Cell-free expression of RNA encoded genes using MS2 replicase

Abstract RNA replicases catalyse transcription and replication of viral RNA genomes. Of particular interest for in vitro studies are phage replicases due to their small number of host factors required for activity and their ability to initiate replication in the absence of any primers. However, the requirements for template recognition by most phage replicases are still only poorly understood. Here, we show that the active replicase of the archetypical RNA phage MS2 can be produced in a recombinant cell-free expression system. We find that the 3′ terminal fusion of antisense RNAs with a domain derived from the reverse complement of the wild type MS2 genome generates efficient templates for transcription by the MS2 replicase. The new system enables DNA-independent gene expression both in batch reactions and in microcompartments. Finally, we demonstrate that MS2-based RNA-dependent transcription-translation reactions can be used to control DNA-dependent gene expression by encoding a viral DNA-dependent RNA polymerase on a MS2 RNA template. Our study sheds light on the template requirements of the MS2 replicase and paves the way for new in vitro applications including the design of genetic circuits combining both DNA- and RNA-encoded systems.


INTRODUCTION
The RNA coliphage MS2 is one of the oldest model systems of modern molecular biology and its detailed investigation has led to numerous fundamental findings and applications. For example, its genome was the first to be completely sequenced, revealing for the first time the genetic organization of a biological entity (1). The small (+) strand 3569 nucleotide (nt) single-stranded RNA (ssRNA) genome encodes for only four proteins: a maturation protein required for adhesion and cell entry into its bacterial hosts, a coat protein for capsid formation and RNA packaging, a lysis gene required for virion release at the end of the infection cycle, and the catalytic replicase ␤ subunit (rep ␤ subunit) required for RNA replication by the replicase heterocomplex ( Figure 1). Further studies led to the discovery of RNA-RNA and RNA-protein interactions that control the precise timing and strength of viral protein expression during the bacteriophage life cycle (2)(3)(4). From these interactions, binding of the coat protein to a 'translational operator' stem-loop containing the start codon of the rep ␤ subunit (4) has become a versatile tool in molecular and cell biology applications such as RNA imaging (5)(6)(7).
Primer-independent replication of MS2 (+) strand genomes proceeds via complementary (−) strand intermediates, which serve as templates for efficient transcription of viral progeny (+) strands ( Figure 1) (8,9). The replicase complex responsible for this process was among the first active viral RNA-dependent RNA polymerases that could be purified and studied in isolation (10,11). However, its limited stability and difficult purification protocol at that time (9,10) prevented a detailed characterization such as the molecular principles that confer specificity during MS2 genome replication. Instead, most of our knowledge about phage RNA replication comes from studies on the replicase from the closely related bacteriophage Q␤. The Q␤ replicase (and presumably also the MS2 replicase) holoenzyme (hereafter referred to only as replicase) form through association of the catalytic rep ␤ subunit with three bacterial host factors: the ribosomal protein S1 (␣ subunit), the elongation factors EF-Tu (␥ subunit), and EF-Ts (␦ subunit) (9,(12)(13)(14)(15). However, even in the Q␤ replicase, the exact roles of all host factors are still under debate. Both ␥ and ␦ subunit appear to act as chaperones for the rep ␤ subunit (16) and are essential for processive RNA elongation (17,18), while the ␣ subunit seems to be necessary for RNA initiation and termination (19).
The Q␤ replicase was central in a series of pioneering molecular evolution experiments that led to the discovery of a small 'RQ' RNAs (20) including the famous 'Spiegelman's monster' (21), which are efficiently replicated in the presence of the Q␤ replicase. Some RQ RNAs can even be used as scaffolds for the amplification and evolution of mR-NAs in cell-free expression systems (22)(23)(24)(25)(26)(27). However, one of the disadvantages of using such small 'selfish' RNA replicators as scaffolds for in vitro gene expression is the general difficulty of designing RNA constructs that are still suitable as efficient replication templates (20,25,(27)(28)(29). Tedious steps of incremental RNA secondary structure optimization are often needed (27,30,31) to prevent formation of longer RNA duplexes unfit for replication and translation (32,33). Moreover, existing scaffolds are not suited to decouple transcription and RNA replication, which complicates experiments in which, for example, RNA amplification is not desired.
In this study we show that the active MS2 replicase can be synthesized in a recombinant, Escherichia coli-based in vitro transcription-translation system (PURE -Protein synthesis Using Recombinant Elements (34)). We find that the de novo synthesized replicase can catalyse transcription of mRNAs using engineered antisense RNA constructs as templates. This DNA-free in vitro transcription activity enables RNA-dependent transcription-translation of nonviral genes in both batch reactions and reactions encapsulated in water-in-oil emulsion droplets. Finally, we demonstrate that the new MS2-based system can be used to link in vitro DNA-and RNA-dependent transcription-translation reactions enabling informational coupling between otherwise orthogonal genetic systems. Thus, in addition to shedding light on the template requirements of the poorly characterized MS2 replicase, the new cell-free RNA expression system could enable novel directed evolution strategies or the design of genetic circuits involving RNA in a secondary genome.

