Directed evolution, DNA library preparation, Rolling Circle Amplification, transformation efficiency, in vivo homology The manipulation of DNA and its subsequent insertion into a desired host cell is a bedrock of modern biotechnology. This ability has advanced both basic science (e.g., the human genome project, metagenomics, gene and protein function assays), as well as applied science (e.g., the directed evolution of proteins for a desired functionality, and the production of proteins and small molecules via fermentation and biocatalytic processes). To achiev e DNA transfer into microorganisms there are three main routes: conjugation, transduction, and transformation, in which DNA is transferred cell-to-cell, virus-to-cell, or extracellular DNA-to-cell, respectively. Transformation is the most routinely used DNA transfer mechanism in laboratory settings due to its relative ease (eschewing any need for viral packaging reactions1 or co-culturing of conjugative cells2) and historical precedent, in which protocols have been established for many different cell types across all kingdoms of life. In particular, many protocols have been established for two of the most common organisms used in biotechnology, E. coli3 and S. cerevisiae4, that are broadly separable into either chemical competent or electrocompetent methods. Briefly, chemical competent approaches typically rely on washing in mixtures of salts followed by subsequent incubation with exogenous DNA that is induced to pass through the cell membrane by a heat-shock step, whilst electrocompetent approaches wash cells in hypertonic solutions to remove any ions and then use exposure to electrical current to drive exogenous DNA through the cellular membrane5. Despite the vast number of reported protocols, transformation is still limited by relatively poor efficiencies, with typically <1% of cells being transformed by exogenous DNA. Particularly in the field of directed evolution, it is desirable to make libraries that are as large and high quality as possible, so that the greatest amount of sequence space can be explored in the search for a desired trait6. Limited transformation efficiencies greatly hinder this work, and are compounded by the discrepancies between advertised transformation efficiencies (measured using picogram quantities of supercoiled plasmid DNA) and those ac hieved during library preparations (using nanogram-microgram quantities of relaxed circular DNA) often resulting in 100-1000-fold loss in efficiency.

In order to overcome limited transformation efficiencies, researchers typically scale-up the process by transforming multiple aliquots of cells using as much DNA as possible. Whilst multiple aliquots of cells can either be purchased (at significant cost) or prepared in-house with rel ative ease, the preparation of large quantities (10-100 micrograms) of library DNA is more difficult. To achieve large quantities of assembled library DNA, one can either scale-up the assembly reactions or amplify the assembled product in a way that it is suitable for transformation into host cells. There are now many different ways to assemble libraries beyond the traditional restriction-ligation approach, many of which offer the benefit of 8scarless9 assemblyi,.e., the absence of any undesired sequence in the final product. The most common scarless approaches are Gibson assembly7 and golden-gate assembly8 encountered in the form of commercial kits, which cause the scale-up of library assemblies to quickly become economically infeasible for many labs (overviewed nicely in Xia et al.9). Whilst much work has gone into non-commercial de rivatives10 or alternatives9,11, these are often non-robust or complex to establish. Perhaps the most established non-commercial scarless cloning approach utilises the native ability of multiple cell types to in vivo repair linear DNA into circular plasmid DNA. Whilst it is widely known that S. cerevisiae is transformable by linear DNA with homologous ends12, it is often overlooked that E. coli possesses the same ability13 (albeit likely occurring through a different nonrecombination-based pathway). As these approaches u nite the plasmid assembly and transformation step in a costfree manner, they are particularly scalable and therefore suitable for large library generation. Thus, a method to generate large quantities of DNA in a suitable format (i.e. linear DNA) is required. Much work on amplifying li brary DNA from small-scale assemblies has already been carried out, typically focussing on Rolling Circle Amplification (RCA) due to its ability to massively amplify circu lar DNA in an isothermal manner. The product of RCA (highlybranched concatomers) is however not directly suita ble for transformation and must be resolved into monomeric units, achieved in the literature through digestion with restriction enzymes and subsequent large-volume low concentration religation14 or Cre recombinase activity15. Whilst both approaches are successful, they are either cumbersome or require non-standard reagents. A recent elegant development utilises a one-pot golden-gate RCA reaction, which simplifies the generation of large quantities of cloned DNA. However, the cloned DNA is generated only in a final circular format and without a quantification of its impact on the generation of large libraries16. To best deliver a method that can easily generate large libraries in multiple different cell types, we thus sought to combine features from multiple different approaches. Notably, we utilised RCA to amplify a small-scale golden-gate assembly reaction ~10,000 fold generating ~100 ug of product, followed by nicking endonuclease (Nb. BbvCI) trea tment (E. coli workflow) to generate linear DNA with single-stran ded homologous ends or blunt-ended restriction endonuclease (FspI) treatment ( S. cerevisiae workflow) to generate linear DNA with double-strand ed homologous ends. Subsequently, these linear DNA molecules can be efficiently transformed into the appropriate cell types.

