A table you can fill step by step, with the evidence and its strength behind each piece
Nearly all living cells read the same table of 64 codons, called the standard code, and that table is older than the last common ancestor of all life. A few organisms and organelles read it with small changes; the best known is UGA, a stop codon in the standard code, which reads Trp in Mycoplasma and in the mitochondria of animals and yeast (see the variants below). Nobody has seen an early form of it. Researchers infer the early forms from chemistry that happens without biology, from meteorites, from the shape of the table, and from the oldest parts of today's proteins and translation machinery.
This page puts those inferences in one place and says how well each one holds up. Every claim carries one of five labels.
The table demonstration below is a model. It takes a published order of amino acids and shows what that order implies under one simple rule. It makes no claim about what happened. The sourced claims on the rest of the page carry the evidence.
Pick one of five published orders in which the amino acids may have joined the code, then step through it. Click any codon to see its history and its one-mutation neighbors.
Click a codon to see its history in this order.
The rule: when an amino acid enters, it claims each box (the first two bases) where it has a codon today. An empty box is claimed whole, so for now the amino acid reads all four codons. In a box that is already held, it takes only its own codons from the holder. The idea that boxes were split after being assigned whole is common in code-growth models (see the hypotheses section), and the particular rule here is a choice made for this page.
The first four are often given as G, A, D and E, or as G, A, D and V. Both versions appear in the literature. G, A, D, E matches the ranking by prebiotic abundance (Higgs and Pudritz). Trifonov's rankings have V in fourth place and E in seventh. A 2024 study of ancient proteins puts V first and D late. The table shows how far the five orders agree.
Serine has six codons in two groups. UCN is in the top row of the table and AGY is in the third row. No single mutation turns one group into the other. Leucine and arginine also have six codons, but their groups are neighbors. Serine is the one amino acid whose blocks are cut apart.
None is settled. They are not exclusive: a block could arise one way and persist for another reason.
Two unrelated enzyme families attach amino acids to tRNAs (class I and class II, above). Rodin and Ohno proposed in 1995 that both families were first encoded by the two strands of a single ancestral gene. This page cannot check that claim. It can show what the proposal means and test one thing about it against today's codon table.
RNA and DNA strands pair base to base: A with U (T in DNA), G with C. The two strands run in opposite directions, so the second strand, read the way a ribosome reads (5′ to 3′), is the first strand's complement written backward. A stretch of code therefore has a hidden twin. Read as a gene, the twin gives different codons and so different amino acids.
Click any codon below. The panel shows its twin codon and the amino acids on both sides, colored by synthetase class.
The twin of a codon keeps its middle base's partner: U becomes A, C becomes G. The middle base is also the best single predictor of an amino acid's chemistry (see the claim on columns above). So the twin of an amino acid with a U in the middle (all hydrophobic) has an A in the middle (mostly polar or charged). That is the sense in which the mirror pairs opposite characteristics. The table is computed from the code.
Five families of explanation exist. Each explains some of the table and has a strong published criticism.
The code predates the last universal common ancestor, so the family tree of living things cannot see behind it. Every piece of evidence is indirect. The table lists the kinds of evidence from the most direct to the least, with a rating of how much each can say about the early code. The ratings are an editorial judgment from the sources.
What did cells look like when the code had a handful of amino acids? The evidence supports a few outlines and leaves the rest open.
Four research passes gathered the claims. Each claim has a status, and a quote only where the page was read. "Full text" means the page body was read, "abstract" means only the abstract was, and "summary only" means a tool's summary of the page was the source. Quotes are at most 40 words and were checked against the pages.
Papers that could not be opened are left out. That includes the original Wong 1975 coevolution paper, Freeland and Hurst 1998 (the "one in a million" estimate, cited here through a review), the Rodin-Ohno 1995 paper, and most Hartman papers. Page last updated .