Dual-gRNA design for removal of human C9ORF72 repeat site
In order to remove the expanded GGGGCC repeats in the intron of C9ORF72, we analyzed the genomic sequences of human C9ORF72 (Fig. 1A). Three times of GGGGCC (repeat site) are found 162-bp downstream of exon 1a and 35-bp upstream of exon 1b, consistent with the previous repeat site description (Fig. 1A) 1,2. By using two online gRNA designers 43,44, we identified two gRNAs, gRNA1 and gRNA3, as upstream gRNAs with low predicted off-target effect in the 162-bp region between the exon 1a and the repeat site (Fig. 1A and Supplementary Table 1). For downstream gRNA, the ideal location is in the 35-bp region between repeat site and exon 1b, which pairs with the upstream gRNA to remove the repeat expansion without disturbing exon 1b. However, the 35-bp region is a low complexity region and all the gRNAs we identified in this 35-bp region have high predicted off-target effect. Among them, gRNAb is scored the best (Fig. 1B and 1C, Supplementary Table 1). In order to avoid the potential off-target effect of the downstream gRNAs, we sought DNA region downstream of exon 1b and identified gRNA2, which is close to exon 1b and has low predicted off-target effect (Fig. 1A and 1C, Supplementary Table 1). Therefore, we ended up with two designs: 1) gRNA1 and gRNA2 pair (gRNA1-2), which removes the repeat site together with the exon 1b with low predicted off-target effect; 2) gRNA3 and gRNAb pair (gRNA3-b), which removes the repeat site and keeps the exon 1b intact with high predicted off-target effect (Fig. 1A).
The dual gRNAs remove C9ORF72 repeat site in human cell
To test our designs, we employed human HEK293 cell that contains 2 ~ 3 GGGGCC repeats (Supplementary Fig. 1A) and ensured that our gRNA target sequences are present in the genomic DNA (Supplementary Fig. 1B). Besides HEK293 cell, the target sequences of gRNA1, 2, and b are present in the other 11 human cell lines we examined (Supplementary Fig. 1C). We noticed a SNP (single nucleotide polymorphism) in gRNA3 target sequences in 3 out of 12 human cell lines we examined (Supplementary Fig. 1C and 1D), suggesting a SNP hot spot in the target sequences, which may need additional design for future precise editing.
To express two gRNAs in the same cell with high efficiency, we employed previously reported dual-gRNA expression and lentiviral delivery systems 45,46 (Supplementary Fig. 2). The HEK293 cells constitutively expressing Cas9 were infected with the lentiviral control, gRNA1-2, and gRNA3-b, respectively. To examine deletion of the repeat site, which and surrounding sequences of which are GC-rich, we used primers flanking the repeat site and performed genomic DNA PCR with high GC PCR buffer (Fig. 1A and 1D). With the buffer, we amplified the repeat site and its surrounding sequences in the HEK293 cells infected with the lentiviral control, gRNA1-2, and gRNA3-b. In addition, a deletion of repeat plus exon1b (E1b) (ΔRepeats + E1b) band appeared in HEK293 cells infected with gRNA1-2 with the similar PCR conditions. Due to a small deletion (ΔRepeats) by gRNA3-b, we were not able to distinguish ΔRepeats with the PCR amplicons with repeat site. However, when we lowered amount of high GC buffer in the PCR reactions, we were able to see the clear ΔRepeats or ΔRepeats + E1b bands in the HEK293 cells infected with gRNA1-2 or gRNA3-b but not in that of control virus, suggesting that a ‘cutting-deletion-fusion’ of repeat site takes place in the cells by gRNA1-2 and gRNA3-b. Indeed, the editing site fusions were evidenced at the Cas9 cutting sites by Sanger sequencing (Fig. 1E and 1F). Therefore, we conclude that our approach is able to remove C9ORF72 repeat site in human cell.
Limited off-target effect of gRNA1-2 measured by GUIDE-seq.
Given that off-target effect introduced by CRISPR/Cas9 is a major risk factor for its application in human disease therapy, limited off-target effect of our gRNAs is essential for their future applications. To identify potential off-target sites of our gRNAs, we employed both computational prediction and experimental identification approaches. In agreement with other online off-target predictors (Supplementary Table 1), Cas-OFFinder, a fast and versatile algorithm 47, predicted much more off-targets in gRNA3-b pair than in gRNA1-2 pair when we allowed mismatches (up to 7 mismatches) in the protospacer region (Supplementary Table 2). To experimentally identify off-target genomic loci of our gRNAs, we employed GUIDE-seq, a unbiased and sensitive genome-wide off-target site identification approach 40. We delivered our dual gRNAs and the oligodeoxynucleotide (dsODN) that incorporates into cutting site induced by CRISPR/Cas9 into HEK293 cells constitutively expressing Cas9 (Supplementary Fig. 3A). Indeed, thousands of unique reads from three individual experiments support on-target integration of dsODN of each one of our gRNAs by GUIDE-seq (Fig. 2A-2D). The on-target integration was further confirmed by genomic DNA PCR (Supplementary Fig. 4).
