Articles

CRISPR gRNA Design: Best Practices and Off-Target Avoidance

Cas9 PAM recognition, seed region mismatches, U6 Pol III promoter rules, and oligo duplex cloning

Molecular Cloning

CRISPR gRNA Design: Best Practices and Off-Target Avoidance

Before you can edit a cell line, you usually have to build a plasmid. Most mammalian CRISPR experiments still start with a vector such as pX330, pX459 or lentiCRISPR v2, which delivers the Cas9 endonuclease and a target-specific single guide RNA together.

The cloning itself is quick: two oligos, one Type IIS reaction, done in an afternoon. What decides the success of your experiment is the 20 nucleotides (nt) you put into those oligos. A badly chosen guide gives you no cleavage at all, or cuts in several places it should not and stops your cells from growing.

This guide covers how Cas9 finds its target, the sequence rules that follow from it, the oligo cloning protocol, and what to do when editing fails in a plasmid that sequenced perfectly.

At a glance

  • Mechanism: Cas9 finds a PAM first, then unzips the guide against the DNA starting from the PAM end. This is why mismatches near the PAM are far more damaging than mismatches at the far end.
  • Design: 40% to 60% GC, a G at the first position for U6 transcription, no run of four thymines, and at least three mismatches against any unintended site.
  • Watch for: A TTTT tract in your spacer, phosphate groups missing from both vector and oligos, and off-target cleavage (which shows up in the culture rather than in your sequencing results).

How Cas9 finds its target

The engineered Type II system has two components: the Streptococcus pyogenes Cas9 endonuclease, and a single guide RNA that fuses the targeting crRNA to the structural tracrRNA.

Cas9 does not scan the genome by testing its guide against every sequence. The complex collides with DNA more or less at random, and what it checks on contact is not your guide sequence at all: it checks for a protospacer adjacent motif, or PAM. For wild-type SpCas9 the canonical PAM is 5'-NGG-3'. Non-canonical NAG and NGA motifs can also support cleavage at low frequency, which is worth remembering when you screen for off-target sites.

Only once a PAM is found does the guide come into play. PAM binding bends the DNA and melts the duplex immediately next to the motif. The 20-nt spacer then invades the melted DNA and forms an RNA-DNA heteroduplex, an R-loop, which zips along in one direction: away from the PAM. When the R-loop has propagated the full twenty bases, the HNH and RuvC nuclease domains move into position and cut both strands, three base pairs from the PAM.

This one-directional zip is the fact to hold on to, because every design rule follows from it. The bases next to the PAM are read first and under the least favorable conditions. A mismatch there usually stops the R-loop before it starts, and no cleavage happens. A mismatch at the far end is encountered when the duplex is already mostly formed and holding, so Cas9 will often cut anyway.

Designing the guide

Finding a 20-base sequence next to an NGG is trivial. Choosing one that cuts efficiently and cuts only where you want it to cut takes a few checks:

  • GC content: Keep the spacer between 40% and 60%. Below 40%, the RNA-DNA duplex is unstable and the R-loop tends to collapse before the nuclease domains reach their active position. Above about 65%, you increase the chance of stable binding at partially matched sites elsewhere, and the spacer itself is more likely to fold into a hairpin that blocks the R-loop from forming.
  • The seed region: The 8 to 12 nt immediately next to the PAM are the seed. Mismatches here are close to disqualifying for cleavage, which is useful: it means you can tolerate similarity to other genomic sites as long as the differences fall in the right place. Cas9 can tolerate three or four mismatches at the PAM-distal end (positions 13 to 20). So when you screen candidates, look for at least three total mismatches against any unintended locus, with at least two of them inside the seed.
  • A guanine at position 1: Most mammalian CRISPR vectors transcribe the guide from the human U6 promoter, which is an RNA polymerase III promoter. Pol III initiates efficiently only when the first transcribed base is a G. If your target does not start with a G, either add one to the 5' end of the spacer (giving a 21-nt guide), or choose a 19-nt spacer that is preceded by a G.
  • The base next to the PAM: Empirical cleavage models, such as Doench Rule Set 2, find that a G or C at position 20 (immediately adjacent to the PAM) improves cleavage, and that a T there reduces it. This is a tiebreaker rather than a hard rule, but it is free to apply when you have several candidates.
  • No run of four thymines: A tract of four or more consecutive thymines is the termination signal for RNA polymerase III. If your spacer contains TTTT, transcription stops inside the guide and you get a truncated RNA with no scaffold, which cannot bind Cas9. Discard the candidate. This failure is invisible in the plasmid sequence, which is why it is worth checking at the design stage rather than finding it later.
  • Which exon to target: For a knockout through non-homologous end joining (NHEJ), target an early constitutive exon, ideally in the first 10% to 50% of the coding sequence, and ideally inside a known functional domain. Avoid the first 50 bp of the first exon, where cells can often start translation from a downstream ATG and rescue the protein. Avoid the final exon, where a frameshift escapes nonsense-mediated decay and may leave a partly functional product.

Cloning the guide into the vector

You do not amplify a guide cassette by PCR. The standard vectors are built so that two annealed oligos drop straight into the right place.

The vector carries the constant tracrRNA scaffold, preceded by two inverted BbsI sites (BsmBI in lentiCRISPR v2). Because a Type IIS enzyme cuts outside its recognition sequence (as detailed in Golden Gate Assembly), digesting the backbone removes both recognition sites and leaves two different four-base overhangs, one at the end of the U6 promoter and one at the start of the scaffold. They are not palindromic and they do not match each other, so the duplex can only go in one way round, and the vector cannot easily close on itself.

