Development of mammalian cell logic gates controlled by unnatural amino acids

Summary Mammalian cell logic gates hold great potential for wide-ranging applications. However, most of those currently available are controlled by drug(-like) molecules with inherent biological activities. To construct truly orthogonal circuits and artificial regulatory pathways, biologically inert molecules are ideal molecular switches. Here, we applied genetic code expansion and engineered logic gates controlled by two biologically inert unnatural amino acids. Genetic code expansion relies on orthogonal aminoacyl-tRNA synthetase/tRNA pairs for co-translational and site-specific unnatural amino acid incorporation conventionally in response to an amber (UAG) codon. By screening 11 quadruplet-decoding pyrrolysyl tRNA variants from the literature, we found that all variants decoding CUAG or AGGA tested here are functional in mammalian cells. Using a quadruplet-decoding orthogonal pair together with an amber-decoding pair, we constructed logic gates that can be successfully controlled by two different unnatural amino acids, expanding the scope of genetic code expansion and mammalian cell logic circuits.


In brief
Logic gates enable regulation of protein function by small molecules, but these are often drug(-like) molecules with intrinsic biological activities. Mills et al. present an alternative approach through genetic code expansion, demonstrating that unnatural amino acids can act as biologically inert switches for effective mammalian cell logic operations.

INTRODUCTION
Logic gates refers to integrated systems where an input controls a desired output. Employing logic gates in mammalian cells is a fast-developing area for intricate reversible cellular control, with significant biotechnological and biomedical applications (Cuthbertson and Nodwell, 2013;Fink et al., 2019;Nguyen et al., 2016;Ross et al., 2016;Wu et al., 2015;Zetsche et al., 2015). Some examples include novel sensors, diagnostics, as well as therapeutics (Brown et al., 2018;Kitada et al., 2018;Scheller and Fussenegger, 2019;Singh, 2014;Zhou et al., 2020). In particular, logic gates responding to small molecules can provide (spatio)temporal control by the user, permitting targeted manipulation of a designated phenotype. Specifically, biologically inert molecules that have no influence on endogenous cellular events are ideal molecular switches to control logic gates, therefore enabling the construction of truly orthogonal circuits and artificial regulatory pathways in cells.
We hypothesized that this can be realized by using biologically inert unnatural (non-canonical) amino acids, which are artificial synthetic molecules and do not produce observable phenotypes or toxicities in vitro or in vivo (Chen et al., 2017;Han et al., 2017;Krogager et al., 2018;Liu et al., 2017;Suzuki et al., 2018). More importantly, such unnatural amino acids can be site-specifically incorporated into proteins in mammalian cells by repurposing the cellular translational machinery through the technique of genetic code expansion (Figure 1). This technique has wideranging applications in protein research (Chin, 2017;de la Torre and Chin, 2021;Dumas et al., 2015;Huang and Liu, 2018;Kato, 2019;Neumann-Staubitz and Neumann, 2016;Nodling et al., MOTIVATION Genetic code expansion employing an orthogonal aminoacyl-tRNA synthetase/tRNA pair for site-specific unnatural amino acid incorporation has a wide range of applications. However, a blank (orthogonal) codon is needed for each orthogonal tRNA. Although the amber stop codon is commonly used as the blank codon, the use of stop codons as the blank codons inevitably limits the number of possible different unnatural amino acids to be incorporated into proteins in a single cell. To overcome this limitation, we evaluated the capability of 11 Pyl tRNA variants for decoding quadruplet codons in mammalian cells. With efficient quadruplet-decoding orthogonal tRNA variants available, it is possible for simultaneous incorporation of more than three different unnatural amino acids in mammalian cells. 