In vivo genome editing using Staphylococcus aureus Cas9

In vitro cleavage by small Cas9s

Type II CRISPR-Cas systems require only two main components for eukaryotic genome editing: a Cas9 enzyme, and a chimeric single guide RNA (sgRNA)⁶ derived from the CRISPR RNA (crRNA) and the noncoding trans-activating crRNA (tracrRNA)^4,20. Analysis of over 600 Cas9 orthologs shows that these enzymes are clustered into two length groups with characteristic protein sizes of approximately 1350aa and 1000aa residues, respectively^19,21 (Extended Data Fig. 1a), with shorter Cas9s having significantly truncated REC domains (Fig. 1a). From these shorter Cas9s, which belong to Type IIA and IIC subtypes, we selected six candidates for profiling (Fig. 1a and Extended Data Fig. 1b). To determine the cognate crRNA and tracrRNA for each Cas9, we computationally identified regularly interspaced repeat sequences (direct repeats) within a 2-kb window flanking the CRISPR locus. We then predicted the tracrRNA by detecting sequences with strong complementarity to the direct repeat sequence (an anti-repeat region), at least two predicted stem-loop structures, and a Rho-independent transcriptional termination signal up to 150-nt downstream of the anti-repeat region. Although a truncated tracrRNA can support robust DNA cleavage in vitro⁶, previous reports show that the secondary structures of the tracrRNA are important for Cas9 activity in mammalian cells^8,9,18,22. Therefore, we designed sgRNA scaffolds for each ortholog by fusing the 3′ end of a truncated direct repeat with the 5′ end of the corresponding tracrRNA, including the full-length tail, via a 4-nt linker⁶ (Extended Data Fig. 1b and Supplementary Table 1). To identify the PAM sequence for each Cas9, we first constructed a library of plasmid DNA containing a constant 20-bp _target followed by a degenerate 7-bp sequence (5′-NNNNNNN). We then incubated cell lysate from human embryonic kidney 293FT (293FT) cells expressing the Cas9 ortholog with its in vitro transcribed sgRNA and the plasmid library. By generating a consensus from the 7-bp sequence found on successfully cleaved DNA plasmids (Fig. 1b), we determined putative PAMs for each Cas9 (Fig. 1c). We observed that, similar to SpCas9, most Cas9 orthologs cleaved _targets 3-bp upstream of the PAM (Extended Data Fig. 2). To validate each putative PAM from the library, we then incubated a DNA template bearing the consensus PAM with cell lysate and the corresponding sgRNA. We found that the Cas9 orthologs, in combination with the sgRNA designs, successfully cleaved the appropriate _targets (Fig. 1d and Supplementary Table 2).

To test whether each Cas9 ortholog can facilitate genome editing in mammalian cells, we co-transfected 293FT cells with individual Cas9s and their respective sgRNAs _targeting human endogenous loci containing the appropriate PAMs. Of the six Cas9 orthologs tested, only the one from Staphylococcus aureus (SaCas9) produced indels with efficiencies comparable to those of SpCas9 (Extended Data Fig. 3a, b and Supplementary Table 3), suggesting that DNA-cleavage activity in cell-free assays does not necessarily reflect the activity in mammalian cells. These observations prompted us to focus on harnessing SaCas9 and its sgRNA for in vivo applications.

SaCas9 sgRNA design and PAM discovery

Although mature crRNAs in S. pyogenes are processed to contain 20-nt spacers (guides) and 19- to 22-nt direct repeats⁴, RNA sequencing of crRNAs from other organisms reveals that the spacer and direct repeat sequence lengths can vary^4,20,23. We therefore tested sgRNAs for SaCas9 with variable guide lengths and repeat:anti-repeat duplexes. We found that SaCas9 achieves the highest editing efficiency in mammalian cells with guides between 21- to 23-nt long and can accommodate a range of lengths for the direct repeat:anti-repeat region (Fig. 2a, b, Extended Data Fig. 4). This notably contrasts with SpCas9, where the natural 20-nt guide length can be truncated to 17-nt without significantly compromising nuclease activity, while increasing specificity²⁴. Additionally, replacing the first base of the guide with guanine further improved SaCas9 activity (Extended Data Fig. 3c).

