Different Cas9 proteins for editing genes and gene activity

We used the CRISPRdisco pipeline (23) to mine bacterial genomes from select Lactobacillales genera that are commonly used in the food supply chain, widely associated with agricultural samples and hypothesized to be compatible with human exposure based on their broad use in food manufacturing and historical healthy occurrence in human microbiomes (24) (Fig. 1A). We also focused on streptococci and lactobacilli because these genetically and functionally diverse genera are enriched in CRISPR-Cas systems in general and Cas9-associated type II systems in particular. For each candidate Cas9, we predicted a PAM through spacer-protospacer matching and designed a single guide RNA (sgRNA) based on CRISPR repeat and tracrRNA predictions (25). The selected Cas9 candidates range from approximately 1,100 to 1,400 amino acids in length and exclusively belong to type II-A CRISPR-Cas systems, except for a single type II-C ortholog (Fig. 1B). Phylogenetic trees, constructed based on protein sequence similarity, reveal that our selected candidates cluster into three main groups, with two clusters predominantly composed of Cas9 from Streptococcus and one cluster exclusively featuring Cas9 from Lactobacillus and Pediococcus. Notable exceptions among our candidates include the type II-C Cas9 from L. backii, which closely resembles reference type II-C Cas9 orthologs, and a Cas9 from L. mali that is clustered with the smaller Cas9 orthologs derived from Streptococcus. Additionally, the alignment of putative tracrRNA sequences associated with these Cas9 proteins revealed similar phylogenetic relationships, underscoring the coevolution of Cas9s and their corresponding tracrRNAs (Fig. 1C). This comprehensive analysis yielded a list of candidate Cas9 orthologs, each paired with a corresponding putative PAM and sgRNA, therefore enabling their preliminary functional characterization as epigenome editors (Datasets S1↗ and S2↗ for sequences of all 47 Cas9s and sgRNA).

The 47 wild-type Cas9 sequences were human codon-optimized and converted to nuclease-deactivated mutants (dCas9) via alanine substitutions at catalytic residues (3, 26) of the predicted RuvC and HNH nuclease domains (27, 28). These dCas9 sequences were fused with the ZNF10 KRAB repressor domain (29) and cloned into a lentiviral vector with an EGFP reporter to create a dCas9-KRAB-2A-EGFP lentiviral construct for epigenetic gene repression in human cells (Fig. 2A). We reasoned that while nuclease activity requires both DNA binding and cutting, transcriptional repression with a dCas9-KRAB fusion primarily relies on DNA binding, potentially expanding the number of functional Cas9 orthologs we could identify. To assess efficacy in repressing an endogenous gene in human cells, we used a K562 cell line containing the HBE gene tagged with an in-frame mCherry reporter protein, facilitating rapid evaluation of HBE knockdown (30) (Fig. 2A). Because repression efficiency is typically dependent on the target sequence, we designed up to five spacer sequences targeting the HBE promoter region for each computationally identified PAM sequence using ChopChop (31, 32) (Fig. 2A and Datasets S1↗ and S2↗). Considering the scarcity of some predicted PAM sequences, we searched for possible spacer sequences within a 1 kb region flanking the HBE transcription start site (Fig. 2A). We designed five spacers for each Cas9 except for Pediococcus damnosus, where the predicted PAM (NGATACA) was found only once (Datasets S1↗ and S2↗). As a control, we designed five spacers targeting the same TSS-flanking region for Streptococcus pyogenes Cas9 for the NGG PAM, using the same settings in ChopChop as for all other candidate Cas9 orthologs. A validated S. pyogenes sgRNA which targets the HBE enhancer and mediates potent downregulation of HBE was used as a positive control (30).