2019), including the proposed use of unnatural amino acids as logic gate inputs. To incorporate an unnatural amino acid into a protein during translation, an orthogonal aminoacyl-tRNA synthetase (aaRS)/tRNA pair that selectively decodes a blank codon is required. The amber stop codon (UAG) is often used as the blank codon because it does not encode an amino acid and is the rarest codon in many organisms (Ivanov et al., 2001). The use of the amber codon to encode an unnatural amino acid is known as amber suppression ( Figure 1A). To date, various amber suppressor orthogonal pairs have been developed (Chin, 2017;Dumas et al., 2015;Huang and Liu, 2018;Neumann-Staubitz and Neumann, 2016;Nodling et al., 2019). Within prokaryotic and eukaryotic systems, pyrrolysyl (Pyl)-tRNA synthetase (PylRS) and its cognate tRNA from the archaeal Methanosarcina species is arguably one of the most widely utilized and developed orthogonal pairs for unnatural amino acid Charged tRNA molecules are generated by aminoacylation of tRNA molecules by their respective synthetases with the respective amino acid or unnatural amino acid. Charged tRNA molecules then enter the ribosome and interact with the complementary codon on the mRNA template before formation of the peptide bond with the brought-in amino acid. Abbreviations are as follows: AA, amino acid; UAA, unnatural amino acid. (B) Graphical representation of aaRS/tRNA pair orthogonality: (i) aaRS can only charge their paired tRNA with the respective amino acid, supplementation of a different amino acid results in no aminoacylation of tRNA; (ii) tRNA can only be aminoacylated by their respective aaRS, regardless of amino acid supplementation. (C) Principles imposed when using quadruplet decoding. Alongside the introduction of the quadruplet codon to the gene, a further point mutation is introduced downstream to the codon to ensure that, if the first three bases of the quadruplet are translated by the canonical triplet tRNA, an in-frame stop codon is generated downstream to the incorporation site enabling premature truncation. Moreover, if the quadruplet is correctly decoded, this point mutation remains silent and does not alter the overall amino acid sequence of the protein.
incorporation. Nevertheless, various Escherichia coli aaRS enzymes (e.g., EcLeuRS, EcTyrRS, EcTrpRS) and their cognate tRNAs have also been exploited and specifically engineered to function as amber suppressor orthogonal pairs in mammalian cells. It is possible to employ several orthogonal pairs for simultaneous incorporation of multiple unnatural amino acids in a single mammalian cell (Meineke et al., 2018(Meineke et al., , 2020Xiao et al., 2013;Zheng et al., 2017Zheng et al., , 2018. However, such applications are often thwarted as these orthogonal pairs were originally engineered for amber decoding. To attain site-specific incorporation of multiple unnatural amino acids within a single cell, orthogonal pairs need to decode unique blank codons and have no crossover interactions ( Figure 1B). An emerging strategy to generate additional blank codons is to use quadruplet (four-base) codons. Employing quadruplet codons can greatly expand the genetic code, providing the possibility for simultaneous incorporation of three or more distinct unnatural amino acids within a single cell (Dunkelmann et al., 2020).
Here, we recognized the potential applications of genetic code expansion in the development of orthogonal logic circuits and engineered novel AND and OR logic gates that can be controlled by two distinct unnatural amino acids. As this proposed system design required two mutually orthogonal aaRS/tRNA pairs to respond to the logic inputs, we set out to identify a quadrupletdecoding aaRS/tRNA pair that is orthogonal to an amber-decoding aaRS/tRNA pair in mammalian cells. Specifically, we investigated the quadruplet-decoding efficiency of 11 Pyl tRNA XXXX variants previously tested in E. coli and found that all selected variants decoding AGGA (Niu et al., 2013;Wang et al., 2014) or CUAG (Wang et al., 2014) were functional in human embryonic kidney (HEK293) cells, whereas none of the tested variants were able to decode UAGN codons efficiently. We then confirmed the orthogonality of an amber-decoding pair to an AGGA-decoding PylRS/tRNA pair, enabling incorporation of two different unnatural amino acids in the same mammalian cell. After a series of system optimizations, we constructed AND and OR logic gates by using a split GFP reporter assay. To the best of our knowledge, this is the first example exploiting genetic code expansion in mammalian cell logic gates. This study not only expands the applications of genetic code expansion but also provides an alternative approach to engineer mammalian cell logic circuits.