Preparation of DNA/RNA constructs
The preparation of all DNA and RNA constructs is described in the Supplementary Methods. Primers used in this study are listed in Supplementary Table S1, ribosome binding sites (RBS) are listed in Supplementary Table S2, and   the final sequences of all gene constructs are listed in Supplementary Table S3.

In vitro protein synthesis
In vitro protein synthesis was performed using the PURExpress ® system (NEB), which utilizes two solutions: solution A (tRNAs, rNTPs, amino acids and other small molecules) and solution B (ribosomes and proteins including T7 DNA-dependent RNA polymerase) (35). The total volume for a standard PURE reaction was 12.5 l, consisting of 5 l Solution A, 3.75 l Solution B and other added components as indicated in the corresponding sections. All reactions were set up on ice. If necessary, the final reaction volume was adjusted with nuclease-free water (Thermo Fisher Scientific).

Fluorescent labelling and imaging of in vitro expressed MS2 replicase ␤ subunit
A standard PURE reaction programmed with 17.5 nM (250 ng) of a linear Rep DNA template (under control of a T7 promoter) was supplemented with 0.6 l FluoroTect™ Green Lys tRNA (FluoroTect™ Green Lys in vitro translation labelling system, Promega). Template DNA was omitted in the negative control reaction. Samples were incubated for 2 h at 37 • C in a nuclease-free PCR tube (Thermo Fisher Scientific) using a ProFlex PCR System (Thermo Fisher Scientific) and subsequently treated with 0.6 l RNase Cocktail™ Enzyme Mix (0.5 U/l RNase A and 20 U/l RNase T1, Thermo Fisher Scientific) for 15 min at 37 • C to degrade non-incorporated Green Lys tRNA. 7.5 l sample were mixed with an equal volume 2× Laemmli sample loading buffer (incl. 200 mM DTT) and denatured for 2.5 min at 65 • C. Samples were analysed by conventional discontinuous SDS-PAGE (10% gel) run at 4 • C (100 V for 10 min, then 200 V) on a Midi-format electrophoresis system (Atto). The fluorescent signal of the de novo expressed rep ␤ subunit was imaged on a fluorescence laser scanner (Typhoon FLA 9000, GE Healthcare) at either 473 nm (blue LD laser/510LP filter) or at 532 nm (green SHG laser/575LP filter). Total protein and the molecularweight size marker (PageRuler™ Unstained Protein Ladder, Thermo Fisher Scientific) were visualized after SYPRO Ruby (Bio-Rad) staining using the same instrument (473 nm, blue LD laser/575LP filter).

