We set out to explore the potential of short RNA oligonucleotides as substrates for RNA-catalyzed RNA replication. To do this, we required a ribozyme capable of general, iterative RNA-templated oligonucleotide ligation. Previously-described RNA polymerase ribozymes such as the ‘Z’ RPR (Wochner et al., 2011) can use NTPs to iteratively extend a primer hybridized to an RNA template, but do not accommodate oligonucleotides bound downstream of the primer or accept them as substrates. However, we detected a weak templated ligation activity in a truncated version of the Z RPR comprising its catalytic core domain (Zcore) (Figure 1a), which supported incorporation of oligonucleotide substrates as short as three nt (Figure 1—figure supplement 1) when incubated in the eutectic phase of water ice (Attwater et al., 2010; Mutschler et al., 2015).

Figure 1 with 2 supplements see all

Download asset Open asset

Figure 1—source data 1

Selection conditions of rounds 1–7.

https://doi.org/10.7554/eLife.35255.006

Download elife-35255-fig1-data1-v2.xlsx

Figure 1—source data 2

Triplet concentration-dependent extension values.

https://doi.org/10.7554/eLife.35255.007

Download elife-35255-fig1-data2-v2.xlsx

To be able to properly examine such RNA trinucleotide triphosphates (triplets) as replication substrates, we first sought to convert Zcore into an effective triplet polymerase ribozyme using in vitro evolution. We devised a selection strategy that required iterative templated triplet ligation by ribozymes to achieve their covalent linkage to a tagged primer (Figure 1—figure supplement 2). This enables their recovery, amplification and mutagenesis before further rounds of selection to enrich the selection pool in improved triplet polymerase ribozyme variants.

We initiated selections from a library of 1.5 × 10¹⁵ Zcore variants with a new random 3’ N₃₀ region under eutectic phase conditions that increase RNA half-life and enhance ribozyme activity (Attwater et al., 2010;Attwater et al., 2013b). After 7 rounds of in-ice evolution, one-quarter of the selection pool comprised an improved ribozyme (type 0). Its core domain (0core, Figure 1a) could catalyse the iterative polymerization of multiple triplets allowing us to begin to investigate the properties of triplet-based RNA replication.

Significantly, we found that 0core could catalyze triplet polymerisation on a series of structured templates, which had proven intractable to the parental Z RPR (Figure 1b). Here, primer extension exhibited a steep sigmoidal dependence upon triplet concentrations (Figure 1c), suggestive of a cooperative invasion and unraveling of template secondary structures by the triplet substrates themselves. Although still inefficient, the fact that the nascent activity of the 0core ribozyme could already copy templates that had confounded an established RPR encouraged us to continue to seek improved triplet polymerase ribozymes to leverage this substrate behaviour.

We continued selections for a further 14 rounds. At this point, the type 0 ribozyme had gone extinct, replaced by six new types of RNA each characterised by a unique 3’ domain (Figure 2a, Figure 2—figure supplement 1). Type 1 RNAs were the most abundant, comprising ~50% of pool sequences, but mysteriously were catalytically inactive with diverse mutations in their core domains. In contrast, the type 2–6 RNAs all displayed triplet polymerase activity, but fell short of the polyclonal activity of the selection pool (Figure 2—figure supplement 2). To attempt to explain this discrepancy, we explored potential interactions among the different pool lineages, and found that addition of an equimolar amount of type 1 RNA substantially enhanced triplet polymerase activity of all the other ribozyme types 2–6 (Figure 2b).

Figure 2 with 2 supplements see all

Download asset Open asset

Figure 2—source data 1

Selection conditions of rounds 8–21.

https://doi.org/10.7554/eLife.35255.011

Download elife-35255-fig2-data1-v2.xlsx

Dissecting type 1 RNA function, we found that 5’ truncation of the region that previously contacted the primer/template duplex (Shechner et al., 2009) did not affect its cofactor activity (Figure 3a, Figure 3—figure supplement 1). As judged by gel mobility shift (Figure 3b) and activity enhancement (Figure 3—figure supplement 1), type 1 RNA appears to form a 1:1 heterodimeric complex directly with active triplet polymerase ribozymes. Our attention was drawn to their selection construct-derived 5’ hairpin elements, which differed between active triplet polymerases (‘cap+’, Figure 3a, Figure 2—figure supplement 1) and the most common type 1 variants in the selection pool where this hairpin had acquired a mutation (yielding ‘cap–’, Figure 3a). ‘cap–’ was dispensible for type 1’s cofactor activity, but when replacing ‘cap+’ in active triplet polymerases it abolished both their activity enhancement by type 1 (Figure 3—figure supplement 1) and complex formation (Figure 3b). This points to the ‘cap+’ hairpin as the critical site of interaction with type 1; ‘cap–’ in type 1 presumably served to deter its homodimerisation during selection.

Figure 3 with 2 supplements see all

Download asset Open asset

Figure 3—source data 1

Relative intensities of shifted bands when varying type 1 equivalents.

https://doi.org/10.7554/eLife.35255.015

Download elife-35255-fig3-data1-v2.xlsx

Indeed, transplanting the ‘cap+’ element could make the parental ribozymes (Zcore and Z RPR) receptive to activity enhancement by type 1 RNA (Figure 2b, Figure 3—figure supplement 2). The catalytically inert type 1 RNA thus represents a general, mutualistic RNA species. This molecular symbiont appears to have emerged spontaneously during in vitro evolution by forming a heterodimeric holoenzyme with triplet polymerase ribozymes, enhancing their activity to boost the recovery prospects of both complex components.