HBE-mCherry K562 cells were transduced with each dCas9-KRAB-2A-GFP construct and subsequently transduced with the corresponding pool of sgRNA lentiviruses (up to five sgRNAs per CRISPR system) (Fig. 2A). Nine days posttransduction, we assessed mCherry expression by flow cytometry. Out of 47 samples of dCas9-transduced cells, 34 exhibited a sufficient number of GFP-positive cells for analysis, indicating that low lentiviral expression of these uncharacterized Cas9 constructs is a common failure mode. Several dCas9-KRAB and sgRNA combinations markedly decreased mCherry expression (Fig. 2B). From those showing significant repression, we selected six dCas9s for further validation, prioritizing shorter Cas9 protein length, robust gene repression, and systems that have not otherwise been characterized. This validation process included individual testing of each of the five sgRNAs from the sgRNA pool used for HBE-mCherry repression. Notably, dCas9s from S. iniae, S. gallolyticus, S. lutetiensis, S. parasanguinis, and S. uberis, fused with KRAB, were successfully validated, demonstrating significant downregulation of HBE-mCherry with at least one sgRNA (Fig. 2C and SI Appendix, Fig. S1↗).

To evaluate the specificity of repression of the three dCas9-KRAB species with the strongest gene repression effects, we performed RNA sequencing analysis. We transduced K562 cells with lentivirus encoding dCas9-KRAB and a sgRNA targeting the HBE promoter and compared differentially expressed genes between samples transduced with a nontargeting sgRNA or the HBE-targeting sgRNA (Fig. 3). All three dCas9-KRAB proteins from S. uberis, S. gallolyticus, and S. iniae showed highly specific gene repression, with HBE1 being the most significantly and strongly downregulated gene. All other downregulated genes with a log₂(fold-change) greater than 0.5 were in the globin locus, such as HBA2. Modulation of additional globin genes is expected due to the known interdependency of gene expression at the globin locus (33–35). We used our previously validated HS2 enhancer-targeting S. pyogenes sgRNA (30) as a positive control, which showed similar levels of specific HBE1 downregulation, as well as an expected downregulation of HBG1 (Fig. 3). These differentially expressed genes were also validated by RT-qPCR, which confirmed significant changes in HBE1 and HBG1 in the expected direction (SI Appendix, Fig. S2↗).

We next set out to understand the relationship between Cas9 orthologs previously reported to be functional in mammalian cells and these Cas9 orthologs, narrowing our selection to four Cas9s based on their PAM flexibility and smaller size for potential AAV delivery. Sequence alignment between S. iniae (Sin), S. gallolyticus (Sga), S. parasanguinis (Spa), and S. uberis (Sub) Cas9s with those from Staphylococcus aureus (Sau) and S. pyogenes (Spy) highlighted distinct similarities and differences. SgaCas9, SpaCas9, and SubCas9 showed greater sequence identity to SauCas9 (32 to 35%), featuring a shorter REC domain and an extended wedge domain compared to SpyCas9 (Fig. 4 A and B). SgaCas9, SpaCas9, and SubCas9 also show 60 to 65% identity to each other. Conversely, SinCas9 demonstrated 70% sequence identity to SpyCas9, contrasting with an 18% identity to SauCas9 (Fig. 4B). Structural analyses, leveraging data from the AlphaFold database (36), revealed close structural alignment (1 to 3 Å RMSD) of SauCas9 with SgaCas9, SpaCas9, and SubCas9, despite sequence identities of only 32 to 34% (Fig. 4 B and C). Comparison to additional commonly used Cas9 proteins with evidence of activity in human cells, including those from N. meningitidis (Nme), C. jejuni (Cje), and S. thermophilus (Sth), revealed that SgaCas9, SpaCas9, and SubCas9 share 20 to 23% sequence identity with NmeCas9 and CjeCas9 (Fig. 4B) and a structural alignment RMSD of 4.7 to 9.7 Å (SI Appendix, Fig. S3↗). SgaCas9, SpaCas9, and SubCas9 share only 16% sequence identity (Fig. 4B) and a structural alignment RMSD of 11 to 21 Å with SthCas9 (SI Appendix, Fig. S3↗). Conversely, SinCas9 was more closely related to SthCas9, with 53% sequence identity, compared to only 15 to 16% identity to NmeCas9 and CjeCas9 (Fig. 4B). Comparison of predicted sgRNA structures reveals a similar grouping, in which S. iniae more closely resembles S. pyogenes with two 3’ hairpins. By contrast, sgRNAs from S. gallolyticus, S. parasanguinis, and S. uberis more closely resemble the S. aureus sgRNA, with a single 3’ hairpin, although differences between these orthologs are seen in the nexus and 3’ hairpin structure (SI Appendix, Fig. S4↗).

