Rewriting Life's Dictionary: Harvard Lab Dislodges the CCA Rule from Its Stone Tablet

Every protein of every organism on Earth is translated from the same dictionary: 64 triplet codons mapping to 20 amino acids, unchanged for over four billion years.

On September 15, C&EN reported work from Harvard Medical School and the Wyss Institute under the headline: custom genetic codes are now easy to make. The key word is tRNA—the binder that holds that dictionary's pages.

The machine first

Protein translation in the cell works like this: every three bases on mRNA form a codon; a tRNA carries the matching amino acid on one end and presents the anticodon on the other; the aminoacyl-tRNA synthetase (aaRS) is the loader that attaches the correct amino acid to the correct tRNA; the ribosome does final inspection and assembly. Every entry in the dictionary is maintained by this three-way relationship.

Textbooks state one iron rule: every tRNA's 3' end must be the three bases CCA, where the amino acid is attached, and the ribosome grips it via two bases of the 23S rRNA large subunit, G2251 and G2553—like a bolt catching a door frame. In an interview, Church said he had firmly bet that CCA was the most conserved sequence on the planet.

That bet lost.

How the dogma was overturned

The results, published in Nature (online August 26, "Automated prototyping of genetic codes," first author Felix Radford, corresponding author Church): when the CCA tail is snipped off tRNA and replaced, E. coli synthetases still attach amino acids. Among the non-natural tails tested, CGA worked best—about 62% charging efficiency in a purified system, versus about 57% for a CUA tail and 42% for CAA. The natural CCA tail averaged about 67% in cell lysate. The difference exists, but it is nowhere near an absolute "lock."

The gatekeeper changed. The real checkpoint sits on those two ribosomal bases—engineer G2251 and G2553 to pair with the new tail, and the resulting ribosome recognizes only the new tRNA. Mass spectrometry found no crosstalk from wild-type ribosomes in the products. Two ribosomes, two different dictionaries, neither crossing the line.

They named the workflow AGENTEX. It runs in test tubes with robot arms—an ordinary open-source Opentrons OT-2 pipetting workstation with open-source software. To try a new codon-assignment scheme, you no longer need to touch a genome: assemble a tRNA pool, pair it with engineered ribosomes, drop it into lysate, and the robot works through the combinations automatically.

The headline 34, versus the 3 actually achieved

"34 reassignable codons" is the most widely circulated number, and it needs unpacking. Of the 64 codons, 20 sense codons plus 14 open slots that can be freed give a theoretical ceiling of 34. That is the area of the design space, not this harvest.

What the paper actually demonstrated: reassignment of 3 codons, incorporation of one non-natural amino acid (pazF), and a separate 22-tRNA compressed-code system. In reporter-gene assays, the new system produced roughly six times more full-length product than the control. All experiments were done in cell-free systems. Does it work in living cells? C&EN states plainly that the team is "working on it." In progress, not done.

A timeline worth reading closely

• 2013 — Church and Isaacs free the first codon (UAG); it took nine years.
• 2019 — the UK MRC Laboratory synthesizes genomically recoded E. coli syn61, compressing 64 codons to 61.
• 2021 — a syn61 derivative gains complete resistance to a cocktail of viruses.
• 2026 — AGENTEX moves the battlefield into test tubes, robotically prototyping entire codes.

The real information hides in that timeline. Freeing the first codon took nine years; Church's group says the second took another ten. All of that work cut directly into living genomes—design, synthesis, replacement—at a pace measured in years. AGENTEX moves "defining a code" from the genome level to the tRNA level in a test tube, compressing a decade-scale project into pipeline-scale work. The "easy" refers to the pipeline; the dictionary itself did not get simpler.

Why the dictionary is worth rewriting

Three uses, each with precedent. First, viral immunity: the 2021 recoded bacterium simply could not read viral genes—when a virus cocktail attacked, it was wiped out. Second, new chemistry: writing non-natural amino acids into proteins to build polymers, materials, and catalysts nature cannot make; the authors say the goal is "as many monomers as we can stuff in." Third, containment: cell-free components lose activity outside the test tube, which the paper cites as built-in biocontainment.

What the coverage omitted is more notable than what it included. The C&EN article contains no mention of biosecurity—a conspicuous absence for a programmable genetic code coming out of the Church lab. The paper's official answer is "process safety": no living cells are modified, no viable engineered organisms are produced. But the team's stated next step is transplantation into living cells—at which point, the antiviral firewall and the new chemical space become two sides of the same coin. That dual-use discussion is missing from the English coverage; consider it supplied here.

A footnote on the people involved: Church's name is attached to GRO Biosciences and Pearl Bio, and the team has filed a provisional patent on the method. If a textbook took forty years to turn a page, industry's page-turn appears to be moving much faster.

A dictionary finalized four billion years ago got its appendix opened this September. The first entry is modest: a CGA tail, sixty-percent efficiency. With dictionaries, the hard part was never writing the first word.

---

*Sources: C&EN (2026-09-15); Nature paper 10.1038/s41586-026-10949-y (online 2026-08-26); Wyss Institute press release; Science 342 (2013) and Nature 569 (2019); Science 10.1126/science.abg3029 (2021).*