The ribosome installs whatever an aminoacyl-tRNA synthetase loaded onto the tRNA that pairs with the codon in the A site. Both routes to a non-canonical amino acid (ncAA) follow from that. Either you persuade one of the cell's own synthetases to accept an analogue of its substrate, making every codon for that residue a site of replacement, or you supply a synthetase/tRNA pair the host ignores and hand it a codon nothing else uses. The first is residue-specific and global, the second site-specific and positional. They fail differently, and the mass spectrometry that confirms them asks different questions.
Orthogonality is a selected ratio, not a designed switch
An orthogonal system requires a suppressor tRNA that is uniquely acylated with the desired ncAA by an orthogonal aminoacyl-tRNA synthetase [1]. Two cross-reactions have to be suppressed at once: the imported synthetase must not charge host tRNAs, and the imported tRNA must not be charged by host synthetases. Neither is guaranteed by importing a pair from a distant organism, which is why the imported pair described below was finished by selection.
Wang and Schultz built a library of amber suppressor tRNAs derived from Methanococcus jannaschii tRNATyr and ran selections in both directions. A negative selection based on suppression of amber nonsense mutations in the barnase gene deleted from the pool those tRNATyrCUA variants that are substrates for endogenous E. coli synthetases; survivors were then selected for the ability to suppress amber codons in the β-lactamase gene in the presence of the cognate M. jannaschii tyrosyl-tRNA synthetase [1]. Four mutant suppressor tRNAs emerged that are poorer substrates for E. coli synthetases than the parent tRNATyrCUA but can still be charged efficiently by M. jannaschii TyrRS [1]. Poorer, not incapable: orthogonality delivered by selection is a ratio between cognate and non-cognate charging, and residual non-cognate charging is one route by which a canonical amino acid could reach your reassigned codon — something to measure rather than assume away.
An early amber suppression of this kind in E. coli incorporated O-methyl-L-tyrosine in response to an amber nonsense codon, with fidelity of translation greater than 99% as determined by analysis of dihydrofolate reductase containing the unnatural amino acid [2]. Keep that number with its referent: greater than 99% was determined for that amino acid, in that protein, in that host. It is not a constant of the method and does not transfer to another synthetase variant or another site.
The second workhorse pair was found rather than evolved. In Methanosarcina barkeri, pyrrolysine — a lysine derivative — is encoded by UAG in methylamine methyltransferase genes; the pylT gene encodes an unusual tRNA with a CUA anticodon, and the adjacent pylS gene encodes a class II aminoacyl-tRNA synthetase that charges the pylT-derived tRNA with lysine but is not closely related to known lysyl-tRNA synthetases [3]. Srinivasan and colleagues describe charging a tRNACUA with lysine as a likely first step in translating UAG as pyrrolysine in certain methanogens, and read their results as indicating pyrrolysine is the 22nd genetically encoded natural amino acid [3]. A synthetase sitting that far from the canonical lysyl-tRNA synthetases is what orthogonality needs.
Amber suppression is a race against release factor 1
Reassigning a sense codon would corrupt every protein that uses it, so site-specific incorporation is hung on a stop codon, usually UAG. That puts the charged orthogonal tRNA in direct competition with termination. Johnson and colleagues frame it with a hedge worth preserving: stop codons have been exploited for genetic incorporation of unnatural amino acids in live cells, but their low incorporation efficiency, possibly due to competition from release factors, limits the power and scope of the technology [4]. Competition is the leading explanation for poor suppression, not something separately measured in every system that suffers from it.
Removing the competitor tests that explanation directly. Johnson and colleagues showed that RF1, reportedly essential, can be knocked out from E. coli by "fixing" release factor 2; the resulting strain JX33 is stable and independent, and allows UAG to be reassigned from a stop signal to an amino acid once a UAG-decoding tRNA-synthetase pair is introduced, with unnatural amino acids efficiently incorporated at multiple UAG sites in the same gene without translational termination [4]. Incorporation at endogenous UAG codons was found to depend on RF1 and on mRNA context, which the authors offer as the reason E. coli tolerates apparent global suppression of UAG [4]. The genomically recoded route removes the collateral damage too: Lajoie and colleagues replaced all known UAG stop codons in E. coli MG1655 with synonymous UAA codons, permitting deletion of release factor 1 and reassignment of UAG translation function, and the recoded organism exhibited improved properties for incorporation of nonstandard amino acids that expand the chemical diversity of proteins in vivo [5].
At the bench that buys a clean split. When suppression is poor in a conventional strain, moving the identical construct into an RF1-deleted or recoded background changes only the termination term, so the size of the improvement tells you whether you were fighting RF1 or an underpowered pair.
Suppression efficiency, and the truncated protein you also made
Multi-site incorporation in a conventional strain has a measured ceiling. The pUltra suppressor plasmid incorporates up to three unnatural amino acids within the same protein at levels up to 30% of the level of wild-type protein expression [6]. Because it carries a different origin of replication (CloDF13) and resistance marker (spectinomycin), pUltra can be run alongside the previously reported pEVOL suppressor plasmid, each encoding a distinct tRNA/synthetase pair, which is how ncAAs carrying keto and azido side chains were placed into ketosteroid isomerase and derivatised with two distinct fluorophores capable of a Förster resonance energy transfer interaction [6].