In complex with type 5 (the fastest enriching triplet polymerase ribozyme in the final selection pool), type 1 boosts polymerization of triplets (or longer oligonucleotides) to enable synthesis of long RNAs (Figure 3c). Here, it became apparent that type 1 also obviates the need for ribozyme-template tethering. Due to their poor affinity for primer/template duplex (Lawrence and Bartel, 2003), RPRs generally depend upon such tethering to template (Attwater et al., 2010; Wochner et al., 2011; Horning and Joyce, 2016), which enhances local ribozyme concentration and promotes formation of the RPR-primer/template holoenzyme (Attwater et al., 2010; Attwater et al., 2013a). In contrast, the triplet polymerase heterodimer appears to have a capacity for true intermolecular, sequence-general interaction with primer-template duplexes, which enables holoenzyme formation and copying of RNA templates without requiring specific ribozyme-template hybridization sites.

We performed an additional five rounds of in vitro evolution to further evolve the type 5 triplet polymerase ribozyme (now in the presence of truncated type 1 RNA), diversifying the previously-fixed 3’ domain reverse transcription primer binding sequence. This reselection yielded a shorter final heterodimeric triplet polymerase holoenzyme, hereafter termed ‘t5⁺¹’ (Figure 4). This robust triplet polymerase activity now proved suitable for exploring the scope and potential of triplet-based RNA replication.

Figure 4 with 1 supplement see all

Download asset Open asset

Figure 4—source data 1

Selection conditions of rounds 1–5 of the reselection.

https://doi.org/10.7554/eLife.35255.017

Download elife-35255-fig4-data1-v2.xlsx

As a first examination of t5⁺¹ activity, we revisited triplet-based RNA synthesis on structured templates. To provide a stringent test of template structure inhibition, we now examined hairpin-containing templates (4S, 6S, 8S) with increasing RNA hairpin stability and estimated T_Ms of up to 93˚C (8S). The latter had previously strongly arrested even the most advanced mononucleotide RPRs at higher temperatures (Horning and Joyce, 2016). However, using triplets as substrates t5⁺¹ robustly copied all of these (Figure 5a), even when templates were pre-folded allowing RNA secondary structures to form prior to triplet addition (Figure 5—figure supplement 1). The triplet concentration-dependent cooperative structure invasion and unraveling (previously observed with the simpler 0core domain and partly wobble-paired RNA template structures [Figure 1b,c]) was recapitulated with t5⁺¹ and the highly stable 8S hairpin template (Figure 5—figure supplement 2). In contrast, dinucleotide triphosphate substrates yielded extension only up to the structured region (Figure 5—figure supplement 2).

Figure 5 with 2 supplements see all

Download asset Open asset

We began to explore whether triplet-based RNA synthesis by t5⁺¹ might exhibit the generality required not just for synthesis of arbitrary structured sequences, but for replication of functional sequences (requiring synthesis of both ‘+’ and ‘−’ strands). Encouragingly, t5⁺¹ could synthesise both a functional fluorescent ‘+’ strand of the 52 nt Broccoli RNA aptamer (Filonov et al., 2014) and its encoding ‘−’ strand template from their 13 (+) and 12 (−) different constitutive triplets (Figure 5b).

We next turned to the critical test of generality: could triplet substrates allow self-synthesis? As t5⁺¹ currently lacks the efficiency to synthesise RNAs its own length, we divided the catalytic t5 ribozyme into five segments α, β, γ, δ and ε. This segmentation strategy (akin to that used by some RNA viruses e.g. influenza) could reduce tertiary structures (Doudna et al., 1991; Mutschler et al., 2015) and ease product separation during RNA replication (Szostak, 2012). Starting from ~8 nt RNA primers, t5⁺¹ achieved synthesis of the β⁺, γ⁺, and δ⁺ segments from their constitutive triplets as well as all of the ‘−’ strand segments α^-, β^-, γ^-, δ^- and ε^-, but required some triplets pre-linked (as e.g. hexa- or nonanucleotides) for synthesis of full-length α⁺ and ε⁺ segments (Figure 6a).

Figure 6 with 5 supplements see all

Download asset Open asset

Operating across 70 distinct ligation junctions in these reactions including AU-rich sequences, t5⁺¹demonstrates the sequence generality for self-synthesis using triplet substrates. Notably, the average extent of ligation per junction during synthesis of t5 ‘+’ and ‘−’ strands (78%) was similar to that observed when t5⁺¹ used an unstructured model template (74%, Figure 3c) upon which the parental Z and other RPRs excel (Attwater et al., 2013b; Horning and Joyce, 2016).

At this point, we tested whether the broad oligonucleotide ligation capacity of t5⁺¹ (Figure 3c) might allow assembly of synthesised ‘+’ strand segments. Indeed, t5⁺¹ could assemble these into αβ⁺ and γδε⁺ fragments, guided only by partially overlapping ‘−’ strands (Figure 6b, Figure 6—figure supplement 1). Through non-covalent association (Vaish et al., 2003; Mutschler et al., 2015), the ribozyme-synthesised αβ⁺ and γδε⁺ fragments spontaneously reconstituted a new catalytically active triplet polymerase ribozyme (with in vitro transcribed type 1 RNA). We found that this synthesis product could regenerate fresh δ^- segment using t5⁺¹ ribozyme-synthesised δ⁺ (left over from ribozyme assembly) as a template (Figure 6c), recapitulating elements of a self-replication cycle. However, while the t5⁺¹ ribozyme displays a nascent capacity for templated synthesis of its own catalytic domain ‘+’ strands (and ‘−’ strands), efficiency of both segment synthesis and assembly will need to be increased significantly to realise a full self-replication cycle (which would also require synthesis and replication of the type 1 subunit).

Templated ‘+’ strand self-synthesis is a central element of ribozyme self-replication. However, a limitation of our above strategy in the context of triplet-based self-replication is the continued requirement for some pre-synthesized longer oligonucleotides to act as primers and occasional substrates (together providing here the equivalent of ~25% of triplet junctions pre-ligated). In particular, some specific oligonucleotide substrates were required for efficient synthesis of α+ and ε+ segments to compete out inhibitory mutual hybridisation between ‘−’ strand template and corresponding ‘+’ strand unstructured elements in the t5 ribozyme (Figure 6—figure supplement 2). In vitro selections that stabilise the ribozyme tertiary structure (Figure 6—figure supplement 3) may contribute to attenuating this requirement. Additionally, more concentrated triplet substrates can successfully compete with ribozyme unstructured elements for hybridization to ‘−’ strand templates (Figure 6—figure supplement 2).