Quadruplet-decoding analysis
Pyl tRNA variants that can decode UAGN (N = A/U/C/G), CUAG, or AGGA codons have been engineered for unnatural amino acid incorporation in E. coli (Chen et al., 2018;Niu et al., 2013;Wang et al., 2014Wang et al., , 2016. However, the translation machineries in E. coli and mammalian cells are different (Melnikov et al., 2018), thus tRNA optimized in one system might not work well in the other. In light of this, we evaluated the performance of 11 quadrupletdecoding Pyl tRNA variants in mammalian cells (Figure 2A). These tRNA variants consisted of the simple replacement of the CUA anticodon to NCUA (for decoding UAGN), CUAG (for decoding CUAG), or UCCU (for decoding AGGA), as well as previously identified tRNA variants that were evolved in E. coli and carry additional mutations in the anticodon stem loop. Pyl tRNA UCUA(Ev1) , Pyl tRNA CUAG(Ev1) , and Pyl tRNA UCCU(Ev1) were chosen for their high decoding efficiency in E. coli (Wang et al., 2014), whereas Pyl tRNA UCUA(Ev2) (Chen et al., 2018) and Pyl tRNA UCCU(Ev2) (Niu et al., 2013) have been used for unnatural amino acid incorporation in mammalian cells.
Nε-Boc-L-lysine (BocK) ( Figure 2B) is a substrate of the wildtype PylRS (Yanagisawa et al., 2008). To perform a systematic comparison, we constructed dual-fluorescence reporters (Figure 2C) encoding the wild-type Methanosarcina mazei PylRS, a Pyl tRNA variant and an mCherry-P2A-eGFP(150XXXX) reporter, where the 150 th amino acid residue in eGFP corresponds to a quadruplet codon. P2A is a self-cleavage sequence, leaving only a Pro residue in the C-terminal fragment after cleavage (Kim et al., 2011). In this reporter, mCherry is produced constitutively and thus serves as the transfection control. On the other hand, green fluorescence will only be observed upon successful decoding of the quadruplet codon and production of full-length eGFP. If the first three bases of a quadruplet codon are decoded as a triplet codon, this will lead to a translational frameshift and premature termination ( Figure 1C). Thus, quadruplet-decoding efficiency can be calculated from the ratio between the two fluorescence intensities in the presence and absence of BocK (Bartoschek et al., 2021;Gautier et al., 2010;Monk et al., 2017;Potts et al., 2020;Schmied et al., 2014).
Experimentally, HEK293 cells were transiently transfected with a reporter vector ( Figure 2C) and cultured in the presence or absence of 1 mM BocK, followed by flow cytometry analysis. Interestingly, in contrast to the observations in E. coli where all variants can decode the corresponding quadruplet codon (Niu et al., 2013;Wang et al., 2014Wang et al., , 2016, only Pyl tRNA variants decoding CUAG or AGGA appear to be functional ( Figure 2D). Particularly, Pyl tRNA CUAG and Pyl tRNA UCCU(Ev2) outperformed the other variants, and the functional Pyl tRNA variants remained orthogonal in mammalian cells, manifesting as a lack of detectable eGFP in the absence of BocK by flow cytometry, fluorescence microscopy, or immunoblotting ( Figures 2E and S1). Interestingly, a higher production of eGFP was observed with Pyl tRNA UCCU(Ev2) compared with Pyl tRNA CUAG when using a single fluorescence eGFP(150XXXX) reporter ( Figure S1), although these findings could be due to a higher translation frequency or better transfection efficiency of the vector. Nevertheless, we chose to use tRNA UCCU(Ev2) for subsequent experiments.

Identification of a second orthogonal pair
To utilize genetic code expansion to control logic gate inputs, two orthogonal aaRS/tRNA pairs that selectively incorporate different unnatural amino acids are required. Several E. coli aaRS enzymes (e.g., EcTyrRS) and their cognate tRNA  (Italia et al., 2018;Nodling et al., 2019). As PylRS/ tRNA can be used as an orthogonal pair in E. coli, it should remain orthogonal to engineered EcTyrRS/tRNA CUA pairs. To verify this, an EcTyrRS variant (referred to hereafter as TyrRS*) containing Y37V, D182S, F183M, and G265R mutations was chosen. TyrRS* was reported to be able to incorporate a range of unnatural amino acids, including O-methyl-L-tyrosine (OMeY) and 4-azido-L-phenylalanine (AzF), into proteins in mammalian cells .