Real-time fluorescence measurements of MS2 RdTT
Design and synthesis of individual readout constructs (Supplementary Table S3 ,5% (w/v) PEG 4000 and ∼1 U RNase inhibitor (moloX) per l reaction. Note that the latter reaction was slightly diluted (13.5 l versus 12.5 l) due to the high amount of additives required to perform the reaction.
All reactions were prepared in MicroAmp Fast 8-Tube Strips (Thermo Fisher Scientific) and incubated at 37 • C in a StepOne Real-Time PCR System (Thermo Fisher Scientific). Fluorescence signals were recorded every 60 s (total incubation time was 4 h). If not stated otherwise, all experiments were performed in technical triplicates from single master mixes. All samples and the standards were analysed by native TBE-PAGE (5% Mini-Protean ® TBE Gel, Bio-Rad) run at 100 V in 1× TBE on a Mini-format electrophoresis system (Bio-Rad). RiboRuler Low Range RNA Ladder (Thermo Fisher Scientific) was used as molecular weight standard. Following electrophoresis, gels were washed 3 × 5 min with ultrapure H 2 O and then stained for 20 min in 20 ml DFHBI-1T staining solution (10 M DFHBI-1T, 40 mM HEPES pH 7.4, 100 mM KCl, 1 mM MgCl 2 ). The fluorescent signal of the de novo transcribed [F30-Bro] MS2 (+) RNA was imaged on a fluorescence laser scanner (Typhoon FLA 9000, GE Healthcare) at 473 nm (blue LD laser/510LP filter). Gel band intensities were quantified using ImageQuant TL 1D 7.0 (GE Healthcare). Brightness and contrast adjustments of the displayed gel image were applied homogenously.

Microfluidic device fabrication, droplet generation and microscopy
Droplet nozzles were fabricated through soft lithography with final chips cast from SU8 masters as 10:1 base:curing agent poly-(dimethylsiloxane) (PDMS, Sylgard), bonded to glass using oxygen plasma and surface treated with cytop 809M (AGC Chemicals Europe) to be fluorophilic, as previously described (37). Monodisperse droplets with a diameter of about 50 m were produced at room temperature by flow focusing the aqueous phase with a fluorinated oil phase (Novec 7500, 3M) and stabilized against coalescence with a biocompatible surfactant (Pico-Surf 1, 2% (w/w) in Novec 7500; Sphere Fluidics Limited). Before encapsulation in water-in-oil emulsion droplets, PURE reactions were prepared in PCR tubes and stored on ice. F30-Broccoli transcription reactions and ␤-gal expression reactions by MS2 RdTT (Supplementary Figure S6) were prepared the same way as the batch reactions described above with the only difference being, for the latter, the antisense RNA tem-Nucleic Acids Research, 2019, Vol. 47, No. 20 10959 plate was replaced by [lacZ-RBS 2 ] MS2 (−) RNA (containing a weaker ribosome binding site) at a concentration of 60 nM. For compartmentalization, samples were aspirated from the PCR tubes into the PTFE tubing by operating syringe pumps (neMESYS, Cetoni) in withdrawal mode at 4000 l/h. Subsequently, the filled sample tubes were connected to the inlets of the PDMS chip and injected at a flowrate of 400 l/h for the aqueous phase and the oil phase each. The resulting emulsion droplet creams were loaded into rectangular 50 × 500 m glass capillaries (VitroTubes) by capillary action. For imaging, the filled glass capillaries were placed on a glass microscope slip and capillary inlets were sealed with 5 Minute-Epoxy to prevent evaporation. All images were taken on a LSM 780 confocal laser scanning microscope equipped with a custom environmental chamber pre-heated to 37 • C and a Plan-Apochromat 10×/0.45 M27 objective (Carl Zeiss). Green fluorescence of the droplets was excited using the 488 nm Argon laser on the first channel (Fluorescence) with the corresponding filter ( em = 559 nm), while transmission bright-field images were also collected to access overall droplet shapes throughout the experiment. Time series for both reactions were acquired with 3 min imaging intervals for 80 cycles. The images in Figure 6, Supplementary Figure S6 and the Supplementary Movies show a merge of both fluorescence and bright-field channels. The microscope images and the Supplementary Movies were processed with ZEN (Carl Zeiss) and analysed using Fiji v1.52j. To assess the fluorescence change of individual droplets during the experiment, the mean fluorescence of eight random droplets was plotted over time. Brightness and contrast adjustments of the displayed images or movies were applied homogenously.