To fully characterize the SaCas9 PAM and the seed region within its guide sequence²⁵, we performed ChIP using catalytically mutant forms of SaCas9 (dSaCas9, D10A and N580A mutations, based on homology to SpCas9) or SpCas9 (dSpCas9, D10A and H840A mutations) and their corresponding sgRNAs. We _targeted two loci in the human EMX1 gene with composite NGGRRT PAMs, which allow _targeting by both Cas9 variants. A search for motifs containing both the guide region and PAM within 50-nt of the ChIP peak summits revealed seed sequences of 7–8 nt for dSaCas9 (Fig. 2c). In addition, NNGRRT and NGG PAMs were found adjacent to the seed sequences for dSaCas9 and dSpCas9, respectively (Extended Data Fig. 5). Although the 6^th position of the PAM is predominantly thymine, we did observe low levels of degeneracy in both the biochemical and ChIP-based PAM discovery assays (Fig. 1c and Extended Data Fig. 5a). We therefore tested the base preference for this position and determined that although SaCas9 cleaves genomic _targets most efficiently with NNGRRT, all NNGRR PAMs can be cleaved and should be considered as potential _targets, especially in the context of off-_target evaluations (Fig. 2d, Extended Data Fig. 6, and Supplementary Table 4).

Unbiased profiling of Cas9 specificity

As advances in Cas9 technology promise to enable a broad range of in vivo and therapeutic applications, accurate, genome-wide identification of off-_target nuclease activity has become increasingly important. Although a number of studies have employed sequence similarity-based off-_target search^22,26–30 or dCas9-ChIP^31,32 to predict off-_target sites for Cas9, such approaches cannot assess the nuclease activity of Cas9 in a comprehensive and unbiased manner. To directly measure the genome-wide cleavage activity of SaCas9 and SpCas9, we applied BLESS (direct in situ breaks labeling, enrichment on streptavidin and next-generation sequencing)³³ to capture Cas9-induced DNA double-stranded breaks (DSBs) in cells. We transfected 293FT cells with SaCas9 or SpCas9 and the same EMX1 _targeting guides used in the previous ChIP experiment, or pUC19 as negative controls. After cells are fixed, free genomic DNA ends from DSBs are captured using biotinylated adaptors and analyzed by deep sequencing (Fig. 3a). To identify candidate Cas9-induced DSB sites genome-wide, we established a three-step analysis pipeline following alignment of the sequenced BLESS reads to the genome (Extended Data Fig. 7a, Supplementary Discussion). First, we applied nearest-neighbor clustering on the aligned reads to identify groups of DSBs (DSB clusters) across the genome. Second, we sought to separate potential Cas9-induced DSB clusters from background DSB clusters resulting from low frequency biological processes and technical artifacts, as well as high frequency telomeric and centromeric DSB hotspots³³. From the on-_target and a small subset of verified off-_target sites (predicted by sequence similarity using a previously established method²² and sequenced to detect indels), we found that reads in Cas9-induced DSB clusters mapped to characteristic, well-defined genomic positions compared to the more diffuse alignment pattern at background DSB clusters. To distinguish between the two types of DSB clusters, we calculated in each cluster the distance between all possible pairs of forward and reverse-oriented reads (corresponding to 3′ and 5′ ends of DSBs), and filtered out the background DSB clusters based on the distinctive pairwise-distance distribution of these clusters (Extended Data Fig. 7b, c). Third, the DSB score for a given locus was calculated by comparing the count of DSBs in the experimental and negative control samples using a maximum-likelihood estimate (MLE)²² (Supplementary Discussion). This analysis identified the on-_target loci for both SaCas9 and SpCas9 guides as the top scoring sites, and revealed additional sites with high DSB scores (Fig. 3b–d).

