A compact, high-accuracy Cas9 with a dinucleotide PAM for in vivo genome editing

Three clades of Neisseria meningitidis Cas9 orthologs share greater than 98% amino acid identity outside their PAM-interacting domains (PIDs) yet recognize distinct PAMs: Nme1Cas9 (the ortholog previously known simply as NmeCas9) recognizes N4GATT, Nme2Cas9 (1,082 aa, ~52% PID identity to Nme1Cas9) recognizes an N4CC dinucleotide PAM with a modest additional preference for a 3′ A, and Nme3Cas9 (1,081 aa, ~86% PID identity to Nme1Cas9) recognizes an N4CAAA PAM.

Nme1Cas9, Nme2Cas9, and Nme3Cas9 share identical tracrRNA sequences and equivalent 24-nucleotide crRNA repeat and guide lengths despite their divergent PAM specificities, so the same single-guide RNA scaffold sequence guides all three orthologs — only the spacer sequence needs to change to retarget each enzyme.

Nme2Cas9 is a compact, 1,082-amino-acid Cas9 ortholog from a distinct Neisseria meningitidis strain that recognizes a simple N4CC dinucleotide PAM immediately 3′ of the protospacer. This PAM affords an average target site density of one site every 8 bp of genomic DNA — identical to the density afforded by the NGG PAM of Streptococcus pyogenes Cas9 (SpyCas9) — while remaining compact enough for single-AAV vector packaging alongside its guide RNA and promoters.

In the integrated TLR2.0 reporter locus in HEK293T cells, Nme2Cas9 edits all 15 tested target sites bearing an N4CC PAM, including sites with each of the four possible nucleotides at PAM position 7, showing that the modest preference for an adjacent A residue observed in vitro does not reflect an actual PAM requirement in cells. Nme2Cas9 shows no detectable editing at N4DC (D = A, T, or G) or N4CD sites in the same reporter, indicating both cytosines of the N4CC PAM are required for efficient activity.

Across 40 endogenous human genomic sites tested in HEK293T cells by transient transfection (29 bearing an N4CC PAM, 11 bearing an N4CD PAM), Nme2Cas9 edits most sites at efficiencies that vary by locus, and it also induces indels — albeit less consistently and at lower levels — at several N4CD sites despite the strict N4CC requirement established at the TLR2.0 reporter. Increasing the amount of Nme2Cas9 plasmid delivered significantly improves editing efficiency at these endogenous sites, and pairing two Nme2Cas9 guides against the same locus extends the platform to precise segmental deletion.

Nme2Cas9 tolerates crRNA spacer lengths of 21 to 23 nucleotides (G21–G23 format, including the 5′-terminal G required for U6 promoter transcription) with comparable editing efficiency at a given target site, but truncating the spacer further to 20 or 19 nucleotides significantly decreases activity.

Nme2Cas9 uses a PAM-proximal seed sequence within its guide RNA to determine target specificity, a feature it shares with previously characterized Cas9 orthologs.

Nme2Cas9 carrying the HNH-inactivating mutation D16A functions as a nickase that supports both NHEJ- and HDR-mediated editing in TLR2.0 reporter cells when paired with a second guide targeting a site 32 bp or 64 bp away on the same DNA molecule, in the presence of a double-stranded GFP donor. Nme2Cas9 carrying the RuvC-inactivating mutation H588A instead, tested under the identical paired-guide conditions, produces no detectable editing at either spacing, showing that only the HNH nickase is functional for Nme2Cas9-based paired-nicking genome editing.

A single plasmid co-expressing Nme2Cas9 and its sgRNA edits the Rosa26 and Pcsk9 loci by transient transfection in mouse Hepa1-6 hepatoma cells. Human K562 leukemia cells engineered to stably express Nme2Cas9 from an SFFV promoter via lentiviral transduction show no detectable difference in growth or morphology relative to untransduced cells, and electroporating sgRNA-expressing plasmids into these stable cells achieves greater than 50% indel formation at all three tested sites.

Purified, recombinant Nme2Cas9 assembled with in vitro-transcribed sgRNA as a ribonucleoprotein (RNP) edits target sites in HEK293T cells by electroporation, and the same RNP delivery method also edits genomic sites in primary human dermal fibroblasts (HDFa) and IMR90 lung fibroblasts — cell types that are otherwise recalcitrant to DNA transfection-based editing. Nme2Cas9 RNP delivery also supports HDR-based precise editing at endogenous loci, with efficiencies similar to those previously reported for SpyCas9.

