928 VOLUME 33 NUMBER 9 SEPTEMBER 2015 NATURE BIOTECHNOLOGY Eric Westhof is at Architecture & Reactivity of RNA, University of Strasbourg, Institute of Molecular and Cellular Biology of the CNRS, Strasbourg, France. e-mail: e.westhof@ibmc-cnrs.unistra.fr approaches allow experimental probing of accessible RNA regions and determination of RNA structure (essentially at the secondary- structure level) on a transcriptome-wide scale 4,5 . However, although they identify nucleotides or regions that form helical domains or are engaged in some sort of base pairing, they cannot identify the paired strand or nucleotide. This is why additional data—such as phyloge- netic analysis, free-energy minimizations and computational modeling—are needed. If all the proximity and pairing information in a given RNA structure were available at sufficiently high resolution, in principle the structure could then be determined by computational methods. The approach of Ramani et al. 1 , called RNA proximity ligation (RPL), is a first step toward determining the missing three- dimensional connectivity. The authors use deep sequencing and proximity ligation to obtain information on the spatial proximity of nucleotides in a complex mixture of nonde- natured RNA transcripts extracted from yeast cells stripped of their cell wall or from cultured human cells (Fig. 1). In the first step, RNases (endogenous RNases for yeast cells, exogenous RNases for human cells) are allowed to cleave RNA. Next, exog- enous T4 RNA ligase I is added to randomly ligate free RNA ends, and the chimeric mol- ecules resulting from ligation are submitted for deep sequencing. The sequencing results are compared to the known primary structure of an RNA of interest. The authors find that the vast majority of ligations are intra- rather than inter-molecular. The data are noisy, with random ligations seen between termini that are close or far in sequence and in space. To correct for PCR errors, ligations between termini closer than 10 bases are not counted. Ligations that restore the original sequence are not detected. Although the ligation data are noisy, the authors can improve the signal-to-noise ratio by analyzing the data with 21-nucleotide long windows, which reveals an enrichment of Our knowledge of RNA structure derives from studies using X-ray crystallography and nuclear magnetic resonance spectroscopy. However, many RNAs and RNA-protein complexes are not amenable to these time-consuming methods for a variety of reasons, such as a requirement for large quantities of starting RNA or understand- ing of the optimal solubility and crystallization conditions that preserve molecular integrity. This has led to the development of alternative approaches for inferring RNA structure. Phylogenetic analysis, which looks for patterns of nucleotide co-variation across conserved RNA sequences, is very useful for predicting RNA secondary structure. Homologous sequences are expected to yield similar folds and maintain the same number and lengths of core helices. However, this method requires sufficient sequence variation data and careful sequence alignments. Another type of in silico approach relies on experimentally derived energies of base- paired stacks to compute the secondary struc- tures of a given RNA sequence with the lowest free energy and maximum number of base pairs. Although these methods are constantly improving, they are mathematically complex and can yield several potential solutions with accuracies of 75–80% (ref. 3), making the true structure difficult to determine. Experimental methods involving chemi- cal and enzymatic probes have been used to determine the accessibility and dynamics of RNA structures by measuring the susceptibil- ity of specific nucleotides to modification or cleavage. The chemicals are chosen according to their reactivity with specific atoms on the base or sugar-phosphate backbone, whereas the endonucleases cleave folded RNA mole- cules with a preference for unpaired or paired regions. The resulting data can be used to con- strain the number of RNA secondary structures predicted by computational methods. In recent years, chemical and enzymatic probing methods have been adapted for use with deep sequencing technologies. These Solving the three-dimensional structure of an RNA molecule means laborious study by X-ray crystallography or nuclear magnetic resonance spectroscopy. However, faster complementary methods are on the horizon, many involving deep sequencing and sophisticated compu- tational analysis. In this issue, Ramani et al. 1 use deep sequencing and proximity ligation to identify nucleotide regions that interact in folded RNA molecules. The method provides an entirely new source of information on intra- molecular RNA interactions that, with further improvement, may enable accurate prediction of RNA structure. Single-stranded RNA molecules have a strong tendency to fold back on themselves, locally and globally, creating complex spatial architectures. Folding relies on stacking hydro- gen bonds between nucleobases. All base-base interactions that involve at least two ‘stan- dard’ hydrogen bonds can be classified into 12 families. Each family is a 4 × 4 matrix of the four RNA bases—U, C, A, G 2 . The com- mon Watson-Crick pairs belong to one of these families, and the other 11 families are made up of non-Watson-Crick pairs. Watson-Crick pairs form the double- stranded hairpins of RNA secondary struc- ture. The remaining families are involved in the formation of RNA modules—the building blocks of tertiary structure—and long-range intramolecular contacts. RNA architecture can therefore be viewed as the hierarchical assembly of preformed hairpins defined by Watson-Crick base pairs and RNA modules maintained by non-Watson-Crick base pairs. Computational approaches to solving RNA structure often follow this model, determin- ing secondary structure before building up tertiary structure. RNA structure from deep sequencing Eric Westhof A method to identify interacting regions in a folded RNA is a step toward solving RNA structures from sequencing data. NEWS AND VIEWS npg © 2015 Nature America, Inc. All rights reserved.