🧬 Genomics
Life Has Written Its Code in 4 DNA Letters for 4 Billion Years. A Bacterial Enzyme Just Read 8, and the Codon Table Jumps From 64 to 512.
E. coli RNA polymerase transcribed an 8-letter genetic alphabet called hachimoji, and the arithmetic nobody ran says a 20-nucleotide aptamer library just got a million-fold bigger while the codon table grew 8x.
Three bits per base pair: that is what DNA becomes when its alphabet doubles from four letters to eight. In early September, a team led by Dong Wang at UC San Diego showed that a natural bacterial enzyme can read and transcribe all eight letters, which means one of biology's oldest machines just accepted a second alphabet. Here is the part nobody calculated: the same molecule now carries 50 percent more information, the codon table jumps from 64 to 512, and the search space for DNA-based diagnostics expands a million-fold.
Chemists call the expanded alphabet hachimoji, Japanese for "eight letters," introduced in a 2019 Science paper by Steven Benner's group at the Foundation for Applied Molecular Evolution. Its four synthetic letters, P, Z, B, and S, form two new base pairs, P:Z and B:S, designed to sit in the double helix exactly where natural pairs sit. Chemists rearranged the weak links that hold base pairs together, moving them to new positions so the new pairs behave like the originals without disturbing them.
Copying hachimoji was already possible in 2019, but only with T7 RNA polymerase, a simple single-subunit engine borrowed from a bacterial virus and engineered for the job. What Wang's team tested is the real thing: E. coli's own RNA polymerase, a large multi-subunit machine that evolution tuned for exactly four letters. Combining biochemical assays with cryo-electron microscopy at 2.42 to 2.75 angstrom resolution, finer than the width of a single atom, the researchers captured structural snapshots of the enzyme recognizing and incorporating both synthetic pairs. Wang said the result surprised his group, a reaction that says something about how settled the four-letter dogma had become: the enzyme treats the unnatural pairs "in the exact same manner as native base pairs," using the same biochemical checkpoints it applies to natural DNA.
Three bits instead of two
Four letters carry log base 2 of 4, or 2 bits per base pair; eight letters carry log base 2 of 8, or 3, which is 50 percent more information in the same number of letters, an arithmetic step the coverage of this story skipped entirely.
Apply it to something concrete. The E. coli genome is about 4.6 million base pairs. Written in the natural alphabet, its information capacity is roughly 9.2 megabits; written in hachimoji, the same number of base pairs holds 13.8 megabits. Same molecule, 575 extra kilobytes for zero additional letters, arithmetic nobody in the coverage ran. The number sets the ceiling for what an expanded genome could one day carry.
From 64 codons to 512
Biology reads DNA in triplets, which gives the natural code 4 to the third power, or 64 codons: 61 encode 20 amino acids, and 3 are stop signs. An 8-letter alphabet reads 8 to the third power, or 512 codons, eight times as many, and because codons are read in triplets, the vocabulary of the code grows geometrically rather than linearly, which is why doubling the letters multiplies the codons by eight. Keep the 64 that nature uses, and 448 slots remain for non-canonical amino acids, synthetic building blocks with chemistries evolution never tried. Divide 448 by nature's 20: the expanded code offers 22.4 times more ingredients than the entire protein chemistry of the living world, which means everything alive today was assembled from roughly one twenty-second of the parts this alphabet makes available.
If natural proteins are a piano with 20 keys, hachimoji adds 448. Transcription was the first gate, and it just opened. The reader comes next.
| Property | Natural alphabet (4) | Expanded alphabet (8) | Gain |
|---|---|---|---|
| Bits per base pair | 2 | 3 | 1.5x |
| E. coli genome capacity (4.6 Mbp) | 9.2 Mb | 13.8 Mb | +575 KB |
| Possible codons | 64 | 512 | 8x |
| Slots for non-canonical amino acids | 0 (in use: 20) | 448 | 22.4x nature's 20 |
| 20-mer aptamer sequences | ~1.1 trillion | ~1.15 quintillion | ~1,048,576x |
A million-fold bigger search space
Those numbers are theoretical; this one lands sooner. Earlier expanded-alphabet work already produced synthetic DNA molecules capable of recognizing liver cancer cells. That is the UC San Diego announcement's own example. Such molecules, called aptamers, are short DNA strands that fold into shapes that grab specific targets, and their power comes from screening enormous libraries of candidate sequences. A 20-nucleotide aptamer has 4 to the 20th power possible sequences in the natural alphabet, about 1.1 trillion, versus 8 to the 20th power in hachimoji, about 1.15 quintillion, a ratio of exactly 2 to the 20th power, or 1,048,576: a million shapes for every one. For the diagnostic screens that are the nearest real application, the space of shapes to search just got a million times bigger.
