Generation of a tyr loss-of-function mutant by CRISPR-Cas9
The efficiency of genome editing is highly variable between loci making comparable assessments of different published methodologies difficult. To address this, we decided to compare the efficiencies of different HDR-mediated gene editing strategies at a single locus. To do this easily and efficiently, it was necessary to create a suitable animal model. The tyr gene encodes tyrosinase which converts tyrosine into melanin, and a mutation in tyr results in an albino phenotype in zebrafish embryos; therefore we chose the tyr mutant as a quick visible read-out [6]. We designed several single guide RNAs (sgRNA) targeting the tyr locus and selected one with high efficiency for the following experiments (Fig. 1a, b). Co-injecting this tyr guide RNA (gRNA) and Cas9 mRNA into one-cell stage zebrafish embryos caused reduction of pigmentation in more than 96% of injected embryos (108/112), some of which totally lacked pigmentation. A T7E1 mutagenesis assay demonstrated a ~ 80% efficiency of indel mutation at the locus (Fig. 1b). After screening several founders that transmitted targeted indels to F1 progeny, we established a stable line named tyr25del/25del that has a frameshift mutation caused by the deletion of 25 bp (Fig. 1c). The homozygous tyr25del/25del adult fish and their embryos developed normally but lacked body pigmentation (Fig. 1d). To verify the reliability of the tyr25del/25del mutant line, tyr transcripts were measured using quantitative real-time PCR (qRT-PCR). At 3 days post fertilization (dpf), tyr transcripts were significantly downregulated compared with sibling control embryos (Additional file 1: Figure S1). Since melanophores of tyr25del/25del are unable to produce melanin, this feature was used as a quick visible read-out for quantitatively comparing multiple repair template donors for HDR because of the correlation between phenotypic rescue and knock-in efficiency.
Comparison and optimisation of DNA template donors for HDR mediated knock-in efficiency
The 25 bp deletion in the tyr25del/25del genome created a new CRISPR-Cas9 site, which itself showed 73% efficiency of generating indels, here named tyr25del/25del gRNA (Fig. 1a, b, Additional file 2: Figure S2). To compare the HDR efficiencies of different strategies, we designed 12 different DNA donors (Fig. 2a). For the circular dsDNA (cdsDNA) donor, the targeted genomic locus of tyr was amplified from wild type genomic DNA, and cloned into a pMD19-T vector both with and without two CRISPR target sites at both ends of the homologous arms. We used a symmetrical 105 nt ssODN or an asymmetrical 129 nt ssODN both synthesised by Sangon Biotech. For the zLOST donors, a 299 nt or 512 nt lssDNA containing exon 1 was generated, with the 25 nt flanked by symmetrical left and right homology arms using the following protocol [17]. The dsDNA donor fragments were generated by PCR and purified. Since tyr25del/25del gRNA could not target the wild type tyr sequence, we directly injected zCas9 mRNA, tyr25del/25del gRNA and the different donors into homozygous albino embryos and checked the rate of pigmentation recovery (Fig. 2a).
If the injected embryos gained pigmentation, we named these individuals “pigmented embryos”. Scoring the pigmented embryos in a tyr25del/25del-HDR assay indicated that variations in length, single- vs. double-stranded DNA, linear vs. circular templates, and symmetrical vs. asymmetrical template donors all affected the HDR efficiency. Our observations suggest that HDR efficiency was maximal across zLOST, ssODN, and cdsDNA donors with two gRNA sites at both ends of the homologous arms (Fig. 2a). However, among dsDNA fragment donors, the length and symmetry of homology arms (≤3%) did not show an order of magnitude difference.
