Generating efficient Methanomethylophilus alvus pyrrolysyl-tRNA synthetases for structurally diverse non-canonical amino acids

Abstract

Genetic Code Expansion (GCE) technologies commonly use the pyrrolysyl-tRNA synthetase (PylRS)/tRNA^Pyl pairs from Methanosarcina mazei (Mm) and Methanosarcina barkeri (Mb) for site-specific incorporation of non-canonical amino acids (ncAAs) into proteins. Recently a homologous PylRS/tRNA^Pyl pair from Candidatus Methanomethylophilus alvus Mx1201 (Ma) was developed that, lacking the N-terminal tRNA-recognition domain of most PylRSs, overcomes insolubility, instability and proteolysis issues seen with Mb/Mm PylRSs. An open question is how to alter Ma PylRS specificity to encode specific ncAAs with high efficiency. Prior work focused on “transplanting” ncAA substrate specificity by reconstructing the same active site mutations found in functional Mm/Mb PylRSs in the Ma PylRS. Here, we found that this strategy produced low efficiency Ma PylRSs for encoding three structurally diverse ncAAs: acridonyl-alanine (Acd), 3-nitro-tyrosine and m-methyl-tetrazinyl-phenylalanine (Tet3.0-Me). On the other hand, efficient Ma PylRS variants were generated by a conventional life/death selection process from a large library of active site mutants: for Acd encoding, one variant was highly functional in HEK293T cells at just 10 μM Acd; for nitroY encoding, two variants also encoded 3-chloro, 3-bromo- and 3-iodo-tyrosine at high efficiency; and for Tet-3.0-Me, all variants were more functional at lower ncAA concentration. All Ma PylRS variants identified through selection had at least two different active site residues when compared with their Mb PylRS counterparts. We conclude that the Ma and Mm/Mb PylRSs are sufficiently different that “active site transplantation” yields suboptimal Ma GCE systems. This work establishes a paradigm for expanding the utility of the promising Ma PylRS/tRNA^Pyl GCE platform.

Graphical Abstract

INTRODUCTION

Genetic code expansion (GCE) enables site-specific incorporation of noncanonical amino acids (ncAAs) into proteins during translation. Ideal GCE platforms should be (1) easily adapted or evolved to accept a wide range of ncAA structures, (2) efficient in encoding ncAAs into proteins with high yields at a re-purposed codon of choice, (3) orthogonal to natural translational components from prokaryotes and eukaryotes and to currently existing GCE platforms (for dual ncAA incorporation), and (4) soluble and stable for in vitro characterization and cell-free protein synthesis applications. The original Mathanocaldococcus jannaschii (Mj) Tyr-RS/tRNA^Tyr platform has proven highly successful for a variety of GCE applications, including cell-free synthesis, but it is not orthogonal to eukaryotic translational systems, and the ncAAs which it can be evolved to accept as substrates are largely limited to tyrosine-like derivatives.¹ The most widely used GCE platforms are derived from the closely related Methanosarcina barkeri (Mb) and Methanosarcina mazei (Mm) pyrrolysyl-tRNA synthetase (PylRS)/tRNA^Pyl pairs, in part because they are specified by the amber codon UAG, they are orthogonal in both prokaryotic and eukaryotic translational systems, and they have amino acid binding pockets that can be evolved to accept chemically diverse ncAAs.² Shortcomings of the Mb and Mm PylRSs pairs are that they frequently express as insoluble proteins, they are susceptible to proteolysis, they tend to localize to the nucleus and they are of limited utility in yeast expression hosts.^3–7 These challenges can limit the efficiency of in vivo ncAA incorporation and the utility of the platforms for cell-free protein synthesis and kinetic characterization. More Pyl-based GCE platforms that overcome these issues would expand our capacity to incorporate multiple, structurally diverse ncAAs into target proteins.

The Candidatus Methanomethylophilus alvus (Ma) PylRS/tRNA^Pyl pair is a recently developed GCE platform that is orthogonal in both prokaryotic and eukaryotic cells,^8–10 and was recently engineered to be orthogonal to the Mb/Mm PylRS platforms, enabling dual ncAA encoding with the two Pyl platforms in the same cell.^{8, 11, 12} This GCE platform was also found to be highly functional in yeast.⁷ A distinctive feature of Ma PylRS is that it lacks the critical N-terminal domain found in the Mb/Mm enzymes (Fig. 1A) that binds to the T-arm and variable loop of their cognate tRNA. The simpler architecture of the Ma PylRS is presumably key to the increased stability and solubility observed for this enzyme, making the Ma GCE platform an attractive alternative to the Mb/Mm system for both in vivo and in vitro ncAA-protein expression systems.⁵

Figure 1.

The M. alvus pyrrolysine synthetase. (A) Comparison of the Mm/Mb and Ma PylRS/tRNA_CUA architectures showing the N-terminal domain of the Mm/Mb synthetase is required for recognition of its cognate tRNA, which is absent in the M. alvus system. (B) The wild-type M. alvus amino acid binding pocket (PDB 6jp2) with pyrrolysine docked based on a structural alignment with M. mazei PylRS. Residues targeted for saturation mutagenesis in this study are colored with maroon carbon atoms. Residues with teal or gray carbon atoms are first or second shell residues, respectively, that when mutated to smaller residues expand the active site to better accommodate ncAAs larger than pyrrolysine⁵. (C) Modeling of the Ma Y126A active site transplant mutation shows enlargement of the amino acid binding pocket, which allows the synthetase to accept larger amino acids (Z-lys, BCN-Lys and TCO-Lys) than wild-type does, which accepts pyrrolysine and Boc-Lys (shown in panel D).