The majority of specific oligonucleotides, however, were provided as primers to initiate syntheses, as required by all RPRs akin to the activity of replicative polymerases in biology. As a consequence of this, the priming sequence would effectively be excluded from evolution during replication. Furthermore, RNA oligonucleotides able to act as specific primers are unlikely to be prevalent in prebiotic substrate pools, and their depletion during successive replication cycles could lead to loss of sequence at genome ends. This ‘primer problem’ has previously been noted in the context of nonenzymatic replication (Szostak, 2012) as one of the fundamental obstacles to RNA self-replication.

Unexpectedly, triplet substrates provide a route to bypass the ‘primer problem’. We observed that t5⁺¹ can extend primers bidirectionally, in both the canonical 5’−3’ as well as the reverse 3’−5’ directions (Figure 7a). This allows not only completion of RNA synthesis from either template end but also initiation from anywhere along a template, potentially allowing non-classical hierarchical or distributive RNA replication schemes as previously proposed (Szostak, 2011). Given this flexibility, we wondered if t5⁺¹ even had a requirement for a primer oligonucleotide. Indeed, this triplet polymerase could achieve ‘primer free’ RNA synthesis (whereby synthesis is presumably initiated by ligation of adjacent triplets anywhere on the template), as exemplified here for the β⁺ segment (Figure 7b), as well as ‘primer free’ RNA replication as shown for the ‘+’ and ‘−’ strands of the γ segment, which can be replicated using triplets alone (Figure 7c).

Figure 7

Download asset Open asset

Thus, the capacity of triplet substrates to pre-organise themselves on a template not only enables replication of structured templates but also allows complete copying of some RNA sequences exclusively from triplet building blocks, suggesting an alternative to the canonical end-primed replication strategies inspired by PCR. Such a ribozyme operating in a more distributive polymerisation mode might be able to replicate RNA sequences directly from the putative pools of short random RNA oligonucleotides furnished by prebiotic chemistry.

Next, we investigated the consequences of using analogues of such prebiotic pools as a source of substrates for the t5⁺¹ triplet polymerase ribozyme. Random sequence triplet pools (‘^pppNNN’, comprising equimolar amounts of all 64 triplets) could be used as substrates by t5⁺¹ in segment syntheses in place of defined triplet sets (Figure 6—figure supplement 4). Furthermore, extension activity remained robust upon pool supplementation with noncanonical dinucleotide and mononucleotide substrates (Figure 6—figure supplement 5).

However, a replicase must incorporate the correct template-complementary substrate from random sequence pools, or genetic information may become irretrievably corrupted during replication (Eigen, 1971). Sequence fidelity is therefore a critical parameter of RNA replication. The fidelity challenge is exacerbated in triplet-based RNA replication by the need to discriminate between 64 distinct substrates; indeed, a previous investigation into the incorporation of individual trinucleotides indicated that misincorporations could outstrip cognate incorporation for some triplets (Doudna et al., 1993).

In order to assess the fidelity of triplet polymerase ribozymes of widely differing activity, we identified the triplets incorporated from random ^pppNNN triplet pools using 12 different compositionally representative N′N′N′ triplet sequences as templates. These were examined in a consistent sequence context (5’-GGG-N′N′N′-GGG-3’) and collated, which allowed an estimation of ribozyme misincorporation tendencies. On average, the starting Zcore ribozyme exhibited ~91% fidelity per position (Figure 8a), lower than that described for RPRs (92% – 97% [Attwater et al., 2013b; Horning and Joyce, 2016]). Furthermore, its accuracy exhibited a pronounced downward gradient from the first (5’) to the third (3’) triplet position, highlighting escalating risks to fidelity of synthesis founded on longer building blocks.

Figure 8 with 3 supplements see all

Download asset Open asset

Figure 8—source data 1

Analysis of collated errors by ribozymes in the fidelity assay.

https://doi.org/10.7554/eLife.35255.033

Download elife-35255-fig8-data1-v2.xlsx

Figure 8—source data 2

Calculation of residual fidelity phenotypes in Figure 8c.

https://doi.org/10.7554/eLife.35255.034

Download elife-35255-fig8-data2-v2.xlsx

To investigate if ribozymes could exhibit higher triplet incorporation fidelity, we had included a persistent adaptive pressure for fidelity during in vitro evolution, spiking in an excess of mispairing 3’-deoxy ‘terminator’ triplets from round nine onwards, precluding recovery of ribozymes that incorporated these mispairs (Figure 2—source data 1, Figure 4—source data 1, Figure 2—figure supplement 2). This yielded reshaped and improved fidelity profiles in the ‘surviving’ type 2–6 ribozymes (Figure 8a). Notably, the final t5⁺¹ ribozyme achieves an average positional fidelity of 97.4% using ^pppNNN in this sequence context, higher than the best RPR fidelity with NTPs under comparable eutectic conditions (Attwater et al., 2013b). Deep sequencing of internal triplet positions of a defined sequence (β⁺ segment) synthesised by t5⁺¹ using ^pppNNN indicated similar aggregate fidelity could be achieved during longer product synthesis excluding the final triplet (Table 1).