In vitro synthesized MS2 replicase enables general MS2 RdTT of non-viral genes
The MS2 replicase forms in vivo presumably from the association of the catalytic rep ␤ subunit with E. coli host factors similar to the Q␤ replicase (9,12-15) ( Figure 1). To probe whether active MS2 replicase can also be produced in vitro, we first tested conventional DNA-dependent transcription-translation (DdTT) of the rep ␤ subunit in a commercial PURE system, which contains the T7 DNAdependent RNA polymerase (T7 pol). To this end, we engineered a linear DNA construct encoding a minimized version of the MS2 genome devoid of all coding sequences except for the rep ␤ subunit with an upstream T7 promotor (Rep DNA). Using this construct, we could indeed detect de novo synthesis of the 60.8 kD rep ␤ subunit in the PURE system by the Green Lys labelling approach and conventional SYPRO Ruby staining in yields matching the concentration of the translation factors present in the PURE system (Figure 2).
Next, we set out to probe if the expressed rep ␤ subunit can form an active replicase through complex formation with the proposed E. coli host factors, which are present in the PURE system (34). RNA replication during the life cycle of (+) ssRNA viruses is typically highly asymmetric with the genomic (−) ssRNA being the better transcription template (38) to prevent formation of biologically inert RNA duplexes (39,40). Therefore, we sought to detect replicase activity by the conversion of (−) ssRNA to (+) ssRNA, i.e. RNA-dependent RNA transcription. We anticipated that template recognition by the replicase relies on specific 5 and 3 terminal RNA secondary structure elements, similar to other RNA viruses (41)(42)(43)(44). In particular, we expected that the responsible RNA domains are contained in the two untranslated regions (UTRs) of the MS2 genome, each of which folds into a defined secondary structure (45). The UTR leader sequence (129 nt, UTR L ) is located at the 5 end of the (+) strand genome and the UTR trailer sequence (181 nt, UTR T ) overlaps with the end of the rep ␤ subunit gene at the 3 end (Figure 1). We thus wanted to verify if the reverse complements of the UTR L (cUTR L ) and the UTR T (cUTR T ) found in the genomic (−) strand are sufficient for general template recognition and transcription initiation by the MS2 replicase.
To test this hypothesis, we created two types of RNA modules: Readout modules for the detection of RNAdependent RNA transcription and a replicase module (MS2-rep (+) RNA) encoding the open reading frame of the rep ␤ subunit. The readout modules were designed as such that they should serve as synthetic (−) RNA templates for the MS2 replicase and were created by inserting the antisense strands of genes between both cUTRs ( Figure 3A). The expression of the rep ␤ subunit directly from MS2-rep (+) RNA allowed us to bypass conventional T7 DdTT and made the system completely DNA-independent. This not only prevented competition of the T7 pol for NTPs, but also enabled a better control over the expression levels of the rep ␤ subunit. Both modules combined enabled MS2-based RNAdependent transcription-translation (MS2 RdTT) in the PURE system ( Figure 3A). In detail, we monitored (+) strand transcription of differently sized (m)RNAs from their respective (−) strand constructs using de novo synthesized MS2 replicase and, if applicable, their translation into functional proteins. As first readout, we co-incubated the PURE components with the replicase module MS2-rep (+) RNA, the fluorophore DFHBI-1T and the readout module [F30-Bro] MS2 (−) RNA, in which the antisense strand of the F30-Broccoli aptamer (F30-Bro) (46,47) was embedded between both cUTRs. We anticipated that a successful RNA-dependent RNA transcription of the 410 nt F30-Bro aptamer by the MS2 replicase would generate a fluorescent signal upon binding of the otherwise nonfluorescent DFHBI-1T. Indeed, we detected a strong increase in fluorescence just after several minutes of incubation at 37 • C, confirming that the in situ expressed rep ␤ subunit forms the MS2 replicase together with the host factors provided in the PURE system ( Figure 3B). In contrast, F30-Bro aptamer transcription, i.e. fluorescence, was not detectable in presence of an RNA template encoding the catalytically inactive rep ␤ subunit variant (D341S/D342V, MS2-ddrep (+) RNA), in which crucial residues for Mg 2+ coordination in the conserved palm domain of the replicase had been mutated (17,48).
In vitro transcription-translation reactions can be subject to considerable variabilities reflecting in part the complexity of translation, variabilities in template folding and batch-tobatch variations of the transcription-translation machinery (51,52). To assess the reproducibility of the MS2 RdTT system, we performed batch-to-batch variation experiments. To this end, we programmed two different batches of the  In conclusion, we demonstrated that the de novo synthesized rep ␤ subunit and translation factors present in the PURE system form an active replicase whose RNAdependent RNA transcription activity can be detected in situ by MS2 RdTT using various engineered (−) RNA templates.