The predicted PAM sequence for the four validated Cas9 orthologs was further empirically verified using an in vitro cleavage assay with Cas9 proteins and sgRNAs targeting a constant spacer flanked by a randomized PAM of seven base pairs (7 N). This revealed distinct PAM preferences for each Cas9 after deep sequencing of cleaved amplicons, which consistently aligned with predicted PAM sequences (Fig. 4D). Notably, certain orthologs exhibited additional flexibility beyond the predicted PAMs. For example, while the predicted PAM sequence for SubCas9 was NNAATA, empirical determination revealed NNARTA, effectively doubling the potential target space.

To investigate the expanded target space provided by these orthogonal PAM sequences, we analyzed the theoretical base editing target space for each uncharacterized Cas9 protein, as well as SauCas9 and SpyCas9. We based our analysis on all single nucleotide variants (SNVs) that can be addressed by base editing (37). We found that the PAM sequences of our Cas9 proteins could potentially add 6,867 targetable sites to the 20,823 sites targetable by SpyCas9 and SauCas9 base editors, out of a total of 57,810 base-editable SNVs in ClinVar (SI Appendix, Fig. S5↗).

To characterize the in vitro nuclease activity of these systems, each Cas9 protein was expressed in Escherichia coli, purified, and complexed with an in vitro transcribed sgRNA corresponding to the best HBE-targeting protospacer (Fig. 2C) for each Cas9. Incubating these ribonucleoprotein complexes (RNPs) with a DNA PCR amplicon containing the HBE target site demonstrated efficient in vitro nuclease activity in all cases (Fig. 5A). To further evaluate gene editing in human cells, plasmids encoding Cas9 proteins and sgRNAs were transfected into HEK293T cells (Fig. 5B). Deep sequencing of the endogenous HBE target sites in genomic DNA and quantification of indel frequencies demonstrated that all tested Cas9s are effective nucleases in mammalian cells (Fig. 5 C and D). SubCas9 exhibited the greatest cleavage activity, achieving ~35% editing at the best target, comparable to SpyCas9. SgaCas9 showed ~20% editing at both targets, while SpaCas9 and SinCas9 exhibited ~10% editing with at least one sgRNA. To demonstrate the versatility of SubCas9 in targeting various sequences, we designed two additional sgRNAs targeting the open reading frame of the TRAC gene, a therapeutically relevant target to prevent graft-versus-host disease in allogeneic T cell therapies (38), and transfected plasmids encoding Cas9 and these sgRNAs into HEK293T cells. Editing rates of ~35% were also observed for the best TRAC-targeting sgRNA, demonstrating that SubCas9 is a potent and robust nuclease in mammalian cells (Fig. 5C). Although the 20-nucleotide spacers used in these experiments proved sufficient for achieving editing, we sought to determine the optimal spacer length for the strongly active SubCas9 as the optimal length can vary between Cas9 systems (39, 40). We generated versions of the best HBE and TRAC sgRNAs for SubCas9 using spacer lengths ranging from 16 to 24 nucleotides (Fig. 5D). Spacer lengths below 20 nucleotides were found to completely ablate nuclease activity, while the optimal length was found to be 21 or 22 nucleotides, with the 21-nucleotide TRAC protospacer achieving 60% editing.