Step 1: Design the oligos

Order two standard, unphosphorylated single-stranded oligos, 24 to 25 nt each. Do not include the PAM in the oligos.

BackboneForward oligo (5' to 3')Reverse oligo (5' to 3')
pX330 or pX459 (BbsI)CACCG [20-nt target]AAAC [20-nt reverse complement] C
lentiCRISPR v2 (BsmBI)CACCG [20-nt target]AAAC [20-nt reverse complement] C

If your target already begins with a G, leave out the extra G after CACC, and leave out the trailing C on the reverse oligo.

Step 2: Phosphorylate and anneal in one reaction

Synthetic oligos have no 5' phosphate, and T4 DNA ligase needs one. You can order phosphorylated oligos at a premium, or add the phosphate yourself with T4 polynucleotide kinase in the same tube as the annealing.

Use T4 DNA ligase buffer here, not PNK buffer: PNK buffer does not contain the ATP that the kinase needs for the transfer.

Reaction mix (10 µL):

  • 1 µL forward oligo at 100 µM
  • 1 µL reverse oligo at 100 µM
  • 1 µL 10X T4 DNA ligase buffer (supplies ATP)
  • 0.5 µL T4 PNK
  • 6.5 µL nuclease-free water

Thermal cycler program:

  1. 37 °C for 30 minutes (PNK adds the 5' phosphates)
  2. 95 °C for 5 minutes (inactivates the kinase and melts secondary structure)
  3. Ramp down to 25 °C at 5 °C per minute (anneals the duplex cleanly)

This gives you a duplex at roughly 10 µM. Dilute it 1:200 in nuclease-free water (bringing it to ~50 nM) and use the dilution in the ligation.

Step 3: One-pot digestion and ligation

Because the BbsI sites are destroyed when the oligo duplex goes in, you can run the digest and the ligation together rather than gel-purifying a cut backbone first.

Reaction mix (20 µL):

  • 100 ng circular vector
  • 2 µL diluted oligo duplex (1:200 dilution)
  • 1 µL BbsI-HF
  • 1 µL T4 DNA ligase
  • 2 µL 10X T4 ligase buffer
  • Nuclease-free water to 20 µL

Run ten cycles of 37 °C for 5 minutes, then 16 °C for 5 minutes.

Plan it first

Designing CRISPR constructs in a text editor invites mistakes: an oligo copied from the wrong strand, a PAM accidentally included in the spacer, a guide that violates the +1 G rule, or a candidate that happens to sit three mismatches from an exon somewhere else in the genome.

In Photo51, you can find your gene in the built-in Genome Browser, compare candidate guides by their efficiency scores alongside the SNP tracks for the region, and save your chosen guide together with its genomic region to your library, ready for the cloning step (see also: Use CRISPR-Cas Tools in Photo51). Working in genome context rather than in a copied-out sequence is what keeps the guide, the region and the variants in it attached to each other.

Troubleshooting

The plasmid sequenced cleanly and nothing was edited

Symptom: Sequencing confirms your guide is in the vector and correct. Transfection efficiency was good. A T7 endonuclease I mismatch assay, or sequencing of the target locus, shows no indels at all.

Cause: Four or more consecutive thymines in the spacer. TTTT terminates RNA polymerase III transcription, so the guide RNA is truncated inside the spacer and never acquires the scaffold it needs to bind Cas9. The plasmid is exactly what you designed; the RNA is not.

Fix: Discard the guide and pick a new protospacer. There is no way to rescue this one, because the problem is in the twenty bases themselves. Add the TTTT check to your design routine, since nothing downstream of the design will catch it.

Bare plates after the ligation

Symptom: You annealed the oligos, set up the ligation into the digested backbone, and got nothing, or only a trace of background.

Cause: Two things cause this: either nothing in the reaction has a 5' phosphate, or the oligo duplex is in enormous excess.

Fix: For the phosphates, do not treat a Type IIS digested vector with alkaline phosphatase. Because BbsI leaves two incompatible overhangs, the backbone cannot self-ligate efficiently anyway, and removing its phosphates means neither partner can be joined. Leave the vector's phosphates intact so it can ligate to unphosphorylated oligos, or phosphorylate the oligos with T4 PNK as above. For the excess, use the 1:200 dilution. Undiluted duplex at around 10 µM is in vast molar excess over a 10 kb vector and saturates both cut ends with separate oligo molecules, preventing circularization.

The locus is edited, but the cells are dying

Symptom: Your editing worked at the target site, but transfected cells are growing poorly, rounding up or disappearing from the culture, and genome-wide assays show breaks you did not intend.

Cause: Either the guide is cutting at closely matched sites elsewhere, or wild-type Cas9 is being expressed from a strong promoter for days. Several simultaneous double-strand breaks can trigger a p53 damage response and arrest the cell cycle.

Fix: Two options, which can be combined:

  1. Switch to a high-fidelity Cas9 variant such as eSpCas9(1.1), SpCas9-HF1 or HiFi Cas9, which carry mutations that destabilize non-specific contacts and require a closer match before cutting.
  2. Use a Cas9 D10A nickase with two guides targeting opposite strands 10 to 30 bp apart, so that a double-strand break only forms where both guides bind. Single nicks are repaired accurately, and reported off-target rates for paired nickases are substantially lower than for wild-type Cas9.

A third option is a truncated guide of 17 or 18 nt, which lowers the binding energy enough that a single mismatch is usually sufficient to prevent cleavage.