Type IIS Enzymes for Golden Gate and MoClo Assembly: BsaI, BsmBI, BbsI, and SapI Compared

How type IIS enzymes generate a programmable overhang

Classic restriction enzymes like EcoRI or BamHI are type IIP enzymes: they recognize a palindromic sequence and cut within it, so the overhang they leave is fixed and is always a fragment of the recognition sequence itself. Type IIS enzymes work differently. BsaI, BsmBI, BbsI, and SapI all recognize a short, non-palindromic (asymmetric) sequence and cut at a fixed distance downstream of it. The two strands are cut at different offsets from each other, which is what creates a staggered, single-stranded overhang rather than a blunt end. Because the cut site is separated from the recognition site, the bases in between and at the cut itself are not constrained by the enzyme at all; the designer chooses them. That is the entire mechanism Golden Gate exploits: put the same recognition sequence next to any 4 (or 3) bases you want, and the enzyme will release a fragment with exactly that overhang.

The standard shorthand for this is REBASE's N(x/y) notation, e.g. GGTCTC(1/5) for BsaI. The number before the slash is how many bases downstream of the recognition sequence the top strand is cut; the number after the slash is how many bases downstream the bottom strand is cut. The difference between the two numbers is the overhang length. Because the recognition sequences are asymmetric, a site only "faces" one direction — Golden Gate vector maps describe sites as facing inward (cutting toward the insert) or outward (cutting toward the backbone), and getting that orientation backwards in a primer design is a common, silent source of failed assemblies.

The four enzymes: recognition sites, offsets, and overhangs

These are the four families actually used in the standards covered later. Each entry below was checked against NEB's product pages, REBASE, and (for the Thermo Fisher isoschizomers) Thermo's own catalog data.

• BsaI / BsaI-HFv2 — recognition sequence 5'-GGTCTC-3', notation GGTCTC(1/5): a 1-base spacer, then a 4-base 5' overhang (positions 2 through 5 downstream of the recognition site). Reaction temperature 37°C. The original BsaI (NEB #R0535) was discontinued December 31, 2020, and replaced by BsaI-HFv2 (NEB #R3733), which has identical specificity but has been engineered for reduced star activity (off-target cutting at near-cognate sequences) and works in rCutSmart Buffer. This is the default enzyme in most Golden Gate part libraries.
• BsmBI-v2 / Esp3I — recognition sequence 5'-CGTCTC-3', notation CGTCTC(1/5): same spacer and overhang geometry as BsaI (1-base spacer, 4-base 5' overhang), but a different recognition sequence, so it is orthogonal to BsaI in a mixed reaction. NEB's catalog unit-definition for both the discontinued original BsmBI (#R0580, replaced March 31, 2020) and its BsmBI-v2 replacement (#R0739) is specified at 55°C — a genuinely higher temperature than the other three enzymes here, consistent with BsmBI's source organism being a thermophile (a Bacillus/Geobacillus stearothermophilus strain). Esp3I recognizes the same CGTCTC(1/5) site but is a genuinely different protein, isolated from Erwinia sp. RFL3, not an engineered variant of BsmBI; it is catalogued for standalone digestion at 37°C, and NEB sells it as its own product (#R0734) in addition to Thermo Fisher's Esp3I line. NEBridge Golden Gate kits run BsmBI-v2 with T4 DNA ligase as a temperature-cycling one-pot reaction rather than holding a single fixed temperature throughout, and the exact cycling parameters have shifted across kit revisions, so treat 55°C as the citable catalog number for BsmBI/BsmBI-v2 and check your kit's current protocol card for the assembly-specific cycling temperatures rather than assuming one number applies everywhere.
• BbsI / BbsI-HF (isoschizomer: BpiI) — recognition sequence 5'-GAAGAC-3', notation GAAGAC(2/6): a 2-base spacer, then a 4-base 5' overhang. Reaction temperature 37°C. BbsI-HF is engineered for reduced star activity and runs in rCutSmart Buffer; per NEB it is not sensitive to CpG, Dam, or Dcm methylation, which matters if you're cutting genomic fragments that carry native methylation.
• SapI (isoschizomers: BspQI, PciSI) — recognition sequence 5'-GCTCTTC-3' (7 bases, one longer than the other three), notation GCTCTTC(1/4): a 1-base spacer, then a 3-base 5' overhang — the shortest overhang of the group. Reaction temperature 37°C. SapI is not used by any of the four major standards covered below; it shows up instead in gene-synthesis and combinatorial-assembly workflows that specifically want the 3-base overhang.