We also investigated the potential for S. uberis Cas9 to function as an adenine base editor (41, 42), utilizing this validated TRAC sgRNA. We generated a nickase Cas9 (nCas9) version of SubCas9, which creates a single-strand nick instead of a double-strand break in the target DNA and is thus suitable for base editing. S. uberis nCas9 was fused to the tadA adenine base editor domain and cotransfected into HEK293T cells along with TRAC or nontargeting sgRNA. We observed approximately 1% conversion of adenine to cytidine bases at protospacer positions 7, 9, 11, and 14 (SI Appendix, Fig. S6↗). These results confirm that S. uberis nCas9 can indeed function as a base editor and suggest an expanded editing window similar to that observed for SaCas9 base editors (43).

To further explore the epigenome editing capabilities of these dCas9 proteins, we fused each to the catalytic core of histone acetyltransferase p300, which we previously reported as a robust activator of promoters and enhancers (44). We then assessed HBG1 gene activation in HEK293T cells after transfection with plasmids encoding dCas9-p300c and either an HBG1 promoter-targeting sgRNA or nontargeting control (Fig. 6A). The dCas9-p300c fusions from S. uberis and S. gallolyticus successfully upregulated HBG1, while those from S. iniae and S. parasanguinis did not show any activation (Fig. 6B and SI Appendix, Fig. S7↗). Among the 12 sgRNAs tested with S. uberis dCas9-p300c, three significantly upregulated HBG1 expression, with one achieving levels comparable to S. pyogenes dCas9-p300c paired with a previously optimized HBG1 promoter-targeting sgRNA (Fig. 6B).

We selected PCSK9 as an example of a therapeutically relevant target for gene repression, as PCSK9 repression in hepatocytes leads to reduced cholesterol levels and is a promising clinical target for treatment of hereditary hypercholesterolemia (26, 45). Previous studies have demonstrated efficient PCSK9 repression using S. aureus and S. pyogenes dCas9-KRAB systems both in vitro and in vivo (46, 47). For each dCas9 variant, we designed four to six sgRNAs targeting the region −200 to +300 bp relative to the PCSK9 transcription start site, representing all experimentally determined PAM sequences. We included S. aureus dCas9-KRAB as a positive control. In the AML12 mouse hepatocyte cell line, S. uberis dCas9-KRAB significantly repressed PCSK9 with two out of four tested sgRNAs, achieving repression levels comparable to S. aureus dCas9-KRAB using previously optimized sgRNAs (Fig. 6C). S. parasanguinis dCas9-KRAB achieved approximately 50% repression of PCSK9, while the remaining two dCas9-KRAB/sgRNA combinations did not significantly repress PCSK9 expression (Fig. 6C). We further confirmed transcriptome-wide specificity of PCSK9 repression by S. uberis dCas9-KRAB using RNA sequencing analysis. After transduction of AML12 cells with S. uberis dCas9-KRAB and PCSK9-targeting or nontargeting sgRNA, the most significantly repressed gene was PCSK9 (Fig. 6D). Eight other transcripts were also significantly downregulated; however, these genes were also downregulated when PCSK9 is targeted by SauCas9 in AML12 cells (SI Appendix, Fig. S8↗), suggesting that these are downstream consequences of PCSK9 repression rather than off-target activity of S. uberis dCas9-KRAB.

All available prokaryotic genomes were downloaded from the National Center for Biotechnology Information’s Reference Sequence Database (52) (May 2018). CRISPR arrays were identified using the CRISPR Visualizer software package (53). To facilitate more rapid processing of each genome, only the regions 20 kb upstream and downstream of identified CRISPR loci were kept for subsequent processing (genome trimming). Cas genes were annotated using the CRISPRdisco software pipeline (23). The resulting repeat, spacer, and cas sequences were stored in a Postgres database.

