To establish a highly sensitive approach for rare off-target cleavage detection in cells, a previous GFP activation assay was used (Hu et al., 2020). A target sequence (protospacer) with a PAM is inserted between ATG start codon and GFP coding sequence, disrupting GFP expression by frameshift mutation. When Cas9 nucleases cleave the protospacer and generate small insertions/deletions (indels), a portion of cells will restore the GFP reading frame, leading to GFP expression (Figure 1A). The reporter system was delivered into HEK293T cells by lentiviral infection to generate a stable cell line. Background GFP-positive cells caused by plasmid mutations were removed by cell sorting (Figure 1B).

CIRCLE-seq is the most sensitive approach for genome-wide off-target detection to date (Tsai et al., 2017). Tsai et al. identified up to hundreds of off-target sites for a given target by CIRCLE-seq, but they failed to validate a list of off-target sites in cells by using targeted amplicon sequencing (Tsai et al., 2017). It possible that off-target cleavage occurred at these sites but the detection assay was not sensitive enough. To test the power of the GFP activation assay, we first chose two off-target sites (EMX1-OT5 and VEGFA-OT7; Table 1) that can be detected by both CIRCLE-seq and GUIDE-seq (Tsai et al., 2017). GFP-positive cells were observed 2 days after the transfection of SpCas9 with on-target gRNA but not SpCas9 alone (Figures 1C,D; Supplementary Figures S1a,S1b). The GFP-positive cells were sorted out, and the protospacer sequences were PCR-amplified for deep sequencing. Deep sequencing analysis revealed that indels occurred (Figure 1E). These data demonstrated that the GFP activation assay enabled off-target cleavage detection in cells.

We chose 10 additional off-target sites (five for EMX1 and five for VEGFA site 1; Table 1) that were detected by CIRCLE-seq but failed to be validated by targeted amplicon sequencing (Tsai et al., 2017). GFP-positive cells were observed for all of the tested sites 2 days after the transfection of SpCas9 with on-target gRNA but not SpCas9 alone (Figures 1C,D; Supplementary Figures S1a,S1b), indicating that off-target mutations could occur with these sites in cells. Indels were confirmed by deep sequencing (Supplementary Figures S2, S3). These data demonstrated that the GFP activation assay was a highly sensitive platform for DNA cleavage detection.

In addition to NGG PAM, SpCas9 also recognizes noncanonical PAMs, including NAG, NCG, NGA, and NNGG (Jiang et al., 2013; Doench et al., 2016). We investigate whether the GFP activation assay is sensitive enough to identify additional PAMs. For this purpose, a protospacer containing 7nt downstream randomized DNA sequences was inserted between ATG start codon and GFP coding sequence, disrupting GFP expression (Figure 2A). The GFP reporter construct was stably inserted into the genome. Quantitative PCR (qPCR) revealed that there were average 3.82 copies of construct per cells. GFP-positive cells could be observed 2 days after the transfection of SpCas9 with the corresponding gRNA but not SpCas9 alone (Figure 2B). The GFP-positive cells were sorted out, and the protospacer with randomized DNA sequences were PCR-amplified for deep sequencing. Deep sequencing analysis revealed that indels associated with different PAMs could be detected (Figure 2C).

Next, we systematically analyzed the PAM sequences recognized by SpCas9 from deep sequencing data. Only in-frame mutations were considered as novel mutations induced by SpCas9, thus minimizing the background mutations derived from library construction or deep sequencing. Since indels could disrupt the randomized DNA sequences, GCG triple-nucleotide was used to fix the 7nt randomized DNA sequences (Figure 2C). Both WebLogo and PAM wheel captured the canonical NGG as the most enriched PAM, but the sequences other than NGG were also observed (Figures 3A,B). The nucleotide preference at position 1–4 was not random. We analyzed the PAM frequencies for all possible NNNN PAM sequences (Figure 3C). The top 74 PAMs (frequency over 0.1%) included all NNGG, NAGN, NGAN, and GGYN PAMs (Supplementary Figure S4). The most efficient PAM was NGGN, as expected, followed by NNGG, NAGN, NGAN, and GGYN (Figure 3C). Interestingly, some noncanonical PAMs such as GTGG, GCGG, and GAGT displayed comparable efficiency to NGGN PAMs.