Worked example: from recognition sequence to sticky end

Take a real, widely used fusion site so the arithmetic isn't abstract. In the common syntax adopted across MoClo-derived standards, a coding sequence part conventionally begins with the 4-base overhang AATG, chosen so the ATG start codon sits at the end of it with one extra base (A) upstream to fill out the 4-base overhang. Bird, Marles-Wright, and Giachino's 2022 ACS Synthetic Biology review lays out this convention directly: a part beginning with the site written GGTCTCN▼AATG▲ will ligate with a part whose released end is written ▼AATG▲NGAGACC — the second string being simply the reverse complement of the first, both describing the two ends that meet at the same AATG junction.

Laying out the top strand base by base: positions 1–6 are the recognition sequence GGTCTC; position 7 is the 1-base spacer (any base — call it C); positions 8–11 are A, A, T, G. BsaI's notation GGTCTC(1/5) says the top strand is cut 1 base downstream of the recognition sequence — between position 7 and 8 — and the bottom strand is cut 5 bases downstream — between position 11 and 12. The fragment to the right of the cut therefore starts with AATG as a single-stranded, 4-base, 5'-protruding overhang, with double-stranded DNA resuming at position 12. Note that 1 (spacer) + 4 (overhang) = 5, which is exactly the second number in the (1/5) notation — that arithmetic is a useful sanity check whenever you're reasoning about a type IIS site by hand. A part supplying the upstream sequence (e.g., a promoter or 5' UTR module) is designed with its own inward-facing GGTCTC site oriented so that its released fragment ends in the exact complementary single strand, and the two 4-base single strands anneal to fuse the two parts with no scar other than the AATG itself, which in this case is functional (it contains the start codon) rather than an arbitrary filler sequence.

SapI's shorter overhang follows the same logic with different numbers: GCTCTTC(1/4) means a 1-base spacer and a 3-base overhang (1 + 3 = 4). A 4-base overhang enzyme like BsaI can in principle produce 4⁴ = 256 distinct overhang sequences; SapI's 3-base overhang is limited to 4³ = 64. That reduction matters for reading-frame-sensitive junctions in combinatorial CDS assembly, where a 3-base overhang can be positioned to avoid inserting or deleting a base relative to the intended frame — one reason SapI and its relative EarI are sometimes described as enabling "scarless" fusions, though this depends entirely on where the designer places the 3-base junction relative to the reading frame, not on any property of the enzyme itself.

Enzyme comparison table

Summarized for quick lookup. "Overhang" is the number of bases in the single-stranded 5' overhang generated by digestion; "standards" lists which of the four systems covered in this guide use that enzyme, and at which step (see the next two sections for the full level-by-level mapping).

• BsaI-HFv2 (current) / BsaI (discontinued 2020) — recognition GGTCTC, notation (1/5), 4-base overhang, 37°C — used by MoClo (Level 0→1), the plant Golden Gate toolbox (Level 0→1), CIDAR MoClo (parts→cassettes), and GoldenBraid (α level).
• BsmBI-v2 (current) / BsmBI (discontinued 2020), isoschizomer Esp3I — recognition CGTCTC, notation (1/5), 4-base overhang, 55°C by catalog definition for BsmBI/BsmBI-v2 (Esp3I, a distinct protein with the same recognition site, is catalogued at 37°C by both NEB and Thermo Fisher; Golden Gate kit protocols cycle rather than hold one temperature) — used by GoldenBraid (Ω level) and as the alternate second enzyme in some MoClo end-linker rounds beyond Level 2.
• BbsI / BbsI-HF, isoschizomer BpiI — recognition GAAGAC, notation (2/6), 4-base overhang, 37°C — used by MoClo (Level 0 domestication and Level 1→M, referred to as BpiI in the original paper), the plant Golden Gate toolbox (Level 1→M, also as BpiI), and CIDAR MoClo (cassettes→devices, as BbsI).
• SapI, isoschizomers BspQI and PciSI — recognition GCTCTTC, notation (1/4), 3-base overhang, 37°C — not used by MoClo, the plant toolbox, CIDAR MoClo, or GoldenBraid; used in other Golden-Gate-style gene-synthesis and combinatorial-assembly methods that specifically want a 3-base junction.

Why hierarchical standards alternate enzymes across levels

A finished Golden Gate junction contains no copy of the recognition sequence that built it, because the fusion-site overhang the designer chose doesn't happen to reconstitute that sequence. That's true regardless of which enzyme you used. So in principle you could keep reusing the same enzyme at every level of a hierarchical assembly, as long as no part sequence happens to contain a stray, unintended copy of that enzyme's site anywhere else in it — a promoter, CDS, or terminator can, by chance, contain an internal GGTCTC or GAAGAC that has nothing to do with the designed junctions.