Focusing firstly on E. coli, we designed our approach around three key techniques: 1) RCA of small-sca le library assemblies to generate ~100 ug of DNA, 2) nicking e ndonuclease treatment to resolve the RCA product into linear monomeric units, and 3) in vivo assembly of linear DNA by E. coli to generate circular plasmids (Figure 1A). Nicking endonuclease treatment was chosen based on observations from Xia et al.9 that generation of single-stranded homologous ends are sufficient to recapitulate the cloning and transformation efficiencies of more complex Gibson assemblies, and are more efficient than the transformation of linear DNA with double-stranded homologous ends seen in García-Nafría et al.13. Additionally, use of nicking endonucleases for cloning purposes has been successfully developed by Wang et al.17. Our test plasmid was thus modified to include a 8nickase cassette9, in which two Nb. BbvCI sites were added on the top and bottom strand of the plasmid respectively, such that upon nickase treatment a) the concatomeric RCA product will be resolved to monomers and b) the linear monomers will possess s inglestranded 59 and 39 overhangs that are homologous toone another.

As a first test of our approach, we investigated the effect of creating single-stranded homologous ends on transformation efficiency across multiple cell types and homology lengths. Initially, two different homology lengths were tested (10 bases and 15 bases) by adding the a ppropriate sequences through PCR amplification of the template plasmid (a phagemid for a VHH phage display library, with either the pelB or DsbA signal peptides for periplasmic export of the VHH-p3 fusion protein). When subjecti ng the linear DNA products to nickase treatment, a substantial boost in transformation efficiency (10-100 times higher) was observed compared to linear DNA with doub le-stranded homologous ends for both homology lengths. Notably, the 10-base homology length exhibited a slightly superior transformation efficiency relative to the 15-base homology length in the context of the DsbA signal peptide (see Figure 1B). As such, we proceeded with the 10 base homology length and additionally tested the transformation efficiency of a pelB signal peptide containing phagemid in comparison to the DsbA containing phagemids. Little-tono effect was seen, thus we proceeded with the more standard pelB signal peptide in subsequent experiments. We then explored the transformation of different cell types relevant to directed evolution (XL-1 blue for phage display, BL21-DE3 for protein expression, and NEB5³ for cloning) using nickase-treated linear DNA, and observed the same effects as before, i.e., that single-stranded homologous ends enhanced transformation efficiency when compared to double-stranded homologous ends across all cell types (Figure 1C). as the mean of triplicates (with individual data points displayed with green triangles).Error bars rep resent standard deviation of replicates nand face only upwards for simplicity.

