Introduction

Phenotypic analysis of single and multi-gene mutants is critical for assessing the function of individual genes, redundancy between related genes, and genetic interactions. While molecular and cellular phenotypes resulting from genetic disruption in monocellular organisms and cultured cells can be informative, whole animals are required to identify affected tissues throughout development and assess more complex phenotypes. Model organisms used by geneticists include Drosophila melanogaster (fruit fly), Caenorhabditis elegans (nematode worm), Xenopus laevis (African clawed frog) and Danio rerio (zebrafish), but the mouse is the “workhorse” for understanding mammalian genetics and modeling human disease.

Experimental techniques can be divided into two categories: forward genetics (phenotype-driven) and reverse genetics (gene-driven)¹. In mice, forward genetics began with the collection of animals with visible traits by 19^th century “mouse fanciers”. These traits occurred naturally in wild populations or arose spontaneously in laboratory colonies. Subsequently, the mutation rate was enhanced by mutagenising the male germline with radiation (primarily X-rays) or chemicals (primarily N-ethyl-N-nitrosourea, ENU) or by “gene-trapping” using transposons (Sleeping Beauty or PiggyBac). Of these, the “supermutagen” ENU is most frequently used to screen for phenotypic traits². ENU is a DNA alkylating agent that creates small lesions, mostly single base pair changes, creating hypomorphic/hypermorphic/neomorphic/dominant-negative/null/neutral alleles. Dominant traits (e.g. development of cataracts) can be detected in G1 offspring derived from the mutagenised males. Recessive traits (e.g. embryonic lethality) require further breeding of G2 females with G1 males to achieve homozygosity of de novo mutations in G3 offspring^3,4. ENU mutagenesis has also been used to screen for dominant suppressors of genetic diseases (thrombocytopenia, Rett syndrome and Factor V Leiden), revealing key disease pathways and druggable targets^5–8. The greatest challenge in these screens is identifying the causative mutation(s). ENU mutagenesis is random and G1 pups carry ∼40-70 intragenic lesions (and many more intergenic ones). The vast majority of these have no/little phenotypic consequence and mutation(s) underlying observed phenotypes must be determined in pedigrees derived from the G1 animals by linkage analysis or by a combination of whole exome sequencing (WES) and genotyping of G3 animals. If G1s cannot be bred due to fertility or health problems, then identification of causative mutations is not possible. Overall, ENU screening involves several challenges and huge numbers of animals (especially for recessive traits), highlighting a clear need for a more efficient and cost-effective method.

The sequencing and annotation of model organism genomes has revolutionised genetic studies. Of ∼20,000 protein-coding genes in the human genome, 80-90% are conserved in mice⁹. Reverse genetics approaches predominantly involve knocking out individual genes to produce null alleles. Such studies have been carried out at a small scale by individual laboratories on their genes of interest, and at a large scale by the International Mammalian Phenotyping Consortium (IMPC). The IMPC’s aim is to identify the function of every protein-coding gene in the mouse genome by producing a knock-out mouse line for each¹⁰. Initially, mouse lines were created by gene targeting in cultured embryonic stem (ES) cells, which were then used to produce chimeras comprising both wild-type and mutant ES cells. If the mutant cells contributed to the germline, mutant mouse lines could be established upon breeding these chimeras. Over the last decade, CRISPR/Cas9 technology has been adopted to improve the efficiency of gene editing in ES cells¹¹, and – more importantly – to perform gene targeting directly in zygotes^12–14. This has accelerated mouse model production and enabled mutations to be introduced onto different strain backgrounds (circumventing the need for backcrossing). There are several ways to deliver the single guide RNA (sgRNA) and Cas9 nuclease to zygotes, with the best results obtained when recombinant sgRNA and Cas9 protein are combined to form a ribonucleoprotein (RNP), which is introduced into zygotes via electroporation¹⁴. Null alleles can be produced by excising a critical exon with sgRNAs targeting either side; or by creating insertions/deletions (indels) within a critical exon using one sgRNA and relying on imperfect repair via non-homologous end joining (NHEJ) or microhomology-mediated end joining (MMEJ). Founders are bred and quality control assays are performed on N1 animals to verify disruption of the targeted gene before mouse lines are established and analysed¹⁵.