Four Type II-C anti-CRISPR (Acr) protein families — AcrIIC1Nme, AcrIIC2Nme, AcrIIC3Nme, and AcrIIC4Hpa — completely inhibit Nme2Cas9 DNA cleavage in vitro, and each also inhibits Nme2Cas9 genome editing in HEK293T cells, with AcrIIC3Nme and AcrIIC4Hpa producing complete inhibition of editing and AcrIIC4Hpa showing the most potent, dose-dependent effect of the four. AcrIIC5Smu, the most potent characterized inhibitor of Nme1Cas9, fails to inhibit Nme2Cas9 either in vitro or in cells, and also fails to inhibit an Nme1Cas9 chimera carrying the Nme2Cas9 PID, indicating that AcrIIC5Smu acts through the PAM-interacting domain that differs between the two orthologs.

At three genomic dual sites accessible to both enzymes (flanked by 5′-NGGNCC-3′, permitting matched-guide comparisons, and chosen so the Nme2Cas9 guide directs on-target editing similarly to, less efficiently than, or more efficiently than the corresponding SpyCas9 guide), GUIDE-seq detects markedly fewer off-target sites for Nme2Cas9 than for SpyCas9 in HEK293T cells: SpyCas9 guides direct editing at 93, 10, and 118 candidate off-target sites respectively, while the corresponding Nme2Cas9 guides direct editing at only 1, 0, and 1 off-target sites, despite Nme2Cas9's higher genomic target site density creating more opportunities for off-target matches. GUIDE-seq analyses of three additional Nme2Cas9 guides (targeting TS6 and the mouse Pcsk9 and Rosa26 loci) similarly detect 0, 0, and 1 off-target sites, respectively.

Targeted deep sequencing confirms that the single GUIDE-seq-validated Nme2Cas9 off-target site (Rosa26-OT1, which retains a consensus CC PAM dinucleotide and carries three mismatches to the guide outside the seed region) is edited at only ~1% versus ~30% on-target editing with the same guide in Hepa1-6 cells. Independently, predicted off-target sites for two efficiently edited Nme2Cas9 guides (targeting TS25 and TS47), selected by CRISPRseek to carry two to five mismatches to the guide, show no detectable indels by targeted sequencing despite efficient on-target editing.

A single all-in-one AAV8 vector encoding both Nme2Cas9 (driven by a U1a promoter) and its sgRNA (driven by U6) achieves efficient in vivo genome editing in adult mouse liver after tail-vein injection of 4×10^11 genome copies: targeted deep sequencing detects ~38% and ~46% indels in whole liver tissue at the Pcsk9 and Rosa26 target sites, respectively, corresponding to an estimated 54%–58% and 66%–71% editing when normalized to the hepatocyte fraction of the liver. Off-target editing at the Rosa26-OT1 site remains at only ~2.25% overall (~3%–3.5% in hepatocytes), comparable to the off-target level observed in transfected Hepa1-6 cells.

All-in-one AAV8 delivery of Nme2Cas9 with a Pcsk9-targeting sgRNA reduces serum cholesterol by approximately 44% at both 14 and 28 days post-injection relative to Rosa26-targeted control mice, which show no significant change, and renders liver PCSK9 protein undetectable by western blot without producing histological signs of liver toxicity. This ~44% reduction is comparable to the ~40% cholesterol reduction previously reported for all-in-one AAV delivery of SauCas9 targeting the same gene.

Soaking mouse zygotes in culture medium containing an all-in-one AAV6 vector encoding Nme2Cas9 and a Tyrosinase (Tyr)-targeting sgRNA, without microinjection or electroporation, produces live-born mice with albino, light gray, or variegated coat color, consistent with a high frequency of biallelic Tyr mutations since a single wild-type Tyrosinase allele would preserve black pigmentation; on-target editing detected in tail DNA of these mice ranges from 84% to 100%.

A mutation generated by all-in-one AAV6-mediated Nme2Cas9 editing of a mouse zygote transmits through the germline: breeding an albino female mouse carrying a uniform 2-bp Tyr deletion, generated by this zygotic AAV6.Nme2Cas9 editing, to a classical albino male produces albino offspring in all six resulting pups.