To determine spacer identity, all spacer sequences for the genera Lactobacillus, Pediococcus, and Streptococcus were compared against NCBI’s Nucleotide Collection database (nt) via nucleotide BLAST (54) using default parameters with the following exceptions: -e-value 1e-3, -task blastn-short, -dust no. BLAST alignment matches that were viral or plasmid in origin were aligned, and PAMs were determined after visual inspection of the alignments in Geneious (55). A consensus repeat sequence for each CRISPR locus was determined by multiple alignment of all available repeat sequences per array. The consensus repeat was used to predict the putative tracrRNA sequence through the identification of the antirepeat sequence, as previously described (56). sgRNA sequences were generated by joining the repeat and predicted tracrRNA sequences to form a crRNA:tracrRNA duplex with a short nucleotide tetraloop: “CGAA,” as previously described (25).

The Cas9 protein sequences were aligned using MUSCLE 3.8.425 with default parameters in Geneious (57). A consensus tree was generated using the Geneious Tree Builder in Geneious with a Jukes-Cantor model, a neighbor-joining tree build method, and 100 replicates. The tracrRNA sequences were aligned using MUSCLE with default parameters in Geneious. A tree was generated with Geneious by using the RAxML plugin with a GTR GAMMA nucleotide model, a rapid hill-climbing algorithm, and 20 replicates (58). Trees were visualized using iTOL (59). After aligning the protein sequences, each Cas9 ortholog was modified into a dead Cas9 (dCas9) by introducing specific single amino acid changes within the HNH and RuvC catalytic domains. These changes are analogous to the D10A and H840A mutations found in SpyCas9, which are known to deactivate the nucleolytic activity of the RuvC and HNH domains, respectively (3).

To generate dCas9^KRAB plasmids for all candidate dCas9s, we modified a lentiviral expression plasmid (Addgene no. 83925) by removing the U6 promoter and gRNA scaffolds and inserting two AgeI restriction sites separated by a short stuffer sequence, an SV40 nuclear localization sequence, and KRAB-2A between the promoter and GFP-2A-BlastR sequences. The sequences of the candidate Cas9s were supplied to Twist Biosciences for human codon optimization and homology-based cloning into the AgeI site in our custom expression plasmid.

Individual or pooled gRNA plasmids were cloned using Gibson assembly downstream of the hU6 promoter into a plasmid expressing a puromycin resistance gene from a separate, hUbC promoter. gRNA scaffolds and protospacers were ordered as long oligonucleotides (IDT) and extended via PCR (Q5, NEB).

For protein production, E. coli codon-optimized genes encoding nuclease-active Cas9 from S. uberis, S. gallotyticus, S. parasanguinis, S. iniae, and S. pyogenes were cloned into pET29B between the NcoI and PacI sites by Gibson assembly.

For plasmid used in nuclease experiments in HEK293T cells, nuclease active sites were restored in the human codon-optimized DNA encoding Cas9 from S. ub., S. ga., S. pa., S. in., and S. py. and cloned between the NcoI and PmeI sites of Addgene plasmid 61357 by Gibson assembly to generate CMV-driven constructs optimal for transfection of mammalian cells.

K562 HBE-mCherry cells (30) were maintained in RPMI 1640 and used to test gene repression activity of each dCas9-KRAB with sgRNAs targeting the HBE promoter. K562 HBE-mCherry (40,000 cells) were transduced with dCas9-KRAB lentivirus (50µL) corresponding to each Cas9 in 500µL RPMI with 5 µg/mL polybrene. Three days later, the resulting dCas9-KRAB cell line was further transduced with pooled (Fig. 2B) or individual (Fig. 2C) sgRNA lentivirus (sgRNAs in a cassette containing a puromycin resistance gene) and selected with puromycin for 72 h. Then, 9 or 10 d posttransduction, cells were harvested and EGFP-positive cells were analyzed for mCherry repression on a flow Sony SH800 cytometer. For RNA-seq experiments, EGFP-positive cells were first sorted and collected, followed by sgRNA transduction, culture, and RNA extraction using the Qiagen RNeasy kit according to the manufacturer’s instructions.