The complement of that 30% is not nothing; it is protein. Every ribosome that terminated at the amber codon released a chain stopping there, so the lysate holds a truncated species differing from the product by everything C-terminal to the suppression site. Placing the affinity tag downstream of the amber codon, so only full-length chains carry it, is the standard way to keep truncation out of purified material — and the reason a purified sample can look clean while the suppression that produced it was inefficient. Purity after such a purification reports on the tag, not on the codon.
Residue-specific replacement: the whole set or nothing
The other route never touches the codon table. Starve an auxotroph of a canonical amino acid, supply an analogue, and the host's own synthetase charges it at every codon for that residue.
Hendrickson and colleagues established an expression system for incorporation of selenomethionine into recombinant proteins produced from plasmids in E. coli, and demonstrated replacement of methionine by selenomethionine at the level of 100% for both T4 and E. coli thioredoxins, the natural proteins and the selenomethionyl variants crystallising isomorphously [7]. That 100% belongs to those two thioredoxins under those conditions — the best case for the approach, not its default.
Budisa and colleagues give the protocol in enough detail to see where it breaks. For norleucine and ethionine, E. coli B834(DE3) (hsd metB), auxotrophic for methionine, was grown under methionine-limited conditions with an excess of the analogue in the medium, and expression under a T7 promoter was induced after the methionine supply had been exhausted [8]. Conditions were optimised with respect to the frequency of appearance of revertants, high-level replacement and maximal protein yield [8] — revertants matter because an auxotroph that regains methionine synthesis manufactures its own competitor for the analogue. Telluromethionine needed a different schedule: because the factor limiting its high-level incorporation is sensitivity towards oxidation, cells were grown with limited methionine, harvested after its exhaustion, resuspended in fresh media without methionine, and only then given the analogue before induction, so that significant protein could be expressed before the telluromethionine had degraded completely, which happens within hours [8].
The same machinery accepts side chains chosen for chemistry rather than crystallography. Azidohomoalanine, a methionine surrogate, is activated by the methionyl-tRNA synthetase of E. coli and replaces methionine in proteins expressed in methionine-depleted bacterial cultures; the resulting proteins can then be selectively modified in the presence of other cellular proteins by Staudinger ligation with triarylphosphine reagents [9]. Note both halves of the bargain: handles appear wherever methionine was, and you cannot choose which one carries the label. If your question is which residue does something, this method cannot answer it.
What is worth installing site-specifically
Photo-crosslinkers convert a proximity into a covalent bond you can resolve on a gel. Benzophenones are among the most useful photocrosslinking agents in biology, and an evolved orthogonal synthetase/tRNA pair made possible the in vivo incorporation of p-benzoyl-L-phenylalanine into proteins in E. coli in response to the amber codon TAG; it was incorporated with high translational efficiency and fidelity into the dimeric protein glutathione S-transferase, and irradiation resulted in efficient crosslinking, greater than 50%, of the protein subunits [10]. That figure is a subunit-to-subunit yield for one homodimer at the position chosen, under the irradiation conditions used. The authors present the methodology as one that may prove useful for discovering and defining protein interactions in vitro and in vivo [10] — not as a yield to expect elsewhere.
Bioorthogonal handles carry a kinetic requirement that is easy to underestimate. Lang and colleagues note that many bioorthogonal reactions for which a component can be genetically encoded are too slow to effect quantitative site-specific labelling on a timescale useful for studying many biological processes, and report a fluorogenic reaction between bicyclo[6.1.0]non-4-yn-9-ylmethanol (BCN) and tetrazines that is 3–7 orders of magnitude faster than many bioorthogonal reactions [11]. Unlike reactions of strained alkenes — including trans-cyclooctenes and norbornenes — with tetrazines, the BCN–tetrazine reaction gives a single product of defined stereochemistry [11], which matters when you intend to read an intact mass as one labelled species rather than a mixture of adducts. Pairs for incorporating a BCN-containing and a trans-cyclooctene-containing amino acid were found for proteins expressed in E. coli and mammalian cells [11].
Spectroscopic probes report on local environment, and the reporting range is the specification to check first. L-(7-hydroxycoumarin-4-yl)ethylglycine has been genetically encoded in E. coli in response to the amber TAG codon; its high fluorescence quantum yield, relatively large Stokes shift and sensitivity to both pH and polarity are the properties its authors say should make it a useful probe of protein localisation and trafficking, conformational changes and protein–protein interactions [12]. A vibrational probe makes the dependence quantitative. The C≡N stretching vibrations of AlaCN and PheCN shift as much as approximately 10 cm-1 towards higher frequency when tetrahydrofuran is replaced with water, and in a peptide from the calmodulin-binding domain of skeletal muscle myosin light chain kinase carrying a single PheCN, the stretch resembled PheCN in water on the partially water-exposed polar side of the helix and PheCN in THF at a site that becomes buried [13]. That nitrile was introduced synthetically rather than encoded, and the authors say only that the results suggest the C≡N stretch can serve as a local internal environmental marker: a shift of that size tells you the probe moved between water-like and low-dielectric surroundings, not how far or what it contacts.