MS2 UTRs enable stoichiometric MS2 RdTT
After demonstrating successful (−) to (+) ssRNA conversion by the MS2 replicase, we sought to determine the amount of (+) strand that is synthesized during a typical RdTT reaction and whether the de novo synthesized (+) strand would also serve as template for more (−) strand synthesis, thereby initiating an RNA amplification cycle. To quantify the amount and kinetics of the To further probe whether MS2 replicase is capable of synthesizing (−) strands from engineered (+) strands, we initiated RdTT reactions directly with in vitro transcribed [F30-Bro] MS2 (+) RNA. If both (−) and (+) strand could serve as a template, a further increase in F30-Bro levels would have been expected. However, no significant fluorescence increase compared to a negative control was observed during 3 h of incubation, implying that (+) strand readout modules do not serve as templates for (−) strand synthesis under our tested conditions (Supplementary Figure S3).
In  Figure S4B). Higher amounts of input template reduced the yields of overall (+) strand, suggesting that the excess amount of free (−) strand promotes the formation of non-fluorescent dead-end duplexes by enhanced annealing with newly synthesized [F30-Bro] MS2 (+) RNA.
In conclusion, the new MS2 RdTT system enables unidirectional conversion from (−) to (+) strand of various RNAs without further (unregulated) amplification as in most current Q␤ systems (24,27). The lack of amplification in the engineered RNA results presumably from the absence of RNA motifs required for (+) RNA recognition and initiation by the replicase. Indeed, for the related Q␤ phage, (−) RNA transcription from genomic (+) RNA is highly dependent on internal RNA domains, long-range RNA-RNA and specific RNA-protein interactions (9,53).

Only the cUTR L domain is required for MS2 RdTT of (−) strand RNA
In an attempt to further minimize the UTRs required for MS2 RdTT, we compared MS2 RdTT-based ␣complementation with the ␣-peptide mRNA embedded between either full-length cUTRs (Supplementary Figure  S5A) or minimized MS2 cUTRs consisting of only the terminal hairpins of the MS2 genome (Supplementary Figure  S5B). When using these shortened constructs, formation of the active ␤-gal tetramer was strongly reduced (Supplementary Figure S5B), implying that the minimized RNA domains are only poorly recognized by the replicase.
Next, we set out to test if both cUTRs of (−) strand templates are required for MS2 RdTT. For the Q␤ replicase, it has been suggested that both 5 and 3 termini of genuine RNA templates cooperate during and after the initiation step, presumably via a circular configuration through a terminal helix (44). However, when we compared the levels of F30-Broccoli aptamer transcribed from a standard [F30-Bro] MS2 (−) RNA (containing both full-length cUTRs) with transcription from a truncated [F30-Bro] MS2 (−) RNA lacking the cUTR T domain ( Figure 5A), we found that transcription from the truncated template occurred at a higher rate compared to the full-length construct ( Figure  5B). On the contrary, deleting the cUTR T from a longer sfGFP construct ([sfGFP-RBS 2 ] MS2 (−) RNA) had barely any impact on sfGFP expression levels ( Figure 5C). Both findings imply that the cUTR T domain is not essential for MS2 RdTT. The observed differences between 5 -terminally truncated and full-length (−) strand templates are rather suggestive of a context-dependent influence of the 129 nt segment on RNA template structure and/or stability rather than of a direct role of the RNA domain in transcription. For example, deleting the cUTR T from the 410 nt F30-Broccoli construct decreases the overall RNA length and, thus, the synthetic burden for transcription by ∼43%, which may explain the overall increase in synthesis yields by MS2 replicase.

