2.1 Self-replication as a molecular property

Self-replication (at the genetic, cellular and organismal levels) is a defining hallmark of life. However, its beginnings are currently unknown. But self-replication as a system-level property is widespread beyond biology not just in the digital realm, e.g. in the form of computer viruses but in macromolecular and colloidal chemistry. Examples include crystal seeding, as well as colloidal self-organizing systems such as lipidic vesicles, which can display both autocatalytic growth and self-replication (Hanczyc & Szostak, Reference Hanczyc and Szostak2004; Oberholzer et al. Reference Oberholzer, Albrizio and Luisi1995a, Reference Oberholzer, Wick, Luisi and Biebricherb).

Autocatalytic chemical systems capable of self-replication have also been designed based on various components, including small molecules and peptides (Bissette & Fletcher, Reference Bissette and Fletcher2013; Conn et al. Reference Conn, Wintner and Rebek1994; Lee et al. Reference Lee, Granja, Martinez, Severin and Ghadiri1996). However, these systems differ from genetic systems in several crucial aspects. Key differences include the unique ability of nucleic acids (DNA, RNA and XNA) to store information both redundantly (on both strands) and at exceptionally high density (Church et al. Reference Church, Gao and Kosuri2012) using an exclusive double-sided recognition code based on non-covalent interactions by hydrogen bonding. Furthermore, and possibly even more importantly, replication in the autocatalytic chemical systems is by necessity perfect, and a ‘mistake’, i.e. side-reactions, etc. simply dissipate the self-replication cycle and are non-heritable. In contrast, information transfer in nucleic acid replication – while accurate – is imperfect, enabling both faithful transmission of the genetic information to the next generation, as well as generating low-level sequence diversity (i.e. mutations), which is a prerequisite for evolution.

Some autocatalytic systems have been built from synthetic nucleic acid components. These include systems involving palindromic trinucleotide ligations using carbodiimide (EDC) chemistry (Sievers & von Kiedrowski, Reference Sievers and von Kiedrowski1994) either in solution or on longer (24-mer) duplex palindromic polypurine/polypyrimidine DNA (Li & Nicolaou, Reference Li and Nicolaou1994). A common problem of such approaches is product inhibition, which can be overcome by surface tethering and thermocycling to liberate the daughter strands from the template (Luther et al. Reference Luther, Brandsch and von Kiedrowski1998).

Joyce and co-worker repurposed the R3 RNA ligase ribozyme for self-ligation (Paul & Joyce, Reference Paul and Joyce2002) and faced the same problem but overcame product inhibition through an elegant cross-catalytic system, which allowed self-assembly of the two R3 variants from their constituent parts with true exponential growth kinetics (Lincoln & Joyce, Reference Lincoln and Joyce2009). This system has also been optimized for the sensing of ligands (Lam & Joyce, Reference Lam and Joyce2009) as well as for impressive speed (Robertson & Joyce, Reference Robertson and Joyce2014). Similarly, although with much slower growth kinetics, split variants of the Azoarcus self-splicing intron (SSI) can self-assemble both in cis and in trans into active complexes and can form cross-catalytic assembly networks (Hayden et al. Reference Hayden, von Kiedrowski and Lehman2008; Vaidya et al. Reference Vaidya, Manapat, Chen, Xulvi-Brunet, Hayden and Lehman2012). However, although both the cross-catalytic ligase and Azoarcus SSI can form new variants through recombination and network growth, the need to provide pre-fabricated RNA oligomer-building blocks with substantial homology to the ribozyme/SSI core constrains their ability to evolve freely.

2.2 Physicochemical properties and information storage capacity

A strong case can be made that nucleic acids are singularly suited for information storage and transmission (Benner, Reference Benner2004). Beyond the specific base-pairing and redundant double-helical information encoding famously recognized by Watson & Crick, a key feature of the chemistry of nucleic acids is that information content and physicochemical properties are effectively decoupled due to the dominant influence of the polyanionic phosphodiester backbone. In contrast to the behaviour of proteins, where single mutations can have dramatic consequences on folding, structure or solubility, most nucleic acid sequences display identical physicochemical properties. Indeed, without this feature much of recombinant DNA technology, microarrays and sequencing would be technically impossible. Other features include the charge repulsion along the backbone favouring an extended conformation facilitating information readout. Finally, there are the unusual chemical properties of phosphodiester bonds combining thermodynamic instability with an unusual kinetic stability as famously pointed out by Westheimer (Reference Westheimer1987). The kinetic stability of phosphodiesters is in sharp contrast to other esters, including the chemically closely related arsenodiester linkage, which undergoes rapid hydrolysis in aqueous solution due to inefficient charge shielding of the larger arsenic atom (Fekry et al. Reference Fekry, Tipton and Gates2011). In addition, the restricted number of sugar ring conformations provide a stable scaffold for the nucleobases and is essential for duplex formation, stability and the restriction of conformational polymorphism to just two main double-helical structures, A- and B-forms, under physiological conditions (Saenger & Egli, Reference Saenger and Egli1984).

Despite this seemingly ideal ‘Goldilocks’ chemistry, it should be noted that recent work has shown that these fundamental principles are stable to considerable variation in both the canonical sugar and nucleobase chemistry, which in turn give rise to a wide range of structural variation (Anosova et al. Reference Anosova, Kowal, Dunn, Chaput, Van Horn and Egli2016). Building on earlier work from Orgel and Eschenmoser (Eschenmoser, Reference Eschenmoser1999; Kozlov et al. Reference Kozlov, De Bouvere, Van Aerschot, Herdewijn and Orgel1999a, Reference Kozlov, Politis, Van Aerschot, Busson, Herdewijn and Orgelb; Schoning et al. Reference Schoning, Scholz, Guntha, Wu, Krishnamurthy and Eschenmoser2000) nucleic acids in which the canonical (deoxi)ribo-furanose of DNA and RNA is replaced by ring congeners not found in nature, including HNA (1,5 anhydrohexitol nucleic acid), CeNA (cyclohexenyl nucleic acids), LNA (2′ O, 4′-C-methylene-β-D-ribonucleic acids; locked nucleic acids), ANA (arabinonucleic acids), FANA (2′-fluoro-arabinonucleic acid) and TNA (α-L-threofuranosylnucleic acids, based on a tetrose sugar) are capable of genetic information storage and propagation (Pinheiro et al. Reference Pinheiro, Taylor, Cozens, Abramov, Renders, Zhang, Chaput, Wengel, Peak-Chew, Mclaughlin, Herdewijn and Holliger2012). Furthermore, these XNAs support a replication cycle progressing through a DNA intermediate (conceptually similar to retroviral replication) enabling the in vitro evolution of XNA aptamers (Pinheiro et al. Reference Pinheiro, Taylor, Cozens, Abramov, Renders, Zhang, Chaput, Wengel, Peak-Chew, Mclaughlin, Herdewijn and Holliger2012) and catalysts (Taylor et al. Reference Taylor, Pinheiro, Smola, Morgunov, Peak-Chew, Cozens, Weeks, Herdewijn and Holliger2015). So far, no prebiotic synthesis of XNAs has been described, though this argument in itself is insufficient to argue against their inherent plausibility (as prebiotic syntheses of XNAs have not been actively sought).

Similarly, there might also exist alternative patterns of information encoding. Indeed genetic information storage and transfer have been demonstrated for a range of artificial base-pair designs. These expand the genetic alphabet and can be based on alternative hydrogen-bonding patterns, hydrophobic and/or geometric compatibility or even metal ion chelation. Some of these expanded genetic alphabets have also enabled evolution of superior aptamer ligands to protein or cell-surface targets incorporating one or more bases or base-pairs (Benner, Reference Benner2004; Hirao et al. Reference Hirao, Kimoto and Yamashige2012) and have even been integrated into a plasmid in a living organism (Malyshev et al. Reference Malyshev, Dhami, Lavergne, Chen, Dai, Foster, Correa and Romesberg2014). Importantly, both unnatural base-pairs as well as a number of XNA backbones retain their molecular memory function despite deviations from canonical helical conformations (Georgiadis et al. Reference Georgiadis, Singh, Kellett, Hoshika, Benner and Richards2015; Lescrinier et al. Reference Lescrinier, Esnouf, Schraml, Busson, Heus, Hilbers and Herdewijn2000; Nauwelaerts et al. Reference Nauwelaerts, Fisher, Froeyen, Lescrinier, Aerschot, Xu, Delong, Kang, Juliano and Herdewijn2007) and planar base-stacking (Betz et al. Reference Betz, Malyshev, Lavergne, Welte, Diederichs, Romesberg and Marx2013).

In contrast to the comparable tolerance to different sugar/nucleobase chemistries, the design of alternatives to the canonical phosphodiester backbone chemistry that can also support genetic information storage and propagation and allow cross-talk (i.e. helix-formation with natural nucleic acids) has proven challenging (Micklefield, Reference Micklefield2001; Nielsen, Reference Nielsen1995). The only successful designs fulfilling all of the above criteria are isosteric and largely isoelectronic modifications such as phosphorothioates (Eckstein, Reference Eckstein2014) and boranophosphates (Li et al. Reference Li, Sergueeva, Dobrikov and Shaw2007) (in which the non-bridging oxygen is replaced by sulphur or borano-trihydride substituents, respectively). More radical departures from the canonical backbone chemistry such as peptide nucleic acids (PNAs) (Sharma & Awasthi, Reference Sharma and Awasthi2016), in which the ribofuranose-phosphate backbone of DNA/RNA is replaced by N-(2-aminoethyl)-glycine or morpholino nucleic acids (PMO), in which the sugar–phosphate linkage is substituted by a morpholino ring–phosphorodiamidate linkage are among the few exceptions. Both PNAs and PMOs show specific hybridization to target sequences, but currently cannot be replicated enzymatically and hence are not amenable to laboratory evolution. Nevertheless, using reductive amination chemistry (Li et al. Reference Li, Zhan, Knipe and Lynn2002) PNA can be used in information transfer from a DNA template (Brudno et al. Reference Brudno, Birnbaum, Kleiner and Liu2010; Rosenbaum & Liu, Reference Rosenbaum and Liu2003) and indeed it has been proposed that PNA may have been involved in pre-biotic evolution (Nielsen, Reference Nielsen2007; Ura et al. Reference Ura, Beierle, Leman, Orgel and Ghadiri2009).

3.1 RNA catalysis

The first examples of RNA catalysis were discovered by Cech and Altman, in the SSI of Tetrahymena (Kruger et al. Reference Kruger, Grabowski, Zaug, Sands, Gottschling and Cech1982) and the RNA component of RNAse P (Guerrier-Takada et al. Reference Guerrier-Takada, Gardiner, Marsh, Pace and Altman1983) and were followed by the discovery of a wide range of self-cleaving ribozymes in viruses as well as an ever-expanding number of RNA catalysts generated by in vitro selection technologies. Finally and most fundamentally, RNA catalysis was found to be at the heart of both the spliceosome and the peptidyl-transferase activity of the ribosome. The landmark discovery of RNA catalysis also set the starting point for the exploration of the essential regulatory function of RNA in vivo (Cech & Steitz, Reference Cech and Steitz2014). Ribozyme catalysis is based on distinct 3D structures, with stacking, base-pairing and tertiary contacts all contributing to the complex folding of the ribozyme/substrate complex. Ribozyme and more generally RNA folding and dynamics occur in hierarchical order with structural elements forming on timescales ranging from picoseconds to seconds (Mustoe et al. Reference Mustoe, Brooks and Al-Hashimi2014). The folding is generally facilitated by metal ions, due to the highly polyanionic character of the sugar phosphate backbone (Denesyuk & Thirumalai, Reference Denesyuk and Thirumalai2015). Nevertheless RNA folding in vitro (as it has mostly been studied) is often different from the much more crowded natural in vivo conditions (Leamy et al. Reference Leamy, Assmann, Mathews and Bevilacqua2016).

RNA catalysis in vivo can be either solely performed by RNA, as for the small nucleolytic ribozymes, the Hammerhead (HHR), Hairpin (HP), Varkud satellite (VS), Hepatitis delta (HDV), twister and the glmS ribozyme (Lilley, Reference Lilley2011; Wilson et al. Reference Wilson, Liu and Lilley2016b) or aided by proteins forming ribonucleoprotein (RNP) complexes, as for the group II intron (Pyle, Reference Pyle2016), RNaseP (Mondragon, Reference Mondragon2013), the ribosome (Voorhees & Ramakrishnan, Reference Voorhees and Ramakrishnan2013) and the spliceosome (Wahl et al. Reference Wahl, Will and Luhrmann2009), with the RNA component responsible for catalysis and the protein component mainly acting as a scaffold and/or counterion. The principal mechanisms of naturally occurring ribozymes are either based on general acid–base catalysis as for the small nucleolytic ribozymes or on two metal ion catalysis as for group I, group II introns, RNase P and the spliceosome.

All natural occurring ribozymes, with the notable exception of the ribosome (which performs peptidyl transfer), catalyse phosphoryl transfer reactions. This is initiated by nucleophilic attack on the phosphate by the adjacent 2′-oxygen (as for the nucleolytic ribozymes), the 3′-oxygen of an exogenous guanosine (group I intron), the 2′ oxygen of an internal adenosine (group II intron and the spliceosome) or water (RNase P) (Lilley & Eckstein, Reference Lilley and Eckstein2008) (Fig. 2).