To extract reliable off-target events detected by GUIDE-seq, we included the events that were supported by unique reads or from at least two out of our three independent experiments (Fig. 2A-2D and Supplementary Fig. 3B). The off-target rate was estimated by percentage of reliable off-target reads divided by total reads. In contrast to high off-target rate of gRNA3 (3.9%) and gRNAb (38.9%) identified by GUIDE-seq (Fig. 2C-2E), only two off-target events (0.2%) were detected in the cells treated with gRNA1 and no reliable off-target event was detected for gRNA2 (Fig. 2E). For gRNA3 off-target sites, a majority of them (80%) are located in the introns and the rest of them (20%) are in the intergenic region (Fig. 2F). Because the target sequences of gRNAb are GC-rich, many of gRNAb off-target sites are located in the region between promoter and transcription start site (P-TSS, 37.5%) and 5’UTR (16.7%). Some gRNAb off-target sites hit protein coding exons (8.3%), suggesting that the off-target effect by gRNAb leads to functional impairments of these genes.
To confirm the off-target events identified by GUIDE-seq, we performed targeted deep sequencing of Off-T1 and Off-T2 of gRNA1 and 8 potential (P) off-target sites (up to 3-mismatches) of gRNA1 and gRNA2, respectively (Supplementary Fig. 5 and Supplementary Table 2). HEK 293 cells constitutively expressing Cas9 were transfected with gRNA1 or gRNA2 and applied for the targeted deep sequencing. Meanwhile, the cells transfected with GFP were employed as a negative control. Although at least 2.0×105 coverage of each sites were achieved in the cells transfected with gRNA1 or gRNA2, we did not observe differential indel rate between gRNA-transfected and GFP-transfected cells, suggesting that Off-T1 and Off-T2 of gRNA1 and potential off-target sites of gRNA1 (P1-P7) and gRNA2 (P1) are probably not caused by gRNA1- and gRNA2-mediated DNA editing. Taken together, our online prediction and experimental results support limited off-target effect of our gRNA1-2 design, but high off-target risks of gRNAb, which may limit its future clinical application.
Deletion of repeat site together with exon 1b has limited effect on C9ORF72 protein expression
Haploinsufficiency of C9ORF72 in C9-ALS/FTD patients implies that loss-of-function of C9ORF72 may contribute to disease pathogenesis 1,2. In fact, C9orf72 deficient mice showed abnormal immune response 8–10. Given that our gRNA1-2 design removes both the repeat site and exon 1b, we then asked whether the removal affects the C9ORF72 protein expression. Using a validated C9ORF72 antibody (Supplementary Fig. 6), we demonstrated that the relative C9ORF72 level was slightly decreased in HEK293 cells infected with gRNA1-2 (0.9 ± 0.06) compared to non-editing control (1.0 ± 0.06) with no statistical significance (Fig. 3A and 3B).
To examine whether gRNA1-2 affect C9ORF72 protein expression in cells derived from human brain, we employed U251 and SH-SY5Y, the human glioblastoma and neuroblastoma lines. We infected these two cell lines with Cas9-expressing lentivirus and established the U251 and SH-SY5Y cell lines constitutively expressing Cas9. After we infected these two cell lines with gRNA1-2 lentiviral particles, the removal of repeat site and exon 1b was evidenced by genomic DNA PCR with low amount of high GC buffer (Supplementary Fig. 7A), similar to what we observed in HEK293 cells (Fig. 1D). In addition, the editing site fusions were confirmed by Sanger sequencing (Supplementary Fig. 7B and 7C). As we observed in HEK293 cells (Fig. 3A), we did not observe significant change of C9ORF72 protein expression in these two cell lines expressing gRNA1-2 compared to that of control (Supplementary Fig. 7D and 7E).
To further examine the effect of gRNA1-2 on C9ORF72 protein expression, we established three independent HEK293 cell lines derived from a single edited cell by gRNA1-2 infection. Our genomic DNA PCR suggested that both alleles of repeat site and exon 1b were removed by gRNA1-2 in these three single cell clones (Fig. 3C and 3D), which was further confirmed by Sanger sequencing (Fig. 3E). In comparison with the non-editing control, no significant C9ORF72 protein expression level change was evidenced in these single cell clones (Fig. 3F and 3G). Taken together, our results indicated that removal of repeat site and exon 1b by our dual gRNAs has limited effect on C9ORF72 protein expression at least in cultured human cell lines, including HEK293, U251, and SH-SY5Y cells.
