Versatile and robust genome editing with Streptococcus thermophilus CRISPR1-Cas9

St1Cas9 achieves high gene disruption rates in human cells under standard conditions, contrary to earlier reports of weak activity, once its expression and guide are optimized. Flanking the human codon-optimized open reading frame with nuclear localization signals and rebuilding the sgRNA — truncating the repeat:antirepeat region and replacing a wobble base pair in the lower stem with a canonical Watson–Crick pair — produces the best-performing architecture. The same sgRNA changes also markedly improve transcriptional activation by catalytically dead dSt1Cas9 fused to VPR.

St1Cas9 from the LMD9 strain recognizes a core PAM of NNRGAA (R = A or G), functioning efficiently at both NNAGAA and NNGGAA in human and mouse cells, with NNRGAAD (D = A, G or T) as the optimal sequence for consistently high editing; a C at position 7 is tolerated but tends to reduce efficacy. The two-base nonconserved PAM linker cannot be extended to three bases in human cells even though that extension works in bacteria, indicating the mammalian system is more stringent. Guide length shows no obvious correlation with activity, but 20-bp guides are preferred because shorter guides are less mismatch-tolerant in the related SaCas9.

PAM specificity in St1Cas9 is modular and can be reprogrammed wholesale by swapping domains between natural strain variants, which diverge mostly within the C-terminal wedge (WED) and PAM-interacting (PI) domains that sandwich the PAM duplex. Replacing the WED and PI domains of St1Cas9 LMD9 with those of strain LMG18311 or CNRZ1066 redirects targeting to NNGCAA and NNACAA respectively with high efficacy, and swapping sgRNAs between the resulting nucleases produces minimal cross-reactivity. This is true reprogramming toward a new PAM rather than a relaxation of specificity.

Nucleases built on a single St1Cas9 backbone collectively target five distinct A-rich PAMs: LMD9 at NNRGAA, LMG18311 at NNGCAA, CNRZ1066 at NNACAA, TH1477 at NNGAAA and MTH17CL396 at NNAAAA. The last two were predicted by SPAMALOT database mining and built by the same domain-swap strategy, yielding highly active nucleases that retain specificity for their cognate PAM despite differing from one another by only two residues.

St1Cas9 and SaCas9 function orthogonally and can be combined, despite sharing 37% sequence identity and conserved structure. St1Cas9 variants cannot cleave at SaCas9's NNGRRT PAMs in human cells, and the two differ in anti-CRISPR sensitivity: St1Cas9 is inhibited by both AcrIIA5 and AcrIIA6 while SaCas9 is blocked only by AcrIIA5.

PAM readout in St1Cas9 is governed by a small set of PI and WED domain residues, mapped onto a 3.3 Å cryo-electron microscopy structure of St1Cas9 (DGCC7710) bound to sgRNA and a 5′-GCAGAAA-3′ PAM duplex of seven Watson–Crick base pairs. The K1086 side chain hydrogen bonds the guanine at PAM position 4, and variants carrying I1086 in its place lose that specificity. Substituting Q1084 with R1084 confers recognition of a guanine at position 3, analogous to SaCas9 R1015 which occupies the same position relative to its PAM and is anchored by a salt bridge to a residue equivalent to St1Cas9 E1057. K1086 and R1084 are mutually exclusive, since their co-occurrence would clash sterically, and replacing T1048 and M1049 with N1048 and D1049 specifies adenine at positions 4 and 5, mirroring SaCas9 N985 and N986.

A cytosine base editor can be built on St1Cas9 by swapping the nuclease module: replacing SpCas9 D10A with St1Cas9 LMD9 D9A in the SpBE4max architecture, which fuses APOBEC1 cytidine deaminase to a Cas9 nickase along with two copies of uracil DNA glycosylase inhibitor, yields St1BE4max. Its editing window resembles SaBE4's and is wider than SpBE4max's, sometimes extending upstream of the guide.

The PAM-reprogramming domain swaps transfer intact from St1Cas9 nucleases to St1Cas9 base editors. St1BE4max built on the LMG18311, CNRZ1066 and TH1477 variants edits efficiently at NNGCAA, NNACAA and NNGAAA PAMs respectively and is inactive at noncognate PAMs, so the base editors remain orthogonal to one another.

The ABEmax architecture transfers poorly to compact Cas9 orthologs: St1ABEmax built on St1Cas9 LMD9 mediates only moderate A•T to G•C conversion in genomic DNA, matching the moderate efficiency seen for SaABEmax and indicating the architecture is not fully compatible with these shorter Cas9s.

Holo-St1Cas9 — nuclease plus sgRNA — fits within a single AAV vector, and cassette design determines how well it works in vivo. The best-performing design pairs an engineered liver-specific promoter, LP1b, assembled from the human apolipoprotein E/C-I locus control region (ApoE-HCR), a modified human α1 antitrypsin promoter and an SV40 intron, with a synthetic polyadenylation element. This raises cleavage efficacy markedly at low AAV8 dose and produces a vector of approximately 4.7 kb, optimal for viral particle packaging, with in vivo editing levels comparable to the established AAV-SaCas9 system.

A single neonatal injection of AAV8-delivered St1Cas9 can rewire a metabolic pathway and rescue a lethal disease phenotype. Retro-orbital delivery at day 2 of life into nitisone-treated Fah−/− mice, targeting Hpd upstream in the tyrosine catabolic pathway to convert a hereditary tyrosinemia type I phenotype into the milder type III, normalizes excretion of the toxic diagnostic metabolite succinylacetone — near-completely eliminated four months after nitisone withdrawal even at the lower 5 × 10^10 dose. Treatment rescued lethality in all mice and normalized glycemia and weight, whereas saline-treated animals met the weight-loss criterion at about three weeks. Liver indel frequencies rise over the same period, reflecting selective expansion of edited hepatocytes that repopulate the diseased organ.

AAV8-delivered St1Cas9 disrupts a second metabolic gene with a clear physiological consequence. Neonatal injection into wild-type mice targeting Pck1, encoding cytosolic phosphoenolpyruvate carboxykinase 1, produces substantial hepatic gene disruption and, after a 24-hour fast at six weeks of age, markedly increased plasma and hepatic triglyceride content. Circulating free fatty acids are unchanged and hepatic glycogen stores are not depleted, giving a phenotype intermediate to that of a prenatal hepatic knockout — consistent with unedited hepatocytes partially compensating.