To confirm novel PAMs identified here, we isolated four GFP reporter constructs with different PAM sequences (CGCT, TGCT, GACT, and TGTA) from the PAM library and established stable cell lines for each construct (Figure 4A). Transfection of SpCas9 with gRNA induced GFP expression for all of them, indicating that cleavage occurred with these PAMs (Figure 4B). We also isolated a construct with GTTA PAM as a negative control. Consistent with the PAM screening results, GFP-positive cells could not be observed with GTTA PAM (Figure 4B). We further tested 21 endogenous targets with noncanonical PAMs. Two NNGG PAMs displayed very high activity, with indel rates of 64.3 and 57.6% for GCGG and CTGG, respectively (Figure 4C). Three GGYN PAMs (GGCG, GGCT, and GGTC) as well as GTGA and GTGT PAMs also displayed significant activity. Other PAMs only displayed minimal activity. These novel PAMs are useful for in silico prediction of potential off-target sites.

HEK293T cells were maintained in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 10% FBS (Gibco) and 1% antibiotics at 37°C with 5% CO₂. For the PAM library screen, HEK293T cells were plated into 10 cm dishes, and transfected at ∼60% confluency with Cas9-gRNA-expressing plasmid (15 μg) using Lipofectamine 2000 (Life Technologies). For PAM validation at endogenous sites, HEK293T cells were seeded on 48-well plates and transfected with Cas9-gRNA-expressing plasmid (500 ng) using Lipofectamine 2000 (Life Technologies).

The DNA oligonucleotides containing a target sequence followed by random sequences and flanking homologous sequences (for Gibson Assembly) were synthesized from GENEWIZ (Suzhou, China). Full-length oligonucleotides were PCR-amplified using Q5 High-Fidelity 2X Master Mix (NEB), size-selected using a 3% agarose gel EX (Life Technologies, Qiagen), and purified using MinElute Gel Extraction Kit (Qiagen). PCR products were cloned into a lentiviral vector by Gibson Assembly (NEB) and purified with Agencourt AMPure XP SPRI beads (Beckman Coulter). The Gibson Assembly products were electroporated into MegaX DH10B^™ T1^R Electrocomp^™ Cells (Invitrogen) using a GenePulser (BioRad). The bacteria were added into recovery media and grew at 32°C, 225 rpm for 14 h. The plasmid DNA was extracted from bacteria using Endotoxin-Free Plasmid Maxiprep (Qiagen). The plasmid sequence is shown in Supplementary Figure S5. All primers and gRNA sequences used in this study were listed in Supplementary Table S1.

For PAM library packaging, HEK293T cells were seeded in three 10 cm dishes and transfected at ∼40% confluency. For each dish, 12 μg of PAM library plasmid, 9 μg of psPAX2, and 3 μg of pMD2.G were transfected with 60 μL of Lipofectamine 2000 (Life Technologies). Viruses were harvested twice at 48 and 72 h post-transfection. The viruses were concentrated using PEG8000 (no. LV810A-1, SBI, Palo Alto, CA), dissolved in PBS and stored at −80°C. For single PAM reporter construct packaging, HEK293T cells were seeded into 6-well plates, 1.2 μg of PAM reporter plasmid, 0.9 μg of psPAX2, and 0.3 μg of pMD2.G were transfected with 5 μL of Lipofectamine 2000. Viruses were harvested twice at 48 and 72 h post-transfection.