For systematic comparison, a reporter vector following the same design as the reporter vector for Pyl tRNA variants was generated, in which PylRS was replaced by TyrRS* and Pyl tRNA by Tyr tRNA CUA ( Figure 3A). Although TyrRS* was (D) BocK incorporation efficiency of each tRNA XXXX variant. HEK293 cells were transiently transfected with a reporter vector and incubated in the presence or absence of 1 mM BocK. Cells underwent flow-cytometry analysis and the equation displayed was used to calculate incorporation efficiency (Bartoschek et al., 2021). Means and standard deviations calculated from three biological replicates are shown. (E) Representative flow cytometry results used for calculating BocK incorporation efficiency. Events were gated for HEK293 cells and to exclude doublets. Cells were then gated to include only transfected (i.e., mCherry-positive) cells. Fluorescence intensities are given in arbitrary units (a.u.). The geometric mean of fluorescence intensity was used to calculate the incorporation efficiency. See also Figure S1 for fluorescence imaging and immunoblotting results of a single fluorescence eGFP(150XXXX) reporter.
shown to incorporate both OMeY and AzF efficiently , we observed higher incorporation efficiency of AzF than OMeY by TyrRS* ( Figure 3A) and the observable difference seemed to be more prominent in fluorescence microscopy analysis (Figure S2A). Further, AzF incorporation by TyrRS* was greater compared with an EcTyrRS variant (Takimoto et al., 2009) evolved specifically for AzF incorporation ( Figure S2B). We thus continued with AzF and TyrRS*/tRNA CUA to confirm the orthogonality with PylRS/tRNA UCCU(Ev2) , and showed that the two pairs specifically recognized their respective unnatural amino acids ( Figures 3B and S2C). We then transfected HEK293 cells with two vectors, one carrying only the TyrRS*/tRNA CUA machinery, and the other carrying the PylRS/tRNA UCCU(Ev2) pair alongside a double-incorporation reporter eGFP(40TAG,150AGGA) that contains a TAG codon for the 40 th amino acid residue and an AGGA codon for the 150 th residue in the eGFP gene. For 24 h, cells were maintained under four distinct conditions: (1) no unnatural amino acid, (2) BocK only, (3) AzF only, and (4) both BocK and AzF. Fluorescence was only detected when both unnatural amino acids were present ( Figure S3), confirming the orthogonality of the two pairs as well as demonstrating their use for double incorporation into the same protein.

Optimization of unnatural amino acid incorporation
To use unnatural amino acids as logic gate inputs, their incorporation efficiency would directly affect the logic gate responses and so a higher incorporation efficiency is preferable. We first attempted to improve the incorporation efficiency by overexpression of an engineered eukaryotic release factor 1 (eRF1). With an E55D point mutation, this engineered eRF1 was shown to improve amber suppression by Pyl tRNA CUA(U25C) (Li et al., 2020;Schmied et al., 2014). We therefore tested the incorporation efficiency of both aaRS/tRNA pairs here in the presence and absence of this engineered eRF1; however, no significant improvement was observed in either case ( Figure S2D). It has also been reported that increasing the copy number of orthogonal tRNA can improve the unnatural amino acid incorporation efficiency in mammalian cells Schmied et al., 2014). Thus, we constructed vectors carrying four copies of tRNA (n = 1 or n = 4; Figure 3C), and both PylRS/ tRNA UCCU(Ev2) and TyrRS*/tRNA CUA pairs demonstrated a significant increase in eGFP production (p < 0.005) (Figures 3D and 3E). Moreover, by increasing the tRNA and reporter copy numbers, a marked increase in eGFP production was also observed in the double-incorporation experiment, as demonstrated by fluorescence microscopy and immunoblotting ( Figure S3).

Construction of logic gates controlled by two unnatural amino acids
For proof of concept, we then applied PylRS/tRNA UCCU(Ev2) and TyrRS*/tRNA CUA pairs to construct logic gates controlled by two unnatural amino acids, BocK and AzF. The logic gates were designed by using a split GFP system, which requires the concurrent presence of two complementary polypeptides sGFP(1-10) and sGFP(11) to form a fluorescent complex (Cabantous et al., 2005;Kamiyama et al., 2016). When produced individually, the GFP chromophore is unable to assemble, and no fluorescence can be detected (Figures 4A and 4B). In short, the designed logic gates Cell Reports Methods 1, 100073, October 25, 2021 5 Article ll OPEN ACCESS have two inputs (i.e., BocK and AzF) and one output (i.e., GFP fluorescence). To test them, we created a two-vector system based on the dual-fluorescence reporters (Figures 2 and 3) whereby there is an N-terminal mCherry leading to the sGFP fragments via a P2A sequence ( Figures 4A and 4B). These vectors would hence allow for control of sGFP fragment production via selected introduction of TAG or AGGA codon upstream of the sGFP sequences.
