‹

02. Four Letters

EN

02. Four Letters

Something alive, still close to 4.2 billion years old, is already reading. Its alphabet has exactly four letters: adenine and thymine, guanine and cytosine, in the double-stranded archive called DNA, or adenine and uracil standing in for thymine, in the single-stranded working copy called RNA. The letters do not pair at random. Adenine always finds thymine, or uracil, and guanine always finds cytosine, the two strands running alongside each other like a zipper whose teeth fit only one specific partner apiece. That fixed pairing is the whole reason a cell can copy its archive at all: split the zipper down the middle, and each half already spells out exactly what its missing partner must be.

Four letters would be a thin alphabet for spelling out a body, except that the cell never reads one letter at a time. It reads in words of exactly three letters, called codons, run end to end with no punctuation and no gaps between them — an arrangement the Nobel committee's own 1989 press release described in almost exactly these terms: "only four such letters... arranged as a long sentence of three-letter words." Three letters, drawn from an alphabet of four, is not an arbitrary word length. It is the shortest word length able to give the cell enough distinct spellings to name every amino acid it needs, with room to spare.

The arithmetic is worth doing in full, because it turns out to run the whole living world. Four possible letters at the first position, four at the second, four at the third: four times four times four, sixty-four possible three-letter codons, no more and no fewer. Of those sixty-four, sixty-one specify one of the twenty amino acids that proteins are built from. The remaining three — UAA, UAG, and UGA — carry no amino acid at all; they are stop signs, telling the cell's protein-building machinery exactly where a protein ends. One further codon, AUG, pulls double duty: it is both the standard start signal and the codon for the amino acid methionine, a capital letter that also happens to spell a word on its own. This entire structure was cracked experimentally, codon by codon, in the early-to-mid 1960s, chiefly by Marshall Nirenberg, Heinrich Matthaei, and Har Gobind Khorana, and it is taught identically, without local variation, in essentially every genetics classroom since.

Sixty-four codons for twenty amino acids and one stop signal means most amino acids answer to more than one spelling, a property biologists call degeneracy, or redundancy. The pattern has its own precise arithmetic: five amino acids — proline, alanine, threonine, valine, and glycine — each get four different codons, a set biologists call a quartet. Nine amino acids get exactly two codons each, a doublet. Three amino acids — leucine, arginine, and serine — get six codons each, a sextet, the richest spelling of all. One amino acid, isoleucine, gets three codons, a triplet. Two amino acids, methionine and tryptophan, get exactly one codon each, a singlet. Multiply it out and it closes perfectly: five quartets plus nine doublets plus three sextets plus one triplet plus two singlets comes to twenty amino acids, spelled across sixty-one codons, with nothing left over and nothing missing.

That redundancy is not sloppiness. It is insurance, built directly into the alphabet. Because so many amino acids have backup spellings, a large fraction of the copying errors that strike a gene's third codon letter change nothing at all about the resulting protein — insurance the cell gets for free, simply because of how the sixty-four spellings were handed out. The insurance runs deeper still: Francis Crick proposed in 1966 that the pairing between a codon and its matching anticodon, the three-letter tag carried by the transfer RNA molecule that ferries each amino acid into place, is strict at the first two letters but tolerant, or "wobbly," at the third. Because of that wobble, a single transfer RNA can often read more than one synonymous codon, which is why a cell gets by with roughly forty distinct transfer RNA types instead of needing a separate one for every one of the sixty-one sense codons.

The redundancy pattern turns out to do something more than buffer typos: it buffers meaning. In 1998, Stephen Freeland and Laurence Hurst measured the standard code against roughly one million computer-generated alternative codes, all built under the same basic rules, and scored each one for "mutational cost" — how much damage a single-letter copying error tends to do to the resulting protein, given that codons differing by only one letter tend to specify chemically similar amino acids. The real, standard code beat more than 99.99 percent of the alternatives. Freeland extended the finding in 2000 with Knight, Landweber, and Hurst, arguing that this near-optimal arrangement was essentially locked in before LUCA existed. Not every researcher accepts that this reflects deliberate selection for error resistance: a 2010 paper in PNAS, "A Neutral Origin for Error Minimization in the Genetic Code," argues the same low-damage pattern could have emerged as a side effect of how codon assignments accumulated over time, through a process of reducing ambiguity, rather than through direct selection for minimizing errors. The debate remains open in the primary literature.