Fig. 2. First step of phosphoryl transfer reactions of natural occurring ribozymes. The nucleophile (in blue) attacks the phosphorus of the RNA phosphodiester bond.

This rather limited chemical reactivity spectrum raises the question of whether the many diverse chemical transformations necessary to support a putative RNA world could have been performed by RNA alone. It may be that there are more RNA-world molecular fossils (with more diverse chemical capabilities) waiting to be discovered, in particular considering that still only a small section of the ‘RNAome’ of the biosphere has been explored.

There is a strong discrepancy between the occurrence and significance of different ribozymes in the tree of life. The essential reactions catalysed by the more complex RNP structures such as the peptidyl-transferase activity of the ribosome, the RNase P catalysed tRNA maturation and RNA splicing by the spliceosome (or its simpler forerunnner the group II intron) are distinctive and found across all branches of life. On the other hand, the simpler nucleolytic ribozymes are rather sparsely distributed in biology (with the VS ribozyme only found once) and with a narrow biological function only fully explored in viruses. Nevertheless, biochemical experiments and bioinformatic search algorithms identified HHR, HDV and HP sequences in all domains of life, with their precise functions in most cases still to be explored (Jimenez et al. Reference Jimenez, Polanco and Luptak2015; Salehi-Ashtiani et al. Reference Salehi-Ashtiani, Luptak, Litovchick and Szostak2006; Webb et al. Reference Webb, Riccitelli, Ruminski and Luptak2009). This ubiquitous presence of the small nucleolytic ribozymes suggests that either they too might be leftovers from an ancient RNA world (as well as actively participating in modern nucleic acid metabolism, and hence being part of the ‘modern RNA World’) (Cech, Reference Cech2012) or alternatively, that this distribution might be simply a consequence of their comparative structural and functional simplicity. Indeed, the HHR fold, which is particularly ubiquitous (Hammann et al. Reference Hammann, Luptak, Perreault and De La Pena2012), is also the most likely motif for RNA cleavage identified by in vitro selections (Salehi-Ashtiani & Szostak, Reference Salehi-Ashtiani and Szostak2001), presumably due to its small size and relaxed sequence requirements, i.e the ‘tyranny of the small motif’. On the other hand, evolutionary pressure has clearly also led to different outcomes for the same reaction and seemingly to alternative structural and catalytic solutions such as the HDV, Twister, etc. ribozymes (see below). In general, the nucleolytic ribozymes reveal a high sequence specificity and catalytic efficiency with their essential information content encoding catalytic function lower than that suggested by the length of the ribozyme. RNA sequences capable of catalysis, in particular RNA cleavage, are therefore rather common in sequence space. Hence, even a rather modest repertoire of random RNAs should already contain a number of active folds indicating how they could have contributed to the emergence of RNA catalysis from the pools of short RNA oligomers provided by prebiotic chemistry

The direct involvement of divalent metal ions in RNA catalysis (inner sphere coordination) by the small nucleolytic ribozymes has been largely excluded (Murray et al. Reference Murray, Seyhan, Walter, Burke and Scott1998), but outer sphere coordinated divalent metal ions are likely involved in HDV catalysis (Ke et al. Reference Ke, Zhou, Ding, Cate and Doudna2004), and might also play a direct role in HHR catalysis (Mir & Golden, Reference Mir and Golden2016). Apart from their involvement in catalysis, metal ions fulfill a prominent role in the folding process and stabilization of the 3D structure of ribozymes (Lipfert et al. Reference Lipfert, Doniach, Das and Herschlag2014; Sigel et al. Reference Sigel, Sigel and Sigel2012). From an origins perspective, metal ions were abundantly present on the early earth, making them the most likely early interacting partner for RNA, with divalent cations (such as Mg²⁺) more efficiently decreasing the electrostatic repulsion upon folding of the RNA molecule compared with monovalent cations (such as Na⁺ and  K⁺). However, there is a fundamental functional trade-off between the essential functions of divalent metal ions in ribozyme folding and catalysis, and the increased degradation rate of RNA in their presence. This trade-off has to be considered as a major evolutionary driving force both towards the assembly of folded RNA structures – as double-stranded RNA (dsRNA) is much more robust against degradation compared with single-stranded RNA (ssRNA) – and towards the replacement of structural metal ions by peptidic or proteinaceous counterions (see Section 6).

High-resolution structures of examples of all the natural classes of ribozymes are now available, including at least 20 different structures for the HHR and HP ribozymes. Starting with the first crystal structure of an HHR variant (Scott et al. Reference Scott, Finch and Klug1995), the crystal structures of the HDV (Ferre-D'Amare et al. Reference Ferre-D'amare, Zhou and Doudna1998) the HP (Rupert & Ferre-D'Amare, Reference Rupert and Ferre-D'amare2001), the glmS (Klein & Ferre-D'Amare, Reference Klein and Ferre-D'amare2006) and finally also the VS ribozyme (Suslov et al. Reference Suslov, Dasgupta, Huang, Fuller, Lilley, Rice and Piccirilli2015) were solved over the following 20 years. Similarly, high-resolution structures of the more complex RNA structures and RNP complexes were obtained for the group I intron (Adams et al. Reference Adams, Stahley, Kosek, Wang and Strobel2004), the group II intron (Toor et al. Reference Toor, Keating, Taylor and Pyle2008), RNase P (Kazantsev et al. Reference Kazantsev, Krivenko, Harrington, Holbrook, Adams and Pace2005), the ribosome (Ban et al. Reference Ban, Nissen, Hansen, Moore and Steitz2000) and very recently also the spliceosome (Yan et al. Reference Yan, Hang, Wan, Huang, Wong and Shi2015). Recent technical breakthroughs in CryoEM (cryo-electron-microscopy) techniques (Nogales & Scheres, Reference Nogales and Scheres2015; Vinothkumar & Henderson, Reference Vinothkumar and Henderson2016) revolutionized structural biology of large RNP complexes such as the ribosome (Frank, Reference Frank2016) and the spliceosome (Nguyen et al. Reference Nguyen, Galej, Fica, Lin, Newman and Nagai2016) resulting in unprecedented and detailed pictures of RNA catalysis by these complex molecular machines. While RNA catalysis at the heart of the ribosome had been suspected some time ago (Noller et al. Reference Noller, Hoffarth and Zimniak1992) to be confirmed by the structure of the peptidyl-transferase site (Nissen et al. Reference Nissen, Hansen, Ban, Moore and Steitz2000), the conjectured ribozyme catalysis of the spliceosome could only recently be ascertained by a combination of biochemical and structural studies (Fica et al. Reference Fica, Tuttle, Novak, Li, Lu, Koodathingal, Dai, Staley and Piccirilli2013; Nguyen et al. Reference Nguyen, Galej, Bai, Savva, Newman, Scheres and Nagai2015; Wan et al. Reference Wan, Yan, Bai, Wang, Huang, Wong and Shi2016), identifying the U2–U6 snRNA as the catalytic complex and showing, likely ancestral similarities to group II intron two metal ion catalysis.

Mechanistically, RNA undergoes non-enzymatic degradation by an internal transesterification reaction, through nucleophilic attack of the 2′-oxygen on the adjacent 3′-phosphodiester forming a 2′,3′-cyclic phosphate and 5′-hydroxyl. The reaction is catalysed by the deprotonation of the 2′-hydroxyl and is therefore increased at higher pH values. This transesterification proceeds through a concerted S_N2 mechanism, with the 2′-oxygen, the 5′-oxygen and the phosphorus in an in-line geometry. However, the main contribution to cleavage rates is believed to arise from deprotonation events (by a factor of 10⁵–10⁶) with the optimal orientation, i.e. in-line geometry less important and contributing only a factor of around 10² to the observed rate enhancement (Emilsson et al. Reference Emilsson, Nakamura, Roth and Breaker2003; Lilley, Reference Lilley2005) (measured for ribozyme catalysed cleavage reactions but likely similar for the non-enzymatic reaction). The non-enzymatic degradation of RNA phosphodiesters is about 10⁴-fold faster than that of DNA at neutral pH and even more accelerated at basic pH (though slower at acidic pH). This stability divergence is likely one of the functional drivers for the switch from RNA to DNA for information storage in living systems as genomes became larger.

The ‘classical’ (HHR, HP, VS, HDV, glmS) small nucleolytic ribozymes all catalyse phosphodiester cleavage of RNA by general acid–base catalysis along the mechanistic trajectory described above (Fig. 3). The active structures of the HHR, HP and VS ribozyme are formed by multihelix junctions and all three bind their substrate RNA by Watson–Crick base-pairing on both sides of the cleavage site, therefore the reverse ligation reactions are possible according to the principle of microscopic reversibility. The HP ribozyme applies the N1 of G8 and N1 of A39, as general base and acid, respectively (reversed in the ligation reaction) (Kath-Schorr et al. Reference Kath-Schorr, Wilson, Li, Lu, Piccirilli and Lilley2012). Similary, the VS ribozyme uses the N1 of G638 as general base and the N1 of A756 as general acid (Suslov et al. Reference Suslov, Dasgupta, Huang, Fuller, Lilley, Rice and Piccirilli2015). In HHR catalysis the N1 of G12 attracts the proton from the 2′-oxygen nucleophile, acting as general base and the 2′-hydroxyl of G8 is positioned near the 5′-oxygen leaving group, fulfilling the role of the general acid (Martick & Scott, Reference Martick and Scott2006) (Fig. 4). In contrast, the HDV and glmS ribozymes, whose active structure is formed by pseudoknots, only basepair with their substrates 3′ to the cleavage site; therefore the intermolecular reverse ligation reaction is (akin to RNase A) excluded under standard reaction conditions. In the HDV ribozyme, the pK _a shifted N3 imine proton of the catalytic C75 acts as a general acid and a hydrated Mg²⁺ ion as general base (Das & Piccirilli, Reference Das and Piccirilli2005; Nakano et al. Reference Nakano, Chadalavada and Bevilacqua2000), with mainly C75 contributing to the observed rate enhancement. The glmS catalytic riboswitch has an absolute requirement for G40, with the N1 of G40 acting as general base and with the amino group of the glucosamine-6-phosphate substrate in close proximity to the 5′-oxygen leaving group, consistent with its function as general acid (Jansen et al. Reference Jansen, Mccarthy, Soukup and Soukup2006; Klein et al. Reference Klein, Been and Ferre-D'amare2007) (Fig. 4).

Fig. 3. Mechanism of general acid–base catalysis as performed by the small nucleolytic ribozymes. The general base (in green) is attracting a proton from the 2′-hydroxyl in the cleavage reaction or from the 5′-hydroxyl in the reversed ligation reaction. The general acid (in blue) is protonating the 5′-oxyanion leaving group for cleavage or the 2′oxyanion for ligation. The proposed trigonal bipyramidal phosphorane transition state is shown in the centre.

Fig. 4. Proposed cleavage mechanism of the small nucleolytic ribozymes, based on general acid–base catalysis. The general acid is displayed in blue and the general base in green. The general acid is in all cases G, with the exception of the Varkud satellite, where a hydrated metal ion acts as general base.

The small nucleolytic ribozymes are the favourite study objects for RNA catalysis, related to their small size and the fact that they provide different structural and mechanistic solutions. They may also embody independent evolutionary trajectories towards the same chemical problem, therefore representing an example of convergent evolution at the molecular level. In principle, RNA cleavage by the different nucleolytic ribozymes could have been based on the same active site nucleotides arranged on different structural scaffolds. However, detailed biochemical, structural and biophysical methods have elucidated not only different structural arrangements, but also unique constellations of functional groups, pH and metal ions inside the framework of general acid–base catalysis within this group of ribozymes. Furthermore, even different constructs of the same ribozyme can have different structural folds and catalytic rates, as was shown for the HHR, in which the full-length variant (Martick & Scott, Reference Martick and Scott2006) was found to adopt a different structural arrangement compared with a previously crystallized minimal variant (Scott et al. Reference Scott, Finch and Klug1995). This shows that seemingly irrelevant residues distal to the catalytic core can lead to major structural changes, impact catalytic turnover and influence metal ion requirements and overall stability through non-Watson–Crick long-range tertiary interactions. An interesting recent finding in this context was the identification of a minimal HHR variant with a strong increase in catalytic activity, based solely on the interaction of a single AU Hoogsteen base pair, formed by an A residing in the loop region of stem 2 of the HHR and an unpaired U from the 3′-end of the substrate RNA (O'Rourke et al. Reference O'Rourke, Estell and Scott2015).