After confirming that nickase treatment was beneficial, we investigated the best conditions for RCA of plasmids focussing on both amplification yield and specificity based on literature protocols18,19. Importantly, the plasmid of interest now contained the 10 bp nickase cassette that resulted from transformation of the linear DNA products in circle amplification, several parameters were considered and evaluated when test-amplifying the plasmid 8PFNbb102-CAM9, of which some are shown in the graph.Going clockwise, firstly, two ï29 polymerases (at 10 units per 20 µL reaction) were compared using 5 mM random hex amer primers, including three phosphorothioate (PT) bonds on the 83 end, while following the suppliers9 recommendations regarding duration and temperature of the amplification step (orange section). Next, different primers (at 5 mM concentration) were assessed for their suitabi lity (red section). The specific DNA primer (more information in the method section, as well as vector maps in the supplementary information) was designed to one site of the select ed plasmid. Furthermore, we tested whether the presence of phosphatase would benefit the amplification (blue section). In four conditions, a range of three order s of magnitude were covered (0 units, 0.01 unit, 0.1 unit, and 1 unit). ( b) Comparing the output RCA product mass (left y-axi s, highlighted in green bars) to the input mass of pla smid DNA (plotted on the x-axis). Amplification fa ctors (right yaxis, depicted with black/white dots connected by black line) versus input mass of plasmid DNA is also shown. Error bars reflect standard deviation over replicates of three for each condition. (c) Visualisation of nickase-treated RCA product (lane 7 and 8 (left to right)) in compariso n to the presumably supercoiled input plasmids (8PF-Nbb102-CAM9 and 8DF-Nbb102-CAM9 in lanes 2 and 3, respectivelya)nd the non-nickase treated linear RCA-products (lanes 5 and 6). The RCA reaction was performed using random DNA PT- and RNA hexamers (lanes 5 and 7 (PT-DNA) and 6 and 8, respectively). The dashed line indicates tha t the image has been cropped for simplicity. The original version can be found in Figure S3. (d) Beyond establishing the parameters for DNA amplif ication and transformation into E. coli, we sought a better understanding of the transformation efficiencies (electroporation and chemical) o ver the in vitro process of library assembly. The plasmid 8PF-Nbb102-CAM9 was purified from an overnight culture and transformed either directly (condition 1 and 4), oramplified by PCR for subsequent re-insertion of the VHH encoding sequence by golden-gate assembly. The purified golden-gate product was then either transformed directly (condition 2 and 4) or subjected to RCA amplification and nicked (using Nb.BbvCI) and subsequently transformed (c ondition 3 and 6).

Upon confirmation of our workflow in the context of purified plasmid DNA, we next investigated the best conditions for the amplification of DNA in a 8library context,9i.e., amplification of a golden-gate assembly. Amplification of assembled DNA was readily achieved with >18,000-fold amplification yield, generating 9.2 µg of DNA from 0.5 ng of input DNA, corresponding to one golden-gate reaction containing 8.8 x 108 DNA molecules (Table 1). Post nickase treatment and purification a yield of ~3 µg was achieved, representing an input to transformable output amplification yield of ~6,500 fold. In comparison, a similar PCR-based amplification workflow achieved an amplification yield ~10 fold lower than the RCA workflow. The RCA workflow is readily scalable in terms of initial volume and output mass, e.g., parallelised amplification of 40.2 ng golden-gateassembled product resulted in >700 µg purified, nickase-treated plasmid target DNA. We noticed that treatment of the golden-gate product with exonuclease to remove unassembled insert DNA was not necessary (Figure S1).

To establish the applicability of our approach for creating large libraries, we then looked at the transformation efficiencies of circular supercoiled plasmids, relaxed circular golden-gate assembled plasmids, and linear nickasetreated RCA product across both electroporation and chemical transformation of XL-1 Blue E. coli cells. Here, we observed that the chemically competent cells were transformed more efficiently with our linaer nicked DNA than relaxed circular DNA (golden-gate product), yet the opposite was true for electrocompetent cells (Figure 2D). Due to this observation, we explored a variety of parameters around chemical transformation, including cell type, transformation buffers, cell concentrations, storage media, growth media for competent cell production, and growth media for post-transformation selection plates, as well as the potential scalability of chemical transformations through parallelisation, and the use of increased cell volumes (Figure S2). The most efficient in our hands wa s the use of a modified Inoue protocol, XL-1 blue cells grown in SOB media, stored in DMSO and plated on SOC agar supplemented with 2% Glucose and 5 mM MgCl2.

RCA-mediated amplification of plasmid DNA for S. cerevisiae transformation After establishing the use of the RCA nickase workflow for E. coli, we also tested its applicability to other common organisms used for directed evolution. We chose to investigate S. cerevisiae, notably the EBY100 strain for two reasons; firstly, EBY100 is commonly used for yeast surface display, a powerful technique for the ultra-high throughput quantitative screening of binder libraries20,21, and secondly, because yeast is known to possess strong homologous recombination activity enabling transformation with linear DNA molecules, we expected that our homology-based cloning approach may have an effect on transformation efficiencies.