To generate a logic gate that performs AND operation, we inserted TAG and AGGA codons into the reporter genes before the sequences encoding sGFP(1-10) and sGFP(11), respectively ( Figure 4C). In this case, production of sGFP(1-10) and sGFP(11) became strictly dependent on the presence of AzF and BocK concurrently, and only in the presence of both AzF and BocK can the fluorescent complex form.
HEK293 cells were transiently transfected with two vectors ( Figure 4C). One vector encoded the TyrRS*/tRNA CUA pair and mCherry-P2A-TAG-sGFP(1-10), whereas the other vector encodes PylRS/tRNA UCCU(Ev2) and mCherry-P2A-AGGA-sGFP(11). As expected, cells with green fluorescence were only observed in the presence of both AzF and BocK, and the level of GFP fluorescence indicates a promising efficiency of the operation.
Using a similar concept, we constructed an OR logic gate ( Figure 4D). In this case, sGFP(1-10) was constantly produced, but sGFP(11) was only generated in the presence of either AzF, BocK, or both. Thus, green fluorescence is observed when at least one unnatural amino acid is present, and again the GFP fluorescence observed indicates a promising operational efficiency.
Finally, to confirm the specificity of BocK/AzF dependence and continuation of orthogonality, the vectors used for logic testing were individually transfected into HEK293 cells and incubated in the presence or absence of their respective unnatural amino acids. Upon imaging, no green fluorescence was detected regardless of unnatural amino acid supplementation for AND logic vectors ( Figure S4). Indeed, green fluorescence was only detected with the OR logic vector containing TyrRS*/ tRNA CUA and in the presence of AzF.

DISCUSSION
Here, we have demonstrated the applications of genetic code expansion for developing novel mammalian cell logic gates.
We quantified the quadruplet-decoding efficiency of 11 Pyl tRNA variants in HEK293 cells. Unlike most other aaRS enzymes, PylRS does not recognize the anticodon stem loop on the Pyl tRNA during aminoacylation (Suzuki et al., 2017;Tharp et al., 2018), so alteration of the anticodon is theoretically tolerated by the synthetase. Nevertheless, whereas change of the anticodon from CUA to CUAG or UCCU was tolerated, Pyl tRNA variants with the anticodon NCUA did not seem to be functional in mammalian cells (Figures 2 and S1). In addition, although variation in the nucleic acid sequence of the anticodon stem loop can significantly affect the incorporation efficiency, we found that tRNA sequences optimized in E. coli (Niu et al., 2013;Wang et al., 2014Wang et al., , 2016 might not necessarily work well in mammalian cells. For example, Pyl tRNA CUAG outperformed Pyl tRNA CUAG(Ev1) .
Successful incorporation of an unnatural amino acid involves four steps: (1) cellular uptake of the designated unnatural amino acid; (2) aminoacylation of the orthogonal tRNA by the orthogonal synthetase; (3) formation of a ternary complex containing aminoacylated tRNA, eEF1A (or EF-Tu in bacteria), and GTP; and (4) interaction of the ternary complex with the ribosome, followed by peptide bond formation. The high incorporation efficiency of BocK by Pyl tRNA CUA(U25C) indicates that BocK can be readily taken up by mammalian cells. Additionally, all tested Pyl tRNA variants should be able to be aminoacylated by PylRS given that all variants were reported to be functional in E. coli. Thus, the differences in the prokaryotic and eukaryotic translational machinery (Melnikov et al., 2018) are likely the cause of the observed discrepancy in decoding efficiency, although it remains elusive whether formation of the ternary complex or interaction of the ternary complex with the ribosome is the main cause.
Intriguingly, negligible eGFP production was observed with Pyl tRNA UCUA and Pyl tRNA UCUA(Ev2) . These two variants were reported to be functional in mammalian cells (Chen et al., 2018), although under different experimental conditions. Thus, the discrepancy might result from the choice of the cell line (HEK293 versus HEK293T), BocK concentration (1 mM versus 5 mM), unnatural amino acid incorporation site (40 th versus 150 th residue in eGFP), or a combination of these factors.