Recent additions to the above-mentioned nucleolytic ribozymes are the Twister ribozyme (Roth et al. Reference Roth, Weinberg, Chen, Kim, Ames and Breaker2014) (Fig. 4) and related variants (Twister sister, Pistol and Hatchet) (Harris et al. Reference Harris, Lunse, Li, Brewer and Breaker2015; Li et al. Reference Li, Lunse, Harris and Breaker2015; Weinberg et al. Reference Weinberg, Kim, Chen, Li, Harris, Lunse and Breaker2015) that were identified by sequence- and structure-based bioinformatics algorithms. The Twister motif was identified in all domains of life, but its exact biological functions remain to be explored. The Twister ribozyme forms a double pseudoknot structure with its catalytic mechanism recently elucidated by a combination of structural (Eiler et al. Reference Eiler, Wang and Steitz2014; Liu et al. Reference Liu, Wilson, Mcphee and Lilley2014; Ren et al. Reference Ren, Kosutic, Rajashankar, Frener, Santner, Westhof, Micura and Patel2014), biochemical (Wilson et al. Reference Wilson, Liu, Domnick, Kath-Schorr and Lilley2016a) and modelling (Gaines & York, Reference Gaines and York2016) studies, using A and G as general acid and base, respectively (similar to the HP and VS ribozymes). The crystal structures were obtained from different Twister variants, (O. sativa) (Huang et al. Reference Huang, Vazin and Liu2014), an environmental variant (env) (Eiler et al. Reference Eiler, Wang and Steitz2014) and a minimized variant thereof (env22) (Ren et al. Reference Ren, Kosutic, Rajashankar, Frener, Santner, Westhof, Micura and Patel2014), showing the same overall ribozyme fold but with a partially different arrangement at the catalytic site.

As a significant difference to the HP and VS, which are using the N1 of A, the Twister applies the more acidic proton of the N3 of the conserved catalytic A (A1, adjacent to the cleavage site) for protonation of the 5′-oxygen (Fig. 4). This can only be achieved by a specific electrostatic environment causing a strong rise in pK _a towards neutrality (Kosutic et al. Reference Kosutic, Neuner, Ren, Flur, Wunderlich, Mairhofer, Vusurovic, Seikowski, Breuker, Hobartner, Patel, Kreutz and Micura2015). Similarly, a perturbed pK _a of A in the catalytic centre of the lead-dependent ribozyme was previously identified by NMR (Legault & Pardi, Reference Legault and Pardi1997). This not only adds a new mechanism to the repertoire of natural RNA catalysis, but also demonstrates how ribozymes can transcend their limited chemical functionalities, by forming micro-environments resulting in dramatically altered pK _a's of specified functional groups and thereby exploring a much broader array of catalytic strategies. Nevertheless, even though these new ribozyme variants comprise a divergent structural scaffold and a new catalytic mechanism, they all represent variations on the theme of RNA transesterification chemistry.

The advent of deep sequencing technology has not only revolutionized genomics (Koboldt et al. Reference Koboldt, Steinberg, Larson, Wilson and Mardis2013), but also provided a much more detailed picture of the fitness landscape of functional RNAs such as RNA aptamers (Jimenez et al. Reference Jimenez, Xulvi-Brunet, Campbell, Turk-Macleod and Chen2013) and short ribozymes (Ameta et al. Reference Ameta, Winz, Previti and Jaschke2014; Petrie & Joyce, Reference Petrie and Joyce2014; Pitt & Ferre-D'Amare, Reference Pitt and Ferre-D'amare2010). A recently introduced mutation analysis method for ribozymes also relies on an in-depth deep sequencing analysis (Kobori et al. Reference Kobori, Nomura, Miu and Yokobayashi2015). For this approach, the starting sequence compromises 97% of the wild-type bases, doped with 1% of each of the remaining nucleobases, and after the ribozyme catalysed reaction the active and inactive variants are separated and analysed by deep sequencing. Such detailed mutational analyses presents an ideal complement to the previously developed combinatorial NAIM (nucleotide analogue interference mapping) approaches that introduced base or sugar-modified nucleotides, to probe essential nucleoside functional groups in ribozymes and other functional RNAs (Cochrane & Strobel, Reference Cochrane and Strobel2004; Jansen et al. Reference Jansen, Mccarthy, Soukup and Soukup2006).

Deep sequencing analysis of a Twister ribozyme variant delivered a mutational landscape, by probing all single and double mutants, and provided a quantitative insight into the structure–function relationship of this ribozyme (Kobori & Yokobayashi, Reference Kobori and Yokobayashi2016). An interesting outcome of this mutational study was the discovery of its robustness to mutation, with mutations outside the catalytic cleft widely tolerated. These findings are entirely consistent with previous results for other small nucleolytic ribozymes (Kun et al. Reference Kun, Santos and Szathmary2005), where again mutations in the stem regions were widely tolerated, as long as the helix context and hence the overall fold of the ribozyme were not strongly perturbed, demonstrating the relaxed sequence requirements (and low error threshold for replication) of the small ribozymes.

In the context of the origin of life, both simplicity of sequence requirements and robustness to mutations emerge as clear advantages for RNA. Indeed, the seemingly disadvantageous compositional simplicity of nucleic acids compared with proteins (with only four structurally and chemical similar nucleobase building blocks compared with 20 structurally and chemically diverse amino acid side-chains) might in fact be critical for early evolution, enabling both high mutational tolerance as well as rapid adaptive trajectories across a lower complexity sequence space facilitating evolution.

3.2 In vitro selected ribozymes

Why is RNA cleavage by transesterification the only reaction catalysed by natural small ribozymes? A putative RNA world would have required a more diverse range of reactions, but given the narrow range of chemical transformations performed by today's natural ribozymes, it was not obvious that ribozymes would be able to support a putative RNA world metabolism. Following the advent of in vitro selection technologies, the principal capability of RNA catalysing diverse chemical reactions likely necessary in an RNA world could be explored (Chen et al. Reference Chen, Li and Ellington2007; Martin et al. Reference Martin, Unrau and Muller2015; Muller, Reference Muller2015).

Apart from RNA cleavage and ligation, one likely fundamental reaction in an RNA world (as in organic chemistry) would have been the formation of carbon–carbon (C–C) bonds. Accordingly, inspired by current organic chemistry, ribozymes catalysing C–C bond formation by either Diels-Alder cyclo-addition (Seelig & Jaschke, Reference Seelig and Jaschke1999; Tarasow et al. Reference Tarasow, Tarasow and Eaton1997), Michael addition (Sengle et al. Reference Sengle, Eisenfuhr, Arora, Nowick and Famulok2001) or aldol condensation (Fusz et al. Reference Fusz, Eisenfuhr, Srivatsan, Heckel and Famulok2005) were identified. Other reactions catalysed by in vitro selected ribozymes and likely necessary at the onset of the RNA world include pyrimidine nucleotide synthesis (Unrau & Bartel, Reference Unrau and Bartel1998), polynucleotide phosphorylation (kinase activity) (Lorsch & Szostak, Reference Lorsch and Szostak1994) and carbon–nitrogen bond formation (N-alkylation) (Wilson & Szostak, Reference Wilson and Szostak1995) (for a more complete overview see Chen et al. Reference Chen, Li and Ellington2007; Silverman, Reference Silverman2008; Wilson & Szostak, Reference Wilson and Szostak1999).

The transition from an RNA world to the more protein-based biology of today would have required RNA-catalysed amide bond (Wiegand et al. Reference Wiegand, Janssen and Eaton1997) or more specifically peptide bond (Zhang & Cech, Reference Zhang and Cech1997) formation and at a later stage the coordinated execution of all the processes comprising today's translation cycle. While modern day proteinaceous aminoacyl-tRNA synthetases (aaRS) combine activation and amino acid transfer, in vitro selected ribozymes are capable of catalysing amino acid activation in two separate steps. Amino acids can be activated as aminoacyl-guanylates (Kumar & Yarus, Reference Kumar and Yarus2001) chemically similar to natural activation as aminoacyl-adenylates, and the transfer of the activated amino acid to the 2′ or 3′ hydroxyl terminus of an acceptor RNA (aminoacylation) can be rapidly catalysed by in vitro selected ribozymes (Illangasekare et al. Reference Illangasekare, Sanchez, Nickles and Yarus1995; Lee et al. Reference Lee, Bessho, Wei, Szostak and Suga2000), even reduced to the smallest ribozyme ever described (Turk et al. Reference Turk, Chumachenko and Yarus2010) comprising only five nucleotides (nt) reacting with a tetranucleotide substrate (Turk et al. Reference Turk, Illangasekare and Yarus2011). Ribozymes were also selected catalysing the transfer of an amino acid (Met) on their own 5′-hydroxyl or -amino terminus forming either ester or amide bonds using 3′-acylated RNA as amino acid donor (Lohse & Szostak, Reference Lohse and Szostak1996), similar to catalysis in the P site of the ribosome. Finally, a range of ribozymes was developed (Flexizymes) (Morimoto et al. Reference Morimoto, Hayashi, Iwasaki and Suga2011) that are able to couple activated amino-acids to given tRNAs in vitro with applications in e.g. peptide selections by DNA display (Roberts & Szostak, Reference Roberts and Szostak1997). What is, however, lacking, so far, are ribozymes able to charge RNAs with specific amino acids, or otherwise link the identity of the amino acid to a coding triplet (or other) sequence unit to manifest a genetic code. Demonstrating control in implementation of catalytic phenotypes is as important as the catalytic phenotypes themselves when understanding RNA's capacity to form a functional translation system.

In the present-day biochemistry, nucleosides are activated as high-energy triphosphates (NTPs) to be used as substrates for nucleic acid synthesis and replication. Therefore, the in vitro selected RNA polymerase ribozyme (RPR) (see below), a molecular analogue of a postulated RNA replicase, was selected using nucleoside triphosphates as substrates (Ekland & Bartel, Reference Ekland and Bartel1996). Nucleoside triphosphates have some key advantages over more highly activated nucleotides such as phosphor-imidazolides. While the latter are highly reactive, they also hydrolyse readily in aqueous solution and therefore need to be continuously replenished. Nucleotide triphosphates (NTPs) on the other hand, while thermodynamically unstable, show a remarkable kinetic stability at neutral pH and therefore, once synthesized would accumulate. However, currently no prebiotic synthesis of NTPs has been described. This has motivated the search for a triphosphorylating ribozyme, which was recently discovered, using the prebiotically plausible trimetaphosphate as phosphate source (Dolan et al. Reference Dolan, Akoopie and Muller2015; Moretti & Muller, Reference Moretti and Muller2014). The identified TPR1/TPR1e ribozyme catalyses the formation of triphosphorylated RNA from trimetaphosphate, and a 5′-hydroxyl RNA oligonucleotide with a catalytic rate of 6·8 min⁻¹ under optimal conditions. Originally 96 nt long, a recently derived fragmented variant can be constructed from oligonucleotides no longer than 34 nt (Akoopie & Muller, Reference Akoopie and Muller2016), approaching the range of RNA oligomers accessible by non-enzymatic RNA polymerization (Ferris et al. Reference Ferris, Hill, Liu and Orgel1996). However, none of the current variants is capable of directly triphosphorylating nucleoside monomers and relies on attachment as part of a polynucleotide for 5′ positioning; general nucleoside substrate binding may be a challenging trait to evolve due to the tendency for RNA molecules to harness base-pairing for molecular recognition.

Even though proteinogenic amino acids exhibit a broader chemical diversity, more than half of modern day protein enzymes use cofactors with a large variety of functional groups, often based around a nucleoside ‘handle’, in particular adenosine (Chen et al. Reference Chen, Li and Ellington2007), potentially representing remnants from RNA world metabolism (White, Reference White1976). As nucleic acids exhibit high affinity and specificity for binding metal cations and small ligands, there is, in principle, no obstacle to ribozymes recruiting cofactors to broaden their chemical functionality and catalytic potential. Nevertheless, except for the glmS ribozyme, none of the natural ribozymes performs cofactor-assisted catalysis (e.g. by applying one of the typical protein cofactors such as coenzyme A (CoA), nicotinamide adenine dinucleotide (NAD) or flavin adenine dinucleotide (FAD)). However, in vitro evolution experiments have established that there are no functional obstacles to ribozymes utilizing cofactors. Examples include, e.g. an alcohol dehydrogenase ribozyme using NAD⁺ (Tsukiji et al. Reference Tsukiji, Pattnaik and Suga2003) or a ribozyme that decarboxylates a pyruvate-like substrate using thiamin as cofactor (Cernak & Sen, Reference Cernak and Sen2013). In vitro selected ribozymes are also capable of catalysing the synthesis of the common cofactors CoA, NAD and FAD from their precursors 4-phosphopantetheine, nicotinamide mononucleotide (NMN) and flavin mononucleotide (FMN) respectively (Huang et al. Reference Huang, Bugg and Yarus2000).

The RNA 4-base ‘code’ is both informationally and chemically much simpler than the 20 amino acid protein composition. Nevertheless, one may ask if an even simpler ternary or even a binary code could support RNA catalysis. Joyce and coworkers explored this question using ribozyme catalysed RNA ligation as a model system. To perform selection experiments in the absence of C (comprising sequences with only A, G and U), all C residues in the original random RNA library were deaminated to U by sodium bisulphite treatment (Rogers & Joyce, Reference Rogers and Joyce1999). From this ternary code RNA library, functional RNA ligases could be isolated, but reselection with the inclusion of C resulted in an increase of the catalytic rate by a factor of 20 (Rogers & Joyce, Reference Rogers and Joyce2001). Ribozyme selections with only two nucleotides [2,6-diaminopurine (replacing the natural adenine for higher base-pairing stability) and uridine] led to a functional ligase variant, however showing only low catalytic rates and yields (8% ligation yield in 80 h, k _obs = 0·05 h⁻¹) (Reader & Joyce, Reference Reader and Joyce2002). Thus, it seems (at least judging from these three examples) that although catalysts can be isolated from simple binary repertoires, catalytic power seems to scale with informational complexity. Nevertheless, in an early environment without competition by efficient ribozymes or protein enzymes even a small rate enhancement over the uncatalysed reaction might have resulted in a substantial selective advantage.