The MS2 RdTT system is active in cell-sized emulsion droplets
Compartmentalization of in vitro translation systems is a well-established method in synthetic biology and useful in applications such as high-throughput screening, molecular evolution of enzymes, or bottom-up synthesis of artificial cells (54)(55)(56)(57)(58). The MS2 RdTT system could be useful in these types of experiments such as the evolution of DNA-modifying enzymes where the presence of an encoding DNA-template in addition to the selection substrate might be incompatible with the optimal selection strategy. To test if MS2 RdTT can enable DNA-independent transcription also in m-sized water-in-oil emulsion droplets formed from a biocompatible surfactant in a fluorinated oil, we encapsulated ice-cold PURE reaction samples containing [F30-Bro] MS2 (−) RNA and MS2-rep (+) RNA (or MS2-ddrep (+) RNA as control) using a custom microfluidic setup (for details see Material and Methods) and monitored the fluorescence of individual droplets by fluorescence microscopy during incubation at 37 • C. We observed robust fluorogenic transcription of the F30-Broccoli aptamer in the droplets during the first 90 min ( Figure 6A, B and Supplementary Movie S1). The averaged time traces of eight individual droplets showed similar kinetics as in the batch reaction ( Figure 6C). We could also show MS2 RdTT of active full-length ␤-gal under the same conditions (Supplementary Figure S6A, B and Supplementary Movie S2), further demonstrating that both formation and activity of the MS2 replicase are compatible with emulsion-based compartmentalization and that the overall system is readily compatible with typical microencapsulation-based evolution protocols.

The MS2 RdTT can be used to trigger DdTT
Having shown that the MS2 RdTT system enables complex schemes of coupled genetic/enzymatic information transfer, we wondered if we could use the system to control the otherwise orthogonal in vitro transcription-translation from DNA templates (i.e. DdTT). Such a 'cross-talk' between DNA-and RNA-dependent in vitro transcriptiontranslation would largely expand the repertoire for the generation of synthetic circuits and switches used in synthetic biology. As a proof of concept, we designed an experiment such that the synthesis of active MS2 replicase would lead to MS2 RdTT of SP6 DNA-dependent RNA polymerase   (SP6 pol) using [SP6 pol] MS2 (−) RNA as the template. The de novo expressed SP6 pol should then catalyse SP6 DdTT of sfGFP from a linear DNA template with an SP6 promoter (SP6-sfGFP DNA). We chose the SP6 pol for DdTT to bypass the T7 pol, which is already present in the PURE system, because both polymerases use different promotors. Altogether, the system coupling both MS2 RdTT and SP6 DdTT combines 5 steps of macromolecular synthesis in a one-pot reaction: Three translation events (rep ␤ subunit, SP6 pol and sfGFP) and two transcription events ([SP6 pol] MS2 (+) RNA and sfGFP mRNA) ( Figure 7A). To our delight, we were indeed able to switch on expression of sfGFP via MS2 RdTT of SP6 pol ( Figure 7B). Thus, our new MS2 RdTT system can be directly used to trigger DdTT, thereby providing an additional control layer to the design of synthetic genetic circuits. In the current setup based on the commercially available PURE system, we observed some background sfGFP expression from leaky transcription of the SP6 promotor by the T7 pol present in the kit (Supplementary Figure S7). Such leakage could be omitted by using alternative, tailor-made PURE systems devoid of T7 pol (25,34,(59)(60)(61).