The sequence identity between the Ma PylRS and the Mb/Mm PylRS is only ~35%, yet their overall three-dimensional structures are quite similar (Cα RMSD ~1.2 Å for the catalytic domain) and the residues lining the amino acid substrate binding pockets are well-conserved.⁵ Their structural similarity led to the idea that time consuming directed evolution approaches to engineering Ma PylRSs could be skipped, and instead the ncAA specificity of an engineered Mb/Mm RS could be “transplanted” into Ma PylRS by recreating the same set of active site mutations.⁹ For example, Ma PylRS variants able to incorporate lysine-like derivatives with bulky, hydrophobic head groups were created by transplanting two mutations (Y126A/Y206F) that were known to enlarge the back end of the Mb/Mm PylRS ncAA binding pocket^{5, 8, 10} (Fig. 1C and 1D). To open space in the front end of Ma PylRSs amino acid binding pocket, up to two mutations have been transplanted from the Mm/Mb PylRS to the Ma PylRS (e.g. N166S or N166A/C168A), permitting encoding of singly-substituted phenylalanine derivatives^{5, 13, 14}.

Despite these successes, it is not clear how generally applicable the “active site transplant” (AST) strategy will be for the many previously engineered Mm/Mb PylRS variants and the extent to which it yields the most effective Ma variant (where “effective” refers to not only high encoding efficiency, but also the ability of an Ma RS variant to function at low ncAA concentrations and encode a family of structurally similar ncAAs). One might expect that, given their low sequence identity, sufficient differences between the two PylRSs exist such that relying on the AST strategy will limit the overall utility of the Ma GCE platform. This might be particularly evident when Mm/Mb RS specificity for an ncAA is conferred by a constellation of interactions between the binding pocket and the ncAA made possible by a considerable restructuring of the Mm active site. Indeed, recent attempts to transplant five mutations from an Mm active site specific for acetyl-lysine failed to yield a corresponding functional Ma variant³, perhaps because important interactions between the hydrophilic head groups and the active site pocket that define specificity were not satisfied in the architecture of the Ma active site.

Here, we set out to explore the general efficacy of AST as a means to transfer ncAA specificity to the Ma GCE platform using three structurally diverse ncAA test cases (Fig. 2A): acridonyl-alanine¹⁵ (Acd, a fluorescent amino acid), 3-nitrotyrosine¹⁶ (nitroY, a post-translational modification) and m-methyl-tetrazinyl-phenylalanine, termed Tetrazine v3.0-Me¹⁷ (Tet3.0-Me, a bio-orthogonal labeling handle). We show that while ncAA specificities were transferred to Ma, the resulting AST-derived Ma RSs as expected displayed much lower efficiency than the Mb systems on which they were based. Instead, for all three ncAAs, de novo selection of new Ma variants from a large library of active site mutants yielded systems with much higher efficiency than the ASTs. The differences seen in comparing panels of selected Ma variants with those from Mb PylRSs for a particular ncAA implies that substrate binding modes may be sufficiently different between the two PylRS systems that AST is not a general solution to obtaining high efficiency Ma RSs. In an accompanying manuscript, we used X-ray crystallography to elucidate the structural basis for the superior activity of the top evolved Ma Acd-RS1 compared to Ma Acd-AST ⁴⁴.

Figure 2.

Evaluating Ma AST-RS efficiencies for incorporating ncAAs into sfGFP-150TAG. (A) Structures of Acd, Tet3.0-Me and nitroY amino acids. (B) Left: residue identities of the wild-type pyrrolysine binding pockets for M. mazei, M. barkeri and M. alvus synthetases (which are able to incorporate Boc-Lys). Right: residue identities that confer active site selectivity for Acd, Tet3.0-Me and nitroY in the Mb synthetase platform that were transferred to Ma to generate the ASTs. (C) Incorporation efficiency of Boc-Lys, Acd, Tet3.0-Me and nitroY into sfGFP-150TAG using the pBK/pALS expression system in DH10B cells (see Materials and Methods) and the indicated RS/tRNA pairs. For each ncAA incorporation system, the constellations of RS active site residues are indicated in panel B and are identical for each Mb and Ma AST RS set, except for the wild-type Pyl RSs whose native sequences were preserved. Culture fluorescence values were normalized to cell density (OD₆₀₀), and error bars represent standard deviations from cultures grown in triplicate (see Materials and Methods).

RESULTS

Evaluation of active site transplant Ma-PylRSs for Acd, nitroY and Tet3.0-Me incorporation.

We first examined the AST approach by creating Ma PylRS variants that contained the same constellation of amino acid changes found in top performing Mb RSs selective for Acd¹⁵, Tet3.0-Me¹⁷ and nitroY¹⁶ (Fig. 2A and 2B). We used the Ma tRNA^Pyl variant with a modified variable loop (previously referred to as variant 6) that was evolved to be orthogonal to the Mm/Mb PylRS so any resulting Ma systems from this study could be used for dual ncAA encoding.⁹ We did not observe notable differences in incorporation efficiency using this orthogonalized Ma tRNA^Pyl compared to the genomic Ma tRNA^Pyl (Supporting Fig. 1). Examination of the efficiency and fidelity of incorporating ncAAs at residue 150 of the super folder green fluorescent protein reporter (i.e. sfGFP-150TAG) showed that all of the AST-RSs produced sfGFP in the presence of their respective ncAAs, but their efficiencies were only about 30%, 25% and 10%, respectively, of the parent Mb enzymes (Fig. 2C). Since the wild-type Ma PylRS/tRNA^Pyl pair incorporated Boc-Lys at about 75% the level of the wild-type Mb PylRS pair (Fig. 2C), this means that all of the AST RSs underperformed compared to their Mb counterparts. We also tested the effect of making two additional pairs of mutations in the first- and second-shell of the Ma PylRS active site (Y126A/M129A and H227I/Y228P, respectively, Fig. 1) that had been shown to improve incorporation efficiency for lysine derivatives by expanding the ncAA pocket size⁵; however, these modifications did not improve efficiency (Supporting Fig. 2). We hypothesized that the reasons for the limited efficiency of AST-RSs stemmed from subtle differences between the Mb and Ma active site scaffolds, and that it would be possible to engineer Ma variants with improved activity compared to the AST-RSs by carrying out fresh de novo selections from a large library of Ma PylRS mutants with changes in the ncAA binding pocket.

A life/death selection for the evolution of new M. alvus PylRSs.