Next, we sought to assess whether DSB scores correlated with indel formation. We used _targeted deep sequencing to detect indel formation on the ~30 top-ranking off-_target sites identified by BLESS for each Cas9 and sgRNA combination. We found that only those sites that contained PAM and homology to the guide sequence exhibited indels (Extended Data Fig. 8). We observed a strong linear correlation between DSB scores and indel levels for each Cas9 and sgRNA pairing (r² = 0.948 and 0.989 for the two EMX1 _targets with SaCas9 and r² = 0.941 and 0.753 for those with SpCas9) (Fig. 3c, Extended Fig. 9b–d). Furthermore, BLESS identified additional off-_target sites not previously predicted by sequence similarity to _target or ChIP (Extended Data Fig. 7 and 9, Supplementary Tables 5 and 6). These new off-_target sites include not only those containing Watson-Crick base-pairing mismatches to the guide, but also the recently reported insertion and deletion mismatches in the guide:_target heteroduplex (Fig. 3d)^29,30. Together, these results highlight the need for more precise understanding of rules governing Cas9 nuclease activity, a requisite step towards improving the predictive power of computational guide design programs.

In vivo genome editing using SaCas9

Following in vitro characterization, we incorporated SaCas9 and its sgRNA into an AAV vector to test its efficacy and specificity in vivo. The small size of SaCas9 enables packaging of both a U6-driven sgRNA and a CMV- or TBG-driven SaCas9 expression cassette into a single AAV vector within the 4.5kb packaging limit. Using hepatocyte-tropic AAV serotype 8, we _targeted the mouse apolipoprotein (Apob) gene (Extended Data Fig. 10a). One week after intravenous administration of virus into C57BL/6 mice, we observed ~5% indel formation in liver tissue; after four weeks, the liver tissue showed characteristic hepatic lipid accumulation from Apob knockdown following histology analysis using oil red staining^34–37 (Extended Data Fig. 10b, c).

We next _targeted proprotein convertase subtilisin/kexin type 9 (Pcsk9), a therapeutically relevant gene involved in cholesterol homeostasis³⁸. Inhibitors of the human convertase PCSK9 have emerged as a promising new class of cardioprotective drugs after human genetic studies revealed that loss of PCSK9 is associated with a reduced risk of cardiovascular disease and lower levels of LDL cholesterol^39–41. We designed two Pcsk9-_targeting sgRNAs and validated their activity in vitro. Each sgRNA was packaged into AAV-SaCas9 and injected into mice (2E11 total genome copies) (Fig. 4a). One week after administration, we observed greater than 40% indel formation at either locus in whole liver tissue, with similar levels two and four weeks post-injection (Fig. 4b). To determine the effect of Pcsk9-_targeting AAV-SaCas9 dosage on serum Pcsk9 and total cholesterol levels, we administered a range of AAV titers from 0.5E11 to 4E11 total genome copies. With all titers, we observed a ~95% decrease in serum Pcsk9 and a ~40% decrease in total cholesterol one week after administration, both of which were sustained throughout the course of four weeks (Fig. 4c, d).

Given the importance of _targeting specificity in a therapeutic context, we next considered SaCas9 off-_target modifications in vivo. To identify candidate off-_target cleavage sites for the two Pcsk9-_targeting guides, we transiently transfected an AAV-CMV::SaCas9 vector into mouse Neuroblastoma-2a (N2a) cells and applied BLESS to detect Cas9-induced DSBs in the genome. For both guides, we found very low levels of DSB signal across the genome except at the on-_target locus (Fig. 4e). _targeted deep sequencing of the candidate off-_target sites identified by BLESS in N2a cells did not reveal appreciable levels of indels in either N2a cells or liver tissue (4 weeks post injection of 2E11 total genome copies) (Fig. 4e, f and Supplementary Table 8). We additionally sequenced off-_target sites predicted by _target sequence similarity, and likewise did not detect indel formations (Supplementary Table 9).

Finally, we examined the titer-matched Pcsk9-_targeting and TBG-GFP cohorts as well as naïve animals for signs of toxicity or acute immune response. At 1 week post-injection, necropsy and gross examination of liver tissue of the cohorts revealed no abnormalities; further histological examination of the liver by hematoxylin and eosin (H&E) staining showed no signs of inflammation, such as aggregates of lymphocytes or macrophages (Fig. 5a). Throughout the time course of the experiment, there were no elevated levels of serum ALT, albumin, and total bilirubin in any of the cohorts. We observed a slight trend in AST increase across all cohorts at four weeks, including the un-injected animals. The elevated levels did not exceed the upper limit of normal and is not indicative of hepatocellular injury (Fig. 5b). However, a larger cohort study should be conducted to further evaluate the effects of in vivo toxicity.

