
Golden Gate is the method to use when you have many parts to join. A part is a piece of DNA with a defined four-base overhang at each end, usually kept in its own plasmid so you can use it again in other assemblies. It is the standard word used in Golden Gate and in modular systems such as MoClo. Gibson and NEBuilder HiFi join a handful of fragments well; Golden Gate routinely puts 10 or more parts into a vector in a single tube, in an order you choose.
It is also the basis of the modular systems that much of synthetic biology runs on, including MoClo and GoldenBraid, and it needs no proprietary master mix: a Type IIS restriction enzyme and T4 DNA ligase will do.
What you pay for instead is design work. The four bases at each junction are yours to choose, which means they are also yours to get wrong, and a part with an unnoticed restriction site inside it will be cut in half along with the flanks. This guide covers how Type IIS enzymes make the method work, the overhang rules, and what to do when a one-pot reaction stalls.
At a glance
- Mechanism: A Type IIS enzyme cuts outside its recognition site, so you choose the four-base overhang yourself. Correct ligation destroys the site, so the finished product is no longer a substrate.
- Design: Every junction needs its own overhang, none of them palindromic, and no part may contain an internal copy of the enzyme's recognition site.
- Watch for: Internal sites you did not scan for, a cycling temperature that does not match your enzyme, and too much DNA in a many-part reaction.
How Type IIS enzymes make this work
Traditional cloning uses Type II enzymes such as EcoRI, which recognize a palindromic sequence and cut inside it. The site survives in the product as a scar, and you are limited to whichever sites your sequences happen to contain (see Restriction Enzymes Explained).
Golden Gate uses Type IIS enzymes, most commonly BsaI, BsmBI or BbsI. These recognize an asymmetric sequence and cut at a fixed distance outside it.
You place the recognition sites so they face inwards, flanking each part. When the enzyme cuts, the recognition sequence leaves with the flanking DNA. What remains is your part, carrying a four-base single-stranded overhang whose sequence you specified in the primer.
Because the overhang is independent of the recognition sequence, you decide which part joins to which. And because a correct join leaves no recognition site behind, the enzyme cannot cut the assembled product again. Everything else in this guide follows from these two facts.
The one-pot cycle
Since the correctly ligated product is no longer a substrate, you can put the enzyme and the ligase in the same tube.
The reaction runs in a thermal cycler, alternating between a digestion temperature and 16 °C for ligation, typically for 25 to 30 cycles. Parts that ligate in the wrong arrangement reconstitute a recognition site and get cut apart again on the next digestion step. Parts that ligate correctly destroy the site, and the junction becomes permanent. The reaction therefore accumulates the product you wanted and keeps retrying the rest. Most protocols finish with a longer digestion step to cut any remaining parental vector, then a heat inactivation.
The digestion temperature depends on your enzyme, and this is where protocols get reused carelessly:
- BsaI protocols: cycle at 37 °C.
- Wild-type BsmBI: works best nearer 55 °C, which is too warm for T4 ligase, so BsmBI reactions are usually cycled at 42 °C as a compromise, or run with an engineered variant such as BsmBI-v2.
- Esp3I: an isoschizomer of BsmBI, prefers 37 °C.
Check the temperature for the enzyme you actually have, including the final digestion step, rather than reusing a protocol written for a different one.
Overhang design rules
The traditional rules are straightforward:
- Never use the same overhang twice in one assembly.
- Avoid palindromic overhangs such as ATAT or CGCG. A palindrome is its own complement, so the part can close on itself and drop out of the reaction.
- Avoid an overhang that is the reverse complement of another overhang in the set.
- Keep at least a two-base difference between any two overhangs, avoid three identical bases in a row, and avoid 0% or 100% GC.
These rules will give you reliable assemblies of roughly 10 to 12 parts. Past this point they start to work against you, because they rule out a great deal of sequence space while still letting through pairs that mismatch.
For larger sets, use an empirical design tool such as the NEBridge Ligase Fidelity Viewer or GoSet, which select high-fidelity orthogonal overhang sets from measured ligation data. Orthogonal here means that no overhang in the set pairs with any other. These tools will let you break some of the traditional rules and still assemble 24 or more parts in one pot.