Five positions within the amino acid binding pocket of Ma PylRS were targeted for saturation mutagenesis: L125, N166, V168, A223 and W239 (colored maroon in Fig. 1B), generating a theoretical library size of 20⁵ = 3.2 million protein variants. These sites were picked based on the structures of the targeted ncAAs, and also because the above tested Ma AST variants are theoretically present in this library, so that if indeed they are among the top functioning Ma variants for encoding their respective ncAAs, they should be identified in the ensuing selections. Before building the library, we compared Boc-Lys-dependent suppression efficiencies of wild-type Ma PylRS expressed from an E. coli and a human codon optimized gene. Negligible differences were observed (Supporting Fig. 3), allowing use of the human codon optimized gene for simplified transfer into human cell lines.

Next, positive, negative and fluorescence-based selection plasmids were engineered for use with Ma PylRS libraries (Supporting Fig. 4). The positive selection plasmid contained a TAG-interrupted chloramphenicol resistance gene, while the negative selection harbored an arabinose inducible 2x-TAG codon-interrupted toxic barnase gene, and the fluorescence selection plasmid featured an arabinose inducible sfGFP-150TAG reporter. Each selection plasmid also constitutively expressed the orthogonalized Ma Pyl- tRNA^Pyl (6) (see Supporting Figure 1). The selection plasmids were paired with a pBK plasmid expressing Ma PylRS variants so that cells receiving RSs specific for an ncAA survive in positive and negative selections in the presence and absence of the ncAA, respectively. The fluorescence-based selection offers a direct readout of TAG codon suppression efficiency and fidelity, and was used to identify the most efficient RSs producing full-length sfGFP only in the presence of the ncAA.¹⁸

Ma PylRS selections for variants that incorporate Acd.

We performed a round of positive selection in the presence of Acd at a low concentration of chloramphenicol (25 μg/ml), followed by a negative selection against canonical amino acids. Surviving clones able to express sfGFP-150TAG in the presence of Acd were identified, and ninety-two were screened for their ability to suppress the sfGFP-150TAG gene. The top 19 clones were sequenced, revealing 7 unique RS protein sequences (Supporting Fig. 5A). One of these variants accounted for 11 of the 19 clones, though interestingly 10 of these 11 contained the same genetic sequence (D4) while one had a different codon for Leu125 (D2), even though these two codons should have been in equal population in the library (see Materials and Methods). These two clones (D2 and D4) coding for the same protein but with different codons for Leu125 were therefore evaluated separately. To further explore the parameters of the selection process, a second experiment was performed using greater selective pressure (100 μg/mL chloramphenicol) in the positive selection. Of the top ten clones sequenced, eight were identical in gene sequence to the A7 clone, and two were identical to D4 (Supporting Fig. 5). Thus, increasing chloramphenicol concentrations in the positive selection step narrowed the pool of selected variants, but did not reveal new ones. The Acd-AST variant was not among any selected protein sequences, even though it was theoretically present in the starting library. Further, none of the 7 unique Ma AcdRSs identified here possess the same constellation of active site mutations found in any of the top 13 Mb Acd-RSs previously selected.¹⁵ These observations imply that, even though we only characterized one Ma Acd-AST, none of the other possible Ma Acd-ASTs would have been as efficient as those obtained through the de novo directed evolution process. Given the lack of overlap among the top selected Ma hits here and the previously identified Mb hits, we conclude that there exist practically impactful differences in the architectures of their respective active sites.

As an initial characterization of the seven top Ma Acd-RS hits, we evaluated their abilities to suppress sfGFP-150TAG as a function of Acd concentration. As we have done before¹⁹, we fit the normalized cell fluorescence as a function of Acd concentration to a Michaelis-Menten-like model to yield both the UP₅₀ (concentration of ncAA at half maximal fluorescence) and the theoretical maximal suppression efficiency (Fmax) achievable by the RS. The most desirable new enzymes are those exhibiting low UP₅₀ and high Fmax values, qualities that would mitigate the potential toxicity of Acd and minimize background fluorescence. Four top AcdRSs were identified based on these criteria (A7, D2, D4 and E4, referred to hereafter as Acd-RS1, Acd-RS2, Acd-RS3 and Acd-RS4, respectively; Fig. 3A–3D, and Supporting Figure 5). These top four Acd-RSs were about six-fold more efficient than the Ma Acd-AST, about three-fold more efficient than the top two performing previously characterized Mb PylRS variants, and were similar in efficiency to the previously identified Mj Acd variant A9²⁰ (Fig. 3B). Ma Acd-RS1 had the lowest UP₅₀ among those tested (Fig. 3C and 3D). The structural basis for the improved activity of Ma Acd-RS1 compared to Ma Acd-AST is reported in detail in an accompanying manuscript ⁴⁴. The Acd-RS2 and Acd-RS3, coding for the same protein sequence, displayed indistinguishable expression trends except at 1.0 mM Acd, leading to slightly different UP₅₀ and Fmax best-fit values (Fig. 3C).

Figure 3.

Characterization of de novo selected Ma Acd-RS variants. (A) Sequence analysis of the active site residues for the top four selected Ma variants with selectivity for Acd. Asterisks indicates a unique Leu codon was observed for Acd-RS2 (CTG) compared to Acd-RS3 (CTT). (B) Efficiency and permissivity of the selected Ma Acd RSs compared with the AST, the previously reported Mb 41 and 82 RSs¹⁵ and the Mj A9²⁰ variant. Error bars represent standard deviations from cultures grown in triplicate (see Materials and Methods). (C) UP₅₀ analysis of the top four selected Ma Acd RSs compared with the AST and the previously reported Mb RS 41. (D) Extrapolated UP₅₀ and F_MAX values from the expression data shown in panel C (see Materials and Methods). (E) Purified wild-type sfGFP and sfGFP-150Acd expressed using the top four Ma Acd RSs indicating full-length protein production. Similar protein quantities were loaded on the gel and do not reflect expression efficiency. (F) Whole-protein mass spectrometry of sfGFP wt (gray) and sfGFP-150Acd expressed with Acd-RS1 (red) demonstrate accurate incorporation of Acd in place of Asn at position 150. Theoretical masses are written in parenthesis. Mass spectra of sfGFP-150Acd produced by the other three Ma Acd RSs are shown in Supporting Figure 6.

Fidelity and permissivity of Acd incorporation.