To construct double-input logic gates, another orthogonal pair, in addition to PylRS/tRNA UCCU(Ev2) , is required. We chose the E. coli-derived TyrRS*/tRNA CUA as it had been optimized for incorporating a range of unnatural amino acids in mammalian cells . Given that PylRS/tRNA can be used as an orthogonal pair in E. coli, it should remain orthogonal to TyrRS*/tRNA CUA in mammalian cells. Indeed, each pair specifically recognized their respective unnatural amino acids (Figure 3B). Furthermore, they can function orthogonally in the same cell, as demonstrated in the double-incorporation experiment ( Figure S3). It is noteworthy that PylRS and TyrRS* can mediate incorporation of different unnatural amino acids, including cyclopropene lysine and propargyl tyrosine (Nodling et al., 2019) for bioorthogonal labeling, so the two orthogonal pairs can further be used for dual modification of proteins.
To use the two pairs in controlling logic gate inputs, higher incorporation efficiency of the two employed unnatural amino acids is ideal. Co-expression of an engineered eRF1 was previ-ously shown to increase the unnatural amino acid incorporation efficiency by PylRS/tRNA CUA(U25C) in mammalian cells (Schmied et al., 2014). However, co-expression of the engineered eRF1 did not appear to improve AzF incorporation by TyrRS*/tRNA CUA or BocK incorporation by PylRS/tRNA UCCU(Ev2) ( Figure S2D). This discrepancy could be due to the difference in the employed orthogonal pairs. In addition, the beneficial effect was most prominent with genes containing multiple UAG codons in the original study (Schmied et al., 2014), whereas we only used single UAG or AGGA codons in our study. Nevertheless, improvement in AzF and BocK incorporation was observed when increasing the tRNA copy number from one to four ( Figures 3D  and 3E), a strategy that has been shown to work previously Schmied et al., 2014) Finally, we employed the two orthogonal aaRS/tRNA pairs to construct logic gates and demonstrated selective AND and OR logic operations dependent on AzF and BocK supplementation. These designs demonstrate that unnatural amino acids can be used as molecular switches in synthetic biology logic gates. The use of biologically inert unnatural amino acids (Chen et al., 2017;Han et al., 2017;Krogager et al., 2018;Liu et al., 2017;Suzuki et al., 2018) is ideal for constructing orthogonal, artificial pathways for synthetic biological applications with no identifiable side effects or non-specific activities. Moreover, the wide substrate scope of PylRS and TyrRS* (Nodling et al., 2019) enables fine-tuning the logic gate response for different applications.
To the best of our knowledge, this is the first example of unnatural amino acid-based logic gates in mammalian cells. Further exploration of quadruplet-decoding orthogonal tRNA in eukaryotic cells, as well as the underlying mechanism for their different performance in bacterial and eukaryotic systems, will provide new opportunities and applications for genetic code expansion. This will facilitate not only our understanding of protein structure and function but also the discovery and development of new diagnostics and therapeutics to target human diseases.

Significance
Mammalian cell logic gates remain heavily limited by the current approaches that involve the use of drug(-like) molecules that carry undesirable biological activities. In light of this, here we developed an alternative approach to mammalian cell logic gates via the simple process of genetic code expansion. By repurposing the cell translational machinery, we demonstrated the use of biologically inert unnatural amino acids as input switch molecules for effective and efficient mammalian cell logic operations, enabling selective and intrinsic control. The results shown here simultaneously expand the scope of genetic code expansion, while providing an exciting approach to mammalian cell logic gates.

Limitations of the study
Although this study expands the scope of mammalian cell logic circuits, it is important to note that this investigation employed human embryonic kidney (HEK293) cells and the tRNA variants might perform differently in other cell lines. Additionally, as with all genetic code expansion applications, amber and quadruplet suppression might not be limited to the gene of interest and

DECLARATION OF INTERESTS
The authors declare no competing interests.

