Enhanced proofreading governs CRISPR–Cas9 targeting accuracy

SpCas9-HF1 and eSpCas9(1.1) improve cleavage specificity without weakening target binding. Both variants cleave on-target DNA at rates similar to wild-type SpCas9 while cleaving mismatched substrates significantly more slowly, yet their dissociation constants for both on-target and PAM-distal mismatched substrates match wild-type SpCas9. Specificity in these variants therefore does not come from reduced binding affinity, contrary to the expectation that substituting their charged DNA-contacting residues with alanine would destabilize target binding.

High-fidelity SpCas9 variants suppress off-target cleavage by raising the threshold for conformational activation of the HNH nuclease domain. Measured by smFRET between dyes at S355C in the stationary REC1 domain and S867C in the mobile HNH domain, wild-type SpCas9 stably populates the HNH active state (EFRET = 0.97) on both on-target and mismatched substrates, whereas only about 32% of SpCas9-HF1 molecules reach that state even on an on-target substrate, the remaining 68% being trapped in an inactive intermediate (EFRET = 0.45). A single PAM-distal mismatch at position 20 abolishes stable HNH docking in SpCas9-HF1 entirely, and its inactive-to-active HNH transition is about 8-fold slower than wild-type SpCas9's.

The non-catalytic REC3 domain of SpCas9 recognizes the RNA/DNA heteroduplex and acts as the allosteric effector that licenses HNH nuclease activation. Deleting REC3 slows on-target cleavage roughly 1000-fold while leaving on-target binding affinity near wild-type, and supplying the isolated REC3 domain in trans rescues the cleavage rate about 100-fold in a concentration-dependent manner. The HNH domain of SpCas9ΔREC3 occupies its active state only when REC3 is supplemented, and REC3's own conformational states grow more heterogeneous as PAM-distal mismatches increase.

The REC2 domain mechanically couples heteroduplex sensing in REC3 to HNH activation in SpCas9. REC2 undergoes a large opening from the sgRNA-bound state (EFRET = 0.96) to the target-bound state (EFRET = 0.43), bulk FRET changes in REC2 and HNH are reciprocal across DNA substrates, and in SpCas9-HF1 bound to a single PAM-distal mismatch REC2 stalls at an intermediate state (EFRET = 0.63). REC2 in that position sterically occludes HNH from the target-strand scissile phosphate, holding the nuclease at its conformational checkpoint on off-target substrates.

HypaCas9 is SpCas9 carrying four alanine substitutions in the REC3 domain — N692A, M694A, Q695A and H698A — identified by systematically mutating clusters of conserved residues lying within 5 Å of the RNA/DNA interface. It retains more than 70% of wild-type on-target activity at 19 of 24 endogenous human gene sites, comparable to SpCas9-HF1 at 18 of 24 and eSpCas9(1.1) at 23 of 24.

The REC3 residues that confer fidelity contribute unequally to activity and specificity: N692A and Q695A combined with the Linker 2 substitution Q926A significantly compromise on-target editing in human cells, but restoring either N692 or Q695 alone to its wild-type identity restores robust on-target efficiency.

HypaCas9 discriminates single guide-target mismatches more stringently than either SpCas9-HF1 or eSpCas9(1.1) in the middle and PAM-proximal regions of the spacer, across positions 1 through 18 at the endogenous human FANCF sites 1, 4 and 6. Mismatch discrimination remains sequence-context dependent: single mismatches along FANCF site 2 are tolerated by every SpCas9 variant tested, including HypaCas9.

By GUIDE-seq across six sgRNAs — FANCF sites 2 and 6, VEGFA sites 2 and 3, and DNMT1 sites 3 and 4 — HypaCas9 detects dramatically fewer genome-wide off-target sites than wild-type SpCas9 and matches or improves on both SpCas9-HF1 and eSpCas9(1.1) for every sgRNA examined, at comparable on-target editing and GUIDE-seq tag integration efficiency among all four nucleases.

HypaCas9's specificity gain tracks its raised HNH activation threshold rather than a general loss of catalytic power: against the FANCF site 1 sequence its in vitro on-target cleavage rate is only slightly reduced relative to wild-type SpCas9 and SpCas9-HF1, while its cleavage of internally mismatched substrates is considerably slower than either.

The REC3 residues whose substitution raises Cas9 fidelity are strongly conserved, implying that protospacer sensing and HNH activation run through the same residues across Cas9 orthologues and that specificity can be tuned in other Cas9s by the same route. Wild-type Cas9's tolerance of mismatches is therefore better read as a native balance optimized for host immunity than as a limit on achievable precision.