What Decides a Golden Gate Assembly Order

The part is the fragment with no recognition site left on it

A domesticated level-0 part has two Type IIS sites facing inwards at the insert. Because a Type IIS enzyme cuts outside its recognition sequence, the site itself goes out with the donor backbone and the released part carries none — which is also why the assembled product cannot be re-cut, and why the reaction can run as a single digestion-ligation.

Three part plasmids built that way, cut with BsaI, release fragments of 124, 94 and 154 bp, each carrying its own left and right overhang: AATG/GCTT, GCTT/CGCT and CGCT/AATG.

If two fragments come off without a site, or none does, the right answer is a refusal rather than a guess — the piece you meant is not identifiable from the sequence alone.

The overhangs chain, and the chain has to close

Those three parts chain head to tail: the promoter's right overhang GCTT matches the CDS's left, the CDS's CGCT matches the terminator's left, and the terminator's AATG matches the promoter's left — closing a circle. The product is 372 bp, exactly 124 + 94 + 154, and it is circular.

List the same three parts in a different order and the result is identical: the same cycle, the same length, the same junctions. Only the starting point of the printed list moves, because a circle has no first element.

Change one overhang so it matches nothing — the terminator ending TTCA instead of AATG — and the assembly does not happen. The message names it: the chain does not close, the terminator ends with TTCA and the promoter starts with AATG, so the assembly stays linear, and no orientation of these parts closes it.

A failure that names the part is the useful output

Most of the work in a multi-part assembly that will not go together is finding which part is wrong. Every excised fragment is reported with its size and both overhangs even when the assembly fails, and that table is usually enough on its own.

A part that comes off backbone-sized is the tell that its two sites face outwards — it was domesticated the wrong way round, and the fragment released is the backbone rather than the insert. A part whose overhang matches nothing is either mis-designed or from a different standard, since fusion-site conventions differ between kits.

Pick the wrong enzyme and the refusal is equally specific: with BbsI on BsaI parts, the first part reports zero sites on each strand, rather than returning an empty assembly.

Overhangs that let a part flip

Two overhang patterns break an assembly without stopping it. A self-complementary overhang such as AATT, GGCC or ACGT presents the same sticky end on both strands, so a neighbouring part can ligate either way round. A reverse-complement twin pair — AATG and CATT in the same reaction — makes two junctions indistinguishable.

In both cases the product shown is one of the molecules the reaction gives, and the junction needs redesigning. These warnings fire even when the assembly succeeds, which is the point: the failure is not that nothing assembles, it is that several things do.

Matching is exact; fidelity is a different question

Two overhangs either agree or they do not, and this is what decides whether a product exists. Whether a set of overhangs ligates cleanly in one tube at scale is about T4 ligase mismatch rates between non-identical overhangs, which is measured data rather than string matching.

So a set that assembles perfectly here can still have poor predicted fidelity, and a four-base overhang set is worth scoring against that data separately before committing to it. Three-base sets, such as SapI's, are refused by the scorer here rather than approximated, because the ligation datasets it reads were measured on four-base overhangs.

Related tools

Related guides

• Golden Gate Overhang Fidelity, Scored Against Real Ligation Data: Three Published Sets vs. Two Naive Designs →
• Gibson vs Golden Gate vs Restriction Cloning →
• Where Did That Extra PCR Band Come From? →