DISCUSSION
While MS2 replicase has been reported to be unstable and difficult to purify (9,10), we could show that the active replicase complex can be readily produced in situ in a recombinant in vitro transcription-translation system. The in situ produced replicase can directly initiate transcription of various RNA templates if its 3 terminus is fused with cUTR L --a short 3 terminal domain of the viral (−) strand. While low levels of RNA transcription can already be achieved with a minimalistic 43 nt version of cUTR L , maximal transcription levels require fusion to the fulllength cUTR L (124 nt, Supplementary Figure S5A, B). This implies that the complete cUTR L domain contains additional structural elements for replicase template recognition and/or initiation.
In contrast to the transcription of (+) from (−) strands, we observed no (−) strand synthesis from synthetic (+) strand RNAs even if both UTRs were present. This finding implies either that the 3 -end of the tested (+) strand construct is only poorly accessible for replicase initiation or that additional sequence elements are required for a complete replication cycle. In agreement with the latter explanation, replication of the Q␤ (+) strand by the Q␤ replicase is crucially dependent on long-distance interactions between UTR T (the 3 terminal UTR of the (+) strand) and internal RNA-sites (9,(62)(63)(64) and similar interactions were predicted to exist in the genomic MS2 (+) strand (64,65), which are missing in our engineered RNA templates. Alternatively, (+) to (−) strand replication might fail due to a potentially inaccessible 3 end, which could prevent replication initiation. An additional yet unlikely possibility is that one or several yet unknown host factors are required for MS2 (−) RNA synthesis, which are missing in the PURE system. In general, the coordination of (−) strand synthesis and translation in ssRNA phages is a topologically complex process as both replicase and ribosomes compete for the same (+) RNA template but proceed with opposite polarities. Therefore, phages have established elaborate mechanisms to ensure that replication or translation are mutually exclusive to prevent collision events (9).
The ability of our system for stoichiometric (+) strand synthesis from an RNA template without further amplification distinguishes the MS2-based RdTT system from similar Q␤-based systems, which are currently the only other bacterial RNA-only in vitro transcription-translation systems described. In these systems, target genes are typically embedded in the (+) strands of small, non-genomic 'parasitic' RNA-scaffolds such as RQ135 (22,66) or MDV-1 (23,24,26). These extremely replication-competent RNAs lack the regulatory motifs required for the controlled timing and strength of gene expression and replication. Instead, they serve as templates for their own unregulated exponential amplification, which is limited only by either the formation of inert duplexes between (+) and (−) strand or the consumption of all available nucleotides. While RNA amplification is desirable in continuous evolution studies (25,27), it can be detrimental in experiments where resources are scarce, which for example is the case for in vitro transcription-translation batch reactions, or where the original coding template should remain unaltered. For example, during RNA amplification by the Q␤ replicase, coding constructs are rapidly replaced by original non-coding parasite and strict compartmentalization and selective conditions are required to maintain the coding RNA pool (25,26). The MS2 RdTT system could therefore be useful for different applications in which RNA amplification is not required. The low fidelity of the MS2 replicase makes it suitable for direct in vivo or in vitro generation of RNA libraries for protein or aptamer selection from clonal RNA templates without the need for DNA mutagenesis. A study describing the in vivo use of Q␤ replicase to generate mRNA libraries showed that the mutational spectrum of phage RNA replicases is close to the ideal (67). Furthermore, the independence of the MS2 RdTT system from DNA might increase the repertoire of selection strategies for DNAmodifying enzymes such as DNA nucleases, ligases, polymerases, recombinases or methyltransferases whose activity can interfere with a conventional DNA construct. As the MS2 RdTT system shows identical activity after encapsulation in water-in-oil emulsion droplets, the use of the system in in vitro selection protocols that are dependent on direct genotype-phenotype linkage through compartmentalization should be straightforward. Finally, we demonstrated that MS2 RdTT can be used to control conventional DdTT. This coupling of the two otherwise orthogonal expression systems could be used to expand the design repertoire of synthetic transcription-based genetic systems and circuits such as switches (68), oscillators (69,70), biosensors (71), or artificial multicellular systems (72).