Codon usage

The reference table every codon number is measured against, the one quantity CAI and its relatives are built on, and why maximising it is not the same as designing for expression.

This page exists because six tools read a codon-usage table and report codon scores, and the same few ideas sit under all of them. Each tool page has its own detail; this page is what they share.

A score is a comparison against a tableCopy Anchor Link

A protein has no single DNA sequence behind it: leucine alone has six codons. Which synonymous codon a gene uses affects how well it is expressed in a given host, so codon scores compare a gene against a reference: how often each of the 61 sense codons is used in some set of genes from some organism.

Every number below is only as good as that reference, so its source is named beside every result. The table can be a published one (CUTG, bundled offline, or Kazusa, which is frozen at the 2007 GenBank release), one counted live from an organism's RefSeq coding sequences, one you count yourself in Codon Tables, or a file someone else counted. A table counted from a handful of genes is too thin to trust, and Back-translate says so beside the CAI it reports.

Relative adaptiveness, the quantity underneathCopy Anchor Link

Within each synonymous family, a codon's relative adaptiveness is its frequency divided by the frequency of the family's most-used codon: w = f / fmax, so the host's favourite codon for each amino acid scores 1.

• CAI (Sharp & Li 1987) is the geometric mean of w over a gene's codons. 1 means every codon is its family's favourite.
• The rare-codon list is the codons whose w falls below a threshold.
• Fop (Ikemura 1981) is the fraction of codons whose w reaches a threshold, 0.80 by default.
• The codon matrix shows w per codon, so a single CAI can be traced to the families that set it.

Because these share one definition, they cannot disagree about the same table. Measures that ask a different question (ENC, tAI, nTE, codon-pair bias, %MinMax and others) are compared on Codon Usage Analysis, which carries nine and says which needs a table, which needs a tRNA pool and which needs only the sequence.

Maximising CAI is not designing for expressionCopy Anchor Link

The obvious design is to give every amino acid its host's single favourite codon. That reaches CAI 1.000 by construction, for any protein and any table, so a CAI from that rule measures the rule, not your sequence; Back-translate says so rather than reporting it. It is also the least like any real gene: uniform codon choice removes the translational pauses a nascent chain may fold around.

Codon bias is also not the strongest predictor of expression. Kudla and colleagues built 154 versions of one GFP gene that differed only at synonymous sites, saw protein levels vary 250-fold, and found that the folding free energy of the message around the start codon (nucleotides −4 to +37) explained more of the variation than codon bias did. So Codon Optimization runs four published strategies side by side rather than picking one:

Each design comes back with its CAI, GC, the folding energy of the initiation window and how many bases changed, so the trade is visible rather than implied.

Which tool answers which questionCopy Anchor Link

RelatedCopy Anchor Link

Codon Usage Analysis · Codon Tables · Back-translate · Codon Optimization · Codon usage matrix