In practice, every published standard covered below alternates between two (sometimes three) enzymes as it moves up levels. The practical reason is that alternating lets a destination vector reuse the same design trick — recognition sites for the current level's enzyme oriented to release the insert cleanly — at every level without worrying about leftover sites from the previous level's vector backbone or cloning history. A part cut out with enzyme A this round can carry incidental leftover B-sites in its backbone from an earlier step, and vice versa; because the reaction only ever contains one of the two enzymes at a time, those leftover sites for the other enzyme simply sit inert. GoldenBraid formalizes this most explicitly: its α- and Ω-level destination vectors carry sites for the same two enzymes (BsaI and BsmBI) in inverted orientation relative to each other, so a composite part that comes out of an α-round of assembly is automatically flanked by Ω-compatible (BsmBI) sites for the next round, and a composite part out of an Ω-round is automatically flanked by α-compatible (BsaI) sites — creating a two-step loop that can repeat indefinitely with only two enzymes and, per the GoldenBraid 2.0 description, eight destination vectors total (pDGBα1, pDGBα2, pDGBΩ1, pDGBΩ2, and their four reverse-orientation counterparts).

MoClo's Level M vectors carry a similar trick with flanking BsaI sites positioned to release the finished multigene fragment intact for excision into a further round of assembly (Level P and beyond); extending the hierarchy past that point requires alternating end-linkers that carry either BsaI or Esp3I sites alongside BpiI — the paper states that cloning beyond level 2i-1 requires "the simultaneous use of two type IIS enzymes: BpiI/BsaI or BpiI/Esp3I." This alternation does not remove the need to check each part for internal copies of whichever enzyme will cut it — that check (usually called domestication: silently mutating out unwanted internal sites, typically via synonymous codon substitution in coding sequence) still has to happen once per part, independent of how many levels the standard has.

Enzyme-to-level mapping in four named standards

The following mappings were checked against the original papers (or, where the paper itself was inaccessible in full text, against a secondary methods description that cites the paper directly).

• MoClo (Weber, Engler, Gruetzner, Werner & Marillonnet, PLoS ONE, 2011). Basic parts (promoters, 5' UTRs, coding sequences, terminators) are PCR-amplified with a BpiI (BbsI) site added by the primers and cloned via BpiI into Level 0 acceptor vectors, which flank the resulting part with inward-facing BsaI sites. Level 1 assembly combines several Level 0 parts into a transcription unit using BsaI. Level 1 vectors in turn flank the transcription unit with BpiI sites, so Level M (multigene) assembly uses BpiI again, combining several Level 1 transcription units plus an end-linker into a single construct. Level M vectors carry an additional pair of BsaI sites so the finished multigene fragment can be excised and moved into a further round of assembly (Level P and beyond), and further multiplexing beyond that alternates end-linkers carrying either BsaI or Esp3I sites — the paper describes cloning beyond level 2i-1 as requiring "the simultaneous use of two type IIS enzymes: BpiI/BsaI or BpiI/Esp3I."
• The plant Golden Gate modular cloning toolbox (Engler, Youles, Gruetzner, Ehnert, Werner, Jones, Patron & Marillonnet, ACS Synthetic Biology, 2014). This toolbox follows the same BsaI/BpiI alternation as the original MoClo paper, adapted with a plant-specific part library: Level 0 parts are flanked by BsaI sites and combined into Level 1 transcription units using BsaI; the resulting Level 1 vectors are in turn flanked by BpiI sites, so Level M (multigene) assembly uses BpiI, and Level M vectors carry both BpiI (consumed during that assembly) and BsaI/BsmBI sites (left in place for release into a further round). Its Level 0 fusion sites follow the community "common syntax" (Patron et al., New Phytologist, 2015), a set of standardized overhangs intended to make Level 0 parts interchangeable across compatible plant Golden Gate toolkits.
• CIDAR MoClo (Iverson, Haddock, Beal & Densmore, ACS Synthetic Biology, 2016). A simplified, E. coli-focused variant with fewer levels: basic parts (promoter, RBS, CDS, terminator) are flanked by BsaI sites and assembled via BsaI into transcriptional-unit "cassettes"; those cassettes are in turn flanked by BbsI sites and assembled via BbsI into multi-cassette "devices," with the BsaI site removed from the cassette in the same way MoClo removes the BpiI/BbsI site during Level 0 domestication. This is the same BsaI-then-BbsI progression as MoClo's Level 0→1→M, just with only two rounds and E. coli-specific parts rather than a deep multigene hierarchy.
• GoldenBraid (Sarrion-Perdigones et al., PLoS ONE, 2011; extended as GoldenBraid 2.0, Sarrion-Perdigones et al., Plant Physiology, 2013). Basic parts ("GBparts") enter the system at the α level and are combined via BsaI. Composite parts move into the Ω level and are combined via BsmBI. Because α- and Ω-level destination vectors (pDGBα1/α2 and pDGBΩ1/Ω2 in the GoldenBraid 2.0 vector nomenclature) carry inverted-orientation sites for both enzymes, a composite part exiting one level is automatically compatible with entering the other, producing the "double loop" that lets assembly continue indefinitely with just two enzymes rather than a new enzyme or vector series at every level.

