.5-mm thickness gels. Under these conditions, 200-nucleotide RNAs subjected to 20 seconds of UV shadowing incurred damage to 16-27% of molecules; and, due to a ‘skin effect’, the molecule-by-molecule distribution of lesions gave 4-fold higher variance than a Poisson distribution. Thicker gels, longer wavelength lamps, and shorter exposure times reduced but did not eliminate damage. These results suggest that RNA biophysical studies should report precautions taken to avoid artifactual heterogeneity from UV shadowing.

Atomic-accuracy structure prediction of macromolecules should be achievable by optimizing a physically realistic energy function but is presently precluded by incomplete sampling of a biopolymer’s many degrees of freedom. We present herein a working hypothesis, called the “stepwise ansatz,” for recursively constructing well-packed atomic-detail models in small steps, enumerating several million conformations for each monomer, and covering all build-up paths. By making use of high-performance computing and the Rosetta framework, we provide first tests of this hypothesis on a benchmark of 15 RNA loop-modeling problems drawn from riboswitches, ribozymes, and the ribosome, including 10 cases that are not solvable by current knowledge-based modeling approaches. For each loop problem, this deterministic stepwise assembly method either reaches atomic accuracy or exposes flaws in Rosetta’s all-atom energy function, indicating the resolution of the conformational sampling bottleneck. As a further rigorous test, we have carried out a blind all-atom prediction for a noncanonical RNA motif, the C7.2 tetraloop/receptor, and validated this model through nucleotide-resolution chemical mapping experiments. Stepwise assembly is an enumerative, ab initio build-up method that systematically outperforms existing Monte Carlo and knowledge-based methods for 3D structure prediction.