To further validate the high HDR efficiency of zLOST, ssODN, and cdsDNA donors, we compared and quantified the effectiveness of the tyr25del/25del rescue assay. For the cdsDNA donor, we found that only 5.4% of the larvae showed small numbers of pigmented cells at 2 dpf. For the ssODN donor with homologous arms, 39.1% of the larvae showed some cells with melanin production, a so-called “low rescue” or “medium rescue” phenotype. Asymmetry of homology arms unexpectedly reduced HDR efficiency (1.3%). However, the zLOST donor resulted in up to 98.5% of injected larvae with observable pigmentation at 2 dpf, significantly more than observed using cdsDNA or ssODN (Fig. 2b, c). While a high percentage of animals had a few cells edited, at those low efficiencies it is unlikely that the mutations will be passed on through the germline. Among these embryos, ~ 10% had extensive pigmentation (“high rescue”, more than 40 pigmented cells per larva) that was never observed in embryos rescued by other strategies (Fig. 2c). To establish that there is a direct “phenotype-genotype” relationship, genome extracts from ten embryos with extensive pigmentation after zLOST were confirmed to contain a correctly repaired tyr gene by Sanger sequencing (Fig. 2d). However, we did not identify such precise HDR-based repair in embryos recused by the less efficient ssODN donor and cdsDNA donor templates (data not shown). That may be because the low HDR efficiency observed with ssODN and cdsDNA occludes identification of the knock-in event by Sanger sequencing. We also designed a 512 nt lssDNA, but did not observe a significant increase in HDR efficiency despite the increased homologous arm length (Fig. 2a). These results indicated that using the tyr mutant as a quick visible read-out model to assess HDR efficiency was efficient, and that zLOST greatly improved HDR efficiency at the tyr locus.
High-efficiency editing of other genomic sites using zLOST
The efficiency of genome editing, regardless of the method used, is highly variable between different loci. Encouraged by the results of the tyr mutant gene rescue, we next investigated whether the relatively high efficiency of the zLOST method to precisely edit the zebrafish genome was generally applicable to other loci. We selected new target sites within three genes (th, nop56 and rps14) to perform specific knock-ins with different templates and confirm that zLOST efficiency is not a site-specific phenomenon. According to Easi-CRISPR and our previous result (Fig. 2a), the distal parts of zLOST were optimally designed to have 150 nt symmetrical homology arms. Details of the target genes, lengths of the ssDNA repair templates, homology arms, and sequencing data are shown in Fig. 3a, c and Additional file 3: Table S1. For each targeted locus, a new restriction site was introduced to identify the positive embryos and to easily screen germline transmission (Fig. 3a). At least 24 embryos per gene were assayed for correct targeting; we randomly selected embryos from the same injection group to perform restriction analysis, and three embryos were pooled per sample to make at least eight technical replicates. For the th locus, four of 9 injected embryo groups contained the introduced XhoI site (“positive embryos”) using zLOST as the repair template (Fig. 3b, top left). However, we did not find “positive embryos” using other knock-in strategies (data not shown). T-A cloning of the zLOST-modified PCR products followed by Sanger sequencing revealed that two out of 14 clones had seamless HDR modification, while three out of 14 clones carried indels (Fig. 3c). It is worth noting that six out of the 14 clones sequenced showed incorrect-HDR knock-in, as they also showed deletions at the target site (KI + indels) (Fig. 3c, where Δ1 and Δ2 are used to represent the indel, which is out of the shown sequence window). We raised mosaic F0 th embryos to adulthood and assayed the rate of germline transmission. Only two of the 21 adult fish that mated produced the desired XhoI identifiable allele and the germline transmitted mutations were confirmed by Sanger sequencing.
Using a similar approach for nop56, two of 8 samples were identified as “positive embryos” (Fig. 3b, top right) of which BamHI site conversion was observed in three of the 16 clones (Fig. 3c). Restriction analysis also indicated that rps14 sites could be efficiently targeted by zLOST (5 of 11 samples, Fig. 3b, bottom), and precision was again confirmed by sequencing (3 out of 16 clones, Fig. 3c). Taken together, these results demonstrate that knock-in zebrafish with specific point mutations can be generated with high efficiency using the zLOST strategy. Finally, we identified 4 nop56 founders (n = 17) with target knock-in mutations in their germline (23.5% germline transmission rate).
Improved assessment of zLOST-mediated HDR efficiency using next generation sequencing
We have developed a series of quantitative phenotype assays, restriction enzyme-based methods and Sanger sequencing to assess the specificity and efficiency of HDR by different donor constructs. However, none of these methods can truly assess the validity of HDR in depth because of the occultation of low frequency events. To address this we repeated the microinjections with different donors and selected 20 embryos at 2 dpf to perform next-generation sequencing. Using Illumina sequencing restricted to the targeted region, we quantitatively compared the editing efficiency of the three strategies, ssODN, cdsDNA and zLOST. The desired edit was a single base substitution only at the designed sites. However, considering that random mutation could also occur in the vicinity of the gRNA site, we decided that random synonymous mutations, which do not change the encoded amino acid, would not preclude a sample being considered as a correct editing event. For all sequenced samples, we divided the editing events into four categories: WT (no editing events happened), Correct_HDR (correct editing events), Incorrect_HDR (editing events happened, but with undesired events such as indels), and Other (other situations, mainly insertions, deletions and unmapped sequence). There was variation in the percentage of correct HDRs with synonymous mutations because of some unknown processes (Additional file 4: Table S2). Imperfect changes (Incorrect_HDR) were uncommon.
