Where Does Cas9 Cut Relative to the PAM?

Cas9 cuts both strands of the DNA target approximately 3 base pairs upstream of the PAM (protospacer adjacent motif), generating a break between the third and fourth nucleotides counting from the PAM toward the guide RNA’s 5′ end. That single sentence is the textbook answer, and it is correct as far as it goes. But the reality of how Cas9 makes that cut, and what the resulting break actually looks like at the molecular level, is messier and more consequential for gene editing than the clean diagram suggests.

The Standard Cut Position

For the most commonly used Cas9, from Streptococcus pyogenes (SpCas9), the PAM sequence is NGG, sitting on the non-target strand immediately downstream of the 20-nucleotide target. Cas9 uses two separate catalytic domains to cut the two DNA strands at slightly different spots. The HNH domain cuts the target strand (the one paired with the guide RNA) at a position 3 bp from the PAM. The RuvC domain cuts the non-target strand at roughly the same position, also around 3 bp from the PAM. In many diagrams, this is drawn as producing a clean blunt-ended double-strand break. That picture is a useful simplification, but not the whole story.

The metal ions coordinated in both active sites are essential for this cleavage. Specific residues in the RuvC domain (like D10, E762, H983, and D986) and in the HNH domain (like H840, D839, and N863) each contribute to catalysis. Mutating any one of these can sharply reduce or eliminate cutting by the affected domain, which is the basis of nickase variants used in paired-nicking strategies.

Blunt or Staggered? What the Break Actually Looks Like

For years, the assumption was that because both domains cut at position 3, the result is a perfectly blunt double-strand break. Molecular dynamics simulations challenged this, showing that the RuvC domain can cut the non-target strand one nucleotide further from the PAM than the HNH domain cuts the target strand, producing a 1-bp 5′ overhang rather than a blunt end.¹ That small offset matters because the cell’s DNA repair machinery reads the break structure when deciding how to fix it.

A more recent experimental approach using a technique called BreakTag directly profiled the structure of Cas9-induced breaks in cells. It found that Cas9 generates blunt breaks about 62% of the time, but roughly 35% of the breaks carry 5′ single-stranded DNA overhangs.² So neither “always blunt” nor “always staggered” is accurate. The real profile is a mix, weighted toward blunt but with a substantial minority of staggered cuts.

This heterogeneity helps explain a well-known editing pattern: when Cas9 cuts produce small insertions, the most common result is a 1-bp insertion that matches the nucleotide on the PAM-distal side of the break. If the cut were always blunt, there would be no template for that insertion. A staggered cut with a 1-nucleotide 5′ overhang, however, naturally provides a single unpaired base. When the cell fills in that overhang during repair, it copies that base, producing the characteristic +1 insertion.³

How the Cut Happens Structurally

Cas9 does not simply land on the PAM and instantly snip both strands. Cleavage is a coordinated sequence of large-scale protein rearrangements. After PAM recognition and R-loop formation (where the guide RNA invades and pairs with the target strand), the HNH domain must physically swing into position to reach the target strand’s scissile phosphate. Cryo-EM structures capturing different stages of the process show that this involves a dramatic movement of the HNH domain, on the order of 34 ångströms, while a recognition domain (REC2) becomes disordered.⁴

The recognition lobe plays a more active role than originally appreciated. Simulations of the full enzyme show that the REC lobe senses the nucleic acid duplex, regulates the HNH domain’s conformational swing, and ultimately locks HNH into a catalytically competent position at the cleavage site.⁵ This gatekeeping mechanism is one reason Cas9’s specificity is as good as it is: imperfect guide-target pairing can stall the HNH domain in a pre-catalytic position, preventing cleavage even when the enzyme has already bound the DNA.

After the Cut, Cas9 Stays Put

One of Cas9’s more surprising behaviors is that it does not let go of the DNA after cutting it. The enzyme remains tightly bound to the broken ends for hours, even under conditions that would normally push it off. Studies probing the stability of the post-cleavage complex found that Cas9 actually binds more tightly after cutting than before, and different release dynamics occur depending on which strand has been nicked.⁶ This persistent binding explains why Cas9 is essentially a single-turnover enzyme in practice: it cuts one target and then sits on it, rather than releasing and cutting another.

Recent cryo-EM work has visualized what happens when Cas9 does eventually begin to release the DNA. The PAM-distal half of the cleaved product lets go first, while the PAM-proximal fragment stays bound longer. Two specific residues (R1333 and R1335) maintain hydrogen bonds with the GG dinucleotide in the PAM even in the dissociated state, which directly interferes with the enzyme’s ability to pick up a new target.⁷ This asymmetric release is consistent across different guide RNAs and helps explain why Cas9 editing kinetics are dominated by the first cut event rather than enzymatic recycling.

