SeqBench

Kozak Sequence vs. Shine-Dalgarno: Two Answers to the Same Question

8 min read · Updated November 11, 2026

Both sequences sit just upstream of a start codon and both are described as controlling translation initiation, which is why they get swapped in people's heads. They are not interchangeable, and the mistake is not cosmetic: a bacterial Shine-Dalgarno in front of a mammalian cDNA does nothing, and a Kozak consensus in front of a bacterial ORF does nothing either.

The difference is mechanical. One is a ribosome-binding site that recruits the small subunit to a specific place on the mRNA; the other is a context that helps a ribosome that has already arrived pick the right AUG.

The mechanism each one addresses

In bacteria, the 30S subunit finds the start codon by base-pairing the 3' end of its 16S rRNA with a purine-rich element in the mRNA — the Shine-Dalgarno sequence. That pairing is what puts the ribosome down at the right place, and the distance from the paired region to the AUG is part of the mechanism, not a detail: the ribosome has to be positioned so that the AUG lands in the P site.

In eukaryotes the small subunit is recruited at the 5' cap and scans 3' along the mRNA until it finds an AUG in a favourable context. There is no base-pairing site to recruit it. What the context does is bias which AUG gets used, which matters because the first AUG is not always the intended one.

So one sequence is an address and the other is a quality signal on the doorway. Combining them is not additive; it is answering two different questions.

The Kozak consensus, and the two positions that carry it

The consensus is usually written gccRccATGG, where R is a purine. Kozak's own work showed the effect is not spread evenly across those ten bases: a purine at position −3 (three bases before the A of the AUG) and a G at position +4 (the base immediately after the AUG) carry most of it. The rest of the consensus matters far less, which is why a construct can match the consensus badly on paper and still initiate well.

That is also why a checker should test those two positions rather than score a ten-base consensus nobody's construct matches exactly. SeqBench's cassette check does exactly that: it reports whether −3 is a purine and +4 is a G, and it is a soft check, because a construct can express without them.

Worked: the four combinations, and which one fails

Take one 60-codon coding sequence and put four different contexts in front of the same ATG, changing only the two bases that matter. The check reports the same verdicts you would derive by hand:

  • GCCACC ATG G — A at −3 and G at +4. Both strong positions met, pass.
  • TCCTTT ATG T — T at −3 and T at +4. Fails both, and the report names both.
  • GCCACC ATG T — a perfect −3 and a T at +4. Still fails, because +4 is G that is wanted.
  • TCCTTT ATG G — G at +4 with a T at −3. Still fails, for the mirror reason.

Worked: Shine-Dalgarno spacing is measured, not eyeballed

The bacterial version of the same exercise is harder to eyeball, because the number that matters is a distance. Put AGGAGG in front of the same ATG with different spacers and the check accepts the ones inside 5–13 bp and rejects the rest. A 5 bp spacer passes, 8 passes, 13 passes, and 16 fails with a sentence saying no run of four or more consecutive bases of the core sits in the window.

The subtlety is what happens when the spacer is too short. Put a 3 bp spacer after AGGAGG and the full six-base core would sit 3 bp out, below the window, so the search falls back to shorter consecutive runs: it finds AGGA at 5 bp and passes with a four-base match. Squeeze it to two bases and nothing four long fits, so it fails. A reported spacing can therefore come from a shorter match than the one you designed, which is the right answer and worth knowing.

Both checks are soft for the same honest reason: a weak SD or an imperfect Kozak still expresses. Neither check is a rate prediction. If you need a number rather than a verdict, that is what the RBS designer is for, and its own documentation says what the number does and does not support.

What neither check establishes

  • That the mRNA is folded in a way that exposes the site. The cassette check looks at the sequence, not at structure; a strong SD can be sequestered in a hairpin.
  • That the start codon is the one the cell uses. An upstream AUG in a better context can take most of the initiation, and nothing in a per-CDS check sees that.
  • That the protein is stable, or folded, or correctly localised. Initiation is the first step, not the outcome.
  • That a Kozak context helps a bacterial construct, or an SD a mammalian one. Each is inert in the other host, which is the whole point of this page.

FAQ

Is a Kozak sequence the same as a ribosome binding site?

No. A bacterial ribosome binding site (the Shine-Dalgarno sequence) recruits the 30S subunit by base-pairing with the 16S rRNA. A Kozak context does not recruit anything; it biases start-codon selection by a scanning eukaryotic ribosome. Each is inert in the other host.

Which positions of the Kozak sequence matter most?

−3 and +4 relative to the A of the AUG: a purine at −3 and a G at +4 carry most of the effect. The rest of gccRccATGG matters far less, which is why checking those two positions is more useful than scoring the whole consensus.

What is the optimal Shine-Dalgarno spacing in E. coli?

5 to 13 nucleotides between the end of the Shine-Dalgarno core and the A of the start codon, with the classic optimum near 7–9. A construct outside that window can still express, which is why the check is soft rather than fatal.

My gene has a perfect Kozak context but expresses poorly in E. coli. Why?

Because a Kozak context does nothing in bacteria. You need a Shine-Dalgarno sequence at a workable spacing instead, and the spacer is a design parameter you can measure rather than guess.

Related references

Related tools

Related guides

Helpful?