CRISPR/Cas9 technology has also been adopted for forward genetic screening of cellular phenotypes in both cultured cells and in vivo in model organisms (including mice) by introducing Cas9 with a genome-wide or bespoke library of sgRNAs at a low multiplicity of infection such that each cell carries one sgRNA and, as a result, its target gene is disrupted^16–18. To assess whole animal phenotypes, recent studies using zebrafish and Xenopus have performed in vivo genetic screens for phenotypes of interest by introducing RNPs into zygotes and phenotyping the resulting F0 founder animals^19–23. These founders are named “crispants”, a portmanteau of CRISPR and mutant. Crispants are genetically mosaic, but they display the null phenotype associated with the targeted gene due to high levels of indel formation. The resulting mutations can be identified by sequencing the sgRNA target sites in the founders. As both copies of the targeted gene are disrupted, recessive phenotypes can be assessed.

The mouse community remains sceptical about whether phenotypic assessment of F0 crispant mice can produce reliable results, with initial studies reporting phenotypic variability due to genetic mosaicism²⁴. Here, we target seven genes associated with recessive visible traits such as coat colour and demonstrate that up to 100% of crispant mice per gene displayed the expected null phenotype despite genetic mosaicism. Our results were due to very high levels of Cas9- mediated double-stranded breaks and the resulting production of loss-of-function mutations. We then asked whether it was possible to target two genes per zygote as this would be desirable when assessing genetic redundancy or genetic interactions. Unfortunately, we found that this decreased the viability to a level that made it not worthwhile. Genes would therefore need to be targeted sequentially in zygotes to introduce secondary mutations. We obtained a relatively high proportion of non-mosaic founders that were compound heterozygous for two null alleles or one null allele and one short in-frame mutation. When compound heterozygous founders carrying a short in-frame mutation displayed the recessive loss-of-function phenotype, this mutation pin- pointed a critical functional protein domain. Overall, our findings demonstrate the validity of phenotypic analysis of crispant mice.

Results

Crispant mice display recessive coat colour phenotypes

Phenotypic assessment of mouse crispant founders is generally avoided due to fears that genetic mosaicism would invalidate results. To determine whether phenotypic assessment of founders is biologically informative, we selected genes that when mutated present with recessive visible phenotypes. Mice with altered appearance first became popular among mouse fanciers and there are now over 300 “named” alleles – named after their phenotypes²⁵. Most of the genes responsible and the causative mutations have now been identified. We chose genes for this study that have a recessive phenotype, no impact on viability or health, and a phenotype visible on a C57BL/6J background. We mimicked previously published null alleles (exon deletion or nonsense mutation) by targeting affected exons with a single sgRNA, introducing insertions/deletions (indels) from the repair of Cas9-induced double stranded breaks via NHEJ/MMEJ. We used strict off-target criteria¹³ whereby all sgRNAs had a least three mismatches to unintended targets, at least one of which was in the 11 bp “seed” region adjacent to the PAM, and all sgRNAs had a predicted frameshift frequency >67% (assessed using inDelphi²⁶). We first selected two coat colour genes, as any phenotypic mosaicism would be clearly visible, allowing us to distinguish between a full, partial or absent knock-out phenotype and link this to the mutations carried by each founder.

Mutations in Tyrp1 (tyrosinase related protein 1) result in brown coat colour on a nonagouti (a/a) background. Although the original “brown” allele, Tyrp1^b, is a SNP (c.329G>A) resulting in the p.C110Y missense mutation²⁷, many other historical “brown” mouse lines contain large deletions at this locus. We therefore decided to knock out Tyrp1 by targeting a constitutive, translated exon located in the first half of the gene and with a length indivisible by three. The gene is composed of 8 exons, with alternative splicing affecting the length of exons 1 and 8. The open reading frame starts in exon 2 and ends in exon 8. We targeted exon 3 (Figure 1A; Table S1), previously deleted by the KOMP project in ES cells²⁸ (Tyrp1^{tm1a(EUCOMM)Wtsi}). We electroporated 19 mouse zygotes with RNP and produced four founder animals (21% viability), all of which had a fully brown coat colour (Figure 1B; Table 1).

Figure 1:

Table 1:

PCR amplification of the targeted locus from tail-tip genomic DNA (gDNA) of three founders (#4083, #4084 and #4085) produced at least one band that differed in size from the wild-type reference, indicating the presence of large indels (Figure 1C-D). Sanger sequencing of the amplicons produced mixed chromatograms that we deconvoluted using DECODR²⁹ and Synthego ICE³⁰ software (Figure S1; Table S2). DECODR was better at detecting longer deletions, so this software was primarily used throughout this study.