Our gRNA1-2 lead to high fusion efficiency at their editing sites in human cell
The removal of repeat expansion depends on the fusion of the gRNA1 and 2 editing sites. Therefore, high fusion efficiency will lead to high chance of repeat expansion removal and better therapeutic outcome. To examine the fusion efficiency, we infected HEK293 cells constitutively expressing Cas9 with gRNA1-2 lentiviral particles and performed Southern blot (Fig. 4A and 4B). The unfusion bands appeared in both cells infected with control or dual-gRNA lentiviral particles. However, the fusion band only appeared in the cells infected with dual gRNA1-2. These results demonstrated that the fusion results from dual-gRNA-mediated editing, which does not always lead to fusion. Next, we calculated the fusion rate by measuring the ratio of fusion to total band intensities, and revealed that the repeat site removal rate reached to 49.5 ± 5.8% in the HEK 293 cells (Fig. 4C). Therefore, we concluded that our designed gRNA1-2 lead to about 50% fusion at their editing sites, at least in the HEK293 cells we examined, which constitutively expressed Cas9 and were infected with our dual-gRNA.
Our dual gRNAs remove C9ORF72 repeat expansion and correct the repeat-produced RNA foci in primary cortical cultures
To examine whether our dual-gRNA strategy is able to remove patient expanded GGGGCC repeats, we cultured the cortical neurons derived from C9ORF72-BAC (C9-Tg) transgenic mouse (line 112), which carries 100–1000 repeat expansion in size 37. The existence of our gRNA1-2 target sequences was confirmed in the C9-Tg but not in the non-transgenic mice, indicating that the target sequences came from the BAC C9ORF72 transgene (Supplementary Fig. 8). We then crossed the C9-Tg mouse with Rosa26-Cas9 knock-in mouse 48 to ensure constitutive expression of Cas9 in the brain and in its derived neurons. Indeed, the editing site fusion appeared in the primary cultures infected with the lentiviral particles expressing the gRNA1-2 but not empty control (Fig. 5A). Like HEK293 cells, the fusion took place at the Cas9 cutting sites, confirming the removal of repeat expansion by our dual gRNAs (Fig. 5B).
We next investigated whether the pathology produced by the C9ORF72 repeat expansion is able to be corrected by our dual-gRNA approach. Similar to a previous report 37, RNA foci were detected in the primary cortical cultures derived from C9-Tg but not from wild type mice (Supplementary Fig. 9). With ~ 50% repeat site removal efficiency by gRNA1-2 (Fig. 4) and our high lentiviral infection rate (94.8 ± 2.7%) in our primary cultures (Supplementary Fig. 10), we expected ~ 50% foci correction efficacy in these infected primary cultures. Indeed, about half of sense foci were corrected by expression of gRNA1-2, which was evidenced by the percentage of cells containing the foci and the numbers of foci in the given cells (Fig. 5C-5E). Therefore, we concluded that our approach is able to remove the C9ORF72 repeat expansion and correct the repeat-produced RNA foci, one of the molecular and pathological hallmarks shown in the patients, in the primary cortical cultures carrying patient repeat expansion.
Our dual gRNAs remove C9ORF72 repeat expansion and correct the repeat-produced RNA foci in C9-Tg brain
With limited off-target effect, limited effect on C9ORF72 protein expression, high fusion rate, and RNA foci correction in vitro, we next asked whether our dual gRNAs are able to correct repeat-produced pathology in vivo. In agreement with a previous report 37, expression of repeat expansion generated widespread sense and antisense RNA foci in hippocampal CA1 pyramidal neurons at both 1- and 3-month old C9-Tg mice but not in that of 3-month old wildtype mice (Supplementary Fig. 11). We then crossed those C9-Tg mice with Rosa26-Cas9 knock-in mice to obtain the offsprings expressing both C9ORF72 repeat expansion and Cas9 (Fig. 6A). In these double positive mice, we injected AAV expressing dual gRNA1-2 driven by H1/U6 promoters and EGFP driven by CamKII promoter, a neuronal promoter, which enabled us to monitor the gRNA-targeted neurons in hippocampal CA1 region (Fig. 6A and Supplementary Fig. 12). The injection of AAV expressing EGFP driven by CamKII promoter (AAV-CamKII-EGFP) alone was used as non-editing control. After two month of injection, we achieved high AAV infection rate (Fig. 6B) and detected the fusion of gRNA1-2 editing sites in hippocampi infected with AAV-gRNA1-2-CamKII-EGFP, but not AAV-CamKII-EGFP control (Fig. 6C). Like our editing site fusion in the primary cortical cultures, the in vivo fusion took place at the same Cas9 cutting sites (Fig. 6D). Remarkably, AAV expressing gRNA1-2 significantly decreased the percentage of EGFP-positive neurons containing the sense and antisense RNA foci and the numbers of RNA foci in the EGFP-positive neurons, compared to the AAV-CamKII-EGFP controls (Fig. 6E-6G). Therefore, our results indicate that our dual-gRNA approach is able to remove the repeat expansion and correct the repeat expansion-generated pathology in vivo.