How to Design a CRISPR Guide RNA (gRNA): PAM, Scoring and Off-Targets

PAM sequences and cut sites: SpCas9, SaCas9 and Cas12a

Every guide design starts with the PAM (protospacer adjacent motif), because the nuclease will not bind or cut without one next to the protospacer. The three enzymes commonly used for genome editing differ in PAM sequence, where the PAM sits relative to the protospacer, and where the cut lands:

Because Cas12a's PAM sits on the other side of the protospacer, a region with no usable SpCas9 site can still have a usable Cas12a site, and vice versa, which is one reason to scan for more than one enzyme in a hard-to-target region.

• SpCas9: PAM is 5'-NGG-3', immediately 3' of a 20-nt protospacer. The cut is a blunt double-strand break about 3 bp upstream of the PAM.
• SaCas9: PAM is 5'-NNGRRT-3' (R = A or G), also 3' of the protospacer. The longer, more restrictive PAM gives higher intrinsic specificity than SpCas9, and SaCas9's smaller size lets it fit into an AAV vector, which is why it shows up in in vivo work where SpCas9 is too big to package.
• Cas12a (Cpf1): PAM is 5'-TTTV-3' (V = A, C or G), located 5' of the protospacer, the opposite side from Cas9. Cas12a makes a staggered cut with 5' overhangs, positioned further from the PAM than Cas9's cut.

On-target design rules of thumb

Once you have a set of PAM-adjacent candidates, most of them can be discarded before you even score them. Keep protospacer GC content in roughly the 40-70% range; sequences well outside that band tend to bind or unwind poorly. Avoid a run of four or more T's immediately before the PAM when the guide will be expressed from a U6 or other Pol III promoter, since Pol III reads a poly-T stretch as a termination signal and will often stop transcription there, producing a truncated guide RNA that never reaches the scaffold sequence Cas9 needs to function. Avoid protospacers that overlap repetitive elements or common SNP positions, since both make the guide's behavior inconsistent across genomic copies or individuals.

For whatever candidates remain, position matters more than base identity: mismatches near the PAM-proximal end of the protospacer disrupt Cas9 binding and cleavage far more than mismatches near the PAM-distal end. That positional sensitivity is the basis for most on-target scoring models, which weight each position along the 20 nt rather than treating the protospacer as a uniform stretch.

How off-target risk gets scored

Off-target screening asks a different question: given a protospacer and PAM, where else in the genome does something close enough to that sequence exist that the enzyme might bind and cut there too? The search allows a handful of mismatches rather than requiring an exact match, since Cas9 and its relatives tolerate some mismatch, especially toward the PAM-distal end.

Each candidate off-target hit is then weighted by how many mismatches it has and, again, where those mismatches fall. A hit with two mismatches both in the PAM-distal region is a more realistic off-target than a hit with two mismatches near the PAM, because PAM-proximal mismatches are far more likely to block binding outright. Ranking candidate guides by the number and severity of off-target hits, rather than just counting hits, is what separates a genome-wide screen from a simple sequence-match search.

A practical design workflow

In practice the steps run in this order:

Doing these in order matters: scoring a guide's off-target risk before you have filtered on GC and poly-T rules just wastes time scoring guides you would have discarded anyway.

1. Scan the target region for every candidate PAM on both strands, for whichever enzyme (or enzymes) you are using.
2. Filter out candidates that fail the GC range, contain a poly-T run before the PAM, or overlap a repeat or known SNP.
3. Rank the surviving candidates by on-target score, using a position-weighted model that penalizes PAM-proximal issues most heavily.
4. Take the top few candidates and check them against the genome for off-target hits, again weighting by mismatch count and position.
5. Pick the guide with the best balance of on-target score and off-target profile before ordering oligos.

Where the CRISPR gRNA Designer fits in

Doing the scanning and filtering steps by hand means checking both strands of a sequence for three different PAM patterns, checking GC content and poly-T runs on every hit, and applying a position-weighted score consistently across dozens of candidates. SeqBench's CRISPR gRNA Designer does that scan for you: give it a sequence and it returns SpCas9, SaCas9 or Cas12a guide candidates with their PAMs and scoring already computed, so you can go straight to comparing a ranked shortlist instead of eyeballing a sequence for NGG triplets.

Once you have picked a candidate, it is worth running the same target region through Restriction Sites to check whether the guide or its flanking sequence overlaps a recognition site you plan to use for cloning the oligo into a vector. If you need a fuller picture of the region, such as reading frame, composition and primer options alongside restriction sites, Sequence Analyzer builds that summary from the same input sequence.

Related references

Related tools

Related guides

• How to Analyze an Unknown DNA Sequence: A Step-by-Step Workflow →
• Restriction Enzymes and How to Plan a Digest →
• How to Find a Motif or Pattern in a DNA Sequence →