RESOURCE AVAILABILITY
Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Yu-Hsuan Tsai (Tsai.Y-H@outlook.com). Data and code availability d All FACS data have been deposited to the Cardiff University data catalogue and are publicly available as of the date of publication. DOIs are listed in the key resources table. Microscopy data reported in this paper will be shared by the lead contact upon request. d This paper does not report original code. d Any additional information required to reanalyse the data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND SUBJECT DETAILS
HEK293 cells were obtained from ECACC (ECACC General Collection, #85120602) and routinely tested for mycoplasma infection. Cells were maintained in T75 flasks at 37 C in a 5% CO 2 high-humidity atmosphere in Dulbecco's Modified Eagle Medium (DMEM) with GlutaMAX supplement (Fisher Scientific, #11574516) supplemented with 10% (v/v) fetal bovine serum (FBS; Fisher Scientific, #11573397). Cells were maintained at a sub-confluent monolayer and split at 80-85% confluency. For splitting, cells were washed with Dulbecco's phosphate buffered saline (DPBS; Sigma, #D1408), detached in 1 mL of 0.25% trypsin-EDTA (Fisher Scientific, #11560626) and 200 mL of the 1000 mL trypsin cell suspension was re-suspended in 12 mL fresh DMEM containing 10% (v/v) FBS in a new T75 flask.

METHOD DETAILS
Vector cloning All Vector stocks were generated and maintained via transformation of chemically competent One Shot TM Stbl3 TM E. coli cells from ThermoFisher (ThermoFisher, #C737303) and isolated from 5 mL cultures using QIAprep Spin Miniprep Kit (Qiagen, #27106) following the manufacturer's protocol. Where appropriate, restriction digests were conducted using FastDigest restriction enzymes (Thermo Scientific) following manufacturer's protocol. PCR was conducted in a thermocycler using PrimeSTAR Max (Takara, #R045A) following the manufacturer's protocol. Restriction digests and PCR products were electrophoresed on 1% agarose gels in TAE buffer (40 mM Tris pH 7.6, 20 mM acetic acid, 1 mM EDTA) and visualised using SYBRSafe DNA stain (Invitrogen, #S33102). Desired bands were excised from the gel and extracted using QIAquick Gel Extraction Kit (Qiagen, #28704) following the manufacturer's protocol. T4 ligations were performed using T4 DNA Ligase (ThermoFisher, #EL0014) and Gibson assembly via NEBuilder HiFi DNA Assembly (New England Biolabs, #E5520) following respective manufacturers protocols. All constructs were confirmed via Sanger sequencing. A list of primers used in this study is provided in Table S1. Vector 1: Pyl tRNA CUA(U25C) This Vector was purchased from GeneArt (ThermoFisher). The full DNA sequence can be found in the supplemental information. Vector 6: eGFP(150CTAG) Cloned by PCR mutagenesis of Vector 4 using Primer 3 and Primer 5. Template Vector 4 removed from reaction mixture via DpnI restriction enzyme digest. Vector 7: eGFP(150AGGA) Cloned by PCR mutagenesis of Vector 4 using Primer 3 and Primer 6. Template Vector 4 removed from reaction mixture via DpnI restriction enzyme digest. Vector 8: eGFP(150TAGT) Cloned by PCR mutagenesis of Vector 4 using Primer 3 and Primer 7. Template Vector 4 removed from reaction mixture via DpnI restriction enzyme digest. Vector 9: eGFP(150TAGC) Cloned by PCR mutagenesis of Vector 4 using Primer 3 and Primer 8. Template Vector 4 removed from reaction mixture via DpnI restriction enzyme digest. Vector 10: eGFP(150TAGG) Cloned by PCR mutagenesis of Vector 4 using Primer 3 and Primer 9. Template Vector 4 removed from reaction mixture via DpnI restriction enzyme digest. Vector 11: Pyl tRNA CUA(U25C) -PylRS-eGFP(150TAGA) Vector generated via digestion of Vector 3 with the restriction enzymes BglII and XbaI. Insert generated via digestion of Vector 5 with BglII and XbaI. The vector and insert fragments were assembled by T4 ligation. Vector 12: Pyl tRNA CUA(U25C) -PylRS-eGFP(150CTAG) Vector generated via digestion of Vector 3 with restriction enzymes BglII and XbaI. Insert generated via digestion of Vector 6 with BglII and XbaI. The vector and insert fragments were assembled by T4 ligation. Vector 13: Pyl tRNA CUA(U25C) -PylRS-eGFP(150AGGA) Vector