Having established that accurate triplet-based copying is possible (in at least some sequence contexts), we sought to understand how the triplet polymerase ribozyme achieves it. Investigating the fidelity contributions of different t5⁺¹ ribozyme components, we found that the type 1 RNA cofactor did not contribute; rather, fidelity gains appeared to be mediated by the newly-evolved t5 ‘ε’ 3’-domain, as its deletion (yielding the truncated ‘αβγδ’ ribozyme) reverted the fidelity profile towards that of Zcore (Figure 8—figure supplement 1). Presence of the ε domain did not uniformly increase fidelity, but selectively reduced the most acute errors at the second and third triplet positions (with over 10-fold reductions for some errors, Figure 8b, Figure 8—figure supplement 2). Overall error rates at the second and third triplet positions were reduced by 4-fold and 9-fold compared to Zcore (Figure 8a), though increased (1.3-fold) at the first triplet position due to a localised asymmetric tolerance of G:U wobble pairing (Figure 8b). The ε domain fidelity function is contingent upon the presence of a downstream triplet, operating only with basal fidelity for final triplet incorporation (Figure 8—figure supplement 3).

Dissecting the molecular determinants of the fidelity phenotype, we found that using triplet substrates modified at the third position with a 2-thiouracil in place of a uracil (disrupting minor groove hydrogen bonding capabilities) rendered the ε fidelity domain unable to discriminate mismatches (Figure 8c, Figure 8—figure supplement 3). Previously, a similar replacement of a uracil 2-keto group with a 2-thio modification had been shown to impair Z RPR activity when present upstream in the primer/template region (Attwater et al., 2013a), where Z is thought to rely upon sequence-general minor groove contacts through an ‘A-minor’ motif (Shechner et al., 2009). Modification at the third triplet position reverts ε’s divergent effects on fidelity at the adjacent second and the distal first triplet positions (Figure 8—figure supplement 3); disruption of this minor groove contact site thus abolishes overall ε fidelity domain operation. ε sensitivity to minor groove composition may be critical to its recognition of cognate Watson-Crick base pairs, reminiscent of Tetrahymena group I intron folding (Battle and Doudna, 2002) and the decoding centre of the ribosome (which also tolerates wobble pairing at the analogous (5’) triplet position) (Ogle et al., 2001).

An important contribution to triplet fidelity also appears to arise from unexpected behaviours of the triplet substrates themselves. We observed that in some direct pair-mispair triplet contests, inclusion of their complementary triplets caused a striking (~3 fold) drop in misincorporation errors (Figure 9). A potential explanation may arise from differential formation of triplet:anti-triplet dimers in the reaction: for example, more extensive ^pppGCC:^pppGGC (than ^pppACC:^pppGGU) dimer formation would selectively reduce the effective concentration of free ^pppGCC vs. ^pppACC upon inclusion of their complementary ^pppGGC and ^pppGGU.

Figure 9 with 1 supplement see all

Download asset Open asset

Figure 9—source data 1

Collated error rates and ratios at different substrate concentrations.

https://doi.org/10.7554/eLife.35255.038

Download elife-35255-fig9-data1-v2.xlsx

These pairwise reductions were recapitulated in the presence of random ^pppNNN substrate pools (Figure 9). Indeed, counterintuitively, raising ^pppNNN concentrations from 0.5 to 5 μM each almost halved the overall error rate (Figure 9—figure supplement 1). Although diverse effects upon individual misincorporations were observed, this fidelity enhancement was driven by pronounced reductions in errors where the mismatched triplet has a high GC content compared to the cognate triplet, including common G-U wobble mispairs (Figure 9—figure supplement 1). Dimer formation among ^pppNNN substrate pools would be expected to selectively buffer the free concentrations of the more strongly-pairing GC-rich triplets, which could promote both fidelity and sequence generality through normalization of triplet availability against template (and complementary triplet) binding strength.

Indeed, more efficient, higher fidelity segment synthesis was observed when partially mimicking this outcome using an ^pppNNN pool formulated with a reduced G content (Figure 6—figure supplement 4, Table 1). In a prebiotic scenario, substrate pool composition would have been determined by the abundance and nontemplated polymerization tendencies of the different nucleotides; large biases in these could skew triplet compositions or deplete a triplet (resulting in mismatch incorporation). However, the potential for replication to proceed in different triplet registers may provide a degree of resilience towards such biases.

Standard ribozyme activity assays (modified where specified) comprise 5 pmol of each ribozyme annealed in 2.5 μl water (80˚C 2 min, 17˚C 10 min), with 2 μl of 1 M MgCl₂ and 0.5 μl of 1 M tris•HCl pH 8.3 (at 25˚C, pH raised to 9.2 at −7˚C) then added on ice, and left for >5 min to ensure folding. This was added to 5 pmol each of primer and template and 50 pmol of each triplet pre-annealed in 5 μl water, then frozen on dry ice (10 min) and incubated at −7˚C in a R4 series TC120 refrigerated cooling bath (Grant (Shepreth, UK)) to allow eutectic phase formation and reaction.

Final pre-freezing concentrations of components are displayed throughout (in this example, yielding 0.5 μM ribozyme/primer/template, 5 μM each triplet, 200 mM MgCl₂, 50 mM tris•HCl pH 8.3). Supercooled reactions (Figure 1c) remained liquid by omitting the dry-ice freezing step, maintaining these concentrations. Ice crystal formation upon eutectic phase equilibration, however, concentrates all solutes ~4–5 fold (Attwater et al., 2010) to their final operational levels and cooling elevates tris-buffered pH to ~9.2.

Some substrate mixes (e.g. ^pppNNN) led to a higher final reaction volume, but eutectic phase equilibration restored standard operational concentrations, also applicable to the four-fold-diluted extensions with the fragmented ribozyme (Figure 6c). These used 2 pmol each ribozyme/fragment annealed in 3.25 μl 62 mM MgCl₂, 15 mM tris•HCl pH 8.3 (37˚C 5 min, ramped to 4˚C at 0.1˚C/s, 4˚C 10 min), with pre-annealed primer/template/substrates (0.5/0.5/5 pmol) added in 0.75 μl water. These reactions, and preparative syntheses (Figure 6—figure supplement 1, Figure 7c), were supercooled at −7˚C followed by ice crystal addition for quick freezing and optimal activity.

