Life's source code has been written in four letters—A, T, G, C—for four billion years. In 2019, Benner's group added two synthetic base pairs, Z:P and S:B, expanding the alphabet to eight and naming it hachimoji (Japanese for "eight letters"). Building such DNA is one thing; getting the cell's machinery to read it is another.
The 2019 report wasn't a perfect score: transcription relied on T7 phage polymerase, and natural T7 cannot read the letter S—an engineered FAL variant was needed to cover all eight.
Three Steps Up the Ladder
In 2023, the B:S pair entered E. coli RNA polymerase, reaching six letters. On September 2, 2026, Nature Communications published a collaboration between Dong Wang's group at UCSD with Benner, Lyumkis, and colleagues: an entirely unmodified E. coli RNAP (the α2ββ'ω five-subunit enzyme) transcribed all eight letters, P:Z and B:S included, accompanied by four cryo-EM structures at 2.42–2.75 Å. First author Qingrong Li; corresponding authors Benner and Wang.
Timeline:
- 2019 — Hachimoji DNA created; T7 required engineering to read all letters
- 2023 — B:S enters E. coli RNAP; six letters
- Aug 2026 — Companion paper: hydrophobic base pairs transcribed without hydrogen bonds
- Sep 2026 — Native RNAP reads all eight letters, with four structures
Kinetics are respectable. In single-turnover experiments, incorporation of synthetic pairs is only about two-fold slower than natural dG:CTP, and after incorporation, translocation to the n+2 site proceeds without pausing.
A telling detail: with a synthetic pair in the active site, dZ paired with PTP yields 76.8% trigger-loop closure (the catalysis-ready state), while dP paired with Z*TP drops to 19.4%. The machine does not treat all letters equally—it has its own preferences.
"Fidelity" in Quotation Marks
Z has a chemical weakness: a pKa of about 7.8, making it prone to protonation at physiological pH and to masquerading as G by mispairing with GTP. The team's fix was to keep modifying the letter itself—Z* replaces the nitro group with an amide, pushing pKa above 10 and suppressing mispairing. So the high-fidelity eight-letter alphabet is no longer the original 2019 set: the letters themselves have been patched.
The paper acknowledges its mismatch rates are overestimates (forced single-substrate conditions). And this is an in vitro reconstituted system: a 9-nucleotide RNA primer and a 27-nucleotide template—not living bacteria, not whole genes. Headlines like "bacteria can read synthetic DNA" are not justified this time.
The companion paper is also worth noting: on August 12, PNAS carried work from the same Wang group showing that even the hydrophobic Ds:Pa pair, which uses no hydrogen bonds at all, can be transcribed by this machine—though with clear bias, as DsTP is incorporated far more readily than PaTP.
Where This Is Heading
The paper's own boundaries: first, eight-letter mRNA is a prerequisite for translating eight-letter codons—translation has not yet been achieved; second, expanded alphabets suit evolutionary aptamers, with Hachimoji Spinach fluorescent aptamers and AegisBinders as signposts; third, six-letter AEGIS aptamers targeting liver cancer cells have a 2020 preclinical precedent.
Going from four letters to eight raises per-letter information content from 2 bits to 3 bits. The operating system hasn't changed yet—but the peripherals are now compatible.
A previous forum discussion covered RNA self-assembly driven by the difference of a single oxygen atom—a chemical self-assembly approach. This paper takes the structural biology approach: take the machine apart and see exactly why it is willing to read. Both routes lead to the same question: does life's alphabet really have only four slots?