The PAM sequence for each Cas9 protein was determined empirically as previously described (60). Individual 12 µL TXTL reactions were assembled consisting of 9.375 µL myTXTL Linear DNA Master Mix (Daicel Arbor Biosciences), 0.5 mM IPTG, 0.2 nM pTXTL-P70a-T7rnap (Daicel Arbor Biosciences), 2 nM Cas9 linear DNA containing a T7 promoter and 2 nM linear sgRNA expression gBlock, and 0.5 nM 7 N plasmid library. TXTL reactions were incubated at 29 °C for 16 h. The DNA was purified from the TXTL reactions with the Monarch PCR & DNA Cleanup Kit (NEB). DNA libraries were then amplified by PCR and subjected to Illumina sequencing to determine counts of each sequence.

All SNVs from ClinVar which are addressable by base editing (37) were analyzed for the presence of a PAM for each Cas9 which would place the variant within the editing window. For SpyCas9, the editing window used was protospacer positions 4 to 8. Protospacer positions 3 to 11 were considered to be the editing window for SauCas9 (43), as well as for each of SpaCas9, SgaCas9, and SubCas9. This window was used based on both the high degree of structural homology of SpaCas9, SgaCas9, and SubCas9 with SauCas9, as well as our experimental verification of the editing window with an SubCas9 adenine base editor (SI Appendix, Fig. S6↗).

BL21 E. coli cells (Millipore EMD; MilliporeSigma, Burlington, MA) were transformed with Cas9 expression plasmid and plated on plates containing carbenicillin. Liquid cultures were then inoculated and allowed to grow at 37 °C until the OD600 was 0.6 to 0.8, and then induced with 0.5 mM IPTG and grown overnight at 18 °C. Cultures were then pelleted by centrifugation (10 min at 4,000×g). Cells were resuspended in lysis buffer, lysed by sonication, and spun at 20,000×g to remove cell debris. The lysate was then flowed over a 2 mL bed volume of Ni-NTA agarose (Qiagen), washed twice with wash buffer, and then once with wash buffer without triton. Protein was eluted by the addition of 10 mL elution buffer. Concentration was determined by qubit. The buffers used for protein purification included the following: Lysis Buffer (20 mM Tris-HCl pH 8.0, 500 mM NaCl, 10 mM imidazole, 5% glycerol, 1 mg/mL lysozyme, 1 tablet Complete protease inhibitor, EDTA-free); Wash Buffer (20 mM Tris-HCl pH 8.0, 500 mM NaCl, 20 mM imidazole, 0.5% triton x-100); Elution Buffer (20 mM Tris-HCl pH 8.0, 500 mM NaCl, 500 mM imidazole).

For in vitro cleavage reactions, sgRNA were produced by in vitro transcription using the Megashortscript kit (ThermoFisher, Waltham, MA) according to the manufacturer’s instruction for each sgRNA.

In a 50 µL reaction, 100 nM Cas9 protein and 100 nM sgRNA were combined with 6 nM target DNA in 150 mM NaCl, 20 mM tris pH 8, and 10 mM MgCl₂. Reactions were incubated at 37 °C for two hours and then terminated by the addition of proteinase K (500 µg/mL final concentration) and RNase A (10 µg/mL) followed by a 10-minute incubation at 37 °C. DNA was then purified using Ampure XP DNA cleanup beads and analyzed on a 1% agarose gel.

HEK293T cells were maintained in DMEM supplemented with 10% FBS and 1x penicillin/streptomycin. The day before transfection, 200,000 HEK293T cells were plated in 24-well plates. Plasmids encoding nuclease-active Cas9 (200 ng) and sgRNA (300 ng) were transfected into HEK293T cells using lipofectamine 3000 according to the manufacturer’s instructions. For base editing, plasmids encoding S. uberis TadA-nCas9 (300 ng) and sgRNA (200 ng) were used. The S. uberis TadA-nCas9 plasmid was based on ABE8e (42). After 72 h, gDNA was extracted from transfected cells. Extracted gDNA was then subjected to PCR to generate amplicons flanking each target site and sequenced on an Illumina Miseq. The editing efficiency after cotransfection with SubCas9 and sgRNA was measured by next-generation sequencing (NGS). Custom python scripts were used to assess the number of reads containing insertions and deletions or direct A-to-G conversions in the case of base editing, relative to the wild type sequence at each target site.