For anyone designing editing experiments, this lingering occupancy has practical consequences. It slows down the rate at which cellular repair machinery can access the break, which influences the spectrum of repair outcomes. It also means that flooding cells with more Cas9 protein does not linearly increase the number of cuts, because each enzyme molecule is effectively consumed after one cleavage event.

How the HNH and RuvC Domains Compare

Because each domain handles one strand, you can disable either one individually to create a “nickase” that cuts only the target strand (D10A mutation knocks out RuvC) or only the non-target strand (H840A mutation knocks out HNH). This is the basis for paired-nickase strategies, where two nickases guided to nearby sites on opposite strands generate a staggered double-strand break with improved specificity.

Experiments comparing the two nickases found that D10A (which retains the HNH domain for target-strand cutting) is sometimes more efficient at gene disruption than H840A (which retains the RuvC domain).⁸ This difference suggests that the HNH domain has inherently higher catalytic activity than RuvC. For practical guide design, it means that the D10A nickase is generally the preferred choice when using paired nicking to reduce off-target damage.

The key residues in the RuvC active site (D10, E762, H983, D986) coordinate metal ions required for phosphodiester bond cleavage. Mutating any of these significantly weakens or abolishes RuvC activity.⁹ Similarly, N863 in the HNH domain is so critical that its alanine substitution completely eliminates HNH cleavage. These catalytic requirements underscore that the cut position is not just a matter of geometry: the chemical environment within each active site determines whether cleavage actually occurs, and disruptions to either domain produce fundamentally different nicking outcomes.

Guide RNA Length and Cleavage Efficiency

The standard guide RNA (sgRNA) for SpCas9 uses a 20-nucleotide spacer complementary to the target. But researchers sometimes extend the spacer to 30, 40, or even 50+ nucleotides, hoping to boost specificity. The effect on cleavage efficiency is inconsistent and depends heavily on the target site. At one site targeting the GGTA1 gene, a 20-bp spacer achieved about 76% cleavage, but extending to 30 bp dropped efficiency to 18%, and 40 bp dropped it further to about 11%. At a different PAM site in the same gene, 20-bp and 30-bp spacers performed similarly (around 63–76%), and 40 bp stayed above 69%. At yet another site, a 19-bp spacer hit 92% efficiency, and even a 53-bp spacer maintained 82%.¹⁰

The takeaway is that there is no universal “longer is better” or “shorter is better” rule for guide length. The local sequence context, the PAM site, and likely the secondary structure of the extended guide all interact to determine whether a longer spacer helps or hurts. The cut position relative to the PAM stays the same regardless of guide length, but whether the enzyme actually reaches that position and completes cleavage is another matter.

Chromatin Gets in the Way

In living cells, DNA is wrapped around histone proteins to form nucleosomes, and this packaging dramatically affects where Cas9 can cut. Structural studies examining Cas9’s interaction with nucleosomal DNA found that the enzyme can cleave linker DNA (the stretches between nucleosomes) and DNA near the edges of the nucleosome where the DNA transiently unwraps. But Cas9 cannot even bind to the core nucleosome region (roughly the central superhelical positions SHL 0 through SHL ±5), let alone cut there.¹¹

This means that even if a PAM and a perfectly matching 20-nucleotide target exist inside a tightly wrapped nucleosome core, Cas9 may fail to cut. Regions where the DNA occasionally peels off the histone surface (around SHL ±6 to ±7) are accessible during those transient unwrapping events. The practical implication is that chromatin accessibility at your target site can be as important as the guide sequence itself when predicting editing efficiency. Open chromatin regions, promoters, and actively transcribed genes are generally easier targets than silent, densely packaged heterochromatin.

How Cas9’s Cut Compares to Cas12a

Cas12a (formerly Cpf1) is the other widely used CRISPR nuclease, and its cut position relative to the PAM differs from Cas9 in ways that have practical consequences. Cas12a recognizes a T-rich PAM upstream of the target (opposite orientation to Cas9’s downstream NGG) and cuts further from the PAM, typically 18–23 nucleotides away, leaving staggered ends with 5′ overhangs of 4–5 nucleotides. A comparison in rice found that Cas9 editing produced small indels of 1–2 bp or larger deletions of 20–30 bp that included loss of the PAM site, whereas Cas12a editing produced deletions of 2–20 bp that preserved the PAM.¹²

The PAM-destruction pattern matters for experimental design. When Cas9 creates an indel that disrupts the PAM, the site can no longer be re-targeted by the same guide, which is sometimes desirable (prevents re-cutting after successful editing) and sometimes not (prevents iterative editing at the same locus). Cas12a’s tendency to leave the PAM intact means the site remains targetable, which can be an advantage or a problem depending on the application. The cut-to-PAM distance also positions the break differently within the target gene, influencing which codons or regulatory elements are disrupted by the resulting repair.