In vitro selected ribozymes not only show a broad spectrum of different reaction parameters depending on the chemical transformation they catalyse, but also on the applied selection conditions, including strong variations in catalytic rates and yields, catalysis in a cis- and/or trans format and the ability for multi-turnover catalysis. Reaction conditions are also often prebiotically implausible including high concentrations of reactants and/or high Mg²⁺ concentrations detrimental to the half-life of ribozymes. However, the selected ribozymes represent at best a fraction of the potential prebiotic sequence and phenotype space, and therefore should simply be considered as proof-of-principle for the potential of ribozyme-catalysed reactions. Nevertheless, lack of efficient reaction rates and yields with ideally multi-turnover catalysis remain one of the main shortcomings of many in vitro selected ribozymes.

Although substrate selectivity is an essential requirement for catalysts, a degree of substrate promiscuity would provide a mechanism to evolve new ribozyme functions rapidly as has been observed for protein enzymes (Khersonsky & Tawfik, Reference Khersonsky and Tawfik2010). A related question is whether ribozymes have to adopt different structural folds to catalyse different chemical transformations. Bartel and co-workers (Schultes & Bartel, Reference Schultes and Bartel2000) explored this question using a RNA sequence derived from the HDV self-cleaving ribozyme and the class III self-ligating ribozyme (catalysing 2′−5′ linked bond formation from 5′-triphosphorylated and 2′,3′-diol substrate RNAs) (Ekland & Bartel, Reference Ekland and Bartel1996) by a number of iterative mutational steps reaching a ‘hybrid sequence’, which is 42 and 44 mutational steps away from the parent ligase or HDV sequence, respectively. This hybrid sequence was able to fold into two distinct folds, catalysing either RNA cleavage or ligation, but with reduced catalytic rates compared with the original variants that fold into only one catalytic active fold. On the other hand, the conversion of a self-aminoacylating ribozyme, that aminoacylates its 3′ terminus using adenylated phenylalanine (Illangasekare et al. Reference Illangasekare, Sanchez, Nickles and Yarus1995) into a self-kinase ribozyme that phosphorylates its own 5′-end using GTPγS (Lorsch & Szostak, Reference Lorsch and Szostak1994) by in-vitro evolution required on average only 14 mutations, with an increased likelihood to find catalytic activity for the new substrate the more distant the RNA moved from the original fold, indicating the necessity to escape the parent fold (Curtis & Bartel, Reference Curtis and Bartel2005).

The application of deep sequencing technology has allowed a more in-depth analysis of the adaptive fitness landscapes of functional RNAs and therefore also the distribution of a specific catalytic function in RNA sequence space (Pitt & Ferre-D'Amare, Reference Pitt and Ferre-D'amare2010). A recent in vitro selection experiment starting from two different ligase ribozymes, the class I ligase (Ekland & Bartel, Reference Ekland and Bartel1996) and the DSL ligase (Ikawa et al. Reference Ikawa, Tsuda, Matsumura and Inoue2004), both catalysing 3′−5′ bond formation between 5′-triphosphorylated RNA and 2′,3′-hydroxyl RNA substrates, resulted in variants clustered around each parent sequence, indicating a RNA fitness landscape with isolated fitness peaks (Petrie & Joyce, Reference Petrie and Joyce2014). At least for these ribozymes this study deemphasizes the function of neutral drift as primary source of genetic change, but rather as a provider of a reservoir of sequences on which selective adaptation can be based.

While high-resolution structures for all currently known natural ribozymes are available (see above) only few crystal structures of in-vitro selected ribozymes, such as the leadzyme (Wedekind & McKay, Reference Wedekind and Mckay1999) and the Diels-Alder ribozyme (Serganov et al. Reference Serganov, Keiper, Malinina, Tereshko, Skripkin, Hobartner, Polonskaia, Phan, Wombacher, Micura, Dauter, Jaschke and Patel2005) have been determined. The latter adopts a fold that forms a binding pocket for enantioselective catalysis with a combination of different factors such as shape complementarity, electronic effects, stacking interactions (in particular to the anthracene substrate) and hydrogen bonding (mainly to the maleimide substrate) all contributing to the catalysed C–C bond formation.

To expand the chemical functionality beyond the four standard ribonucleotides, modified nucleotides, in particular with modifications to the C5 position of uracil, have been introduced. Substituents attached to the C5 position project into the major groove and cause minimal steric clashes with the polymerase and are therefore well tolerated by most DNA/RNA polymerases and reverse transcriptases. Furthermore, there is a reasonably facile chemical synthesis of C5-modified U-triphosphates. The selection of a Diels-Alder ribozyme (Tarasow et al. Reference Tarasow, Tarasow and Eaton1997) was one of the first ribozyme selections including a base-modified nucleoside triphosphates (5-pyridylmethyl-carboxamide-UTP), with the pyridine contributing to increased stacking interactions. A later selection without modified triphosphates resulted in another Diels-Alder ribozyme variant (Seelig & Jaschke, Reference Seelig and Jaschke1999), with a likely different catalytic fold (Serganov et al. Reference Serganov, Keiper, Malinina, Tereshko, Skripkin, Hobartner, Polonskaia, Phan, Wombacher, Micura, Dauter, Jaschke and Patel2005). Other selections performed with nucleobase-modified triphosphates include, e.g. an amide synthase ribozyme (Wiegand et al. Reference Wiegand, Janssen and Eaton1997) with 5-imadozolyl-UTP and an RNA ligase ribozyme with N6-aminohexyl modified adenine residues (Teramoto et al. Reference Teramoto, Imanishi and Ito2000). Nevertheless none of the modifications are per se essential for the catalysed chemical transformation and other ribozymes without modifications are not inferior in their catalytic activity.

3.3 DNA catalysis

For a long-time RNA was only seen as an information carrier from genes to proteins, while the role for DNA was manifested in its function for long-term storage of genetic information. The capacity of DNA for information storage and the possibility of catalytic activity were considered mutually exclusive. Indeed, DNA is generally depicted in the famous double helical form (Watson & Crick, Reference Watson and Crick1953) which, with its rigid linear structure, seems unlikely to support catalysis. It therefore came as a surprise, when, in 1994, the first deoxyribozyme/DNAzyme was identified by in vitro selection by Breaker and Joyce (Breaker & Joyce, Reference Breaker and Joyce1994). This first deoxyribozyme catalysed the Pb²⁺ assisted cleavage of a single ribonucleotide linkage inside an all DNA substrate strand with a rate enhancement of ~10⁵-fold over the uncatalysed reaction.

So far no bona fide deoxyribozymes have been found in nature and therefore the question of whether catalytic DNA has functions in vivo remains unanswered. Recently, a short Zn²⁺ dependent DNA cleaving deoxyribozyme was identified (Gu et al. Reference Gu, Furukawa, Weinberg, Berenson and Breaker2013), and sequence comparison with natural genomes yielded a number of hits with consensus sequences showing DNA cleavage activity under the selection conditions. Further studies will be needed to establish, if this is merely a fortuitous sequence similarity or, if it reflects true in vivo functionality.

Since this first example, the catalytic potential of DNA has been explored by in vitro selection and many DNAzymes identified that catalyse a diverse range of chemical reactions similar to their RNA counterparts (Hollenstein, Reference Hollenstein2015; Silverman, Reference Silverman2009, Reference Silverman2016). Indeed, it seems that in a number of ways DNA is not catalytically inferior compared to RNA, despite the absence of the 2′-hydroxyl functionality that can assist in acid/base catalysis or act as a nucleophile in RNA (Silverman, Reference Silverman2008). Rather, deoxyribozymes come with a number of (technical) advantages including easier (and less costly) synthesis and greater resistance to chemical and enzymatic degradation. Nevertheless the deoxyribose in DNA leads to a preferential C3′ endo sugar pucker versus a C2′ endo pucker for ribose in RNA, which also results in a preferential B- versus A-form helical conformation for double-stranded DNA compared with RNA. This, together with altered base-pairing energetics prevents the direct conversion of ribozymes into deoxyribozymes (or vice versa) leading instead to inactive variants. However, using in vitro evolution, one ribozyme could be transformed into the corresponding deoxyribozyme (Paul et al. Reference Paul, Springsteen and Joyce2006) requiring only seven mutations suggesting that active ribo- and deoxyribozymes may be proximal in sequence space at least in some cases. Indeed, even HHR variants with a mixed ribo-/deoxyribonucleotide backbone can be catalytically active (Perreault et al. Reference Perreault, Wu, Cousineau, Ogilvie and Cedergren1990).

Similar to ribozymes, DNA catalysts show a strong preference for phosphodiester transfer reactions and for nucleic acids substrates in general. Rather than the true catalytic potential, this may again reflect the biases introduced by selection strategies, which are facilitated by the easy positioning of substrates through Watson–Crick base-pairing.

Mechanistic analysis of ribo- and deoxyribozymes suggests that the catalytic potential of RNA and DNA is realized by comparable catalytic strategies. However, while the 3D arrangement of catalytic residues and aspects of the catalytic mechanism of many naturally occurring ribozymes are known in some detail due to high-resolution structures (see above), deoxyribozymes so far lag behind in structural understanding. Nevertheless, there is hope that this might change in the near future. The recent landmark publication of the first atomic resolution structure of a deoxyribozyme (Ponce-Salvatierra et al. Reference Ponce-Salvatierra, Wawrzyniak-Turek, Steuerwald, Hobartner and Pena2016) paves the way for a more detailed understanding of deoxyribozyme catalysis. The crystal structure was obtained of the 44 nt (of which 31 nt form the catalytic core) comprising minimal RNA-ligating 9DB1 deoxyribozyme (Purtha et al. Reference Purtha, Coppins, Smalley and Silverman2005; Wachowius et al. Reference Wachowius, Javadi-Zarnaghi and Höbartner2010) bound to its 15 nt RNA substrate in the post-catalytic state. The structure resembles the Greek letter λ with the two DNA–RNA duplexes of the binding arms forming an angle of 120° to each other and both lying above and tightly attached to the catalytic core. The catalytic domain consists of a 4 and a 2 nt base-pair stem and two nucleotides in the catalytic core (dT29 and dT30), which directly base pair with the RNA nucleotides A1 and G1 at the ligation junction leading to a double pseudoknot structure of the deoxyribozyme RNA substrate complex (Fig. 5).

Fig. 5. Mechanism and structure of the RNA-ligating deoxyribozyme 9DB1. (a) Secondary structure of the minimized 9DB1 variant, displaying the catalytic core in blue, the RNA binding regions in orange and the RNA substrates 5′ and 3′ of the ligation junction in red and green, respectively. (b) Chemical mechanism of 9DB1 catalysed 3′–5′-RNA ligation. The nucleophilic attack of a 3′-hydroxyl of a 2′,3′-diol terminated RNA on a 5′-triphosphorylated RNA substrate generates regioselective 3′–5′-RNA phosphodiester linkages. (c) Secondary structure of the 9DB1 crystal structure illustrating the double pseudoknot interactions, red marked nucleotides in the catalytic core are sensitive to mutations. (d) Ribbon representation (including the nucleobases) of the crystal structure of the 9DB1 deoxyribozyme bound to its ligated RNA substrate (PDB: 5cck). The colour code corresponds to (a).

The original 9DB1 sequence shows a strong preference for purines (A, G) at the 5′ end of the triphosphorylated RNA substrates. Interestingly, as a result of the observed base-pairing between the two DNA nucleobases in the catalytic core with the RNA nucleotides at the ligation junction, a single mutation in the catalytic loop of dT29 to either dG29 or dA29 allows an exchange of the nucleobase at the 5′ position of the triphosphorylated RNA substrate to C or U respectively. This enables ligation of substrates with all 4 RNA nucleobases and demonstrates how structural data may allow the reengineering of deoxyribozymes.

The structure also provides a first glimpse of how DNA compensates for its ‘missing’ 2′-OH to perform with comparable catalytic efficiency as RNA. This appears to be achieved by the broad range of the pseudorotation phase angles of nucleotides in the DNAzyme. In particular, the DNA nucleotides in the catalytic loop of 9DB1 show a much broader flexibility of the sugar phosphate backbone compared with ribozymes. There are 20 (out of 31) forming south (S)-type and eight north (N)-type sugar puckers, with the remaining three nucleotides adopting sugar conformations outside typical N-/S-conformations enabling positioning of active residues for catalysis.

The most prominent and widely used deoxyribozymes are RNA cleaving deoxyribozymes (Silverman, Reference Silverman2005). Almost all RNA cleaving deoxyribozymes catalyse RNA cleavage by a transesterification mechanism similar to the small nucleolytic ribozymes, involving an intramolecular attack of the 2′-hydroxyl on the adjacent phosphodiester linkage forming a 2′, 3′-cyclic phosphate and a 5′-hydroxyl terminus. Interestingly, other catalytic mechanisms are possible. Recently, a deoxyribozyme was selected that catalyses RNA cleavage by the normally disfavoured hydrolysis mechanism, e.g. attack of a water molecule on a phosphodiester linkage forming either a 5′-phosphate and 3′-hydroxyl or a 5′-hydroxyl and 3′-phosphate (Parker et al. Reference Parker, Xiao, Aguilar and Silverman2013).