sgRNAs for all activation and repression assays were designed using ChopChop version 3 (31, 32, 61) with the following settings differing from default: Custom PAM, G20 efficiency score, no repair profile predictions, no 5’ requirements for sgRNA, and custom sgRNA backbone. Sequences targeted the gene promoter and TSS-proximal regions, but searches were sometimes expanded to within 1,000 bp of the TSS depending on PAM frequency. See Dataset S1↗ for all sgRNA sequences, and Dataset S2↗ for sequences for all 47 Cas9s.

AML12 cells were obtained from ATCC (cat no.CRL-2254) and cultured in DMEM/F12 (Thermo cat no.11320033) with 10% FBS, ITS-G (Gibco 41400045), and 40 ng/mL dexamethasone. To generate stable polyclonal dCas9-KRAB cell lines, AML-12 cells were transduced with lentivirus encoding dCas9-KRAB-2A-GFP-2A-BlastR and selected for 3 d with 3 µg/mL blasticidin. After antibiotic selection, GFP-expressing cells were isolated via cell sorting (Sony SH800Z in semipurity mode). To assay PSCK9 repression in each dCas9-KRAB cell line with sgRNAs, cells were plated at 1.82E4 cells/cm² and immediately transduced with sgRNA virus. Approximately 30 to 48 h after transduction, puromycin was added to fresh media at 1 µg/mL. Cells were selected with puromycin for 4 d. Cells were harvested with Trypsin 0.05% (Gibco cat no. 25300054) 9 d after transduction. RNA was extracted from frozen cell pellets with the Norgen Total RNA Purification Plus kit (Norgen cat no. 48300). Then, 500 ng RNA per sample was reverse-transcribed using the Superscript VILO cDNA synthesis kit (Invitrogen cat no. 11754050) and analyzed by RT-qPCR.

K562 or AML12 cells were transduced as described above with vectors encoding dCas9-KRAB and appropriate sgRNAs. Cells (1 to 2 million) were pelleted and RNA was extracted using the RNeasy kit according to the manufacturer’s instructions. Standard RNA sequencing with polyA selection and ribosomal RNA depletion was performed by Azenta. Raw reads were processed into counts for each gene using the ENCODE standard RNA sequencing pipeline (62). Counts were then compared between HBE-targeting or PCSK9-targeting and nontargeting sgRNA samples using DEseq2 with independent filtering off and shrunken residuals.

G.L.B., D.R., A.R., M.A.N., N.I., R.B., and C.A.G. designed research; G.L.B., D.R., A.R., M.A.N., A.J.R., D.J.M., L.G., and N.I. performed research; G.L.B., D.R., A.R., M.A.N., A.J.R., N.I., R.B., and C.A.G. analyzed data; and G.L.B., D.R., A.R., M.A.N., N.I., R.B., and C.A.G. wrote the paper.

R.B. is a cofounder of Intellia Therapeutics, Locus Biosciences, TreeCo, CRISPR Biotechnologies and Ancilia Biosciences, and a shareholder of Caribou Biosciences, Inari Ag, Tune Therapeutics, and Eligo Biosciences. C.A.G. is a cofounder of Tune Therapeutics and Locus Biosciences, is an advisor to Sarepta Therapeutics and shareholder of TreeCo. G.L.B., D.R., A.R., M.A.N., N.I., R.B., and C.A.G. are inventors on a related patent application. R.B. and L.A.M. co-authored a perspective in 2022 (https://doi.org/10.1016/j.chom.2021.12.013↗).

RNA seq data have been deposited in NCBI (PRJNA1137952↗) (63). All study data are included in the article and/or supporting information↗. Plasmids have been deposited with Addgene (Plasmids 227005↗–227012↗) (64–71).