Crystal structure of a DNA catalyst

Both DNA and RNA molecules have been shown to exhibit catalytic activity, but only the structure of catalytic RNAs has previously been determined; here the structure of an RNA-ligating DNA in the post-catalytic state is solved. Both DNA and RNA molecules are foldable and can adopt conformations that exhibit catalytic activity. While the structures of various catalytic RNAs — or ribozymes — have been determined, DNA enzymes have proved more difficult. Claudia Höbartner and colleagues have now solved the crystal structure of the synthetic single-stranded DNA deoxyribozyme 9DB1 at 2.8 Å resolution. 9DB1 is an RNA ligase catalysing phosphodiester bond formation between 3′-hydroxyl and the 5′-triphosphate termini of two RNA strands. The structure reveals three-dimensional complexity comparable to that adopted by RNA, but with differences reflecting the specific properties of deoxyribose. Catalysis in biology is restricted to RNA (ribozymes) and protein enzymes, but synthetic biomolecular catalysts can also be made of DNA (deoxyribozymes)1 or synthetic genetic polymers2. In vitro selection from synthetic random DNA libraries identified DNA catalysts for various chemical reactions beyond RNA backbone cleavage3. DNA-catalysed reactions include RNA and DNA ligation in various topologies4,5, hydrolytic cleavage6,7 and photorepair of DNA8, as well as reactions of peptides9,10 and small molecules11,12. In spite of comprehensive biochemical studies of DNA catalysts for two decades, fundamental mechanistic understanding of their function is lacking in the absence of three-dimensional models at atomic resolution. Early attempts to solve the crystal structure of an RNA-cleaving deoxyribozyme resulted in a catalytically irrelevant nucleic acid fold13. Here we report the crystal structure of the RNA-ligating deoxyribozyme 9DB1 (ref. 14) at 2.8 Å resolution. The structure captures the ligation reaction in the post-catalytic state, revealing a compact folding unit stabilized by numerous tertiary interactions, and an unanticipated organization of the catalytic centre. Structure-guided mutagenesis provided insights into the basis for regioselectivity of the ligation reaction and allowed remarkable manipulation of substrate recognition and reaction rate. Moreover, the structure highlights how the specific properties of deoxyribose are reflected in the backbone conformation of the DNA catalyst, in support of its intricate three-dimensional organization. The structural principles underlying the catalytic ability of DNA elucidate differences and similarities in DNA versus RNA catalysts, which is relevant for comprehending the privileged position of folded RNA in the prebiotic world and in current organisms.

| Global architecture of the DNA catalyst. a, DNA-catalysed ligation between 3′-OH and 5′-phosphate. nt, nucleotides. b, Proposed secondary structure of the minimal 9DB1 deoxyribozyme 20 . RNA nucleotides are in blue, ligation junction in light blue, DNA-binding arms in grey, and core nucleotides in black and red. Nucleotides in red do not tolerate any mutation. Blue and green lines indicate double pseudoknot interactions. c, Solvent-flattened electron density map contoured at 1.0σ level with orange trace of backbone and nucleobases. d, Secondary and tertiary structure with base pairs denoted in Leontis-Westhof presentation. DNA nucleotides in black have N-type ribose pucker, those in green have S-type conformations; grey nucleotides lie outside typical N or S conformations. P1-P5 are colour coded with respect to the structure shown in e. Inset: schematic overview of the double pseudoknot. e, Cartoon representation of the deoxyribozyme in complex with the ligated RNA product. f, A 60° rotation of the image shown in e.  Fig. 1). Within the catalytic domain, two stacks of two and four base pairs are formed, called P2 and P3, respectively. The paired regions P1, P2 and P3 are positioned on top of one another in a non-coaxial manner (Fig. 1f). In addition, the DNA residues dT29-dT30 base pair with the RNA nucleotides A−1-G1 (denoted P5), which makes the entire DNA-RNA complex fold as a double pseudoknot (Fig. 1d).