The most prominent and best-studied representatives of RNA cleaving deoxyribozymes are the 10–23 and 8–17 deoxyribozymes (Santoro & Joyce, Reference Santoro and Joyce1997) that catalyse RNA cleavage by transesterification, with multiple-turnover capability (Fig. 6).

Fig. 6. Deoxyribozyme catalysed RNA cleavage. (a) The nucleophilic attack of the 2′-hydroxyl on the adjacent phosphorus of the phosphodiester bond generates 2′,3′-cyclic phosphate and 5′-hydroxyl termini. (b) Secondary structure of the most prominent RNA cleaving deoxyribozymes 10-23 and 8-17. The catalytic core is shown in blue, the substrate-binding arms in orange and the RNA strand 5′ and 3′ of the cleavage junction (arrow) are displayed in red and green, respectively.

Variants of the 8–17 motif have been selected independently a number of times (Schlosser & Li, Reference Schlosser and Li2010), making the 8–17 sequence motif the most likely solution for RNA cleavage in DNA sequence space, similar to the HHR in RNA space (Salehi-Ashtiani & Szostak, Reference Salehi-Ashtiani and Szostak2001). The catalytic mechanism of deoxyribozyme-catalysed RNA cleavage is likely similar to that of ribozymes involving one or a combination of the following four catalytic strategies: (a) in-line nucleophilic attack, (b) deprotonation of the 2′-hydroxyl group, (c) neutralization of the negative charge at a non-bridging phosphate or (d) at the 5′ oxygen (Emilsson et al. Reference Emilsson, Nakamura, Roth and Breaker2003). The preference for divalent metal ions may also reflect their availability during the in vitro selection process with their identity having a strong impact on the catalytic rate of deoxyribozymes. Indeed, not only are some deoxyribozymes very selective concerning identity and concentration of the metal ion (whereas others are more relaxed), but also different metal ions can lead to different DNA folding arrangements and reaction rates as demonstrated for the 8–17 deoxyribozyme using FRET (Kim et al. Reference Kim, Rasnik, Liu, Ha and Lu2007). 8–17-catalysed RNA cleavage in the presence of Zn²⁺ and Mg²⁺ proceeds via DNA folding followed by catalysis (i.e. the cleavage reaction), but in the presence of Pb²⁺ the cleavage reaction occurred without a folding step, rationalizing the fast rate of the Pb²⁺ assisted cleavage. This points towards a prearranged structural DNA scaffold of 8–17 in the presence of Pb²⁺ ions, but not for Zn²⁺ and Mg²⁺ ions (Kim et al. Reference Kim, Rasnik, Liu, Ha and Lu2007; Liu & Sen, Reference Liu and Sen2010). An interesting recent finding is the influence of trivalent lanthanide ions on deoxyribozyme catalysis (Dokukin & Silverman, Reference Dokukin and Silverman2012; Huang et al. Reference Huang, Vazin and Liu2014; Javadi-Zarnaghi & Hobartner, Reference Javadi-Zarnaghi and Hobartner2013). A number of lanthanide-dependent RNA-cleaving deoxyribozymes were recently reported (Liu, Reference Liu2015), including variants depending on two metal ions (Torabi & Lu, Reference Torabi and Lu2015; Zhou et al. Reference Zhou, Zhang, Huang, Ding and Liu2016b).

The recent finding of a deoxyribozyme independent of divalent metal ions with a fast catalytic rate (k _obs = 0·1 min⁻¹ in 400 mM Na⁺, 20 °C) and additionally with an astonishing selectivity for Na⁺ over competing monovalent cations (Torabi et al. Reference Torabi, Wu, Mcghee, Chen, Hwang, Zheng, Cheng and Lu2015) underlines the similarity between ribozyme and deoxyribozyme catalysis and points towards the possibility of nucleobase assisted general acid–base catalysis also for deoxyribozymes. This is similar to earlier findings of RNA-cleaving deoxyribozymes that perform catalysis independent of divalent metal ions (Carrigan et al. Reference Carrigan, Ricardo, Ang and Benner2004; Faulhammer & Famulok, Reference Faulhammer and Famulok1997; Geyer & Sen, Reference Geyer and Sen1997). The Na8 deoxyribozyme has a k _obs = 0·007 min⁻¹ (0·5 M M⁺, pH 7 and 25 °C), where the identity of the monovalent cation (M) is largely irrelevant (Geyer & Sen, Reference Geyer and Sen1997). Another deoxyribozyme shows divalent metal independent RNA cleavage at pH3 (Liu et al. Reference Liu, Mei, Brennan and Li2003). As the N1 of adenine, N3 of cytosine and N7 of guanine are expected to be protonated at pH3 (Blackburn et al. Reference Blackburn, Gait, Loakes and Williams2006), the positive charge from the protonated bases likely fulfills the function of the divalent metal ions.

DNA catalysis is also possible with a reduced set of nucleotides, albeit with a substantial decrease in activity. A RNA cleaving deoxyribozyme consisting of only C and G showed a ~10⁴ times reduced cleavage activity compared with the parent one with all four nucleotides, but still with an increase by a factor of ~5000 over the uncatalysed background reaction (Schlosser & Li, Reference Schlosser and Li2009). This parallels findings for ribozymes with a reduced nucleobase composition (Reader & Joyce, Reference Reader and Joyce2002; Rogers & Joyce, Reference Rogers and Joyce1999).

Apart from RNA cleavage, DNA-catalysed RNA ligation represents another important reaction type, mainly pursued by Silverman and co-workers. Initial selection efforts identified deoxyriboyzmes catalysing non-native 2′–5′ ligation using Mg²⁺ as cofactor (Flynn-Charlebois et al. Reference Flynn-Charlebois, Wang, Prior, Rashid, Hoadley, Coppins, Wolf and Silverman2003). Interestingly, using Zn²⁺ instead of Mg²⁺ during the selection process yielded deoxyribozymes catalysing the formation of native 3′–5′ linkages (Hoadley et al. Reference Hoadley, Purtha, Wolf, Flynn-Charlebois and Silverman2005), illustrating the important contribution of the metal ion cofactor, not only to catalytic rates but to regioselectivity. Another selection strategy led to Mg²⁺ dependent 3′–5′ RNA-ligating deoxyribozymes with a broader sequence generality and good catalytic efficiencies (Purtha et al. Reference Purtha, Coppins, Smalley and Silverman2005). In addition to linear RNA ligation, the 5′-end of one RNA substrate could be ligated to an internal 2′-hydroxyl forming a 2′,5′ branched RNA or as a special case of branch formation a lariat RNA, where the RNA reacts on itself in an intramolecular fashion, forming a closed loop. This reaction type is naturally catalysed by group II introns and the spliceosome. The first RNA 2′,5′ branch-forming deoxyribozymes were identified using the 5′-triphosphate/2′,3′-diol RNA substrate combination, albeit with a rather strong sequence requirement at the ligation junction (Wang & Silverman, Reference Wang and Silverman2003). Further selection efforts identified the 7S11 deoxyribozyme, that catalyses 2′,5′-branch formation by ligating a 5′-triphosphorylated G to an internal A residue, which is flanked by Watson–Crick duplex regions, in a similar fashion as the first step of natural RNA splicing (Coppins & Silverman, Reference Coppins and Silverman2004). 7S11 and later identified 2′,5′ branch-forming deoxyribozymes (Lee et al. Reference Lee, Mui and Silverman2011) all form a three-helix-junction (3HJ) with their RNA and DNA substrates. This structural arrangement is similar to ribozymes that also frequently include multiple helix junction structures.

Deoxyribozymes are also capable of using DNA as substrates and catalysing DNA cleavage and ligation reactions. However, as DNA is much less reactive compared with RNA due to the absence of the 2′-hydroxyl group, DNA substrates have to be activated for ligation to achieve similar catalytic rates as their RNA counterparts. The first deoxyribozyme that catalysed DNA ligation was reported soon after the initial description of the first RNA-cleaving deoxyribozyme (Cuenoud & Szostak, Reference Cuenoud and Szostak1995). This deoxyribozyme catalyses the ligation of a 5′-hydroxyl DNA substrate with a 3′-phosphoimidazole activated DNA substrate and is an obligate metalloenzyme, requiring Zn²⁺ (or Cu²⁺) and Mg²⁺ for activity. Similarly, a deoxyribozyme was identified that uses a 5′-adenylate/3′-hydroxyl substrate combination for DNA ligation, mimicking the final step of protein T4 DNA ligase catalysed DNA ligation (Sreedhara et al. Reference Sreedhara, Li and Breaker2004). The 5′-adenylate substrate was itself synthesized by a capping deoxyribozyme (Li et al. Reference Li, Liu and Breaker2000) that forms a 5′,5′-pyrophosphate linkage from ATP and a DNA substrate, which is remarkably different to a phosphorylating deoxyribozyme that uses NTPs to catalyse the 5′ phosporylation of DNA (Li & Breaker, Reference Li and Breaker1999).

Due to the absence of an internal nucleophile (as the 2′-OH in RNA) DNA cleavage is much more difficult to achieve. The first DNA cleaving deoxyribozyme described cleaves DNA in a non-specific manner by a Cu²⁺-dependent oxidative mechanism (Carmi et al. Reference Carmi, Shultz and Breaker1996). A completely different mechanism for DNA strand cleavage was achieved by the deoxyribozyme catalysed N-glycosylation of a particular G residue, leading to strand scission at the apurinic site (Sheppard et al. Reference Sheppard, Ordoukhanian and Joyce2000). Later, the 10MD5 bimetallic deoxyribozyme was identified, requiring both Zn²⁺ and Mn²⁺ for activity, that cleaves single-stranded DNA by a hydrolysis mechanism with multi-turnover kinetics and an astonishing rate enhancement of 10¹², albeit with a rather strong sequence dependence (ATG^T) at the cleavage site (Chandra et al. Reference Chandra, Sachdeva and Silverman2009). Only two mutations in the original 10MD5 sequence changed the metal ion requirements from bimetallic Mn²⁺/Zn²⁺ to Zn²⁺ only, suggesting a simple structural role for Mn²⁺ and a catalytic function for Zn²⁺ (Xiao et al. Reference Xiao, Allen and Silverman2011). Further selection efforts identified different DNA cleaving deoxyribozymes with different dinucleotide sequence requirements at the cleavage junction (Xiao et al. Reference Xiao, Wehrmann, Ibrahim and Silverman2012).

Apart from cleavage/ligation reactions of nucleic acid substrates, deoxyribozymes – just like their ribozyme counterparts – are capable of catalysing a diverse array of other reaction types. Nevertheless, due to design of the selection strategies and the selectivity and convenient ease of programming interactions by Watson–Crick base-pairing, almost all reactions occur on substrates tethered to nucleic acids. Exceptions include the Diels-Alder cycloaddition (Chandra & Silverman, Reference Chandra and Silverman2008) and porphyrin metallation, e.g. the deoxyribozyme catalysed insertion of Cu²⁺ and Zn²⁺ into mesoporphyrin (Li & Sen, Reference Li and Sen1996). The Silverman group in particular has been expanding the scope of deoxyribozyme catalysis and their current focus lies on peptide/protein modifying deoxyribozymes (Silverman, Reference Silverman2015). Initially, the first deoxyribozyme that catalysed a RNA nucleopeptide linkage was formed between a 5′-triphosphate RNA and the hydroxyl of a tyrosine residue that was replacing the branch site A in the 7S11 3HJ structural context (Pradeepkumar et al. Reference Pradeepkumar, Höbartner, Baum and Silverman2008). The less reactive aliphatic hydroxyl of serine required a slightly more flexible arrangement by introduction of a tripeptide sequence (Sachdeva & Silverman, Reference Sachdeva and Silverman2010) and for the lysine amino acid side-chain, the more reactive 5′-imidazolide RNA substrate was required (Brandsen et al. Reference Brandsen, Velez, Sachdeva, Ibrahim and Silverman2014).

The initial selection trial for amide bond hydrolysis led instead to DNA-hydrolysing deoxyribozymes (Chandra et al. Reference Chandra, Sachdeva and Silverman2009). The intended deoxyribozyme catalysed cleavage of amide bonds was finally discovered by a clever selection scheme including a 5′-amino oligonucleotide capture tag, capturing the free carboxyl group that is formed by amide or ester cleavage, but not by DNA phosphodiester bond hydrolysis (Brandsen et al. Reference Brandsen, Hesser, Castner, Chandra and Silverman2013). The chemically more favourable cleavage of aromatic amide bonds was achieved with a standard DNA pool, but for the cleavage of an aliphatic amide bond, a selection scheme including modified deoxyuridines with amino acid type side chains at their 5 position (5-aminoallyl, 5-hydroxymethyl and 5-carboxyvinyl) were used, leading to deoxyribozyme variants with amide bond hydrolase activity for all three modifications and demonstrating the principal ability of DNAzymes to cleave peptidic amide bonds (Zhou et al. Reference Zhou, Avins, Klauser, Brandsen, Lee and Silverman2016a).