Cell-free removes the cell's vote
In-cell incorporation makes the host an unavoidable participant: the ncAA has to cross the membrane, survive metabolism, and avoid being a substrate for anything else. Cell-free translation removes those constraints, and the flexizyme route removes synthetase engineering with them. Flexizyme is a de novo tRNA acylation system for preparing acyl-tRNAs with a nearly unlimited selection of amino and hydroxy acids and tRNAs; combined with an appropriate cell-free translation system it allows mRNA-encoded synthesis of proteins and short polypeptides involving multiple non-natural amino acids [14]. You buy chemical scope and pay in scale.
Extract-based systems inherit the release factor problem from the strain the extract came from. Martin and colleagues built a bacterial cell-free platform from genomically recoded E. coli lacking release factor 1, using multiplex genome engineering to enhance extract performance by functionally inactivating negative effectors [15]. Two numbers, each kept to its own referent: the most productive extracts enabled synthesis of 1,780 ± 30 mg/L superfolder green fluorescent protein, a measure of raw extract productivity; separately, the optimised platform introduced 40 identical p-acetyl-L-phenylalanine residues site-specifically into an elastin-like polypeptide with accuracy of incorporation of at least 98% and a yield of 96 ± 3 mg/L [15]. The 40-site result is the one that speaks to ncAA capability, and its yield is 96 mg/L, not the sfGFP figure.
Confirm incorporation by mass spectrometry, not by band position
A band at the expected molecular weight establishes that a full-length chain was made. It says nothing about what occupies the reassigned codon. Aerni and colleagues state the problem plainly: deletion of release factor 1 in E. coli enhances translation of UAG stop codons, yet may also extend protein synthesis at natural UAG-terminated messenger RNAs, so the fidelity of protein synthesis at reassigned UAG codons and the purity of the ncAA-containing proteins produced require careful examination [16]. They also say why an off-the-shelf proteomics run will not settle it: conventional proteomic analyses cannot readily identify the extended proteins or accurately discover multiple amino acid insertions at a single UAG [16].
Their workflow detected UAG readthrough in native proteins in E. coli strains where UAG had been reassigned to encode phosphoserine, and quantified both non-standard and natural amino acid incorporation at UAG in a recombinant reporter; measuring the fidelity and purity of that phosphoserine system is what let them improve it [16]. The sentence to carry away is their own: the results show a surprising diversity of natural amino acids at reassigned stop codons [16]. Scope it correctly — that was the phosphoserine system in release-factor-1-deleted E. coli, and a different synthetase will present a different contaminant profile. What transfers is the obligation to measure, not the identity of the contaminants.
That diversity is why intact mass alone is a weak check. Intact mass returns a total, and a canonical residue substituting for your ncAA shifts that total by the difference between the two side chains, which for some pairs is small and for isobaric substitutions is zero. Localising the residue and identifying what else appears there requires peptide-level analysis with fragment-ion coverage spanning the suppression site. Residue-specific replacement needs the same discipline with different arithmetic: every occurrence of the parent residue is a candidate, so incomplete replacement gives a distribution of intact masses rather than one shifted peak, which is why Budisa and colleagues confirmed replacement by electrospray mass spectrometry alongside amino acid analysis and X-ray structural analysis [8].
Where code expansion stops and synthesis begins
The library's guide to D-amino acids and peptidomimetic backbones treats inversion of an α-carbon as a route to protease resistance and altered conformation. Genetic code expansion is a poor way to build those molecules, and the limit sits in elongation rather than in charging.
Fujino and colleagues re-examined the question in the flexible in vitro translation system, which enables the compatibility of each D-amino acid with elongation to be assessed quantitatively in the absence of potential competitors; efficiencies were determined by Tricine-SDS-PAGE and the full-length peptide detected by MALDI-TOF-MS [17]. Relative to the corresponding L-amino acid, the D-isomers fell into three groups: 40% or higher for D-Ala, D-Ser, D-Cys, D-Met, D-Thr, D-His, D-Phe and D-Tyr; 10–40% for D-Asn, D-Gln, D-Val and D-Leu; and truncated peptides or no detectable full-length peptide for D-Arg, D-Lys, D-Asp, D-Glu, D-Ile, D-Trp and D-Pro [17]. Using the first group consecutively, or alternating those residues with L-amino acids, completely inhibited elongation, whereas inserting two or three L-amino acids between the D-residues restored the double-incorporation efficiency [17].
Read as a specification: isolated single D-residues, from part of the set, in short peptides, in a system with no competing charged tRNAs — with truncation as the signature of failure, the artefact the mass spectrometry above exists to catch. A sequence needing adjacent D-residues, or D-residues from the third group, is a solid-phase synthesis target rather than a translation target. The flexizyme route removes the synthetase's opinion about what may be charged onto a tRNA [14]; it does not change what the ribosome's elongation step will extend [17].