The overall fold of the catalytic domain is stabilized by tertiary interactions that connect the regions labelled J1/2, J3/2, P2 and P3 in four successive layers of multiplets (Fig. 2a). Hydrogen bonding forms the long-range Watson-Crick base pair dC12:dG26 and the non-canonical Watson-Crick base pair dT14:dG25 in P2 (Fig. 2b, c), as well as alternative base pairs involving the sugar edges of dG8 and dG21, the Watson-Crick edges of dG8, dG21 and dA23, and the Hoogsteen edges of dA10, dG22 and dG26. Moreover, the compact DNA architecture is stabilized by stacking interactions of nucleotides in J1/2 and J3/2, which place dT11 and dG24 sandwiched between the non-canonical base pairs dG8:dA10 and dG21:dA23. These interactions are consistent with previous atomic mutagenesis data, which showed that the Watson-Crick edges of dG8, dT14, dG21 and dG25 are essential for the functional integrity of 9DB1 (ref. 21). Moreover, dG26 did not tolerate changes on the Watson-Crick or Hoogsteen edges, consistent with its interactions in a base triple with the dC12 and dG8 nucleotides (Fig. 2b).
The reactive nucleotides A−1 and G1 are positioned in the active site by a scaffold formed by four nucleotides from distinct locations of the primary sequence, corroborated by the large angular opening between the duplexes P1 and P4 (Fig. 3a, b). Thus, the junction J2/3 protrudes from the catalytic domain and extensively contacts A−1 and G1 by base pairing with dT30 and dT29, respectively, as well as by the stacking of dG27 on G1 (Fig. 3a, b). On the opposite side, a 130° kink in the backbone between dT14 and dC16 causes dA15 to form a stacking interaction with A−1 (Fig. 3b). In this way, A−1 and G1 are stacked on one another, sandwiched by dA15 and dG27 and engaged in hydrogen bonding with dT29 and dT30, resulting in a microenvironment reminiscent of a duplex.
The identification of the previously unknown base pairs (P5) at the ligation junction was intriguing, as dT29 and dT30 were shown to tolerate mutations 20 , and RNA substrates with uridine or guanosine at position −1 were also efficiently ligated 14,15 . In contrast, only 5′-purine nucleotides were allowed at position 1 in the donor RNA (that is, 5′-pppG1 or 5′-pppA1) 14 . Therefore, the crystal structure of 9DB1 revealed an effective way of manipulating the catalytic activity by compensatory mutagenesis of the base pair at the ligation junction. Indeed, RNA substrates that were inert to the original deoxyribozyme, such as 5′-pppC1 and 5′-pppU1 RNA, are now readily ligated with mutated 9DB1 enzymes bearing G and A, respectively, in position 29 (Fig. 3c). On the basis of the crystal structure, 9DB1 deoxyribozymes can now be designed for ligation of any RNA substrates, irrespective of their sequence.
The regioselectivity of the ligation reaction is determined by the relative orientation of the reacting nucleotides. For ribozymes 22,23 as well as deoxyribozymes 24 , continuous Watson-Crick base pairing across the ligation junction favours nucleophilic attack by the 3′-OH group onto the 5′-triphosphate of G1. The regioselectivity of 9DB1 is dictated by tertiary contacts and specific interactions at the ligation junction, which differ markedly from a continuous double-helical arrangement. Thus, the orientation of the donor nucleotide G1 is assisted by a hydrogen bond of its 2′-OH to the minor groove of dC12:dG26 in P2 (Fig. 4a). The arrangement of the binding arms induces a shorter distance between the phosphate groups of A−1 and G1 (5.3 Å as compared to 5.9 Å in a regular A-form duplex). The hydrogen bond between the 2′-OH group of A−1 and the O4′ (or O5′) of G1 may sequester the 2′-OH and thereby assist the regioselective formation of 3′-5′ ligated products (Fig. 3b). Consistently, removal or blocking of the 2′-OH group by introducing a 2′-deoxy or 2′-OCH 3 nucleotide at position A−1 prevented efficient ligation (Fig. 3d). Instead, replacement of the 2′-OH with 2′-NH 2 still allowed RNA ligation, although with a 30-fold    Letter reSeArCH slower rate (Extended Data Fig. 2). Interestingly, 2′-fluoro-modified RNA was ligated only 6-7-fold slower ( Fig. 3d and Extended Data Fig. 2), suggesting that the electronegativity of fluorine and the resulting 3′-endo conformation of the ribose compensates for the loss of