In vitro transcription and cleavage assay

Cas9 orthologs were human codon-optimized and synthesized by GenScript, and transfected into 293FT cells as described below. Whole cell lysates from 293FT cells were prepared with lysis buffer (20 mM HEPES, 100 mM KCl, 5 mM MgCl2, 1 mM DTT, 5% glycerol, 0.1% Triton X-100) supplemented with Protease Inhibitor Cocktail (Roche). T7-driven sgRNA was transcribed in vitro using custom oligos (Supplementary Information) and HiScribe T7 In vitro Transcription Kit (NEB), following the manufacturer’s recommended protocol. The in vitro cleavage assay was carried out as follows: for a 20 μl cleavage reaction, 10 μl of cell lysate was incubated with 2 μl cleavage buffer (100 mM HEPES, 500 mM KCl, 25 mM MgCl₂, 5 mM DTT, 25% glycerol), 1 μg in vitro transcribed RNA and 200 ng EcoRI-linearized pUC19 plasmid DNA or 200 ng purified PCR amplicons from mammalian genomic DNA containing _target sequence. After 30 min incubation, cleavage reactions were purified using QiaQuick Spin Columns and treated with RNase A at final concentration of 80 ng/μl for 30 min and analyzed on a 1% Agarose E-Gel (Life Technologies).

In vitro PAM screen

Rho-independent transcriptional termination was predicted using the ARNold terminator search tool^47,48. For the PAM library, a degenerate 7-bp sequence was cloned into a pUC19 vector. For each ortholog, the in vitro cleavage assay was carried out as above with 1 μg T7-transcribed sgRNA and 400 ng pUC19 with degenerate PAM. Cleaved plasmids were linearized by NheI, gel extracted, and ligated with Illumina sequencing adaptors. Barcoded and purified DNA libraries were quantified by Quant-iT PicoGreen dsDNA Assay Kit or Qubit 2.0 Fluorometer (Life Technologies) and pooled in an equimolar ratio for sequencing using the Illumina MiSeq Personal Sequencer (Life Technologies). MiSeq reads were filtered by requiring an average Phred quality (Q score) of at least 23, as well as perfect sequence matches to barcodes. For reads corresponding to each ortholog, the degenerate region was extracted. All extracted regions were then grouped and analyzed with Weblogo⁴⁵.

Cell culture and transfection

Human embryonic kidney 293FT (Life Technologies), Neuro-2a (N2a), and Hepa1-6 (ATCC) cell lines were maintained in Dulbecco’s modified Eagle’s Medium (DMEM) supplemented with 10% FBS (HyClone), 2 mM GlutaMAX (Life Technologies), 100 U/ml penicillin, and 100 μg/ml streptomycin at 37 °C with 5% CO₂ incubation.

Cells were seeded into 24-well plates (Corning) one day prior to transfection at a density of 240,000 cells per well, and transfected at 70–80% confluency using Lipofectamine 2000 (Life Technologies) following the manufacturer’s recommended protocol. For each well of a 24-well plate, a total of 500 ng DNA was used. For ChIP and BLESS, a total of 4.5 million cells are seeded the day before transfection into a 100mm plate, and a total of 20 ug DNA was used.

DNA isolation from cells and tissue

Genomic DNA was extracted using the QuickExtract DNA Extraction Solution (Epicentre). Briefly, pelleted cells were resuspended in QuickExtract solution and incubated at 65 °C for 15 min, 68 °C for 15 min, and 98 °C for 10 min⁸. Genomic liver DNA was extracted from bulk tissue fragments using a microtube bead mill homogenizer (Beadbug, Denville Scientific) by homogenizing approximately 30–50 mg of tissue in 600 μL of DPBS (Gibco). The homogenate was then centrifuged at 2000 to 3000×g for 5 minutes at 4°C and the pellet was resuspended in 300–600 μL QuickExtract DNA Extraction Solution (Epicentre) and incubated as above.

Indel analysis and guide:_target basepairing mismatch search

Indel analyses by SURVEYOR assay and _targeted deep sequencing were carried out and analyzed as previously described^8,22. The methods for identification of potential off-_target sites for SpCas9 based on Watson-Crick base-pairing mismatch between guide RNA and _target DNA has been previously described²², and adapted for SaCas9 by considering NNGRR for possible off-_target PAMs.