Off-Target Cutting and the PAM Relationship

Cas9’s requirement for a PAM is its first layer of target selection. Without an NGG adjacent to the potential target, SpCas9 generally will not engage. But the PAM check is only the beginning. Off-target cleavage can occur when genomic sites resemble the target closely enough for the guide RNA to pair, even imperfectly. One underappreciated mode involves bulges, where the DNA target or the RNA guide contains an insertion relative to the other, creating a loop in one strand. Studies have shown that Cas9 can tolerate such bulges and still cleave, and that nickase pairs are not immune to this problem either.¹³

Large-scale datasets cataloging off-target sites have revealed that bulge-containing off-targets are surprisingly common. One reprocessing effort of public experimental data generated hundreds of novel bulge-type off-target sites, highlighting a gap in predictive models that were originally trained mostly on mismatch-type off-targets.¹⁴ For guide design, this means that checking for mismatches alone is not sufficient. Computational tools that account for potential insertions and deletions between the guide and the genome give a more complete off-target risk profile.

Non-Standard DNA Structures Near the Cut Site

The local structure of DNA around the target can interfere with Cas9 activity independently of sequence matching. G-quadruplexes (G4s), which are stable four-stranded structures that form in guanine-rich sequences, are one example. When a high-stability G4 sits near the target site, Cas9’s cleavage activity drops significantly. The effect is more pronounced when the G-rich strand is the one being targeted, presumably because the G4 competes with R-loop formation by sequestering the strand that needs to unpair.¹⁵

This is worth keeping in mind when targeting promoter regions or telomeric sequences, both of which are enriched for G4-forming motifs. A guide that looks perfect on paper may underperform in practice because the local DNA folds into a structure that Cas9 cannot easily unfold. Shifting the guide even a few nucleotides to avoid the G4-forming stretch can sometimes restore activity.

Base Editing and the PAM-Relative Window

Base editors, which fuse a catalytically dead or nickase Cas9 to a deaminase enzyme, do not cut both strands. Instead, they chemically convert one base to another within a defined window relative to the PAM. For the standard cytosine base editor (CBE), the editing window for SpCas9 typically spans about positions −19 to −13 from the PAM (counting upstream). A near-PAM-less variant called SpRY, which can recognize a much broader set of PAM sequences, shifts this window slightly. Researchers found that SpRY-based CBE can edit cytosines as close as 12 bp from the PAM, expanding the reachable positions.¹⁶

Understanding the cut position matters here because even though base editors do not create double-strand breaks, the geometry of Cas9 binding to DNA dictates which bases the fused deaminase can reach. The deaminase sits on a flexible linker and can only access single-stranded DNA in the R-loop bubble. Since the R-loop extends from the PAM inward along the guide, the editing window is anchored to the PAM position. Changing the PAM requirement (as SpRY does) changes which genomic bases fall into the editable window, which is why PAM-flexible variants are so useful for base editing applications where you need to modify a specific nucleotide and the available PAM options are limited.

Repair Outcomes Are Predictable, to a Point

Because the cut position is fixed relative to the PAM and the local sequence is known, researchers have made progress in predicting what the cell will do after Cas9 cuts. The interplay between the break structure (blunt versus staggered), the surrounding sequence, and the cell’s DNA repair pathways produces patterns that are reproducible across experiments. The tendency toward +1 insertions matching the PAM-distal nucleotide, described earlier, is one such pattern. Deletion sizes and microhomology-mediated repair outcomes also show site-specific signatures that can be predicted computationally from the sequence flanking the cut.³

Tools that predict editing outcomes use the known cut position as their anchor point, then model what the cell’s repair machinery will do to that specific break in that specific sequence context. The accuracy of these predictions has improved substantially as training datasets have grown, but edge cases remain. Sites in unusual chromatin states, near repetitive elements, or in cells with atypical repair pathway activity can deviate from the predicted profile. For applications where a specific outcome is needed, like introducing a frameshift at a particular codon, checking the predicted mutation spectrum before committing to a guide can save significant time.