To assess the fidelity of Acd incorporation by the selected enzymes, sfGFP-150TAG proteins expressed with each of the top 4 Acd-RSs in the presence of Acd were purified and analyzed by whole protein mass spectrometry. From these purifications, we obtained yields corresponding to ~190 mg of sfGFP-wt per liter culture, and about 180 and 120 mg of sfGFP-150Acd purified from cultures expressing Acd-RS1 and RS2–4, respectively. The reported masses confirmed accurate incorporation of Acd at position N150 by all RSs (Fig. 3F, and Supporting Fig. 6). In terms of permissivity (the ability of an RS to incorporate other ncAAs for which it was not selected), none of the four enzymes were active toward the structurally similar ncAA benzoyl-phenylalanine (Bpa), although they were able to incorporate, albeit inefficiently, N-phenyl-amino phenylalanine (NpF) (Fig. 3B). The Acd used in this study is free of contaminating NpF, an undesired by-product in earlier Acd syntheses,²⁰ and indeed no contaminating sfGFP-150-NpF was observed when Acd was the only ncAA present in the media (Fig. 3F, Supporting Fig. 6).

Evaluation of Acd incorporation into calmodulin.

While sfGFP provides for convenient evaluation of RS efficiency, fidelity and permissivity, we wanted to confirm that these GCE systems could also efficiently install Acd into a biologically relevant protein under typical expression conditions. For this we chose calmodulin (CaM), a protein for which the site-specific installation of Acd has proven useful for understanding its conformational changes and dynamics in response to calcium binding.^{21, 22} Also, we moved the Ma Acd-RS/tRNA^Pyl pairs into a single low-copy plasmid (pKW) that is compatible with co-expressing target proteins from traditional pET/pBAD-like vectors. With this system we expressed, in triplicate, CaM-112TAG fused to a C-terminal intein-His₆ purification handle with each of the unique Ma Acd RSs (Acd-RS1, Acd-RS2 and Acd-RS4), and then quantified yields of purified CaM-112Acd by Acd fluorescence. Yields were approximately 3-fold higher with Acd-RS1 (corresponding to ~10 mg per liter culture) compared to Acd-RS2 and Acd-RS4 (Supporting Figure 7), consistent with the above results with sfGFP. Collectively, these data confirm the Ma Acd GCE system is sufficiently robust to produce biologically relevant proteins in E. coli containing Acd.

Kinetic basis for improved efficiency of Acd-RS1.

We next took advantage of the increased stability and solubility properties of the Ma PylRS system to overexpress and purify the Ma Acd-RS1 and Acd-AST enzymes to determine the kinetic basis for the improved amber suppression efficiencies of Ma Acd-RS1. We examined catalytic efficiency under single turnover conditions using a saturating, fixed concentration of ³²P-labeled tRNA and varying concentrations of Acd from 10 to 1000 μM. The fraction of aminoacylated tRNA was monitored over time for each amino acid concentration, and resulting data were analyzed to derive the Acd binding affinity (K_d) and maximum rate of tRNA aminoacylation (k_obs) (Fig. 4, Supporting Figure 8). These experiments showed that the maximum turnover rate for Acd-RS1 was ~1.7 fold faster than the Acd-AST (0.76 min⁻¹ vs 0.45 min⁻¹, respectively), and the overall catalytic efficiency (k_obs / K_d) was ~3-fold higher. Thus, the better in-cell incorporation efficiency of the Acd-RS1 compared to the Acd-AST can be explained, at least in part, by its more efficient kinetic properties.

Figure 4.

In vitro tRNA amino-acylation kinetic analyses of the Ma Acd-RS1 and Acd-AST synthetases. (A) Observed tRNA amino-acylation rates are plotted as a function of Acd concentration and fitted to a best-fit Michaelis-Menton like curve (see Materials and Methods). (B) Extrapolated k_obs and Acd K_d values from the curves shown in panel A indicate ~3-fold higher catalytic efficiency for the Ma Acd-RS1 compared to the Acd-AST.

Acd incorporation in HEK293 cells.

Next, we evaluated Ma Acd-RSs 1–4 for their capacities to incorporate Acd in HEK293 cells, and compared them with Ma Acd-AST as well as the previously reported top performing mammalian Mb Acd-RS¹⁵. All Ma RSs were cloned into the pAcBac1 vector containing four copies of the orthogonalized Ma tRNA^Pyl(6), and then these plasmids were co-transfected into HEK293T cells together with a pAcBac1 vector expressing sfGFP-150TAG. RS efficiency and fidelity were evaluated by sfGFP fluorescence in the presence and absence of 0.3 mM Acd (the concentration previously identified as optimal for Mb-Acd-RS82 incorporation¹⁵). For comparison, a pAcBac1 vector expressing sfGFP-wt was also transfected. All five Ma Acd-RSs produced fluorescence when expressing sfGFP-150TAG in the presence of Acd but not in its absence, confirming faithful incorporation (Fig. 5A). Quantification of cell fluorescence by flow cytometry revealed Ma Acd-RS1 as the top performing enzyme, roughly two-fold and three-fold more efficient than Ma Acd-RS2 and Acd-RS4, respectively, and marginally more efficient than Acd-RS3, Acd-AST and the Mb AcdRS-82 (Fig. 5B). It is interesting to note the Ma Acd-AST performed comparatively well in HEK293 cells at this high concentration of Acd. This observation highlights the differences between E. coli and eukaryotic expression patterns; however, it remains consistent that the de novo selected Acd-RS1 was still the top performing Ma variant.

Figure 5.