Chromatin immunoprecipitation and analysis

Cells are passaged at 24 hours post-transfection into a 150mm dish, and fixed for ChIP processing at 48 hours post-transfection. For each condition, 10 million cells are used for ChIP input, following experimental protocols and analyses as previously described³¹ with the following modifications: instead of pairwise peak-calling, ChIP peaks were only required to be enriched over both ‘empty’ controls (dSpCas9 only, dSaCas9 only) as well as the other Cas9/other sgRNA sample (e.g., SpCas9/EMX-sg2 peaks must be enriched over SaCas9/EMX-sg1 peaks in addition to the empty controls). This was done to avoid filtering out of real peaks present in two related samples as much as possible.

To identify off-_targets ranked by motif or sequence similarity to guide, motif scores for ChIP peaks were calculated as follows: For a given ChIP peak, the 100-nt interval around the peak summit, the _target sequence, and a given sgRNA guide region L, the query, an alignment score is calculated for every subsequence of length L in the _target. The subsequence with the highest score is reported as the best match to the query. For each subsequence alignment, the score calculation begins at the 5′ end of the query. For each position in the alignment, 1 is added or subtracted for match or mismatch between the query and _target, respectively. If the score becomes negative, it is set to 0 and the calculation continued for the remainder of the alignment. The score at the 3′ end of the query is reported as the final score for the alignment. MACS scores = −10log(p-value relative to the empty control) are determined as previously described⁴⁹. For unbiased determination of PAM from ChIP peaks, the peaks were analyzed for the best match by motif score to the guide region only within 50-nt of the peak summit; the alignment was extended for 10-nt at the 3′ end and visualized using Weblogo⁴⁵.

To calculate the motif score threshold at which FDR < 0.1 for each sample, 100-nt sequences centered around peak summits were shuffled while preserving dinucleotide frequency. The best match by motif score to the guide+PAM (NGG for SpCas9, NNGRRT for SaCas9) in these shuffled sequences was then found. The score threshold for FDR < 0.1 was defined as the score such that less than 10% of shuffled peaks had a motif score above that score threshold.

BLESS for DSB detection

Cells are harvested at 24 hours post-transfection, then processed as previously described³³ with the following alterations: a total of 10 million cells are fixed for nuclei isolation and permeabilization, and treated with Proteinase K for 4 min at 37°C before inactivation with PMSF. All deproteinized nuclei are used for DSB labeling with 100 mM of annealed proximal linkers overnight. After Proteinase K digestion of labeled nuclei, chromatin are mechanically sheared with a 26G needle before sonication (BioRuptor, 20 min on High, 50% duty cycle). 20 ug of sheared chromatin are captured on streptavidin beads, washed, and ligated to 200 mM of distal linker. Linker hairpins are then cleaved off with I-SceI digestion for 1 hour at 37°C, and products PCR-enriched for 18 cycles before proceeding to library preparation with TruSeq Nano LT Kit (Illumina). For the negative control, cells mock transfected with Lipofectamine 2000 and pUC19 DNA were parallel processed through the assay.

BLESS Analysis

Fastq files were demultiplexed, and 30-bp genomic sequences were separated from the BLESS ligation handles for alignment. Bowtie was used to map the genomic sequences to hg19 or mm9, allowing for a maximum of 2 mismatches. Following alignment, reads from all bio-replicates for an individual sample were first pooled, and then nearest neighbor clustering was performed with a 30-bp moving window to identify regions of enrichment across the genome. Within each cluster, the pairwise distance was calculated between all forward and reverse read strand mappings (Extended Data Figure 7b, c). Pairwise distance distributions were used to filter out wide and poorly-defined DSB clusters from the well-defined DSB clusters characteristically found at Cas9-induced cleavage sites (see Supplementary Information). Finally, we adjusted the count of predicted Cas9-induced DSBs at a given locus by using a binomial model to calculate the maximum-likelihood estimate (MLE) of peak enrichment in the Cas9-sgRNA treated sgRNAs given BLESS measurements from an untreated negative control. After the MLE calculation, a list of loci ranked by their DSB scores could be obtained and plotted (Figure 3b, Extended Data Figure 8). Additional descriptions can be found in Supplementary Information.