The first two rules hold either way: no duplicates, and no palindromes.
Cost and when to use it
Golden Gate is relatively cheap in terms of reagents. A Type IIS enzyme and T4 ligase cost a fraction of a proprietary assembly mix, and there is nothing in the reaction you cannot buy as a standard catalog item.
For multi-part one-pot reactions, a high-concentration T4 ligase is worth having. The cycling means the correct junction has to be sealed quickly, before the parts come apart again, and a more concentrated ligase does this more often. Standard ligase is usually fine for a simple two or three-part assembly.
The real cost is upfront. Designing and validating an overhang set, and domesticating parts that contain internal sites, takes time that a single-insert clone does not justify. Where Golden Gate pays for itself is in reuse: once your promoters, coding sequences, tags and terminators exist as validated parts with defined overhangs, every new arrangement is one reaction with no new primer design at all.
Plan it first
Golden Gate is easy to get wrong by hand, in four different ways:
- The recognition sites have to face inwards.
- Each junction needs an overhang no other junction uses.
- The reading frame has to survive every fusion.
- No part may contain an internal copy of the recognition sequence.
The internal site is the one that troubles researchers the most. If your gene contains a BsaI site in the middle of its coding sequence, the enzyme will cut your part in half during cycling, exactly as it was designed to cut the flanks. Nothing about the reaction will tell you this has happened, until the colony PCR comes back with the wrong sizes.
In Photo51, you can simulate the digestion and ligation on your actual sequence, inspect each junction, read the frame across it, and scan for internal Type IIS sites in every part before you order primers (or follow our walkthrough: Golden Gate Cloning in Photo51). Include your vector in this check.
Troubleshooting
Colony PCR gives unexpected sizes, and one part contains an internal site
Symptom: Your reaction produced colonies, but colony PCR gives products of several sizes, none of them right. Checking afterwards, one of your parts carries a copy of the enzyme's recognition sequence inside its coding region.
Cause: The enzyme cut your part internally during cycling, in the same way it cut the flanks.
Fix: Domesticate the part before you use it. Introduce a silent point mutation that destroys the restriction site without changing the amino acid sequence, usually with overlapping mutagenesis primers, and sequence-verify the part before it goes into a one-pot reaction. Then scan every other part, and the vector, for internal sites at the design stage.
Mostly uncut or partly cut parts after you changed enzymes
Symptom: You swapped BsaI for BsmBI, kept the 37 °C and 16 °C cycling, and the reaction gives mostly uncut or partially cut intermediates with little or no assembly.
Cause: The digestion temperature no longer suits the enzyme. Wild-type BsmBI cuts slowly at 37 °C, so there is never much substrate available when the cycler drops to 16 °C, and the assembly never gets going.
Fix: Match the cycling to the enzyme. For wild-type BsmBI, cycle at 42 °C and 16 °C, or use a BsmBI-v2 variant at the temperature its protocol specifies. Check the final digestion temperature as well, since it is enzyme-specific too.
Very few colonies on a simple assembly, and the ones you get are empty
Symptom: A three-part Golden Gate gives you a handful of colonies, and all of them contain an empty vector.
Cause: Usually one of two things: your ligation is not keeping up with the cycling, or the vector is not being cut properly and uncut plasmid is carrying through to the transformation.
Fix: Try a high-concentration T4 ligase, which seals correct junctions more reliably under cycling conditions. At the same time, confirm the vector is being cut: run a digest-only control on a gel. If the backbone is largely uncut, check the enzyme, the buffer and the digestion temperature before you change anything else.
A ten or twelve-part assembly gives no correct clones
Symptom: Your overhangs are unique and validated, the enzyme and ligase are right, and a large one-pot reaction still yields nothing correct.
Cause: With many parts, the final step is circularizing one long linear chain, and this is the least likely event in the tube. High DNA concentrations make it less likely still, because each end is more likely to meet another molecule than its own partner.
Fix: Keep the total DNA low and the parts equimolar, and use a validated orthogonal overhang set rather than one you reasoned out by hand. If it still fails, assemble hierarchically: build the first half into one entry vector with BsaI and the second half into another, verify both, then join the two verified halves in a final reaction using a different enzyme such as BsmBI. You spend one extra round of screening and stop guessing.