Figure 1c extensions were set up by adding buffer, then RNAs (preannealed together, 0.1 μM final concentrations) to triplets. RNAs for ε⁺ syntheses were chilled on ice instead of annealing, with ribozyme/MgCl₂/tris•HCl pH 8.3 mixed with the other RNAs at −7˚C. Oligonucleotide substrates were added equimolar to template binding sites in the primer/template/substrate anneal. NTPs, on the other hand, were added with the MgCl₂/tris•HCl pH 8.3 to the ribozyme polymerase.

At the end of standard incubations, reactions were thawed and 2 μl aliquots added to stop buffer (1 μl 0.44 M EDTA (pH 7.4), with urea to a 6 M final concentration and a 10–20 fold molar excess over template of complementary competing oligonucleotide (see Supplementary file 3) to prevent long product/template reannealing). Samples were denatured (94˚C 5 min) and RNAs separated by 8 M urea 1 × TBE denaturing PAGE.

To avoid using potentially confounding competing oligonucleotide when purifying extension products, reactions with a biotinylated primer or template (stopped as above) could be purified by bead capture using MyOne C1 (Invitrogen) streptavidin-coated paramagnetic microbeads (using 5 μg pre-washed beads per pmol biotinylated RNA) in 0.5 × − 0.8 × bead buffer (BB: 200 mM NaCl, 10 mM tris•HCl pH 7.4 (at 25˚C), 1 mM EDTA, 0.1% Tween-20). After washing twice in BB to remove unbound components, beads were incubated (1 min) in 25 mM NaOH, 1 mM EDTA, 0.05% Tween-20 to denature the duplexes (Horning and Joyce, 2016). To recover biotinylated extension products (e.g. Figure 5b left panels, Figure 6—figure supplement 1 αβ⁺/γδε⁺, Figure 7c left panel) the supernatant was discarded, and beads were washed first in BB with 200 mM tris•HCl pH 7.4, then in BB, then heated (94˚C 4 min) in 95% formamide, 10 mM EDTA to release primers for urea-PAGE. To recover extension products bound to biotinylated templates (e.g. Figure 6—figure supplement 1 β⁺, δ⁺, ε⁺, Figure 7c right panel) the supernatant was removed, neutralized with 500 mM tris•HCl pH 7.4, spin-concentrated using Ultracel 3K filters (MerckMillipore, UK), recovered and denatured in 6M urea/10 mM EDTA before urea-PAGE. β⁺ synthesised in Figure 6—figure supplement 1 was not spin-concentrated, leading to a lower recovery yield; δε⁺ synthesis was denatured directly from the ligation reaction in 60% formamide with excess EDTA.

For gel mobility shift assays (Figure 3b), ribozymes were mixed at 0.5 μM, pre-annealed and buffer added on ice as for extension reactions, then mixed with 5 × loading buffer (50% glycerol, 250 mM tris•HCl pH 8.3, 125 mM MgCl₂) for separation by native PAGE (0.5 × TB, 8% 59:1 acrylamide:bisacrylamide, 25 mM MgCl₂, run in a Hoefer SE600 Chroma (ThermoFisher, Waltham, USA) (upper chamber: 0.5 × TB 50 mM NaOAc, lower chamber: 0.5 × TB 25 mM Mg(OAc)₂) kept at 4˚C in a circulator bath for 6–8 hr at 10 W), then SYBR Gold stained as below.

Fluorescent primer extension products were detected using the appropriate laser wavelength on a Typhoon Trio scanner (GE Healthcare (GE) (Chicago, USA)); gel densitometry allowed quantification of RNA synthesis efficiency. Gel contrasts in figures were linearly adjusted to optimize display of bands of differing intensities.

The gel in Figure 5b (middle panel) was washed thrice (5 min) in water, incubated with 10 μM DFHBI-1T ligand in buffer for 20 min to fold full-length broccoli aptamer (as in [Filonov et al., 2015]) and scanned. The ligand was then eluted in three 1 × TBE washes (leaving negligible background fluoresence), and stained in 1 × TBE with SYBR Gold (1:10000), washed again, and re-scanned to detect all RNA products (left panel); scans were aligned via an adjacent Cy5-labelled primer extension lane (not shown).

Full-length product yields in the Figure 6—figure supplement 1 plus-strand syntheses were calculated by running samples of bead-eluted products (or raw reaction for δε⁺) alongside known amounts of the positive controls indicated, followed by SYBR-Gold staining. To purify, bead-eluted products were run similarly, and excised using UV shadowing. Products were then eluted from the gel fragments in 10 mM tris•HCl pH 7.4, and Spin-X column filtrate (Costar (Sigma-Aldrich, UK)) precipitated in 75% ethanol with 1 μl 1% glycogen carrier (omitted for β⁺). Recovered full-length product yields were calculated similarly to reaction yields for αβ⁺/δε⁺/γδε⁺, or using A₂₆₀s for β⁺, δ⁺, ε⁺.

Purified ribozyme- and TGK-synthesized αβ⁺/γδε⁺ fragments were sequenced by first ligating a 3’ adaptor (10 U/μl T4 RNA Ligase 2 truncated KQ in 1 × RNA ligase buffer (New England Biolabs (NEB), (Ipswich, USA)) with 15% PEG-8000 and 2 μM AdeHDVLig at 10˚C overnight). These reactions were bound to MyOne C1 microbeads (ThermoFisher (Invitrogen)), washed with BB to remove unligated adaptor, and reverse transcribed (50˚C 30 min) with 1 μM HDVrec primer using Superscript III (Invitrogen). Beads were washed again then PCR amplified (five cycles with a 40˚C annealing step, then 20 cycles with a 50˚C annealing step) using GoTaq HotStart master mix (Promega (Madison, USA)) and 0.8 μM each of primers P3HDV, and P5Xα8 or P5Xγ7, for high-throughput sequencing (Illumina (San Diego, USA) MiSeq or HiSeq) after PCR product agarose gel purification. β⁺ syntheses’ cDNAs were amplified with P3HDV and P5Xβ6.

