Robotic Platform Enables Rapid Prototyping of New Genetic Codes

A cell-free, automated system developed at Harvard Medical School and the Wyss Institute allows researchers to design, build, and test alternative genetic codes without the years-long process of engineering a living genome — with implications for biomanufacturing, biocontainment, and the production of new classes of therapeutics.

Key Takeaways

1. A new cell-free platform tests genetic codes without engineering living organisms. AGENTEX, developed at Harvard Medical School and the Wyss Institute, lets researchers design and evaluate alternative genetic codes in test-tube reactions rather than years-long genome editing.
2. The ribosome, not the charging enzyme, guards the genetic code. Using a new sequencing method called tSCAN, researchers found that natural enzymes tolerate altered tRNA ends, but standard ribosomes reject them, identifying the ribosome as translation’s true fidelity checkpoint.
3. Engineered ribosomes ran a second genetic code in parallel with no interference. Ribosomes carrying two specific mutations accepted altered tRNAs, whereas native ribosomes continued to use the standard code in the same reaction.
4. The team compressed and expanded genetic codes in a single system. Using just 22 tRNAs to encode all 20 canonical amino acids, researchers freed up codons and successfully inserted a non-standard amino acid at a reassigned stop codon.
5. The approach could accelerate therapeutic development while incorporating biocontainment. Because the engineered components don’t function in ordinary cells, the system offers built-in biocontainment and a faster path toward proteins with novel chemical properties for drug design.

Every living organism translates DNA into protein using the same basic dictionary: 64 three-letter codons that specify 20 amino acids, plus start and stop signals. Researchers have long sought to expand or rewrite this genetic code — for example, to insert non-standard amino acids into proteins, or to build organisms with built-in biocontainment and resistance to viruses. Such genetic code expansion has already enabled the incorporation of more than 400 non-standard amino acids into proteins made in living cells, and has been used to build biomaterials, therapeutics, and enzymes.

The bottleneck has been implementation. Reassigning more than a handful of codons across an entire genome requires extensive, labor-intensive genome editing, and testing a new genetic code design has traditionally meant first constructing an engineered organism and only then finding out whether the code works. A team led by Felix Radford and George Church, Ph.D., whose work was published in Nature, describes a system that removes that bottleneck by testing genetic codes entirely outside living cells, using an automated, robotic, cell-free platform. [1]

The Platform: AGENTEX
The new system, called AGENTEX (Automated Genetic tRNA Expansion), builds on a known biochemical feature of the ribosome, the cellular machine that reads messenger RNA and assembles proteins. All natural transfer RNAs (tRNAs) — the adaptor molecules that deliver amino acids to the ribosome — end in the same three nucleotides, CCA. The large ribosomal subunit recognizes this CCA sequence directly, through two specific base-pairing contacts, and this recognition step normally prevents any tRNA lacking a standard CCA end from being used in translation.

The authors used ribosomes engineered with two mutations (G2251C and G2553C) that alter this recognition step, causing the ribosome to accept tRNAs ending in CGA rather than CCA. Because these modified ribosomes still coexist with the natural translation machinery, a cell extract containing both mutant and native components can, in principle, run two genetic codes side by side without one interfering with the other.[1]

To build and evaluate tRNA pools for this system, the team developed a laboratory workflow that produces synthetic tRNA pools by chip-based DNA synthesis and in vitro transcription, introduces them into cell-free bacterial lysates, and tests translation outcomes — all automated on an open-source liquid-handling robot (Opentrons OT-2). This allows many genetic code designs to be built and tested in parallel, rather than one at a time in living cells.

Key Findings

1. The tRNA “CCA rule” is more flexible than assumed. It has long been thought that the CCA sequence at the end of a tRNA is essential for its charging with its amino acid by aminoacyl-tRNA synthetases (aaRSs), the enzymes responsible for this step. To test this directly, the authors developed a new sequencing-based method, tRNA sequencing of charging by automated NGS (tSCAN), which can measure, at scale, whether individual tRNAs in a complex pool have been charged with an amino acid. Using tSCAN, they found that natural Escherichia coli aaRSs were able to charge most tRNAs carrying altered 3′ ends (CGA, CUA, or CAA instead of CCA), particularly when the tRNAs were incubated in a cell lysate, where they received RNA modifications, rather than in a purified, reconstituted system. This indicates that native cellular tRNA modifications — chemical marks added during normal tRNA maturation — help preserve aminoacylation even when the terminal sequence is altered.
2. The ribosome, not aaRS enzymes, is the actual gatekeeper of the genetic code. Although aaRSs proved tolerant of altered tRNA ends, the standard (unmodified) ribosome did not accept these altered tRNAs during protein synthesis. Using mass spectrometry to analyze peptides produced in cell-free reactions, the authors found no detectable “crosstalk” — no evidence that altered tRNAs were being used by native ribosomes to insert the wrong amino acid. This means that under normal conditions, the ribosome — not the charging enzyme — is the primary checkpoint that maintains fidelity of the genetic code.
3. Mutant ribosomes allow a second, parallel genetic code to operate without interference. When the engineered CGA ribosomes were introduced, they specifically accepted the altered tRNAs while natural ribosomes continued to use the standard genetic code, all within the same cell-free reaction. The authors demonstrated this by supplying reporter peptide templates with swapped codon assignments (for example, swapping the codons for alanine and serine) and confirming, by mass spectrometry and a luminescence-based reporter assay, that the correct amino acid was inserted according to which ribosome — standard or mutant — was doing the translating.
4. Compressed and expanded genetic codes were built and tested. Using AGENTEX, the team constructed a genetic code compressed to just 22 tRNAs (21 sense codon tRNAs plus one for the stop/non-standard amino acid codon), while still encoding all 20 canonical amino acids. This compression frees up codons that would otherwise be redundant. Using this system, they successfully incorporated a non-standard amino acid — para-azido-phenylalanine — at a reassigned stop codon, along with two other reassigned codons, thereby producing a peptide with an expanded chemical alphabet. Separately, they showed that all 34 codons could theoretically be used to encode both canonical and non-canonical amino acids, opening a theoretical pathway to using up to 14 additional codons for non-standard building blocks.