generated via digestion of Vector 3 with restriction enzymes BglII and XbaI. Insert generated via digestion of Vector 7 with BglII and XbaI. The vector and insert fragments were assembled by T4 ligation. Vector 14: Pyl tRNA CUA(U25C) -PylRS-eGFP(150TAGT) Vector generated via digestion of Vector 3 with the restriction enzymes BglII and XbaI. Insert generated via digestion of Vector 8 with BglII and XbaI. The vector and insert fragments were assembled by T4 ligation. Vector 15: Pyl tRNA CUA(U25C) -PylRS-eGFP(150TAGC) Vector generated via digestion of Vector 3 with the restriction enzymes BglII and XbaI. Insert generated via digestion of Vector 9 with BglII and XbaI. The vector and insert fragments were assembled by T4 ligation. Vector 16: Pyl tRNA CUA(U25C) -PylRS-eGFP(150TAGT) Vector generated via digestion of Vector 3 with the restriction enzymes BglII and XbaI. Insert generated via digestion of Vector 10 with BglII and XbaI. The vector and insert fragments were assembled by T4 ligation. Vector 17: Pyl tRNA UCUA Cloned by PCR mutagenesis of Vector 1 using Primer 10 and Primer 11. The template Vector 1 was removed from the reaction mixture via DpnI restriction enzyme digest. Vector 18: Pyl tRNA UCUA(Ev1) Cloned by PCR mutagenesis of Vector 1 using Primer 10 and Primer 12. The template Vector 1 was removed from the reaction mixture via DpnI restriction enzyme digest. Vector 19: Pyl tRNA UCUA(Ev2) Cloned by PCR mutagenesis of Vector 1 using Primer 10 and Primer 13. The template Vector 1 was removed from the reaction mixture via DpnI restriction enzyme digest. Vector 20: Pyl tRNA CUAG Cloned by PCR mutagenesis of Vector 1 using Primer 10 and Primer 14. The template Vector 1 was removed from the reaction mixture via DpnI restriction enzyme digest. Vector 21: Pyl tRNA CUAG(Ev1) Cloned by PCR mutagenesis of Vector 1 using Primer 10 and Primer 15. The template Vector 1 was removed from the reaction mixture via DpnI restriction enzyme digest. Vector 22: Pyl tRNA UCCU Cloned by PCR mutagenesis of Vector 1 using Primer 10 and Primer 16. The template Vector 1 was removed from the reaction mixture via DpnI restriction enzyme digest. Vector 23: Pyl tRNA UCCU(Ev1) Cloned by PCR mutagenesis of Vector 1 using Primer 10 and Primer 17. The template Vector 1 was removed from the reaction mixture via DpnI restriction enzyme digest. Vector 24: Pyl tRNA UCCU(Ev2) Cloned by PCR mutagenesis of Vector 1 using Primer 10 and Primer 18. The template Vector 1 was removed from the reaction mixture via DpnI restriction enzyme digest.

QUANTIFICATION AND STATISTICAL ANALYSIS
Flow cytometry data was analysed with FlowJo (version 10.7.2, BD Biosciences). Events were gated to ensure only mCherry-positive, single cells were used for quantification purposes. This was achieved by gating for single cells, excluding debris (FSC-A vs SSC-A) and gating to exclude doublets (FSC-A vs FSC-H and SSC-A vs SSC-H). Cells were gated for transfected cells by excluding mCherry negative cells (mCh-A vs FSC-A). Finally, mCherry fluorescence was plotted against eGFP fluorescence and the mean fluorescence intensities were taken to calculate the incorporation efficiency using the formula (Bartoschek et al., 2021): Where ( ) and [ ] indicate samples incubated with and without the designated unnatural amino acid, respectively. Each condition was performed in triplicate (i.e. transient transfection of cells of different passages and split at different time). Data of each replicate containing over 10,000 mCherry positive cells were used to calculate the incorporation efficiency. Average incorporation efficiency was calculated from the three replicates for each condition and are provided in bar charts in Figures 2D, 3D, and 3E. The standard deviation from the means were also calculated and are represented as error bars on Figures 2D, 3D, and 3E and after ± symbols in Figures 3A and 3B. In the results, when directly comparing incorporation efficiencies between two vectors, an independent two-tailed t-test assuming equal variances was used. Incorporation efficiency, average incorporation efficiency, standard deviation from the mean and t-test calculations were performed in Microsoft Excel.