Apart from the cleavage chemistry, the sequence-specific recognition of amino acids and therefore peptides and proteins has been another challenge. Deoxyribozymes are capable of phosphomonoester hydrolysis; hence, phosphatase activity was established by applying an additional selection step, including a RNA capture oligo and a previously identified deoxyribozyme capable of forming a covalent bond between the free hydroxyl of a tyrosine and the 5′ triphosphorylated RNA capture oligo (Chandrasekar & Silverman, Reference Chandrasekar and Silverman2013). This Zn²⁺-dependent phosphatase deoxyribozyme is capable of sequence-specific dephosphorylation of phosphotyrosine and phosphoserine inside a hexapeptide and most importantly also within a protein context. Deoxyribozymes are also capable of catalysing the reverse (phosphorylation) reaction. Deoxyribozymes with tyrosine-specific kinase activity were identified by again using a capture deoxyribozyme catalysing the ligation of only phosphor-Tyr (and not Tyr) with a 5′-triphosporylated RNA or GTP (Walsh et al. Reference Walsh, Sachdeva and Silverman2013). Another recently described kinase deoxyribozyme is able to catalyse the 3′-phosphorylation of DNA by using 5′-triphosphorylated RNA (Camden et al. Reference Camden, Walsh, Suk and Silverman2016) a reaction not catalysed by natural occurring protein enzymes.

4.1 Prebiotic synthesis of RNA monomers

Self-replication may be considered a specialized form of catalysis coupled to information transfer. The emergence of RNA self-replication has often been considered as a key transition in the origin of life (Gilbert Reference Gilbert1986). However, self-replication in a prebiotic setting requires a template molecule to initiate a replication cycle. Thus, nucleic acid polymers need to be first generated by de novo assembly from activated precursors and such activated precursors need in turn be generated from simple prebiotic feedstock molecules. However, a convincing prebiotic synthesis of RNA nucleosides or preferably suitably chemically activated nucleotides had proven elusive for a long time. While individual nucleobases could plausibly be assembled from prebiotic building blocks such as HCN, urea or cyanoacetylene, their linkage to ribose or phosphoribose sugars or indeed the synthesis of such sugars in reasonable yield and purity proved challenging with the most plausible reaction, the so-called formose reaction from formaldehyde, yielding mostly indescribably complex mixtures. Nevertheless, the simple presence of borate salts can selectively stabilize 1,2-cis-diol compounds (Ricardo et al. Reference Ricardo, Carrigan, Olcott and Benner2004) demonstrating a possible path to enrich ribose-containing compounds from such mixtures.

The difficulties in describing credible prebiotic syntheses of ribonucleotides and specifically the apparently intractable problem of N-glycosidic bond formation between ribose and nucleobase in an aqueous environment led to investigation of plausible chemical and genetic precursors of RNA. This ‘pre-RNA world’ or ‘proto-RNA’ chemistry is based on alternative genetic polymers with a different backbone chemistry such as TNA (Schoning et al. Reference Schoning, Scholz, Guntha, Wu, Krishnamurthy and Eschenmoser2000) or PNA (Ura et al. Reference Ura, Beierle, Leman, Orgel and Ghadiri2009) or the exploration of completely different sugar nucleobase combinations (Benner et al. Reference Benner, Karalkar, Hoshika, Laos, Shaw, Matsuura, Fajardo and Moussatche2016; Cafferty et al. Reference Cafferty, Fialho, Khanam, Krishnamurthy and Hud2016; Winnacker & Kool, Reference Winnacker and Kool2013) as possible RNA precursors. Both approaches consider the emergence of RNA not as singular abiotic event from simple organic precursors, but instead as the endpoint of a chemical and evolutionary trajectory from more facile, or seemingly prebiotically easier accessible information systems that were gradually transforming into RNA (Hud et al. Reference Hud, Cafferty, Krishnamurthy and Williams2013).

However, the need for direct N-glycosidic bond formation between ribose and pyrimidine nucleobase was elegantly circumvented by the landmark discovery of a prebiotic synthesis of activated RNA pyrimidine nucleotides (C, U) in high yields from simple prebiotically-accessible precursor molecules and inorganic phosphate via amino-oxazolines (Powner et al. Reference Powner, Gerland and Sutherland2009). In a different pathway, a recently described synthesis of the RNA purine nucleosides (A, G) from formamido-pyrimidines and ribose yielded the correct N9 regioisomer and ribose β-anomer, also avoiding the direct coupling of the full nucleobase and ribose (Becker et al. Reference Becker, Thoma, Deutsch, Gehrke, Mayer, Zipse and Carell2016) and its associated problems in yield and stereoselectivity (Fuller et al. Reference Fuller, Sanchez and Orgel1972).

These syntheses provide proof of principle that a prebiotic synthesis of the four RNA building blocks from simple organic precursors is possible and lessens the need for pre-RNA and/or proto-RNA world scenarios. Indeed, one potentially fatal pitfall of pre- or proto-RNA world scenarios concerns the problem of genetic ‘handover’. While genotypes (i.e. base sequence) are readily transferred between different genetic polymer systems as long as base-pairing properties are not massively distorted (as shown for DNA/RNA and some DNA/XNAs), phenotypes (3D structure/folding/function, e.g. catalytic activity) are generally either substantially impacted or non-transferable. The latter is illustrated by the polymer-specific sequence motifs emerging from in vitro evolution experiments and the failure in interconverting active catalysts even between closely related genetic polymer systems, such as DNA and RNA or DNA and ANA (Paul et al. Reference Paul, Springsteen and Joyce2006; Taylor et al. Reference Taylor, Pinheiro, Smola, Morgunov, Peak-Chew, Cozens, Weeks, Herdewijn and Holliger2015). Finally, while TNA, PNA and other proposed pre-RNA systems show in principle similar information storage capabilities compared with RNA, they nevertheless likely exhibit a different catalytic potential compared with RNA, in particular with regards to transesterification and recombination reactions [as with DNA (see above)], which may have been important for early evolution.

Remarkably, the above described pyrimidine RNA nucleobase synthesis yields 2′,3′-cyclic phosphate activated cytidine and uridine (N > ps) as their final products with similar yields (Powner et al. Reference Powner, Gerland and Sutherland2009). Assuming that such 2′,3′-cyclic phosphate ribonucleotides are readily accessible from prebiotic chemistry, they could polymerize into short oligonucleotides under favourable conditions (Verlander & Orgel, Reference Verlander and Orgel1974) (although with preferential formation of the non-canonical 2′–5′ linkages). While a certain amount of sporadic 2′–5′ linkages (within a predominantly 3′–5′ context) are not incompatible with RNA function (Engelhart et al. Reference Engelhart, Powner and Szostak2013) (see below) it is currently unknown if (and how) a predominantly 2′–5′ RNA polymer could evolve and eventually transition to a 3′–5′ RNA polymer while retaining function. Chemoselective acetylation of the 2′ hydroxyl of ribose may provide a solution: such protection mechanisms can lead to the selective formation of canonical 3′–5′ linkages (Bowler et al. Reference Bowler, Chan, Duffy, Gerland, Islam, Powner, Sutherland and Xu2013).

4.2 Non-enzymatic polymerization of RNA

Non-templated polymerization mediated by substrate alignment and concentration in montmorillonite clays or eutectic ice phases, using the more reactive 5′-phosphorimidazole activated ribonucleotides, can yield RNA oligonucleotides between ~17 nts [with mixed base composition] (Monnard et al. Reference Monnard, Kanavarioti and Deamer2003) up to 50-mers (homopolymers) (Ferris et al. Reference Ferris, Hill, Liu and Orgel1996). The prebiotic plausibility of this form of activation is yet to be demonstrated; nucleotide condensation requires phosphate activation arising from either synthesis (e.g. N > ps) or an external electrophile. Oligonucleotide 5′-polyphosphates (including triphosphates) can be formed from polynucleotide mono-phosphates and sodium trimetaphosphate, although given its reactivity the availability and persistence of this agent needs justification. The ideal activating agent or conditions remain to be characterized, but alternative approaches that promote condensation using dehydrating conditions can be imagined. Nucleoside-5′-phosphates can be assembled into polymers by heating and wet/dry cycles in lamellar lipid phases or at acidic pH (Deamer, Reference Deamer2012; DeGuzman et al. Reference Deguzman, Vercoutere, Shenasa and Deamer2014) though the products of apparent 100 nucleotide length that are observed in gel electrophoresis appear to contain a substantial number of abasic sites (presumably caused by depurination during temperature cycling or at low pH) (Mungi & Rajamani, Reference Mungi and Rajamani2015). Furthermore, due to the inherent chemical fragility of RNA, harsh temperature or chemical/pH gradients are unlikely to be compatible with an early RNA genetic system. Milder conditions for polymerization are likely required to build polymers that retain an intrinsic capability of both information storage and propagation as described below.

RNA templates can pre-organize activated mononucleotides for non-enzymatic polymerization as first explored by Orgel and colleagues for nucleotide phosphorimidazolides and 2-methyimidazolides (Fig. 7).

Fig. 7. Non-enzymatic templated polymerization of RNA. (a) A templated primer is extended at its 3′ end by 5′-methylimidazolide activated (or other activation chemistries, see text) RNA nucleotides. Polymerization is facilitated by transient binding of 5′-activated short oligonucleotides (‘helper’ oligomers), coloured in grey, upstream of the template strand. (b) The polymerization reaction is based on the nucleophilic attack of the primer 3′-hydroxyl on the 2-methylimidazolide activated 5′-phosphorus of the incoming RNA nucleotide, resulting mainly in canonical 3′–5′-RNA linkages.

In particular, the polymerization of guanosine 5′-phosphor-2-methylimidazolides on a polyC template is efficient, resulting in extensions up to 50 nt (Inoue & Orgel, Reference Inoue and Orgel1982). Nevertheless, guanosine presents the best-case scenario, by combining the two traits of three Watson-Crick hydrogen bonds and a purine ring system, leading to favourable stacking interactions.

The analogous polymerization reactions with the three other nucleobases are much less efficient and particularly poor for uridine. Activated ribonucleotides can react with higher efficiency when aided by montmorillonite clay catalysts (Ferris et al. Reference Ferris, Hill, Liu and Orgel1996), more reactive leaving groups such as 1-methyladenine (Huang & Ferris, Reference Huang and Ferris2006) or oxyazabenzotriazolide (Deck et al. Reference Deck, Jauker and Richert2011). More substantial boosts come from tuning the substrate milieu, for example by removing inhibitory hydrolysed monomers by repeated substrate exchange (Deck et al. Reference Deck, Jauker and Richert2011) or through promoting monomer binding by stacking with short downstream ‘helper’ oligomers (Fig. 7), which recently resulted in the synthesis of an active strand of the HHR (Prywes et al. Reference Prywes, Blain, Del Frate and Szostak2016a).

Interactions between leaving groups can substantially alter template-binding affinity (Kervio et al. Reference Kervio, Sosson and Richert2016) and polymerization efficiency of nucleotides for example though the local creation of highly reactive intermediates (Walton & Szostak, Reference Walton and Szostak2016). The latter strategy relies upon imidazolium-bridged dinucleotide intermediates between adjacent imidazole-activated nucleotide monomer substrates in non-enzymatic templated primer extension and thus may be specific to this activation chemistry. Replication efficiency can also be increased by altering the chemistry of the monomer building blocks, e.g. by replacing the 2′- (or 3′) -hydroxyl with the more potent NH₂-nucleophile, or UTP with the stronger stacking analogue 5-propargyl-UTP. However, this generates nucleic acids with unnatural chemistries, and with the drawback of a reduced replication fidelity (Zhang et al. Reference Zhang, Zhang, Blain and Szostak2013).

Altered template chemistries that pre-organize conformation to RNA-like C3′-endo conformation such as HNA and Alitrol-nucleic acids (AtNA) render non-enzymatic RNA polymerization more efficient than on RNA templates, but their replication would be problematic as HNA- and AtNA-phosphorimidazolides are inefficient substrates for polymerization on RNA templates despite highly stable duplex formation (Kozlov et al. Reference Kozlov, De Bouvere, Van Aerschot, Herdewijn and Orgel1999a, Reference Kozlov, Politis, Van Aerschot, Busson, Herdewijn and Orgelb, Reference Kozlov, Zielinski, Allart, Kerremans, Van Aerschot, Busson, Herdewijn and Orgel2000). Fidelity of non-enzymatic replication remains one of the main hurdles, though misincorporations may be depleted in the final products as they lead to stalling of extension and non-templated addition (Leu et al. Reference Leu, Kervio, Obermayer, Turk-Macleod, Yuan, Luevano, Chen, Gerland, Richert and Chen2013). Some altered nucleotides can improve fidelity, as is the case for 2-thioU (or 2-thio-T), which due to the steric bulk of the C2 sulphur atom have a much reduced tendency to form G·U wobble pairs both in non-enzymatic RNA synthesis (Heuberger et al. Reference Heuberger, Pal, Del Frate, Topkar and Szostak2015) as well as in single nucleotide incorporations by the b1–233t polymerase ribozyme (Prywes et al. Reference Prywes, Michaels, Pal, Oh and Szostak2016b). Unfortunately, the resulting minor groove modification by the C2 sulphur atom can impact upon downstream synthesis activity by polymerase ribozymes (Attwater et al. Reference Attwater, Tagami, Kimoto, Butler, Kool, Wengel, Herdewijn, Hirao and Holliger2013a). The above described advances in non-enzymatic polymerization starting from the highly activated phosphorimidazolide nucleotides in some cases begin to reach an efficiency (and fidelity) compatible with the templated synthesis and replication of simple ribozymes, therefore closing the conceptual gap between pools of short oligomers created by prebiotic chemistry and the more complex ribozymes thought to have established the RNA world.