Since the trait is recessive, we were surprised that DECODR detected 26.9% wild-type (unedited) sequence in founder #4084, suggesting editing occurred in 3/4 alleles at the 2-cell stage, leaving 1/4 alleles unedited. This would mean that 50% cells carry a functional allele, which is at odds with the absence of black fur in this animal. One possibility was that this founder carried larger “unseen” deletions that abolish one or both primer annealing sites, so the PCR amplicon represented only a subset of the total alleles and therefore overestimated the proportion of unedited sequence. Quantification by qPCR around the genotyping primer annealing sites revealed that this is not the case (Figure 1C,E). Deep amplicon sequencing (DAS) revealed that a similar proportion of reads (20.4%) contained a duplication of an adenine base (c.644dupA; p.H215QfsX48; Figure 1F). We therefore conclude that DECODR (and ICE) mis- assigned this sequence as unedited when it actually contains an out of frame (+1 bp) insertion, consistent with the fully brown coat phenotype.

Surprisingly, founder #4086 appeared to carry a single allele: a short in-frame (12 bp) deletion (c.636_647del), resulting in the loss of four amino acids (p.D212_H215del) (Figure 1G-H, S1; Table S2). It seemed unlikely that this founder would be homozygous for this allele, as the same repair event would had to have happened in both copies of Tyrp1, and this repair product was not predicted by inDelphi. qPCR analysis around the genotyping primer annealing sites found that gDNA abundance was decreased by ∼50% over the 5’ primer (Figure 1E), suggesting that this founder was heterozygous for a “unseen” deletion spanning the 5’ primer annealing site. To identify the “unseen” deletion in founder #4086, we adapted ELF-CLAMP (Enrichment of Long DNA Fragments using Capture and Linear Amplification)³¹ where regions of interest (up to 2 kb) are amplified from viewpoints using in vitro transcription followed by Nanopore long read RNA sequencing (Figure S2A). We detected a large deletion-insertion (1′140insCTT) within exon 3, which would lead to a frameshift mutation: c.794_933delinsCTT; p.I169Lfs*49 (Figure 1G, S2B). This was confirmed by PCR amplification and Sanger sequencing. Although a 140 bp deletion at a single cut site is relatively large, there are reports of on-target CRISPR/Cas9- mediated deletions in mouse zygotes up to 600 bp³².

With the knowledge that both alleles must be disrupted to cause the brown phenotype, both alleles in this compound heterozygous founder (#4086) must be non-functional, indicating that the four amino acid deletion is sufficient to destroy TYRP1 protein function. However, a report that CRISPR/Cas9-mediated mutagenesis in cell lines can cause genome damage in cis with small de novo mutations³³ raises the alternative possibility that undetected damage nearby may be responsible for disrupting Typr1. Using ELF-CLAMP, we sequenced the region flanking this mutation and did not detect any additional mutagenesis, ruling this out (Figure S2B). The 12 bp (four amino acid) deletion is located in the tyrosinase domain (residues 183-417), and therefore demonstrates the functional importance of this domain in vivo. Consistent with this, a chemically induced (ENU) allele with a missense mutation in the tyrosine domain, Tyrp1^b-Btlr; c.887A>G; p.Y296C (Figure 1H), also confers a brown coat colour. Overall, all four founders reliably displayed the null phenotype because we achieved complete editing and all de novo mutations carried by these animals were loss-of-function alleles (including short in-frame indels).

We next targeted a second coat colour gene, Slc45a2 (solute carrier family 45, member 2), which when mutated results in a light grey coat on a nonagouti background – named “underwhite”³⁴.

The gene is composed of seven constitutive exons and the open reading frame spans them all (Figure 2A; Table S1). The Slc45a2^uw allele contains a 7 bp deletion in exon 3, truncating the protein (c.773_779delCGTCGCT; p.S258CfsX44)³⁵. As we were unable to identify an sgRNA in this exon that met our off-target criteria, we instead mimicked Slc45a2^uw-10Btlr, which phenocopies Slc45a2^uw mice and encodes a nonsense mutation in exon 1 (c.39T>A; p.Y13X).

Figure 2:

Electroporation of 39 zygotes resulted in 21 founders (53.8% viability; Table 1). Only two (9.5%) displayed the full underwhite phenotype, six (28.6%) had a patchy grey and black coat, and the remaining 13 (61.9%) appeared wild-type (Figure 2B). This result could not be explained by a lack of mutagenesis as no unedited sequence was detected in any founders (Figure 2C-D; Table S2).

We therefore asked whether we could determine the functional importance of the mutated N- terminal cytoplasmic domain from genotypic analysis. PCR amplification resulted in a single band for three phenotypically wild-type founders (#4062, #4064 and #4079) (Figure 2D).