To estimate RNA synthesis fidelity, ribozymes extended primers using ^pppNNN on templates encoding CCC-XXX-CCC, where XXX were 12 different triplet sequences evenly exploring base composition and distribution (see Supplementary file 3; for XXX = ACC, template encodes CCC-ACC-UCC to avoid a terminal run of Gs).

Each primer/template pair (0.45/0.525 pmol per reaction) was annealed in 4 mM MgCl₂, 1 mM tris•HCl pH 7.4 (80˚C 2 min, ramped to 4˚C at 0.1˚C/s, then kept on ice). The 12 pairs were combined in 0.27M MgCl₂/67 mM tris•HCl pH 8.3 on ice to discourage primer-template assortment (of which sequencing later revealed negligible levels). 36 pmol of each triplet in ^pppNNN (equivalent to 5 μM final concentration after considering eutectic phase equilibration effects upon this more dilute reaction) were added to a reaction vessel in 10.8 μl water, to which 5.4 μl of the primer/template/buffer mix was added followed by 7.2 pmol of ribozyme pre-annealed (80˚C 2 min 17˚C 10 min, ice >5 min) in 1.8 μl water (f.c. equivalent 1 μM, in excess over the 0.875 μM template to which some ribozymes could tether to enhance extension). Reactions were frozen and incubated (7 days at −7˚C) as described above.

Reactions were stopped with 3.6 μl 0.44 M EDTA and 10.5 pmol of each template's competing oligonucleotide (migrating above product, with marker mutations to ensure exclusion), denatured with 6 M urea, and urea-PAGE separated. After alignment with a fluorescence scan of the gel, a region of the sample lane corresponding to primers extended by +4 to +14 nt was excised (encompassing 2–4 triplet additions), and extension products were eluted, precipitated in 77% ethanol with 1 μl 1% glycogen carrier, washed in 85% ethanol and resuspended in water.

These extension products were 3’ adaptor ligated as for fragment sequencing. Products were reverse transcribed (0.2 × adaptor ligation reaction, 1 μM HDVrec primer in Superscript III reaction, 50˚C 30 min) and then PCR-amplified (1/30^th reverse transcription mix, 0.8 μM each of primers P3HDV and P5GGGX) for sequencing as above (yielding 2 × 10⁵ – 4 × 10⁶ sequences per ribozyme assay).

After processing and 3’ adaptor trimming, sequences corresponding to primer extended by CCC +1–3 additional triplets were collated for analysis. Variations in upstream primer sequences (see Supplementary file 3) allowed the partner template to be identified for each sequenced product; the triplet incorporated after the first CCC was counted. Separately, 10 μl extensions by t5⁺¹ of each primer/template alone with its encoded triplet and ^pppCCC (and ^pppUCC for the ACC pair) were combined for purification and sequencing as above, to allow isolation of the ribozyme-mediated errors resulting from inclusion of the other 62 (61 for ACC) triplets in the reaction (versus errors from sequencing, recombination etc.). The counts of cognate triplet (C) and each error triplet (E) in the positive control (p) reduced error counts in the experimental samples (x) to yield ribozyme-mediated error counts (E_r) thusly: E_r = E_x - E_p*(C_x/C_p) (not reducing E_x below 0, and reallocating all reductions to C_r; ^pppCCC counts (and ^pppUCC for the ACC template) remained uncorrected).

For each template, counts were then collated at the first/second/third positions to yield base-specific mutation rates for each position (Figure 8—figure supplement 2, Figure 8—source data 1). Across the 12 triplets, A, C, G, and U were encoded at each position three times; linear averages were calculated to map the position’s error profile (Figure 8b) and geometric means of the four nucleobases yielded the position’s overall fidelity (Figure 8a, Figure 8—figure supplement 1).

Triplets (and some other short oligonucleotides) were prepared from NTPs by T7 RNA polymerase run-off transcription of a 5’ single-stranded DNA overhang downstream of a DNA duplex T7 promoter sequence. In most cases, the 5’ overhang encoded (was the reverse complement of) the desired oligonucleotide. These oligonucleotides were short enough to synthesise during the abortive initiation stage of transcription, attenuating sequence constraints on the first bases of the transcript. However, T7 RNA polymerase exhibited tendencies to skip the first (or even second) base (most severe for U > C > A > G before second position purines: encoding CGU yielded some ^pppGU, encoding UAC yielded just ^pppAC) or use oligonucleotides generated during transcription to re-initiate (e.g. encoding GAG yielded ^pppGAGAG, encoding AAA yielded ^pppA_6-9, encoding UCC yielded ^pppCCC, encoding CGC yielded some ^pppGCGC; this tendency was most severe when the oligonucleotide could be accommodated opposite the final template bases of the promoter).

These tendencies could be subverted by encoding additional first bases (usually without providing the corresponding NTP). This initiated the oligonucleotide at the second position where skipping tendencies were lower (e.g. encoding CUAG without CTP yielded ^pppUAG, encoding UUAC yielded some ^pppUAC), and reduced recruitment as initiators of products with bases not complementary to the introduced first position template base (e.g. encoding CGAG without CTP yielded ^pppGAG, encoding CAA without CTP yielded ^pppAA and ^pppAAA, encoding AUCC without ATP yielded ^pppUCC, encoding UCGC without UTP yielded ^pppCGC).