Non-templated polymerization of nucleotides activated by the prebiotically more plausible 2′,3′-cyclic phosphate chemistry (>p) tends to generate RNA polymers comprising a substantial fraction of non-canonical 2′–5-linkages (Verlander et al. Reference Verlander, Lohrmann and Orgel1973). These linkages also predominate when using 5′-activated nucleotides due to the higher reactivity of the 2′- versus the 3′-hydroxyl group. Non-canonical 2′−5′-linkages are highly destabilizing to canonical 3′–5′ linked RNA helical structure (Sheng et al. Reference Sheng, Li, Engelhart, Gan, Wang and Szostak2014) due to a reduction in both Watson–Crick base-pairing and base-stacking due to a lateral displacement of the base from the helical base-stack and a preference for non-canonical C-2′-endo puckering (Li & Szostak, Reference Li and Szostak2014; Premraj & Yathindra, Reference Premraj and Yathindra1998; Sheng et al. Reference Sheng, Li, Engelhart, Gan, Wang and Szostak2014). Nevertheless, even fully 2′–5′ linked RNA is able to form specific duplexes with complementary 3′–5′ RNA and (although weaker) with complementary 2′–5′ RNA (Wasner et al. Reference Wasner, Arion, Borkow, Noronha, Uddin, Parniak and Damha1998). A modest percentage (<25%) of such 2′–5′ linkages are even compatible with ribozyme function (Engelhart et al. Reference Engelhart, Powner and Szostak2013) and, due to their lower stability to hydrolysis, might over time become depleted in RNA duplex structures; thus, sporadic 2′–5′ linkages have been suggested to reduce product inhibition and aid primordial RNA replication and evolution by transient duplex destabilization (Engelhart et al. Reference Engelhart, Powner and Szostak2013) at least at low substitution levels. However, due to the ability of 2′–5′ linked RNA strands to self-hybridize and form stable helices (although less stable than 3′–5′ RNA), as well as the altered structural and conformational parameters of 2′–5′ RNA, the possibility that a 2′–5′ RNA sequence space might also contain ligands and catalysts cannot be discounted. Engineering of RNA polymerases capable of synthesizing 2′–5′ linked RNA (or DNA) might allow the exploration of such a sequence space and a testing of this hypothesis (Cozens et al. Reference Cozens, Mutschler, Nelson, Houlihan, Taylor and Holliger2015). Nevertheless, it seems unlikely that canonical 3′–5′ RNA catalysts or ligands could emerge from pools of wholly non-canonical 2′–5′ RNA. Nevertheless, a step-wise transition from a mixed population of 3′–5′/2′–5′ RNA to predominantly and wholly 3′–5′ RNA seems more plausible than a wholesale polymer take-over as postulated for a pre-RNA (or protoRNA) world scenario (see above).

4.3 Ribozyme ligases

While non-enzymatic polymerization provides potential avenues for the generation of pools of short RNA oligomers from prebiotic precursor molecules, it is currently unclear, how the longer RNA oligomers likely needed to encode informational functions such as catalysis of ligation or recombination reactions could have emerged from such pools. It is also unknown how frequent such functional sequences are within the RNA sequence space. Indeed, in vitro selection experiments suggest that functional sequences are extremely rare (Szostak, Reference Szostak2003) although some very small RNAs can display catalytic function such as the aminoacylating 5 nt ribozyme (Turk et al. Reference Turk, Illangasekare and Yarus2011). Furthermore, larger ribozymes such as the hairpin ribozyme (Vlassov et al. Reference Vlassov, Johnston, Landweber and Kazakov2004) and a triphosphorylation ribozyme (Akoopie & Muller, Reference Akoopie and Muller2016) can retain function and near wild-type catalytic rates when fragmented into 20–30 nt pieces, which are within the size range accessible from prebiotic chemistry and non-enzymatic replication. Thus, simple ribozymes, may be able to emerge from pools of short oligomers either directly or by non-covalent assembly into functional units and this might allow the bootstrapping of oligomer pools towards the higher compositional and functional complexity needed for self-replication.

So far, enzymatic templated RNA synthesis from mononucleotides appears likely to require quite large catalytic RNAs. This is supported both by theoretical considerations, which suggest a sharp drop off of stable secondary structures (most likely required to form stable active sites) below 30 nts (Briones et al. Reference Briones, Stich and Manrubia2009) and in vitro evolution experiments aimed at generating ribozymes capable of self-replication. RNA catalysts capable of iterative and template assembly reactions with ligase, recombinase and/or polymerase activity isolated from nature or by in vitro evolution are all substantially larger than 20–30 nts. One of the most striking systems is based on two variants of the R3C RNA ligase ribozyme (Lincoln & Joyce, Reference Lincoln and Joyce2009). These are capable of cross-catalytic self-ligation (see below).

Split variants of the Azoarcus SSI can also self-assemble into both covalent and non-covalent active complexes and can form cross-catalytic assembly networks (Hayden & Lehman, Reference Hayden and Lehman2006). Furthermore, both the sunY SSI and a cross-chiral RNA ligase generated by in vitro evolution can assemble their complement/mirror chirality sequences from activated oligonucleotides, but require a preformed template strand (Doudna et al. Reference Doudna, Couture and Szostak1991; Sczepanski & Joyce, Reference Sczepanski and Joyce2014). Finally, RPRs based on the R18 polymerase ribozyme (Johnston et al. Reference Johnston, Unrau, Lawrence, Glasner and Bartel2001) (itself derived from the class I ligase ribozyme) (Bartel & Szostak, Reference Bartel and Szostak1993) are capable of templated synthesis using NTPs as substrates, and some improved variants are able to synthesize other ribozymes, aptamers, tRNAs (Horning & Joyce, Reference Horning and Joyce2016; Wochner et al. Reference Wochner, Attwater, Coulson and Holliger2011) or RNA oligomers exceeding their own size on favourable template sequences (Attwater et al. Reference Attwater, Wochner and Holliger2013b). Therefore, there remains a compositional gap between the short RNA oligomer pools and the larger, phenotypically complex ribozymes likely to be required for self-replication, although recent experiments suggest that catalytic cooperation between small ligase and fragmented polymerase ribozymes might be able to close this gap (Mutschler et al. Reference Mutschler, Wochner and Holliger2015).

However, even these complex ribozymes are (currently) not capable of self-replication. One might therefore ask, if self-replication can be implemented by using RNA components alone as postulated in the original (strong) RNA world hypothesis (Neveu et al. Reference Neveu, Kim and Benner2013) and if not, what further functions might be required to realize RNA self-replication. The dramatic demonstration of cross-catalytic RNA self-assembly by Lincoln and Joyce provides an efficient RNA replication system (Lincoln & Joyce, Reference Lincoln and Joyce2009). Starting from two variants of the evolved R3C ligase ribozyme that were engineered to operate in a cross-catalytic format, each ribozyme variant catalysed the formation of the other by ligating two oligonucleotide substrates together. Thus, given a supply of the four component RNAs, an initial catalytic spike of ligase initiated exponential self-assembly.

This quasibiological growth behaviour in a simple and elegant molecular system might be leveraged to assemble other synthetic system components – but can it evolve? Ligase assembly requires pre-defined oligomer substrates with substantial homology to the ribozyme core that can only be supplied externally and this constrains the ability of this system to explore sequence space. Indeed, when substrates with variation in pairing sites are supplied, new ligase variants with better pairing dynamics for exponential amplification can emerge (Lincoln & Joyce, Reference Lincoln and Joyce2009), but the information transmission and hence adaptation can only occur through direct substrate hybridization at these specific loci, and is thus constrained to these small parts of the ribozyme. Other parts of the substrate – including the future catalytic site – are not interrogated during assembly, and if random sequences were supplied, only a negligible fraction of ligatable substrates would yield ligase activity. An elegant split-and-pool substrate synthesis scheme forcing catalytic and recognition regions to co-vary can restore some selection for activity (Sczepanski & Joyce, Reference Sczepanski and Joyce2012), but the evolutionary scope of the system remains constrained. Fundamentally, emergence of new functions when assembling long sequences is confounded by the nature of such activities: ligases use less information to choose substrates than is required to define the ligase activity itself, so cannot copy themselves (or other components) from sequences lacking that information, i.e. random sequence. Unconstrained evolution is likely to require more complete information transfer between generations, i.e. encoded RNA from smaller oligonucleotide or mononucleotide building blocks using informationally-complete complementary RNA templates.

4.4 RNA polymerase ribozymes

The emergence of replicases in the RNA world cannot be addressed without understanding mechanisms of non-enzymatic replication. Prior to the emergence of a replicase, non-enzymatic replication would have amplified not just individual sequences but diverse nucleic acid pools. Initially such pools of sequences would evolve to maximize their own abilities as templates (Chen & Nowak, Reference Chen and Nowak2012), priming sequence space with sequences (together with their complements) that would likely be amenable to enzymatic replication. Any RNA sequence then able to fold up and catalyse the pre-existing replication process would access new dimensions of selective advantage, without necessarily having to invent a new replication mechanism.

RNA polymerization need not be limited to monomer-building blocks; natural recombinase ribozymes have been harnessed to link together short oligomers in a templated manner extending down to trimers, although with rather low accuracy (Doudna et al. Reference Doudna, Usman and Szostak1993). Similar approaches have also been explored for unnatural nucleic acids like glycerol nucleic acids (GNA) (non-enzymatic template-dependent polymerization of apGNA-dinucleotides (Chen et al. Reference Chen, Cai and Szostak2009) and PNA tetra- and penta-oligomers (Brudno et al. Reference Brudno, Birnbaum, Kleiner and Liu2010), where monomer hybridization is weak. However, all oligomer assembly strategies face a challenge in that, while oligomers are easier to assemble than monomers and require fewer catalytic steps (for a given sequence), energetic differences in template binding between cognate and non-cognate substrates rapidly diminish in significance with increasing oligomer lengths thus limiting fidelity.

For this reason and due to the analogies with extant polymerases, achieving RNA-catalysed templated RNA synthesis from mononucleotide building blocks has been a goal ever since the discovery of the first catalytic RNAs. The recombinase activity of group I introns can be leveraged to assemble functional RNAs on RNA templates (Doudna & Szostak, Reference Doudna and Szostak1989; Green & Szostak, Reference Green and Szostak1992), but the active sites of these natural ribozymes were poorly suited to controlling the identity of the synthesized sequences (Bartel et al. Reference Bartel, Doudna, Usman and Szostak1991; Doudna et al. Reference Doudna, Usman and Szostak1993).

New active sites were needed, and a pioneering in vitro selection experiment (Bartel & Szostak, Reference Bartel and Szostak1993) unearthed these de novo from pools of random RNA sequences by selecting for the ability to seal a nick in an RNA duplex from 5′-triphosphate and 2′,3′-diol. Among an array of novel ribozyme ligases recovered was the class I ligase, which achieved ligation forming the canonical 3′–5′ linkage. An optimized version of the class I ligase exhibited a remarkable k _cat of 100 min⁻¹, still the fastest all-RNA catalyst described. An engineered version of the class I ligase could polymerize a limited number of nucleoside triphosphates (NTPs) on a constrained template (Ekland & Bartel, Reference Ekland and Bartel1996). Further development of this activity through a combination of in vitro evolution and RNA engineering opened up a path towards general ribozyme-catalysed templated RNA replication (Johnston et al. Reference Johnston, Unrau, Lawrence, Glasner and Bartel2001), and resulted in the first true polymerase ribozyme (R18) able to add up to 14 nucleotides on a separate primer/template duplex.

R18 polymerase activity was improved by different evolutionary strategies by selecting for the synthesis of longer sequences (Wochner et al. Reference Wochner, Attwater, Coulson and Holliger2011; Zaher & Unrau, Reference Zaher and Unrau2007) (Fig. 8). In the course of these selections, Holliger and colleagues discovered a mode of template hybridization by the polymerase ribozyme via a cognate hexanucleotide motif, akin to the binding and recognition of mRNAs by the prokaryotic ribosome through interactions with the Shine-Dalgarno sequence. Such a mode of cognate RNA recognition may also suggest the potential for RNA kin recognition and selection in early RNA replication, which may have been able to promote phenotype–genotype linkage and keep replication parasites in check prior to effective forms of compartmentalization (see below).