Founders #4064 and #4079 were compound heterozygotes carrying two short deletions at a 1:1 ratio, one of which was in-frame: c.35_43delCCTATCAAT; p.T12T,Y13_S15del in #4064 and c.32_37delATACCT; p.H11H,T12_Y13del in #4079 (Figure 2E-F, S3; Table S2). We conclude that short in-frame deletions in the N-terminal cytoplasmic domain do not disrupt S45A2 protein function. Founder #4062 was a compound heterozygote carrying a complex deletion-insertion comprising a short deletion and an insertion including 82 bp from intron 1 (c.36_42delCTATCAAins[AA,385+647_385+566]), and a 479 bp deletion that was detected by ELF-CLAMP (c.1-1_385+93del) (Figure 2E,G, S3; Table S2). Since the underwhite phenotype is recessive, only one of these needs to be functional for the mouse to appear wild-type. The deletion-insertion provided an alternative start codon, predicted to enable the translation of a protein containing a short deletion, p.S2_Q14del (Figure 2F, S3). The finding that short-in frame mutations at the target site do not abolish S45A2 protein function explains the phenotypic variability in these founders.

Crispant knock-out mice display recessive visible phenotypes

We selected four more genes linked to recessive visible phenotypes to knock out in crispant founders. Two genes affect the coat (ectoderm): Fgf5 (fibroblast growth factor 5) and Tgm3 (transglutaminase 3), associated with the named alleles “Angora” (long coat) and “Wellhaarig” (curly coat and whiskers), respectively^36,37. The other two genes have roles in the mesoderm: Pax1 (paired box 1) mutated in “undulated” (short wavy tail) and Bmp5 (bone morphogenetic protein 5) mutated in “short ear” mice^38,39. As with Tyrp1 and Slc45a2, we took the gene structures into account and designed an sgRNA for each gene to mimic an existing knock-out allele^36,39–41 (Figure 3A-D; Table S1). We electroporated 20-27 zygotes per gene and produced 5- 12 founders for each, giving a viability of 25-44% per gene. All founders displayed the expected recessive visible phenotypes (Figure 3E-H; Table 1, S2).

Figure 3:

Targeting two genes in crispants is possible but reduces viability

Targeting two genes at once in zygotes could be used to assess redundancy between related genes or to investigate genetic interactions, e.g. screening for modifiers of a chosen null phenotype. We tested four combinations of genes, ensuring pairs were on separate chromosomes to avoid large deletions. To avoid reagent toxicity, the total amount of RNP used was the same as for the single sgRNA experiments, with half the RNP containing each sgRNA. The viability was very low: 6/202 = 3.0% (**** P < 0.0001, Fisher’s Exact Test; Figure 3I; range per experiment: 0-5.4%; Table 1). The surviving founders, however, did display both expected recessive phenotypes (Figure S4).

Compared to experiments targeting single loci, the reduction in viability could be due to toxicity from the increased number of Cas9-induced breaks per cell and/or to synthetic lethality between the pairs of genes. To investigate this, we paired one sgRNA targeting an endogenous gene (Tyrp1/Fgf5/Tgm3/Bmp5) with an sgRNA targeting EGFP in transgenic zygotes carrying an EGFP transgene⁴². From a total of 117 zygotes, we recovered 17 pups at weaning (14.5%). This was significantly lower than when these four genes were targeted alone (*** P = 0.0007, Fisher’s Exact Test; Figure 3I; Table 1). When the viability for each gene was compared to the same gene combined with EGFP (Figure 3J), viability was significantly reduced for Fgf5 (** P = 0.0048) and Bmp5 (* P = 0.035). There was also a downward trend for Tgm3, but no difference for Tyrp1 (which had the lowest viability when targeted alone). Since disrupting EGFP shouldn’t impact health, we conclude that the generation and repair of more Cas9-induced breaks reduced viability.

The transgene was X-linked, allowing us to compare the impact of adding of one extra target site in hemizygous males vs two extra sites in homozygous females. The high percentage of males in the paired sgRNA experiments (12/17 = 71%) was indicative of increased toxicity from four target sites, but this sex ratio wasn’t significantly different from that obtained in the single sgRNA experiments due to low founder numbers (Figure 3K). Assuming equal numbers of males and females in the 117 zygotes used in the paired sgRNA experiments, the viability of male and female founders was 20.5% and 8.5%, respectively (Figure 3L). The viability of female founders was higher than when two endogenous genes were targeted (8.5% vs 3.0%), despite the same number of target sites. This suggests that synthetic lethality was also a factor.

Overall, the more than four-fold decline in viability makes targeting two genes together less efficient than targeting a second gene in zygotes that are homozygous for mutations in a first gene, derived from a heterozygous intercross (Mendelian ratio: 25%).

All 17 founders obtained in the paired sgRNA experiments lacked EGFP fluorescence and displayed the full KO phenotype associated with the targeted endogenous gene, with mutagenesis at both loci (Table 1). The only exception was one Tyrp1 + EGFP founder, #4728, which had brown fur with black patches (Figure 4A, discussed later).