Evaluation of Acd incorporation into sfGFP-150TAG in HEK293 cells using Mb and Ma GCE platforms. (A) Cellular fluorescence of cells expressing sfGFP-150TAG in the absence (bottom row) or presence of 300 μM Acd (top row) with each of the selected four Ma Acd RSs and the Ma Acd-AST. For comparison, also included are sfGFP-150TAG expressions using the previously published Mb AcdRS 82¹⁵. eRF1 was not expressed in these cultures. (B) Total sfGFP cellular fluorescence of sfGFP-150TAG expressing cells in the presence and absence of 300 μM Acd as determined by flow cytometry (see Materials and Methods). The effect of eRF1 E55D was evaluated for the Mb AcdRS 82 and Ma Acd-RS1 as well. The marked increase in suppression efficiency for Mb AcdRS 82 with eRF1 E55D co-expression was consistent was previous reports¹⁵, and was also observed here with the new Ma Acd-RS1 platform. (C) Evaluation of sfGFP-150TAG suppression by the Mb AcdRS 82 and Ma Acd-RS1 systems as a function of Acd concentration in the presence of eRF1 E55D. The red arrow points to the concentration of Acd used for expressing sfGFP-150Acd that was purified for the mass spectrometry analysis shown in panel D. (D) Whole protein mass spectrometry of purified sfGFP-150Acd expressed from HEK293 cells using 50 μM Acd and the Ma Acd-RS1 system. Measured mass is indicated and expected mass is in parentheses (mass calculated with C-term V5 and His₆ tags, as well as demethionylation, and N-terminal acetylation⁴³).

Previously, we showed that Acd incorporation with the Mb AcdRS-82 was improved by co-transfecting a vector expressing the eRF1 mutant E55D, which is designed to attenuate UAG-dependent translation termination in HEK293 cells^{15, 23}. We therefore tested whether a similar improvement was possible for the Ma platform, and satisfyingly, we found a similar 3-fold enhancement with Ma Acd-RS1, reaching about half the level of sfGFP-wt expression (Fig. 5B).

To determine the optimal Acd concentration for expression in HEK293 experiments, we expressed sfGFP-150TAG with eRF1 E55D using Acd concentrations ranging from 10 to 600 μM. Consistent with prior work¹⁵, the optimal concentration for the Mb AcdRS-82 was ~100–300 μM, and while in this experiment it was slightly more efficient than the Ma Acd-RS1, incorporation efficiency dropped sharply below 100 μM (Fig. 5C). The Ma Acd-RS1 displayed maximal suppression at ~100–200 μM, but incorporation was still >50% of maximal levels at Acd concentrations as low as 10 μM. Thus, the Ma Acd-RS1 platform functions robustly at notably lower Acd concentrations compared with the Mb system, making it advantageous for minimizing background fluorescence and toxicity. Lastly, we purified sfGFP-150Acd protein expressed with eRF1 E55D at 50 μM Acd, and confirmed accurate Acd incorporation by mass spectrometry, with no detectable protein adducts from natural amino acid misincorporation (Fig. 5D). Together, these data show that Ma Acd-RS1 is the top performing synthetase for Acd incorporation in HEK293 cells, in that it can function ~5 fold better at low (~10 μM) Acd concentrations compared to the best Mb Acd RS-82. This high efficiency for Ma Acd-RS1 seen at low Acd concentrations is consistent with it having a low UP₅₀ as measured in E. coli (see Fig. 3C, 3D).

Lastly, we asked whether these Ma RSs could produce a biologically relevant protein containing Acd in HEK293 cells. For this we chose, α-synuclein, a disordered and aggregating protein implicated in Parkinson’s disease for which we have previously developed cell-based aggregation models.²⁴ The ability to track α-synuclein aggregation in real-time using Acd as a non-perturbing fluorescence tag would be a valuable improvement over the current use of fluorescent protein tags or fixed cell labeling with antibodies.²⁵ To test Acd incorporation into α-synuclein, co-expressed α-synuclein-114TAG fused with a C-terminal sfGFP fluorescent tag with each of the four Ma Acd RSs and the Mb AcdRS-82, at Acd concentrations ranging from 0 to 200 μM. The trends in expression efficiency as a function of Acd concentration (Supporting Figure 9) recapitulated those observed for Acd incorporation into sfGFP, with the Mb AcdRS-82 being slightly more efficient at 200 μM Acd, but the Ma Acd-RS1 being nearly 10-fold more efficient at 10 μM Acd.

New de novo selected M. alvus PylRS variants for nitroY.

Given the success of making a robust Acd incorporation system with the Ma platform, we next asked whether the same process could be used to generate Ma systems for incorporation other ncAAs, namely Tet3.0-Me and nitroY (Fig. 2A), for which the ASTs were much less efficient than the parent Mb RS (Fig. 2C).

Following the same low stringency strategy used to select Ma Acd-RSs, we identified seven unique Ma nitroY-RSs able to incorporate nitroY in E. coli (Fig. 6A). The mutations W239R and N166A were conserved in all of the sequenced Ma variants, whereas Arg and Ala were not present at these positions in any Mb nitroY-RSs identified previously; they uniformly had Trp at the first position and either Cys, Ser or Gly at the second.¹⁶ Thus, no de novo selected variant had the same constellation of mutations as any reported Mb nitroY-RSs - even though those variants should have been present in the starting Ma library.¹⁶ As with Acd, when we repeated the selection process under higher stringency (100 μg/mL chloramphenicol), the diversity of RSs was narrowed to only three unique hits among those samples (Supporting Figure 10).

Figure 6.

Selected Ma RS variants for the incorporation of 3-nitroY. (A) Amino acid sequence analysis of the top Ma nitroY RS hits. (B) Efficiency and fidelity of these seven new Ma nitroY RS variants compared to the previously reported Mb nitroY variants¹⁶, as well as our 3^rd generation Mj nitroY A7 synthetase²⁶. All expressions used the same pBK/pALS expression system, and error bars represent standard deviations of triplicate cultures (see Materials and Methods). Permissivity profiles for incorporation of other meta-substituted tyrosine ncAAs are shown in Supporting Fig. 10.

The selected Ma-nitroY-RSs were dramatically improved compared with Ma nitroY-AST and displayed similar amber suppression efficiencies to each other, to the previously identified Mb nitroY-RSs, and to our third-generation Mj nitroY-RS (A7)²⁶ (Fig. 6B). UP₅₀ values for the Ma nitroY-RSs ranged from ~40 to 110 μM (Supporting Fig. S10A and S10B). Accurate incorporation of nitroY into sfGFP-150TAG was confirmed by whole protein mass spectrometry on purified protein produced by RS variants D9, G5 and D12 (Supporting Fig. 11). A small population of purified protein had a mass consistent with 3-amino-tyrosine (aminoY) incorporation, even though none of the Ma nitroY-RSs incorporate aminoY (Supporting Fig. 10C). The presence of aminoY in proteins expressed with nitroY has been observed previously to be caused by reduction of the nitro moiety by E. coli cellular reductants during expression (and after incorporation).²⁷ While unable to incorporate aminoY or 3-nitro-Phe, permissivity profiles show that most of the Ma nitroY-RSs efficiently incorporated 3-chloro-, 3-bromo- and 3-iodo-tyrosine, and in fact, the D9 and F5 variants incorporated all 3-haloY ncAAs more efficiently than nitroY (Supporting Fig. 10C). These data show that de novo selections from a library of active site variants yields more efficient Ma nitroY-RSs than the AST strategy.