HEK293T cells were plated into a 15 cm dish at ∼30% confluence. After 24 h, cells were infected with PAM library lentiviruses with at least 1000-fold coverage of each PAM. 24 h after infection, the cells were selected with 2 µg/ml of puromycin for 5 days. To remove plasmid mutations that induce GFP expression, the GFP negative cells were sorted out with a MoFlo XDP machine (Beckman Coulter) and seeded into 10 cm dishes. The residual GFP-positive cells were removed pipette tips under microscope. For the PAM screen, PAM library cells were transfected with Cas9-gRNA expressing plasmid for 3 days. The GFP-positive cells were sorted out by MoFlo XDP machine and the genomic DNA was isolated using TIANamp Genomic DNA Kit (TIANGEN) following the manufacturer’s instructions. DNA target sites were PCR-amplified by nested PCR with Q5 High-Fidelity 2X Master Mix (NEB). First, the target region was PCR-amplified using primers Deep-F1/R1 with 25 cycles; second, 3 μL of PCR products from the first step were used as a template and amplified by primers P5-adapter-F and P7-adapter-R for 15 cycles. The PCR products were purified using the Gel Extraction Kit (Qiagen) and sequenced on Illumina HiSeq X by 150-bp paired-end sequencing.

Twenty base-pair sequences (AAG​CCT​TGT​TTG​CCA​CCA​TG/GTG​AGC​AAG​GGC​GAG​GAG​CT) flanking the target sequence (GAACGGCTCGGAGATCATCATTGCGNNNNNNN) were used to fix the target sequence. Only target sequences with in-frame mutations were used for PAM analysis. GCG and GTG​AGC​AAG​GGC​GAG​GAG​CT sequences were used to fix a 7-bp random sequence. Only intact 7-bp random sequences were used for PAM analysis. The 7-bp random sequences were extracted and visualized by WebLog3 (Crooks et al., 2004) and wheel chart (Leenay et al., 2016) to demonstrate PAMs.

Plasmids containing specific PAM sequences were isolated from the PAM library. Plasmids of EMX1-OT and VEGFA-OT construction: vector backbone of the plasmid containing specific PAM sequences was PCR-amplified using primers EMX1/VEGFA-OT-F/cozak-R, followed by phosphorylation with T4 Polynucleotide Kinase (NEB) and religation with T4 DNA ligase (NEB). Each plasmid was packed into lentiviruses to generate a stable cell line. To remove plasmid mutations that induce GFP expression, the GFP-negative cells were sorted out by the MoFlo XDP machine (Beckman Coulter). The cells were seeded into 24-well and transfected with 800 ng of SpCas9-gRNA-expressing plasmid (PX459, addgene #118632) by Lipofectamine 2000 (Life Technologies). Five days after transfection, the GFP-positive cells were measured on the Calibur instrument (BD). Data were analyzed using FlowJo. Besides, GFP-positive cells of EMX1-OT, and VEGFA-OT were sorted out by the MoFlo XDP machine (Beckman Coulter). The genomic DNA was isolated and target sites were amplified by nested PCR and extracted by Gel Extraction Kit (QIAGEN). The amplicons are prepared for deep sequencing.

HEK293T cells were seeded on 24-well plates and transfected with epiCRISPR(4) constructs expressing SpCas9 and gRNA followed by puromycin selection (2 µg/ml) for 6 days. The genomic DNA was isolated, and the target sites were PCR-amplified by nested PCR for deep sequencing.

The genomic DNA was isolated from GFP activation cells using TIANamp Genomic DNA Kit (TIANGEN) following the manufacturer’s instructions. To generate standard curve, GFP reporter plasmid was serially diluted to final concentrations of 10⁹,10⁸,10⁷,10⁶,10⁵,10⁴,10³,10² copies/μL. Quantitative PCR were performed with CFX Connect Real-Time System (Bio-Rad) and 2X SYBR Green qPCR Master Mix (APExBIO). Use the following parameters: 1 cycle at 95°C for 2 min, followed by 40 cycles of 95°C for 15 s, 55°C for 30 s, and 72°C for 30 s. Each sample was quantified in triplicate. The same parameters were used to quantify GFP copy numbers in the genomic DNA. Absolute transgene copy numbers were calculated with C_q values and dilution factor based on standard curve.

All the data are shown as the mean ± S.D. Statistical analyses were conducted using Microsoft Excel.