the 2′-OH group and orients the 3′-OH for nucleophilic attack onto the 5′-phosphate. Catalysis of phosphoryl-transfer reactions by proteins or ribozymes often involves activation of nucleophilic groups and electrostatic stabilization of the transition state by divalent metal ions or specific chemical groups 25,26 . The active site of an RNA enzyme that catalyses the same reaction as the DNA enzyme 9DB1, is composed of two phosphate groups coordinating a catalytic metal ion, plus an amino group from a nucleobase and a 2′-OH group that exerts its action through intermediacy of a water molecule 27 . There are no equivalent interactions in the core of 9DB1, and we do not observe electron density for a catalytic metal ion. However, a striking feature is the phosphate group of nucleotide dA13, which lies within 3.1 Å distance from the ligation junction (Fig. 4a). To test the role of the dA13 phosphate in catalysis, we replaced either of the non-bridging oxygen atoms with a sulfur atom, and assessed the enzymatic activity of the resulting phosphorothioate-modified DNAs. In the presence of manganese ions, the ligation rate of the deoxyribozyme containing an Sp phosphorothioate was reduced 100-fold, whereas the Rp stereoisomer retained its full activity (Fig. 4b). A similar phosphorothioate interference effect was observed when magnesium was used as metal ion cofactor, and the catalytic activity could also not be rescued by the addition of cadmium ions. The absence of thiophilic metal rescue cannot prove whether metal ions are involved or not in catalysis. Upon substitution of the non-bridging phosphate oxygens with a methyl group, only one of the two resulting uncharged methylphosphonate diastereomers retained catalytic activity (Fig. 4b). Although the role of metal ions remains unclear, these data are consistent with the critical role of the non-bridging pro-Sp phosphate oxygen of dA13 in activating the 3′-OH group to promote RNA ligation.
This work provides the experimental evidence that DNA possesses the ability to fold into compact three-dimensional units, stabilized by intrinsic tertiary interactions in the absence of 2′-OH groups or artificial hydrophobic substituents 28 . In contrast with ribozymes, the sugar-phosphate backbone of the catalytic domain of 9DB1 exhibits much larger conformational diversity, as reflected by the broad distribution of pseudorotation phase angles found in the DNA nucleotides (Extended Data Fig. 3). Thus, of the 31 nucleotides in the catalytic domain, 8 and 20 adopt north (N)-and south (S)-type ribose puckers, respectively (Fig. 1d). The remaining three nucleotides lie outside typical N or S conformations. The less restrictive puckering allows DNA to explore a wide range of conformations, compensating for the lack of the 2′-hydroxyl group, which has an important structural role in ribozymes.
Comparison of 9DB1 with two RNA ligase ribozymes illustrates how the fold of DNA and RNA enzymes underlies the recognition and positioning of the substrates in the catalytic centre (Extended Data Fig. 4). In the class I ligase and L1 ligase ribozymes, the reactive nucleotides are positioned in extended helices (Extended Data Fig. 5) 29,30 . In contrast, the four contributing bases in the 9DB1 ligase are provided in trans from different primary locations of the catalytic domain. Notably, all four DNA nucleotides adopt S-type ribose puckers, highlighting how the specific conformational propensity of deoxyribose moieties supports recognition and positioning of the substrate in the active centre.
The crystal structure of the 9DB1 deoxyribozyme demonstrates that DNA possesses the intrinsic ability to adopt complex tertiary folds that support catalysis, and unveils for the first time the active site of a DNA enzyme in the post-catalytic state. Together with mutagenic analyses, the structure elucidates the basis of substrate recognition and provides first insights for rationalization of the regiospecific bond formation. The ability of DNA to form complex three-dimensional architectures that support catalysis raises questions about the positioning of this biopolymer in the prebiotic evolution of life and may shed light on metabolic events in current organisms, in which single-stranded DNA may adopt functionally important folds.