New de novo evolved M. alvus PylRS variants for Tet3.0-Me incorporation.

To further validate the utility of selections over the AST strategy to generate practically useful RSs, and to explore the versatility of the Ma PylRS GCE platform, we used the same starting library of Ma RSs to select for incorporation of the Tet3.0-Me ncAA. Tetrazine-containing amino acids are of particular interest because of their ability to function as biooorthogonal handles that react rapidly and specifically with trans-cyclo-octene (TCO)-containing molecules, thus enabling site-directed attachment of fluorophores and polymers in vitro and in vivo.^{17, 28–30} After a single round of selection at low stringency, several Ma variants with similar efficiencies of Tet3.0-Me incorporation were isolated, of which the top four are reported here (Fig. 7A and 7B). Interestingly, the constellation of active site mutations in the isolated Ma Tet3.0-RSs were not only different than Tet3.0-AST (Fig. 7B), but in every case, either 4 or 5, or all 5 of 5 of the residues allowed to change were different. Also, all showed 2- to 3-fold higher activity than the Tet3.0-AST, but were still substantially less efficient than the parent Mb Tet3.0-RS (variant R2–84¹⁷).

Figure 7.

Selected Ma RS variants for the incorporation of Tet3.0-Me. (A) Amino acid sequences of the top four Ma Tet3.0-Me RS hits. (B) Efficiency and fidelity of these four new Ma Tet3.0-me RS variants compared to the previously reported Mb Tet3.0 R2–84¹⁷. All expressions used the same pBK/pALS expression system, and error bars represent standard deviations of triplicate cultures (see Materials and Methods). (C) SDS-PAGE gel of purified sfGFP-wt as well as sfGFP-150-Tet3.0-Me proteins produced by the indicated Mb and Ma RS variants. For each Tet3.0-Me containing protein, sTCO-PEG5k was added to confirm incorporation of the reactive tetrazine group and to assess the extent of reactivity. In each case >95% of the Tet3.0 containing proteins reacted with the sTCO-PEG5k as evidence by an apparent mass increase in SDS-PAGE.

Accurate and homogeneous tetrazine incorporation into the selected variants was confirmed by mixing purified GFP-150Tet3.0 with strained TCO (sTCO) conjugated to a 5000 Da polyethylene glycol polymer (PEG-5000). In this assay, reaction of sTCO-PEG-5000 with GFP-150Tet3.0 imparts an upshift in apparent molecular weight by SDS-PAGE (Fig. 7C). Whole protein mass spectrometry further confirmed the incorporation of Tet3.0-Me (Supporting Fig. S12). Lastly, UP₅₀ values of the new Ma Tet3.0 RSs were all much lower than the Mb R2–84 RS. Three of the four had lower UP₅₀ values than the Tet3.0-AST, with the top performing variant (C1) exhibiting a very low UP₅₀ value of 10 μM (Supporting Figure S12). The much lower UP₅₀ values of the Ma Tet3.0-Me indicate they function much better at low Tet3.0-Me concentrations than the Mb R2–84, however, their lower overall activity suggests that a different library, or different Tet3.0 derivative, might yield more overall efficient incorporation of tetrazine functionality with the Ma platform. Nevertheless, these data confirm the Ma GCE platform is sufficiently versatile to incorporate Tet3.0-Me accurately, and as with Acd and nitroY, de novo selections performed better than active site transplantation for creating highly functional variants.

DISCUSSION

GCE technology requires additional platforms to expand our ability to install one or more ncAAs into proteins site-specifically. Efforts to do so have led to the identification of a variety of RS/tRNA pairs suitable for GCE, including those derived from pairs naturally specific for pyrrolysine³¹, histidine, aspartate, arginine, glutamine, serine, tryptophan and tyrosine^{12, 32–34}, while other efforts have led to creating synthetases with increased thermostability that are more tolerant to mutation and directed evolution^{35, 36}. Consequently, available now are a suite of RS/tRNA pairs compatible with suppression of not just amber stop codons, but also other stop, sense and quadruplet codons³⁷, though most have not yet been evolved to incorporate ncAAs.

Among the pool of recently discovered RS/tRNA pairs, the pyrrolysine pair from M. alvus has garnered notable attention because it belongs to a class of Pyl-RSs that does not have the N-terminal domain characteristic of the Mb/Mm Pyl-RS’s, which is required for their function (Fig. 1A). This N-terminal domain causes the Mb/Mm RSs to aggregate and precipitate, and indeed the Ma Pyl-RS which lacks this domain is much more readily expressed, stable and soluble compared to its Mb/Mm counterparts. Yokoyama and colleagues leveraged these properties to reconstitute cell-free expression systems with higher concentrations of Ma Pyl-RS to achieve a high activity amber suppressing system not achievable with the Mb/Mm platforms,⁵ while others have confirmed the Ma platform is orthogonal in eukaryotic cells and orthogonal Mm/Mb Pyl-RSs, enabling encoding of two distinct amino acids into the same protein.¹¹

Yet to be understood with the Ma platform was whether it has the ability to incorporate structurally diverse ncAAs like the related Mm/Mb platforms, and how best to engineer such variants with altered substrate specificity. We show here for three ncAAs that are not structurally similar to Pyl, that while the AST strategy did produce Ma variants functional for their targeted ncAAs, the AST RSs were as expected not as efficient as their Mb counterparts. We confirmed that for all three of our targeted ncAAs, library generation and standard selection methods were superior to AST. In fact, we never once observed an AST emerge as a top hit in our selections, even though they presumably were present in the starting library. These observations are not surprising given that the level of sequence similarity between the Mb and Ma RSs is only ~35%, and they imply that the mechanisms by which Mm/Mb and Ma PylRS variants confer selectivity for ncAAs are sufficiently dissimilar to Pyl that these systems should be independently evolved to achieve the highest performing GCE systems.