Why It Matters
The significance of this work lies less in any single new molecule and more in the method itself. Constructing an organism with a rewritten genetic code — as has been done previously with recoded E. coli strains — is a multi-year undertaking that involves genome-scale editing. AGENTEX allows many candidate genetic codes to be designed, built, and evaluated for function and fidelity in a matter of days, entirely in a test tube, before any decision is made to attempt the equivalent change in a living genome. This design-build-test cycle is intended to de-risk future genome engineering efforts by identifying which codon reassignments and tRNA designs are likely to work before committing to the labor of strain construction.

Because the engineered ribosomes and tRNAs used in AGENTEX are not functional in ordinary cells if released outside the laboratory, the authors note the system carries an inherent biocontainment feature, relevant to concerns about the safe handling of engineered biological components.

For fields adjacent to oncology — including protein-based therapeutics, targeted biologics, and antibody-drug conjugate (ADC) payload chemistry — the ability to reliably and efficiently incorporate non-standard amino acids into polypeptides could eventually support the design of proteins with tailored chemical handles, novel linker attachment sites, or altered pharmacological properties. The authors’ immediate contribution, however, is methodological: two new assays (tSCAN and tSCAN-M) for measuring tRNA charging at scale, and a validated, automated cell-free platform for testing genetic code designs — tools intended to accelerate the broader effort to engineer translation systems beyond those found in nature.[2][3][4][5][6]

Source: Radford, F., Sapers, N., Burgess, H.M., Ort, L., Budnik, B. & Church, G.M. “Automated prototyping of genetic codes.” Nature (2026). https://doi.org/10.1038/s41586-026-10949-y

Note: Agentex is an open-source, enterprise-grade infrastructure layer created by Scale AI for building, deploying, and orchestrating long-running, asynchronous AI agents at scale.

References
[1] Radford F, Sapers N, Burgess HM, Ort L, Budnik B, Church GM. Automated prototyping of genetic codes. Nature. 2026. doi:10.1038/s41586-026-10949-y. [Primary source article]
[2] Lajoie MJ, Rovner AJ, Goodman DB, Aerni HR, Haimovich AD, Kuznetsov G, Mercer JA, Wang HH, Carr PA, Mosberg JA, Rohland N, Schultz PG, Jacobson JM, Rinehart J, Church GM, Isaacs FJ. Genomically recoded organisms expand biological functions. Science. 2013 Oct 18;342(6156):357-60. doi: 10.1126/science.1241459. PMID: 24136966; PMCID: PMC4924538.
[3] Fredens J, Wang K, de la Torre D, Funke LFH, Robertson WE, Christova Y, Chia T, Schmied WH, Dunkelmann DL, Beránek V, Uttamapinant C, Llamazares AG, Elliott TS, Chin JW. Total synthesis of Escherichia coli with a recoded genome. Nature. 2019 May;569(7757):514-518. doi: 10.1038/s41586-019-1192-5. Epub 2019 May 15. PMID: 31092918; PMCID: PMC7039709.
[4] Nyerges A, Vinke S, Flynn R, Owen SV, Rand EA, Budnik B, Keen E, Narasimhan K, Marchand JA, Baas-Thomas M, Liu M, Chen K, Chiappino-Pepe A, Hu F, Baym M, Church GM. A swapped genetic code prevents viral infections and gene transfer. Nature. 2023 Mar;615(7953):720-727. doi: 10.1038/s41586-023-05824-z. Epub 2023 Mar 15. PMID: 36922599; PMCID: PMC10151025.
[5] Chin JW. Expanding and reprogramming the genetic code. Nature. 2017 Oct 4;550(7674):53-60. doi: 10.1038/nature24031. PMID: 28980641.
[6] Ostrov N, Landon M, Guell M, Kuznetsov G, Teramoto J, Cervantes N, Zhou M, Singh K, Napolitano MG, Moosburner M, Shrock E, Pruitt BW, Conway N, Goodman DB, Gardner CL, Tyree G, Gonzales A, Wanner BL, Norville JE, Lajoie MJ, Church GM. Design, synthesis, and testing toward a 57-codon genome. Science. 2016 Aug 19;353(6301):819-22. doi: 10.1126/science.aaf3639. PMID: 27540174.

Featured image © 2022 – 2026. Licensed under the Unsplash+ License

DOI