For gene nop56, there were 11,391,197 reads, 10,293,322 reads and 12,240,742 reads obtained from the ssODN, cdsDNA and zLOST samples, respectively. After assembly using FLASH, 97.44, 94.92 and 93.50% of these reads, respectively, were retained. Through analysis of nop56 editing, the percentage of correct editing events (the Correct_HDR) in zLOST was 11.82%, which was 22-fold higher than in ssODN (0.54%), and 7-fold higher than in cdsDNA (1.62%) (Fig. 4b and d, Additional file 4: Table S2). Similar results were observed for the targeting of the th and rps14 loci (Fig. 4a, c and d, Additional file 4: Table S2). For th, Correct_HDR events significantly improved from 0.09% in ssODN-treated embryos to 5.11% in those subjected to zLOST. Similarly for rps14, Correct_HDR events were found to be 0.60% in cdsDNA samples, which increased to 17.86% with zLOST. However, unexpected mutations were also found using zLOST, including other point mutations and indels (Incorrect_HDR). Despite this, in the nop56 loci modified by zLOST, Correct_HDR was still observed twice as frequently as Incorrect_HDR (11.82% vs. 5.64%). As such, the higher percentage of Correct_HDR suggests that our method, zLOST, overall showed a 22 to 57-fold higher editing efficiency than the other strategies.
zLOST enables precise modelling of human disease mutations in zebrafish
Base editing for a single amino acid is crucial to study gene function and model human disease. To this end, we tested the potential of zLOST to introduce human disease-related mutations in zebrafish. In many cases, simple loss-of-function mutations generated by targeted mutagenesis are not sufficient to recapitulate human genetic disorders, particularly diseases arising from gain-of-function point mutations. Clinical studies report specific mutation of TWIST2 is observed in patients with Ablepharon macrostomia syndrome (AMS) and Barber–Say syndrome (BSS). Both diseases are rare congenital ectodermal dysplasias with similar clinical features, but arise from different mutations: a lysine at TWIST2 residue 75 results in AMS, whereas a glutamine or alanine at the same site yields BSS [18]. We previously used the BE system to induce an amino acid conversion of p.E78K, precisely mimicking the mutation giving rise to AMS in humans [15]. However, the BE system cannot be used to generate a glutamic acid to glutamine change (p.E78Q) as observed in BSS. Instead, we used zLOST to create a p.E78Q mutation in zebrafish (Fig. 5a, b). After co-injecting the twist2 gRNA, zCas9 mRNA and a lssDNA donor into zebrafish embryos, we found that 4 out of 12 injected embryo sets harboured the desired conversion of G to C (data not shown). Sequencing of the positive embryos successfully detected G to C conversion in 6 out of 15 clones (Fig. 5b). We then went on to identify 7 founders (n = 22) with a p.E78Q knock-in mutation in their germline (31.8% germline transmission rate).
We further tested zLOST to generate the mutation of another human disease, Diamond-Blackfan anaemia (DBA), an inherited bone marrow failure syndrome (IBMFS) characterised by erythroid hypoplasia. Recent genetic studies reported that a heterozygous pathogenic non-synonymous variant (p. L51S) of the rpl18 gene is associated with DBA [19]. To test whether this point mutation directly results in DBA directly, an animal model with the precise point mutation needs to be established. To this end, we successfully used zLOST to achieve the conversion of CTC to TCG in the zebrafish rpl18 gene, thus inducing a p.L51S amino acid change in this protein (Fig. 5c, d). Further phenotypic analysis will be carried out on the zebrafish as they grow to adults. However, these results provide a clear demonstration of the ability of zLOST to achieve HDR, and the utility of this to transmit precise knock-in alleles through the germline.