This work also emphasizes the value of performing UP₅₀ and permissivity analyses on panels of selected RS variants since maximal encoding efficiency at high ncAA concentration need not always be the most desirable property of a GCE system. These types of analyses previously helped us identify Mb nitroY RSs highly functional in both E. coli²⁶ and mammalian cells¹⁶, and indeed here we found that several of our selected Ma nitroY-RSs were also able to incorporate 3-halo-tyrosine amino acids even more efficiently than nitroY. In the case of Acd, the de novo selected Ma Acd-RS1 was more efficient than our previous top Mb Acd RS-82 in E. coli, and for HEK293 expressions UP₅₀-like analyses showed it can be used with Acd concentrations as low as 10 μM. In the case of Tet3.0-Me, the reported Ma PylRS variants all function at much lower concentrations of ncAA than the top performing Mb R2–84 RS, though overall efficiency in E. coli needs improvement. We also show the Ma platform, unlike the Mm/Mb platform, is readily amenable to in vitro kinetic characterization and that such studies are mechanistically informative, in this case revealing better catalytic efficiency of Ma Acd-RS1 compared to the Acd-AST and in an accompanying paper by Gottfried-Lee et al.⁴⁴, the high-resolution structures of both the RS1 and AST enzymes. In summary, our strategy allowed us to we create new Ma RS variants suitable for Acd, 3-nitroY, all three 3-halo-Tyr derivatives, and Tet3.0-Me incorporation that are also more efficient than their respective ASTs. Because these Ma GCE systems are expected to be orthogonal to Mb/Mm GCE systems, they can be leveraged for dual ncAA incorporation. The availability of this Ma PylRS/tRNA^Pyl selection tool box should also provide easier access for those looking to add new ncAA functionality to prokaryotic and eukaryotic systems.

MATERIALS & METHODS

Expanded methods are provided in the Supporting Information

Strains.

DH10b and BL21(DE3) were purchased from Thermo Fisher Scientific. The PPY strain used to generate SLiCE³⁸ cloning extract was a gift from Yongwei Zhang (Albert Einstein College of Medicine).

Molecular biology reagents and amino acids.

Oligonucleotide primers and double stranded DNA fragments were synthesized by Integrated DNA Technologies (Coralville, IA). Molecular biology reagents including restriction enzymes, T4 ligase and polymerases were purchased either from Thermo Fisher Scientific or New England Biolabs. DNA Miniprep, Midiprep, PCR cleanup and gel extraction kits were purchased from Machery Nagel. Nε-Boc-L-lysine, 3-nitro-L-tyrosine, 3-chloro-tyrosine, 3-bromo-tyrosine, 3-iodo-tyrosine and 3-amino-tyrosine were purchased from Sigma. Tet3.0-methyl and sTCO-PEG5k were synthesized as previously described¹⁷. Acd was also synthesized as described²¹. All amino acids were solubilized at 100 mM stock concentration in water by adding 2 molar equivalents of NaOH, except Tet3.0-methyl which was solubilized at 30 mM stock concentration in water.

Evaluating aminoacyl tRNA-synthetase TAG codon suppression efficiency.

Evaluation of amber codon suppression efficiency using the sfGFP fluorescent reporter protein.

pBK plasmids expressing the desired Ma PylRS variants were simultaneously co-transformed with the pALS2 fluorescent reporter plasmid into chemically competent DH10b cells and transformants were grown overnight on LB/agar plates containing kanamycin (50 μg/mL) and spectinomycin (100 μg/mL) at 37 °C. Similarly, to evaluate Mb PylRS variant suppression efficiencies, the corresponding pBK plasmids were co-transformed with pALS (expressing the Mb tRNA^Pyl) and grown with kanamycin (50 μg/mL) and tetracycline (25 μg/mL). After overnight growth, a single colony was picked and used to inoculate 0.5 mL of defined non-inducing media (NIM, Supporting Table 1) containing the appropriate antibiotics in a 96-well block, and grown for ~16 hrs at 37 °C with shaking at 300 RPM. The next day, 10 μL of a NIM culture was used to inoculate three 0.5 mL defined autoinducing media cultures (AIM, Supplementary Table 1) lacking ncAA and also three separate cultures containing the desired ncAA concentration (typically 1 mM, or ranging from 0 to 1 mM for UP₅₀ analyses) and grown at 37 °C with shaking at 300 RPM. Cell fluorescence and OD₆₀₀ readings were taken at 24 hr and 48 hr time points on a plate reader. Thus, the reported culture fluorescence values are the average of three independently grown expression cultures but originating from the same colony. Error bars represent standard deviations. For UP₅₀ analyses, normalized fluorescence values were fit as a function of ncAA concentration using a Michaelis-Menten like equation to extrapolate F_MAX and UP₅₀ values: F = F_MAX * [ncAA] / (UP₅₀ + [ncAA]). Curves were fit using GraphPad Prism 6.0.

Selection of new Ma synthetase variants.

Cloning Ma positive and negative selection plasmids.

The Ma positive selection plasmid was made by replacing the Mb tRNA^Pyl in pREP (tetracycline resistance, p15a origin) with the Ma tRNA^Pyl (variant 6). Similarly, the Ma negative selection plasmid was made by replacing the Mb tRNA^Pyl in pYOBB2 (chloramphenicol resistance, p15a origin) with the Ma tRNA^Pyl (variant 6).

Cloning Ma active site library.

The Ma library (called the “Ma-Lib” library) was generated using the 22-c codon trick³⁹ at residues L125, A223 and W239, and degenerate NNK codons at sites N166 and V168. The library was assembled into the pBK plasmid using a MEGAWHOP cloning procedure.⁴⁰ Greater than 99.9% coverage of the library was confirmed by having a 10-fold excess of transformants over the size of the genetic library. Additionally, ten individual colonies were sent for sequencing to validate each colony contained a unique and intended mutant of the Ma PylRS. Expanded methods for assembling the library are included in the Supporting Information.