Choosing the right catalog version: HF, v2, and discontinued originals

If you're working from an older protocol (roughly pre-2020), it may specify plain BsaI (NEB #R0535, discontinued December 31, 2020) or plain BsmBI (NEB #R0580, discontinued March 31, 2020). Both were replaced by engineered versions with identical recognition specificity — BsaI-HFv2 (#R3733) and BsmBI-v2 (#R0739) — that reduce star activity (off-target cutting at sequences similar to, but not exactly matching, the canonical site) and, for BsmBI-v2 specifically, improve performance in one-pot digestion-ligation reactions run with T4 DNA ligase. "HF" and "v2" in NEB's naming both signal an engineered variant with the same recognition sequence and cut geometry as the original; they are not indicating a different enzyme family.

NEB's NEBridge Golden Gate Assembly Kits (E1601 for BsaI-HFv2, E1602 for BsmBI-v2) pre-mix the restriction enzyme with T4 DNA ligase for a single-tube digestion-ligation reaction, typically run as many cycles alternating a digestion-favorable and a ligation-favorable temperature rather than one static incubation. If you're assembling a construct with parts that require different enzymes at different levels (as every standard above does), you run each level as its own separate reaction — mixing BsaI and BbsI/BpiI parts together in one pot only works within a single level of a standard that is explicitly designed to use both simultaneously at that step (as MoClo's Level M assembly does).

Common mistakes and what enzyme choice doesn't guarantee

Knowing an enzyme's recognition sequence and reaction temperature gets a digestion working. It does not, by itself, guarantee a correct assembly. The following are the specific gaps worth knowing about before troubleshooting a failed Golden Gate reaction.

• Correct cutting does not mean correct ligation fidelity. The choice of overhang sequence itself matters: not all 4-base overhangs ligate with equal fidelity when several fragments with similar overhangs are present in the same pot. Potapov et al. (ACS Synthetic Biology, 2018) profiled T4 DNA ligase fidelity across all possible 4-base overhangs and found real, sequence-dependent differences in mis-ligation rates. Before finalizing junction sequences for a multi-fragment assembly, it's worth checking candidate overhangs against that kind of fidelity data — SeqBench's Golden Gate overhang fidelity tool is one way to do this — rather than assuming any four bases will perform identically to any other.
• "HF" branding refers to reduced star activity in digestion, not to ligation fidelity. These are two unrelated concepts that share the word "fidelity" informally in conversation. An HF or v2 enzyme cutting exactly at its intended sites doesn't tell you anything about whether the overhangs you chose will ligate cleanly.
• Enzyme choice says nothing about whether your part sequence contains an unintended internal copy of that enzyme's site. A promoter, CDS, or terminator can contain a stray GGTCTC, CGTCTC, GAAGAC, or GCTCTTC that has nothing to do with the designed junctions; if it does, the enzyme will cut there too, fragmenting the part. This has to be checked and, if necessary, removed (domestication, typically via synonymous codon substitution in coding sequence) before the part enters any standard that uses that enzyme — regardless of how many hierarchy levels that standard has.
• BsmBI, BsmBI-v2, and Esp3I are not drop-in interchangeable at a fixed protocol temperature, and this isn't just a supplier-labeling quirk: BsmBI/BsmBI-v2 and Esp3I are different proteins from different source organisms that happen to recognize the same site. NEB's own catalog specifies 55°C for BsmBI/BsmBI-v2 and 37°C for its own separately sold Esp3I (#R0734); Thermo Fisher's Esp3I is likewise specified at 37°C. Running the wrong temperature for whichever of the two you actually have on the bench can silently reduce cutting efficiency rather than failing outright, which is easy to misdiagnose as a design problem.
• A 3-base-overhang enzyme (SapI) is not interchangeable with a 4-base-overhang standard. Existing MoClo, plant toolbox, CIDAR MoClo, and GoldenBraid part libraries are built around 4-base junctions; swapping in a SapI site does not just change the enzyme, it changes the number of bases available for junction design and requires redesigning every fusion site involved.
• Neither the enzyme's cut nor the overhang's ligation tells you whether the finished construct is biologically correct — reading frame across junctions, correct part orientation, and absence of unintended internal sites are all things Golden Gate's mechanism is silent on. Sequence verification of the finished construct is still necessary; a clean, single-band gel after assembly confirms fragments joined, not that they joined in the intended arrangement.

Related references

Related tools

Related guides

• Gibson Assembly vs. Golden Gate vs. Restriction Cloning: Choosing a Strategy →
• Cloning a Gene From Scratch: A Complete Workflow From Sequence to Verified Construct →
• Golden Gate Overhang Fidelity, Scored Against Real Ligation Data: Three Published Sets vs. Two Naive Designs →