The Code's Oldest Argument

A second, older argument concerns not how well the code resists damage but why these particular letters ended up paired with these particular amino acids at all. Francis Crick proposed one answer in 1968: the "frozen accident." Once a translation system and a great many proteins already depended on a given set of codon meanings, reassigning even one codon would have been catastrophic to everything built on top of it, so the assignments locked in place by historical accident and simply never moved again. Carl Woese proposed a rival answer starting in 1965: the stereochemical theory, which holds that codon or anticodon sequences carry direct, if weak, chemical affinities for their own amino acids, so the code's shape reflects real molecular binding preferences rather than pure historical lock-in. Laboratory binding assays that Woese ran in 1966 and 1967, and again with Saxinger and Ponnamperuma in 1971, found only what the record describes as "weak and relatively non-specific interactions" — evidence for the idea, but not strong evidence. Most current reviews, including a 2017 survey by Eugene Koonin, treat the honest answer as a blend of both: some early stereochemical shaping while the code was still forming, followed by frozen-accident lock-in once code and genome had grown too tightly interdependent to safely change.

Interdependent, though, has never meant unchangeable everywhere. A small number of living lineages run non-standard versions of the code today. Many animal mitochondrial genomes reassign some of their stop or sense codons; certain ciliates reassign UAA and UAG from stop signals to the amino acid glutamine; Mycoplasma bacteria reassign UGA from a stop signal to the amino acid tryptophan. Human mitochondria are one of the living examples, reassigning codons including AGA and AGG. Far from undermining the frozen-accident idea, these exceptions support it: they show up specifically in small, gene-poor genomes, like mitochondria, where few enough proteins depend on the affected codon that reassigning it is survivable rather than catastrophic — exactly the condition frozen-accident logic predicts should make change possible.

Step back from the mechanism entirely and look at the code as a channel carrying information, and a different kind of number appears. Each of the four bases, treated as one symbol in a four-symbol alphabet, carries a theoretical maximum of two bits of information — log base two of four, a mathematical identity, not a measurement. Real genomic DNA, measured empirically, carries somewhat less: approximately 1.9 bits per base, because real sequences are not perfectly random. Biased base composition and codon-usage bias both pull the realized information down below the theoretical ceiling, and non-coding stretches of the genome measure lower still than coding stretches do. Two bits, in other words, is the speed limit; 1.9 is closer to what life actually drives.

Set beside any human information system, the comparison is almost unfair. The standard code, cracked in the 1960s and barely modified since except in a handful of small, late, low-cost corners, is still running, essentially unchanged, in a bacterium and in every cell of the organism reading about it now. It is very likely the oldest continuously operating data format on the planet, never rewritten from scratch, only patched at the margins, carrying the same twenty-amino-acid, sixty-four-codon logic across more than four billion years of unbroken use.

An alphabet, on its own, is inert. It needs something to read it, and something to act on what it reads — and by the time LUCA's genome held this code, both of those already existed, doing a job neither DNA nor protein can do alone.

[[FIGURE: genetic-code | The genetic code, counted out.]]

Overturned if: a large, ancient, gene-rich genome — not a small, reduced one like a mitochondrion — is found to reassign a sense codon's meaning at negligible fitness cost, showing the code's near-universality reflects easy interchangeability rather than a catastrophic, frozen lock-in.

Sources

บทถัดไป · 3 / 1403. The Molecule That Copied Itself
Size
Page
Font

AMINO — The Immortal Thread

❦
101. The First Choice202. Four Letters303. The Molecule That Copied Itself404. The Frozen Code505. The Upgrade to DNA606. The Membrane and the Machine707. The Great Merger808. Messengers Before Minds909. The First Animals1010. The Bonding Peptides1111. The Lobster's Posture1212. The Vertebrate Bargain1313. Big Brains, Ancient Molecules1414. The Unbroken Thread
‹ Prev2 / 14Next ›