The top-ranking ~30 sites from the list of Cas9 induced DSB clusters were sequenced for indel formation (Extended Data Figure 8; validated _targets in Figure 3d). Within these loci, PAMs and regions of _target homology were identified by first searching all PAM sites within a ±50 bp window around the DSB cluster, then selecting the adjacent sequence with fewest mismatches to the _target sequence.

Code Availability

BLESS analysis code is available upon request.

Virus Production and Titration

For in-house viral production, 293FT cells (Life Technologies) were maintained as described above in 150mm plates. For each transfection, 8 ug of pAAV8 serotype packaging plasmid, 10 ug of pDF6 helper plasmid, and 6 ug of AAV2 plasmid carrying the construct of interest were added to 1mL of serum-free DMEM. 125 μL of PEI “Max” solution (1mg/mL, pH = 7.1) was then added to the mixture and incubated at room temperature for 5 to 10 seconds. After incubation, the mixture was added to 20 mL of warm maintenance media and applied to each dish to replace the old growth media. Cells were harvested between 48h and 72h post transfection by scraping and pelleting by centrifugation. The AAV2/8 (AAV2 ITR vectors pseudo-typed with AAV8 capsid) viral particles were then purified from the pellet according to a previously published protocol⁵⁰.

High titer and purity viruses were also produced by vector core facilities at Children’s Hospital Boston and Massachusetts Eye and Ear Infirmary (MEEI). These AAV vectors were then titered by real-time qPCR using a customized TaqMan probe against the transgene, and all viral preparations were titer-matched across different batches and production facilities prior to experiments. The purity of AAV vector was further verified by SDS-PAGE.

Animal Injection and Processing

All mice cohorts were maintained at animal facility with standard diet and housing following IRB-approved protocols. AAV vector was delivered to 5–6 week old male C57/BL6 mice intravenously via lateral tail vein injection. All dosages of AAV were adjusted to 100 μL or 200 μL with sterile phosphate buffered saline (PBS), pH 7.4 (Gibco) before the injection. Animals were not immunosuppressed or otherwise handled differently prior to injection or during the course of the experiment except the pre-bleed fasting as noted below. The animals were randomized to the different experimental conditions, with the investigator not blinded to the assignments.

To track the serum levels of Pcsk9 and total cholesterol, animals were fasted overnight for 12 hours prior to blood collection by saphenous vein bleeds (no more than 100 μL or 10% of total blood volume per week). Multiple bleeds were made prior to tail vein delivery of AAV vector or control to collect pre-injection samples and to habituate the animals to handling during the procedure. After the blood was allowed to clot at room temperature, the serum was separated by centrifugation and stored at −20°C for subsequent analysis. For terminal procedures to collect liver tissue and larger serum volumes for chemistry panels, mice were euthanized by carbon dioxide inhalation. Subsequently, blood was collected via cardiac puncture. Transcardial perfusion with 30 mL PBS removed the remaining blood, after which liver samples were collected. The median lobe of liver was removed and fixed in 10% neutral buffered formalin for histological analysis, while the remaining lobes were sliced in small blocks of size less than 1×1×3mm³ and frozen for subsequent DNA or protein extraction.

Histology and serum analysis

Following tissue harvesting as described above, flash-frozen mouse liver samples were embedded in O.C.T. compound (Tissue Tek, Cat # 4583), snap-frozen, and stored at −80°C prior to processing. Frozen tissues were cryosectioned at 4-micron in thickness and stained with Oil Red O following manufacturer’s recommended protocol. Liver histology was assessed by H&E staining sections of 10% neutral buffer formalin fixed liver sections.

Serum levels of Pcsk9 were determined by ELISA using the Mouse Proprotein Convertase 9/PCSK9 Quantikine ELISA Kit (MPC-900, R&D Systems), following the manufacturer’s instructions. Total cholesterol levels were measured using the Infinity Cholesterol Reagent (Thermo Fisher) per the manufacturer’s instructions. Serum ALT, AST, albumin and total bilirubin were measured by an Olympus AU5400 (IDEXX Memphis, TN).