Each 30 μl transcription reaction contained 72 nmol of each desired product base as an NTP (Roche) (e.g. for ^pppUCC, 72 nmol UTP, 144 nmol CTP) in 1 × MegaShortScript kit buffer with 1.5 μl MegaShortScript T7 enzyme (ThermoFisher). Also present were 15 pmol of each DNA oligonucleotide forming the transcription duplex target (see Supplementary file 2). The reactions were incubated overnight at 37˚C, stopped with 3 μl 0.44 M EDTA and 17 μl 10 M urea, and separated by electrophoresis (35 W, 4.5 hr) on a 35 × 18 × 0.15 cm 30% 19:1 acrylamide:bis-acrylamide 3 M urea tris-borate gel. Products were identified through their relative migrations (reflecting overall composition, fastest to slowest: C > U≈A > G) by UV shadowing. Triplet bands were excised and eluted overnight in 10 mM tris•HCl pH 7.4, and filtrate (Spin-X) precipitated with 0.3 M sodium acetate pH 5.5 in 85% ethanol. Pellets were washed in 85% ethanol, resuspended in water, and UV absorbances measured with a Nanodrop ND-1000 spectrophotometer (ThermoFisher). Oligocalc (Kibbe, 2007) was used to calculate sequence-specific concentrations and yields. ^pppNNN was generated by combination of equal amounts of each of the 64 triplet stocks in a lo-bind microcentrifuge tube (Eppendorf (Hamburg, Germany)).

3’-deoxy triphosphorylated ‘terminator’ triplets were transcribed as above but using a 3’ deoxynucleoside 5’ triphosphate (Trilink biotechnologies) for the last position, migrating faster during PAGE than the equivalent all-RNA triplet. Triplets with 2-thiouridine residues were transcribed as for their corresponding U, replacing UTP with U^2STP (Jena Bioscience (Jena, Germany)); incorporation and migration were similar between the two, and their concentrations were calculated from A_260nm by comparison to the A_260nm of mixtures of the component ribonucleotides with UTP vs. U^2STP. Triplets with 2’-fluoro, 2’-deoxy positions could also be transcribed, with lower efficiency, by substituting the corresponding triphosphate (Trilink Biotechnologies). The biotinylated ^pppGAU–Bio triplet used in γ segment synthesis (Figure 7c) was transcribed as for ^pppGAU, replacing UTP with biotin-16-aminoallyluridine-5’-triphosphate (Trilink Biotechnologies (San Diego, USA)), quantified via by comparison to the A_290nm of mixtures of the component ribonucleotides.

Longer triphosphorylated oligonucleotides used in ribozyme self-synthesis were generated similarly, but using ~200 ng of fully double stranded DNA as a template. Candidate product bands were purified and the desired oligonucleotide identified by ribozyme-catalysed in-frame incorporation and, for some, fragment sequencing.

Transcriptions were performed on ~15 ng/μl dsDNA using MegaShortScript enzyme and buffer (ThermoFisher) with 7.8 mM of each NTP, or, to yield a 5’ monophosphate on the product to avoid aberrant ligation, 10 mM GMP (guanosine monophosphate) and 2 mM of each NTP (‘GMP transcription’).

dsDNA templates for some of these (in Supplementary file 3) were generated (‘fill-in’) using three cycles of mutual extension (GoTaq HotStart, Promega) between the associated DNA oligonucleotide and 5T7 (or, where indicated, HDVrt for defined 3’ terminus formation [Schürer et al., 2002]) followed by column purification (QiaQuick, Qiagen).

Some 5’ biotinylated RNAs were synthesized using the TGK polymerase (Cozens et al., 2012) (56 μg/ml, in 1 × Thermopol buffer (NEB) supplemented with 3 mM MgCl₂) to extend 5’ biotinylated RNA primers (0.75 μM) on DNA templates (1 μM) using 2.5 mM of each NTP (94˚C 30 s, 45˚C 2 min, 65˚C 30 min, 45˚C 2 min, 65˚C 30 min, then all repeated). Biotinylated products were bead-purified as above.

3’ biotinylation of RNAs was achieved in two stages: 3’ azidylation (at 2 μM with 25 U/μl yeast poly-A polymerase (ThermoFisher) and 0.5 mM 2’-azido-2’-deoxycytidine triphosphate (Trilink Biotechnologies) for 1 hr at 37˚C) with subsequent acidic phenol/chloroform extraction and 75% ethanol precipitation, then copper-catalysed biotin-(PEG)₄-alkyne (ThermoFisher) cycloaddition (Winz et al., 2012) with subsequent 75% ethanol precipitation followed by resuspension and buffer exchange in Ultracel 3K filters (Amicon) to remove residual biotin-alkyne.

Round one libraries were synthesised by mutual extension of 4 nmol of oligonucleotides 1baN30 and 1GMPfo or 1GTPfo at 1 μM each in 1 × isothermal amplification buffer (NEB) with 250 μM each dNTP, annealed (80˚C 3 min, 65˚C 5 min) before addition of 0.4 U/μl Bst 2.0 (NEB) and 30 min 65˚C incubation.

After purification, 375 μg of each DNA (~1.5 × 10¹⁵ molecules) were transcribed in 5 ml transcription reactions (36 mM tris•HCl pH 7.9 (at 25˚C), 1.8 mM spermidine, 9 mM DTT, 10.8 mM MgCl₂, 2 mM each NTP, 1% 10 × MegaShortScript buffer, 2% 1:9 MegaShortScript:NEB T7 RNA polymerase, 37˚C overnight). These were treated with DNase, acid phenol/chloroform extracted and 73% ethanol precipitated prior to urea-PAGE purification, elution, filtering (Spin-X) and re-precipitation, yielding the 1GTP Zcore selection construct (Supplementary file 1). 10 mM GMP was present in transcriptions of the 1GMP construct, and for future transcriptions of the GMP construct selection branch and rounds 8–18; round 19–21 and reselection libraries were transcribed without GMP. Most subsequent selection rounds were transcribed in 1/10^th scale transcriptions with 15 μg of DNA (~6 × 10¹³ molecules) derived from amplification of recovered PCR products (see later).