Fig. 8. Ribozyme RNA polymerase (RPR) development. The in vitro selected class I ligase catalyses the regioselective formation of canonical 3′−5′-RNA linkages. The addition of an accessory domain at the 3′ end of the class I ligase generated the R18 RNA polymerase. Further in vitro selection experiments resulted in the B6·61, tC19Z, tC9Y and 24-3 ribozyme RNA polymerases; the latter three variants include a short tag sequence (ss19) at their 5′ end complementary to the 3′ end of the template sequence. Residues in red are indicating mutations in comparison with R18 for B6·61 and tC19Z or in comparison to tC19Z for tC9Y and 24-3.

Further evolutionary refinement (based on an in-ice evolution strategy) yielded the tC9Y polymerase ribozyme, which, on a favourable template sequence is able to synthesize RNAs >200 nts long, creating RNA polymers longer than itself (Attwater et al. Reference Attwater, Wochner and Holliger2013b). tC9Y demonstrates the potential synthetic power of ribozymes, but is currently restricted to favourable RNA template sequences; long extensions remain inefficient upon templates comprising challenging or structured sequences, including those encoding the ribozyme itself. Recently Horning & Joyce described a new polymerase ribozyme variant with improved sequence generality and efficiency, particularly on purine-rich templates, culminating in its ability to perform simple ‘Ribo-PCR’ reactions (Horning & Joyce, Reference Horning and Joyce2016). This shows the capability of RNA to catalyse exponential amplification at least of short sequences. The new polymerase ribozyme 24-3 (evolved in 24 rounds of in vitro selection from the R18-derived Z RPR as a starting point) also displays an increased ability to read through short template hairpin structures, although at the cost of reduced fidelity of 92%. The increased ability of 24-3 to cope with template secondary structures may be both due to increased speed and efficiency on a wider range of templates.

For RNA templates exhibiting more stable secondary structures alternative strategies may be needed or be helpful. These may include auxiliary factors such as helper strands or helicase ribozymes. However, although the evolution of auxiliary ribozymes like a RNA helicase ribozyme may be possible, it is likely to be challenging and such ribozymes would also need to be replicated, increasing the synthetic burden on the replicase. A more parsimonious approach may be to engineer/evolve a strand-displacement activity in the polymerase ribozyme akin to some proteinaceous polymerases by coupling the energy released from NTP incorporation to strand invasion. Alternatively, one may seek to define conditions or media that would promote a (partial) unfolding of template secondary structures while maintaining ribozyme structure.

Physicochemical cycles (Budin & Szostak, Reference Budin and Szostak2010) including thermal, pH, ionic strength as well as wet–dry and freeze–thaw cycles (Mutschler et al. Reference Mutschler, Wochner and Holliger2015) or episodic exposure to high concentrations of denaturants might be able to effect such unfolding – although both thermal and pH cycles harsh enough to disrupt RNA structures would also be likely to accelerate RNA degradation especially in the presence of divalent metal cations. It may be possible to lessen the destructive impact of necessary thermal and pH cycling by reducing the Mg²⁺ requirements of the polymerase ribozymes. Different denaturing cycles such as denaturants and heat or pH and freezing could also be combined in order to lessen the harshness of each individual treatment. Yet, another interesting approach involves the addition of molecular factors that selectively destabilize the duplex form of RNA (or stabilize ssRNA). Indeed, RiboPCR combines high concentrations (0.9 M) of tetrapropyl-ammonium chloride (TPA) to reduce RNA duplex stability with thermocycling (Horning & Joyce, Reference Horning and Joyce2016). In another approach, an arginine decapeptide (R10) (Jia et al. Reference Jia, Fahrenbach, Kamat, Adamala and Szostak2016) selectively binds to ssRNA upon denaturation of a RNA duplex and may aid RNA replication cycles by facilitating repriming. Finally, while it is not clear how severe a problem + and – strand cross-inhibition presents, a possible solution involved a cross-chiral ligase system, wherein a D-RNA ligase assembled its L-RNA equivalent on an L-template (and vice versa) (Sczepanski & Joyce, Reference Sczepanski and Joyce2014). As enzyme and substrate (i.e. replicase and replicase template) are of opposing chirality and thus cannot form complementary RNA duplexes, + strands of opposing chirality can be assembled from supplied oligonucleotides (although in any full replication scheme each chiral enzyme would still be exposed to its homochiral template).

A critical strategy towards self-replication by an RNA replicase involves fragmentation of the replicase template at the replication stage. Shorter template strands are not only more accessible to ribozyme-catalysed synthesis (or non-enzymatic replication) due to a lower tendency to contain secondary structure, but, if sufficiently short (i.e. <30 nt long), can be more easily separated into product and template strands after replication. While some simple ribozymes are able to self-assemble from RNA fragments in this size range (Akoopie & Muller, Reference Akoopie and Muller2016; Vlassov et al. Reference Vlassov, Johnston, Landweber and Kazakov2004), this does not appear to be generally the case, in particular for more complex ribozymes. Indeed, fragmentation and non-covalent assembly of the R18-derived RPR into multiple fragments dramatically reduces activity, and therefore the covalent assembly through a ligase (or recombinase) ribozyme would be required. Recently the assembly of the full-length polymerase ribozyme from seven fragments by an itself fragmented hairpin ligase ribozyme could be demonstrated. The assembly process was performed in the eutectic phase of water-ice in the absence of divalent metal ions and was driven by freeze–thaw cycles, which were found to increase assembly yields by an order of magnitude (Mutschler et al. Reference Mutschler, Wochner and Holliger2015).

5.1 Compartmentalization without membranes

Several forms of ‘membrane-less’ compartmentalization are conceivable and some may have played a role in the context of early evolution. Of particular interest are porous or layered minerals (e.g. clays such as montmorillonite), eutectic ice phases or porous rocks (Fig. 9). Montmorillonite clays and eutectic ice have furthermore been shown to promote both the formation of RNA oligomers from activated nucleotide-building blocks as well as vesicle assembly. It is conceivable that some of these were important in supporting pre-cellular RNA replication. Alternatively, porous rocks in combination with temperature gradients (such as might occur close to hydrothermal systems) have been shown to be able to promote extreme solute concentration (Baaske et al. Reference Baaske, Weinert, Duhr, Lemke, Russell and Braun2007) as well as drive DNA ligation and replication through thermophoresis (Kreysing et al. Reference Kreysing, Keil, Lanzmich and Braun2015). Thermophoretic systems are of particular interest as they promote the selective concentration of large molecules, i.e. longer RNA oligomers over shorter ones thus providing an unique way of overcoming the ‘tyranny of the shortest’ in replication. Such a size sorting mechanism could also provide some protection against the (generally) small replication parasites, even in the absence of complete compartmentalization.

Fig. 9. Possible modes of compartmentalization for RNA at the origin of life. (a) Compartmentalization could occur (a) in the eutectic phase of water-ice, (b) at the bottom of temperature convective pores, (c) inside micelles generated by water/oil emulsions or (d) inside protocells generated from lipid bilayers.

Formation of liquid–liquid demixing phases and/or coacervates with highly crowded and charged interiors, which occurs spontaneously at critical concentrations of small biologically relevant cations and anions has been shown to promote RNA catalysis (Jia et al. Reference Jia, Fahrenbach, Kamat, Adamala and Szostak2016; Strulson et al. Reference Strulson, Molden, Keating and Bevilacqua2012). Of particular interest are the interactions and the resulting membrane-free microdroplets formed between RNA and simple peptides due to molecular simplicity of the components and the prebiotic context. Indeed, the importance of these phase separation mechanisms is echoed in modern biology, where liquid–liquid demixing gives rise to membrane-free fluidic intracellular compartments rich in DNA, RNA and proteins that are molecularly distinct from the surrounding cytoplasm or nucleus. However, the effects of liquid–liquid demixing and compartment formation on preserving, activating or enhancing RNA activity are still poorly understood.

Another potentially attractive system for both reagent concentration and compartmentalization is the eutectic phase of water-ice. An eutectic phase is formed when aqueous solutions comprising ions, RNA or other solutes are cooled below their freezing point. As freezing proceeds, solutes are excluded from the growing ice crystals and concentrated in an interstitial brine: the eutectic phase. Eutectic phase formation also goes hand in hand with reduced water activity (i.e. dehydration), solute concentration (up to 200-fold) and temperature reduction all of which promote synthetic (over degradative) processes. Indeed, ice phases have been shown to promote some chemical reactions and the formation of RNA oligomers by non-enzymatic polymerization of activated nucleotides (Monnard & Szostak, Reference Monnard and Szostak2008; Monnard et al. Reference Monnard, Kanavarioti and Deamer2003). Eutectic ice phases have also been found to stabilize RPR structure and activity (Attwater et al. Reference Attwater, Wochner, Pinheiro, Coulson and Holliger2010) and enable RPR evolution and adaptation (Attwater et al. Reference Attwater, Wochner and Holliger2013b). In addition, freeze–thaw cycles have been shown to act akin to modern-day RNA chaperones in promoting refolding of kinetically trapped misfolded RNAs to allow assembly of a complex polymerase ribozyme from small fragments (Mutschler et al. Reference Mutschler, Wochner and Holliger2015).

Although not widely considered as likely forms of prebiotic compartmentalization, emulsions provide an efficient model system to explore the linkage of genotype and phenotype (Fig. 9). Emulsions are formed from mixtures of immiscible liquid phases (e.g. an aqueous and a hydrocarbon oil phase), leading to the dispersion of one of the phases in the other as droplets of microscopic size. Although thermodynamically unstable, emulsion phases can be kinetically stable and persist for long periods of time (even at high termperatures) if stabilized by surfactants.

Of particular interest are water-in-oil (W/O) emulsions, in which the disperse phase forms a suspension of, aqueous cell-like droplets within an inert oil phase. W/O emulsions are experimentally easily tractable model compartments, and have been used for exploring the evolutionary behaviour of model systems of self-replication such in polymerase evolution approaches (Ghadessy et al. Reference Ghadessy, Ong and Holliger2001) and to explore the evolutionary impact of compartmentalization in the Qβ replication system; indeed, the Qβ replicase phenotype can only outlast fast-replicating parasites when replication is compartmentalized within the compartments of a W/O emulsion (Ichihashi et al. Reference Ichihashi, Usui, Kazuta, Sunami, Matsuura and Yomo2013).

5.2 Compartmentalization with membranes: protocells

Protocellular compartments formed from amphiphilic lipids assemble spontaneously under the right conditions (Fig. 9). These are of paramount importance because of their clear connection to extant biology. As with other forms of compartmentalization the confinement of macromolecules inside membrane-bound vesicles guarantees coupling between genotype and phenotype, while containing the spread of replication parasites. In addition, the physico-chemical properties of the fluid membranes may influence localization and organization of encapsulated polynucleotides and could alter both folding and higher order RNA functions such as RNA catalysis and replication. Membrane properties such as curvature and permeability to solutes as well as vesicle volume, growth and stability may itself be modified in turn by such interactions.

The past decade has seen detailed study of potential host vesicles formed from simple fatty acids (FA), which are moderately permeable, can grow and divide independently, support template non-enzymatic nucleic acid synthesis and maintain stability at high temperatures (Mansy & Szostak, Reference Mansy and Szostak2008; Mansy et al. Reference Mansy, Schrum, Krishnamurthy, Tobe, Treco and Szostak2008).

Yet, incompatibilities remain. FA vesicles have a low tolerance for the divalent cations needed by many ribozymes and required for non-enzymatic replication. Such ions, specifically Mg²⁺, cause FA membrane destabilization, leakage and ultimately FA precipitation. Potential solutions include adaptation of ribozymes to operate without such cations, the inclusion of chelators such as citrate to buffer free Mg²⁺ (Adamala & Szostak, Reference Adamala and Szostak2013) and the modification of membrane compositions to cope with divalent cations (Namani & Deamer, Reference Namani and Deamer2008). Furthermore, membranes are poorly permeable to some replicase substrates and highly charged species such as NTPs are unable to passively diffuse across such membranes. Potential solutions may be found by studying physicochemical cycling of protocells between a permeable and impermeable state (e.g. thermal or freeze–thaw cycles), inclusion of membrane permeability modifiers or the use simpler permeable building blocks that are activated inside the protocell (e.g. by a separate ribozyme such as a triphosphorylating ribozyme) (Moretti & Muller, Reference Moretti and Muller2014).

Finally, an enclosed dynamic system must contend with a build-up of potentially inhibitory replication products (pyrophosphate, misextended primers or degraded ribozyme fragments). Nuclease processing would enable clearing of monomers from the protocell by diffusion, but it may be more profitable to recycle such products. Mg²⁺-catalysed RNA degradation yields 2′,3′-cyclic phosphate termini, and these are potentially directly amenable to religation by the right catalyst, or through regioselective activation chemistry. As a result, degraded ribozymes as well as incomplete extension products could be fed back into synthesis. This would circumvent the need to synthesize full-length ribozymes faster than any backbone breaks occur, and therefore would only require individual ligation synthesis rates to outperform occurrence of backbone breaks, a far more favourable proposition. It might therefore be beneficial to endow protocells with a simple metabolism of substrate activation (Martin et al. Reference Martin, Unrau and Muller2015) or RNA repair and ligation. Indeed, metabolism need not be constrained to mimicking extant biology (Adamala & Szostak, Reference Adamala and Szostak2013; Rasmussen et al. Reference Rasmussen, Constantinescu and Svaneborg2016).