S. cerevisiae cells are often transformed with linear DNA possessing double-stranded homologous ends; thus, we created a new homology cassette in the yeast plasmids alongside our previously described nickase-single strand cassette. This new cassette contained an FspI site in the centre of a 30 or 90 base repeat sequence, such that upon FspI treatment the RCA product would resolve to monomeric units with double-stranded homologous ends (Figure 3A). We observed that the linear FspI treated doubl e-stranded homology DNA was most efficient in chemical transformation of yeast cells, outperforming both circular supercoiled DNA and the linear nickase treated singlestranded homology DNA by at least 10-fold (Figure 3B). This result established the utility of our appr oach for creating large libraries in yeast, as well as providing data confirming that linear DNA transforms yeast at higher efficiency than circular DNA. As we had observed contradictory results in E. coli between electroporation and chemical transformation, we also tested electrocompetent yeast and observed that linearisation significantly improved transformed efficiency versus circular DNA (Figure 3C). EBY100. (a) Schematic of optimised post-assembly library amplification for transformation into S. cerevisiae. (b) Using chemical transformation: Transformation efficiencies in S. cerevisiae across different formats (circular (orange), linear single-stranded (red), and linear double-str anded (purple)) and two different homology site len gths (30 and 90 base pairs) of similar plasmids (8pCT-antiGFP9). Lki e transformations in E. coli strains, we tested the utility of linearised DNA for chemical transformation in S. cerevisiae strain EBY100. While the treatment of Nb.BbvCI resulted in a linear format with single-stranded 39 overhangs, the double-stranded homology sites restricted by treatment with

FspI leaves blunt ends. Vitally for the double-stranded homology, the according plasmid contains a repeat of the homology site which enables in vivo processing and assembly. (c) Transformation of RCA product (green) in chemical and electroporation transformation using 2 µg and 8 µg per sample, respectively, of linearised plasmid with double-stranded 90 bp homology sites.

The creation of large libraries is a difficult, resource and time-consuming process that significantly limits the accessibility of directed evolution to the wider scientific community. Whilst novel technologies, such as in vivo hypermutation (e.g., in E. coli as PACE22/PRANCE23 or MutaT724, B. thuringiensis as BacORep25 and S. cerevisiae as OrthoRep26), can circumvent the need to physically transform as many cells as you want your library size to comprise, it is still likely that physical transformation will remain the method of choice for many researchers due to specific library designs, or the choice of screening technologies which are not compatible with the mentioned technologies. In addition, an increased transformant yield is also useful in other fields, where subsequent in vivo mutation is undesirable, such as basic biology, sequencing, DNA storage, etc. As such, any methods that can simplify the creation of large libraries are of great use. Interestingly, it has been suggested that transformation of E. coli is not as clonal as typically considered, thus in the creation of large libraries it is likely that the diversity screened is actually greater than the number of transformants27. Such effects can be resolved after candidate identification through plasmid extraction and retransformation at lower DNA:cell ratios, and in the case of phage display would be resolved after the output of the first round of panning is used to infect cells (presuming a multiplicity of infection <1). In this work, we have developed a simple protocol for the creation of large libraries in E. coli through chemical transformation. Whilst it was an unintended consequence of the interesting molecular biology that we uncovered, (that linear DNA does not transform efficiently via electroporation), the outcome of focussing on chem ical transformation is elegant as it is the least limited by accessibility to specific pieces of equipment. The observed difference in efficiency between linear DNA transformation into electrocompetent cells and chemically competent cells can likely be explained by the unique mechanisms by which DNA passes into the cells of each approach. However, as the exact mechanisms of either chemical competency or electrocompetency are not fully understood, it is difficult to posit much beyond this.

To further extend the utility of our approach, we confirmed its use for large library creation in S. cerevisiae, a highly useful organism in biotechnology and synthetic biology for which RCA-mediated library amplification has not been explored previously. We expect that the product of our protocol could be of benefit in organisms other than E. coli or S. cerevisiae, such as the rapidly developing biotechnology chassis organism V. natriegens28 for which efficient and scalable natural transformation techniques have been established29. Additionally, the linear DNA product of our protocol could be of use for genome integration rather than plasmid formation.

In conclusion, our workflow can readily yield ~6,500- fold amplification of purified transformable DNA, such that large libraries (>108) can easily be achieved from even single small-sca le (nanogram) library assembly reactions in both E. coli and S. cerevisiae. This is achieved in as simple a manner as possible, with a focus on reduction of both cost and resource usage.