Life/death double sieve selections.

Selections were performed as previously described²⁶ using the Ma positive, negative and fluorescence plasmids described above, and the “Ma-Lib” pBK library. For the positive selection and fluorescence reporter assessments, 1 mM ncAA was used. Expanded methods for RS selections are included in the Supporting Information.

sfGFP protein purification.

E. coli DH10b cells expressing pBK/pALS2 plasmids for incorporating a targeted ncAA at position N150 of sfGFP were cultured at 50 mL scale in defined AIM (see Supporting Table 1) at 37 °C for 24 hrs with 0.5 mM ncAA. Cells were harvested and lysed in Wash Buffer (50 mM Tris, 500 mM NaCl, 5 mM imidazole pH 7.4) by microfluidizing. Insoluble cell debris was pelleted by centrifugation at 25,000 rcf for 30 min, and sfGFP protein in the soluble fraction was bound to TALON metal affinity resin. The resin was washed with 20–50 column volumes of Wash Buffer, and eluted with Wash Buffer + 300 mM imidazole. Proteins were desalted using a PD-10 column (GE Healthcare) into 50 mM Tris pH 7.4, 150 mM NaCl (or for mass spectrometry into 50 mM ammonium bicarbonate), frozen in liquid N₂ and stored at −80 °C.

Expression and purification of Calmodulin containing Acd in E. coli.

Plasmids for each RS (Ma Acd RS-1, RS-2 and RS-4) in a pKW plasmid were transformed into BL21(DE3) cells with pTXB1_CaM-L112TAG_Mxe-His₆ plasmid (previously described)⁴¹. The pKW plasmid (spectinomycin resistance, p15a origin of replication) expresses the Ma RS under the control of a constitutive GlnS promoter and the Ma tRNA^Pyl with a constitutive lpp promoter. Proteins were expressed in AIM at 30 °C for 20 h, at which point the cells were harvested by centrifugation, resuspended in buffer and lysed by sonication. Insoluble debris was pelleted and then the soluble supernatant fraction was collected and the proteins purified with a Ni²⁺-NTA column following standard procedures. After elution, the protein was subjected to intein cleavage with β-mercaptoethanol. Protein production and cleavage yields were analyzed by SDS PAGE using a Syngene G:Box Mini gel imager. Protein yields were based on Acd fluorescence of the lysate via long-UV excitation and scaled to yields of CaM expressed with the Mj Acd RS A9 which was performed in parallel.¹⁵

Mass spectrometry.

Purified proteins were buffer exchanged into LC-MS grade water or 50 mM ammonium bicarbonate with PD-10 columns, diluted to 50 μM and analyzed via Waters Synapt G2 Mass Spectrometer at the Mass Spectrometry Facility at Oregon State University. The deconvoluted masses were obtained by using Waters MassLynx MaxEnt1 software.

Cloning Ma pAcBac1 plasmids.

The previously described Mb pAcBac1 plasmid expresses the tRNA synthetase (or the target protein) under a CMV transcriptional promoter along with four copies of the amber suppressor Mb tRNA^Pyl.⁴² To create the Ma pAcBac1 plasmids, first the four copies of Mb tRNA^Pyl_CUA were replaced with the Ma tRNA^Pyl(6), and then the Mb PylRS was replaced with either Ma PylRS variants or the target protein (sfGFP wt, sfGFP-150TAG or α-synuclein-112TAG-sfGFP). All cloning for the pAcBac1 plasmids were performed using NEB Stable cells to ensure plasmid stability, and for every propagation the integrity of the plasmid DNA was confirmed by analytical restriction digests. Expanded methods for assembling the Ma pAcBac1 plasmid are included in the Supporting Information.

Expression of proteins with Acd in HEK293T cells.

Transfection and expression.

HEK293T cells were plated in a 24-well plate at about 40% confluency so that they reach 70% ~ 90% confluency at the time of transfection. Cells were transfected with plasmid DNA using JetPrime reagent (PolyPlus). For Acd incorporation, the Ma pAcBac1-GOI plasmid (where GOI was either sfGFP-wt, sfGFP-150TAG, or α-synuclein-114-sfGFP) and the Ma pAcBac1-RS (where RS was any of the Ma Acd RS’s) were mixed prior to transfection in a 3:1 molar ratio, respectively, for a total of 600 ng total plasmid per transfection. For incorporation of Acd using the Mb platform, the Mb pAcBac1-GOI plasmid and Mb pAcBac1-NES-Acd 82¹⁵ plasmid were used. For co-expressing eRF1 E55D²³, 150 ng of plasmid was co-transfected with 337.5 ng pAcBac1-GOI and 112.5 ng pAcBac1-RS so that their ratio was maintained at 3:1, respectively, and 600 ng total plasmid DNA was transfected. Acd was added at the indicated concentrations from a 30 mM stock 2–3 hours after transfection. To prepare this stock solution, an initial 60 mM Acd stock solution was prepared by mixing Acd with water and an equal molar ratio of NaOH, and then diluting it 1:1 with 1 M HEPES pH 7.5 to produce a 30 mM Acd stock solution at physiologic pH. Cells were incubated for 24 to 48 hrs.

Flow Cytometry Assessment.

HEK293T cells were transfected as described above and grown for 24 h. sfGFP expression was confirmed by fluorescence microscopy. Adherent cells were washed once with PBS and dissociated into single cells by 0.05% trypsin/0.53 mM EDTA incubation. Cells were washed with PBS and 10,000–20,000 events were collected and analyzed by flow cytometry using CytoFLEX flow cytometer and CytExpert software version 2.2 (Beckman Coulter). sfGFP signal was collected on FL1-H FITC channel. The basal level of GFP signal was determined from untransfected cells or no ncAA control cells. Any cells that have higher level of GFP than the background level were defined to be a sfGFP expressing cell. The mean fluorescence intensity (MFI) of sfGFP was calculated using CytExpert software.

In Vitro Aminoacylation Analysis.

Ma PylRS kinetic analyses were carried out largely as described previously.¹⁹ Expanded methods are included in the Supporting Information.