Fidelity is the part that did not come for free
Accuracy still lags: mispairing happens, the new work confirms it, and the team reports that a modified synthetic base called Z* sharply reduced one major type. That single asterisk is arguably the most important character in the story: expanded alphabets live or die on error rates, and Z* shows those rates are engineerable rather than fixed by chemistry.
A second study from the same lab, published in PNAS on August 12, pushed the idea further: Qingrong Li and colleagues showed that RNA polymerase can recognize another synthetic pair held together by hydrophobicity instead of hydrogen bonds, and still drive the trigger loop closure that catalyzes RNA synthesis. Base recognition, in other words, does not even require the classic hydrogen-bond pattern, which widens the palette of synthetic letters beyond hachimoji itself and reframes what polymerase fidelity is actually checking: geometry and chemistry, not the specific bonds the textbooks emphasize. The bonds were never the point.
The strongest case against
Nature had four billion years to expand the alphabet and never did, a fact that deserves full weight, because four letters may be close to optimal: every new letter adds new possible mispairing modes, and the error cost compounds across billions of replications.
More important, transcription is only the first of three machines.
Meanwhile the ribosome and the tRNA system, which would have to translate 512 codons into proteins, lag years behind the polymerase result. Until a ribosome reads hachimoji messenger RNA and installs non-canonical amino acids into a protein, the bigger book has no reader, and the 448 extra codon slots are arithmetic, not chemistry.
There is also the dual-use ledger. An expanded alphabet widens the design space for engineered biological systems of every kind, including harmful ones, which is why the biosecurity conversation around any alphabet expansion now has to assume adversaries get the same bigger toolbox. That concern is not hypothetical this week: on September 10, Anthropic reported blocking requests that used its models to advance potential bioweapons research, including orthopoxvirus work. Expanding the letters of life and expanding the reach of AI assistance arrive together, and the safety conversation has to cover both.
Limitations
Everything here happened in test tubes. No living cell has transcribed hachimoji, and none has run an 8-letter genome. Error rates are the caveat: Z* reduces one major mispairing mode, not all of them, and per-base fidelity is still far from natural polymerase error rates. Note that the PNAS hydrophobic pair is separate chemistry from hachimoji, not a drop-in upgrade, and the genome-capacity arithmetic should be treated as a theoretical ceiling rather than a demonstrated storage result, since DNA synthesis cost still dominates any storage application. And the codon-table math assumes translation works, which it does not yet. Sourcing note: the 2.42 to 2.75 angstrom resolution and the Z* mispairing reduction come from secondary reporting on the papers rather than the papers themselves; everything else traces to the UC San Diego announcement, the journals, or the papers' own press coverage.
What you can do
For aptamer and diagnostic developers, the near-term play is concrete: track Z* chemistry and per-base fidelity as it publishes, because a million-fold bigger library only pays off if error rates stay low.
For synthetic biologists, cell-free experiments can now carry synthetic letters through transcription using native enzymes, work that can start this year, since the transcription barrier that used to require engineered viral polymerases has effectively fallen for in vitro work.
For everyone else, honestly, nothing to do yet. Watch two milestones: a ribosome translating hachimoji mRNA into protein, and Z* fidelity approaching natural error rates, because those two results are what turn a 512-codon table from arithmetic into a parts list.
The Bottom Line
A natural enzyme read eight letters, and the new arithmetic is 3 bits per base pair, 512 codons, and an aptamer library a million times bigger.
Honestly: the polymerase is one machine of three, the ribosome has not caught up, and nothing has happened inside a living cell. But the machinery of life just accepted a second alphabet without complaint. Of all the steps on the road to engineered biology, that is the one that used to sound like science fiction.