For round 8, 700 pmol DNA was formed, with Tri3CUUQ amplifying round seven merged output (50 pmol), round seven merged output recombined by StEP (Zhao and Zha, 2006) (200 pmol), and 0core ribozyme with the starting 3’ N₃₀ library domain added (450 pmol, but extinct at the end of selection). DNA encoding type 5^s amplified with AACAt5s was used to generate reselection libraries by PCR amplification using primers TriGAA7GAAM and T5ba13N/T5ba20N/T5ba28N; 5 pmol of the three dsDNA products were transcribed to generate reselection constructs.

An outline of the selection strategy is shown in Figure 1—figure supplement 2, with detailed lists of selection oligonucleotides and extension parameters in Figure 1—source data 1, Figure 2—source data 1, Figure 4—source data 1 and Supplementary file 3. First, selection construct was annealed with equimolar dual-5’ biotinylated primer in water (80˚C 2–4 min, 17˚C 10 min), then chilled extension buffer and triplets were added before freezing and −7˚C incubation.

At the end of incubation the reaction was thawed on ice. To link the primer 3’ hydroxyl to the 5’ monophosphate of GMP constructs, selection constructs were buffer-exchanged directly after thawing using a PD-10 column (GE) in a cold room, into 3 ml ligation mix (optimised to prevent ligation over gaps) (2 mM MgCl₂, 50 mM tris•HCl pH 7.4, 0.1 mM ATP, 1 mM DTT, 2 μM ^HOGCG (Rounds 1–7) or 2 μM ^HOCUG (Rounds 8–18) with 30 U/ml T4 RNA Ligase 2 (NEB)). After incubation at 4˚C for 1 hr, these were stopped with 2.2 mM EDTA and acid phenol/chloroform treated.

Constructs were then precipitated with glycogen carrier and 0.3 M sodium acetate in isopropanol (55%) before resuspension and denaturation (94˚C 4 min, in 6M urea 10 mM EDTA with a 3 × excess of competing oligonucleotide against the primer). For the reselection rounds, constructs were then treated with polynucleotide kinase (NEB) before denaturation to resolve the HDV-derived 2’, 3’-cyclic phosphates and allow later adaptor ligation.

Constructs were urea-PAGE separated alongside FITC-labelled RNA markers equivalent to successfully ligated constructs. The marker-adjacent gel region in the construct lane was excised, excluding the bulk unreacted construct (judged by UV shadowing). Biotinylated (primer-linked) constructs were eluted overnight into BB with 100 μg MyOne C1 beads. After 30 μm filtering (Partec Celltrics(Wolflabs (York, UK))) of the supernatant to remove gel fragments, the beads were washed in BB then denaturing buffer (8 M urea, 50 mM tris•HCl pH 7.4, 1 mM EDTA, 0.1% Tween-20, 10 μM competing oligonucleotide, 60˚C 2 min) to confirm covalent linkage of construct to primer, before further BB washing and transfer (to a fresh microcentrifuge tube to minimize downstream contamination). At this stage in the reselection, 3’ adaptors were then ligated to bead-bound constructs as above for 2 hr (with buffer/enzyme added after bead resuspension in other reaction components including 0.04% Tween-20), and beads BB washed and transferred again.

Bead-bound constructs were now reverse-transcribed using 1 μM RTri (or HDVRec for the reselection) by resuspension in a Superscript III reaction with added 0.02% Tween-20 (50˚C 30 min). Beads were BB washed and the RNA-bound cDNA 3’ end blocked by incubation with terminal deoxynucleotidyl transferase (ThermoFisher) and 0.2 mM dideoxy-ATP (TriLink) with 0.02% added Tween-20 (37˚C 30 min), and beads were BB washed and transferred again.

cDNAs were eluted (10 μl 0.1 M NaOH 0.1% Tween-20 20 min), neutralized and plus strands regenerated with 0.2 μM rescue oligonucleotide in an IsoAmp II universal tHDA kit (NEB) reaction (65˚C 60 min) to read through the structured product region. Whilst at this temperature, reactions were stopped with 5 mM EDTA and one volume of BB with 50–100 μg beads to bind the nascent biotinylated plus strands at room temperature. These beads were then BB washed, NaOH washed again to discard cDNAs (and recover only correctly-primed plus strands), and washed and transferred again.

Each 50 μg of beads were then subjected to plus strand recovery PCR in a 100 μl GoTaq HotStart reaction with 0.5 μM each RTri (or HDVrec for reselection) and RecInt (rounds 1–5)/RecIntQ (rounds 6–9)/RecIntL (rounds 10–14, 19–21 and reselection)/RecIntQL (rounds 15–18). The product was agarose size-purified, A_260nm quantified and added to construct synthesis PCR in 3 × molar amount of the anticipated recovered RNA (judged by test extensions) that yielded it.

This final PCR for construct transcription in the subsequent selection round used 1 μM of the indicated construct synthesis primer, plus 1 μM RTri (or HDVrt for the reselection), in GoTaq HotStart reactions or (where indicated in source data) the GeneMorph II kit for mutagenesis (Agilent (Santa Clara, USA)).

Conditions for selections are included as source data 1 for Figures 1, 2 and 4. Numerical data for Figures 1c and 3b are included in Figure 1—source data 2 and Figure 3—source data 1 respectively. Numerical data and calculations for Figure 8 and its Figure 8—figure supplement 1 and 2 are supplied as Figure 8—source data 1. Numerical data and calculations for Figure 8—figure supplement 3 are supplied as Figure 8—source data 2. A more extensive selection of substrate concentration-dependent error rates is supplied in Figure 9—source data 1. Sequences of ribozymes, triplet synthesis templates, and oligonucleotides used in this study are supplied in Supplementary files 1, 2 and 3 respectively.

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

We thank S Thomsen, M McKie and M Sharrock for important discussions, and BT Porebski and SL James for assistance with sequencing analysis. This work was supported by a Homerton College junior research fellowship (JA.), a Harvard-Cambridge summer fellowship (AR.) and by the Medical Research Council (JA., EG., AM., PH.; programme no. MC_U105178804).

© 2018, Attwater et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.