Figure 4:

Crispants carry high levels of genetic editing and variable indel size in all genes

From our experiments detailed above, we generated a total of 60 founders with one targeted gene, 17 with one targeted endogenous gene plus EGFP, and six founders with two targeted endogenous genes. In total, we mutated the endogenous genes 89 times. Genotypic analysis of all five Pax1 founders using DNA derived from tail (largely ectoderm), spleen (mesoderm) and liver (endoderm) showed high concordance between the three germ layers, with variation limited to the less abundant mutagenesis products (Figure S5; Table S3). This validated the use of tail tip DNA for genomic analysis in all other founders.

Phenotypic variability could result from incomplete editing. This is easily detectable in coat colour mutants (as black patches), but may not be observable in founders with the other visible phenotypes. Only one Tyrp1 founder, #4728, had black (wild-type) patches (Figure 4A).

DECODR predicted 3.8% of its DNA was unmodified, which we confirmed by DAS (3.6%) (Figure 4B; Table S4). Since the phenotype is recessive, up to 7-8% of cells could retain TYRP1 protein function in this animal. DECODR predicted the presence of unedited sequence in one other Tyrp1 founder (#5486). Like #4084 (discussed above), this animal displayed the full brown coat colour phenotype, and DAS revealed that it carried same 1 bp duplication that had been mis- assigned as unedited (Figure 1F, 4B, S1, S6A-D; Table S2, S5). DECODR predicted unedited sequence in two Tgm3 founders (#4136 and #4723). DAS confirmed the presence of 10.7% unedited sequence in #4723 (Figure 4B, S6E-H; Table S4). In #4136, however, the deletion of one of a run of three cytosines (c.739delC; p.P247QfsX47) had been misassigned as unedited (Figure 4B, S6E-H; Table S4). Finally, a low level of unedited sequence was predicted in tail tip DNA in three Bmp5 founders. As the short ear phenotype affects the mesoderm, we repeated this analysis using splenic DNA where no unedited sequence was detected in any of the founders (Figure 4B; Table S2). Overall, incomplete editing was only detected in two founders in this study.

We predominantly used DECODR to deconvolute Sanger sequencing chromatograms as it better predicted longer indels, but we found it grouped together two or more indels of the same length, assigning them as the most abundant sequence. In contrast, ICE separated these distinct repair products, as illustrated by Pax1 founder #4088 (Figure S5D; Table S2). At all loci, we obtained a large number of short indels whose relative abundance reflected inDelphi predictions (Table S2-S5; Figure S5C). This was most clear in the Tgm3 founders since this cut site had high microhomology strength, predicting an 8 bp deletion to occur with a frequency of 44.5%. This allele was detected in 10/20 Tgm3 founders (Figure S6E-H, S7; Table S2-S5). Despite the preference for this repair product, no founders were homozygous for this Tgm3 allele. We did not generate any homozygous founders (i.e. with the same mutation on both alleles).

Large on-target deletions were common. The longest deletion within the genotyping amplicons was 1′598 bp (554_683+368del in Bmp5 founder #3722; Figure 4C; Table S2). ELF-CLAMP enabled us to sequence deletions that extended beyond the genotyping amplicons, the largest of which was 1′822 bp (c.485-634_672 in Bmp5 founder #4727; Figure 4C; Table S4). The presence of grey patches on Bmp5 (short ear) founder, #3725, indicated changes in the Myo5a gene, 700 kb away, in which recessive mutations cause a “dilute” coat colour (Figure 4D-E). We performed qPCR to determine whether deletions could extend to the Myo5a locus. There was a reduction at Myo5a in #3725 compared to wild-type controls, with lower levels in gDNA extracted from skin in the grey patches than from the spleen (Figure 4F). The only large insertion we detected was the 82 bp from intron 1 in Slc45a2 founder #4062 (Figure 2G, S3A, mentioned above). Other types of previously reported genome damage (large inversions or mutations in cis with short indels) were not detected.

Compound heterozygous crispant founders enable genotype – phenotype association

Several founders were compound heterozygotes (Figure 4G, S5, S7-9; Table S2-S5). Most of these (11-27% all founders per gene, except Slc45a2) carried two null alleles: two short frameshifting indels or one short frameshifting indel and one large deletion. For all genes except Pax1, we obtained at least one founder that was heterozygous for an in-frame indel (combined with a null allele). Since all traits are recessive, the phenotype of these founders reveal whether a short in-frame indel is sufficient to abolish protein function and therefore inform whether the mutated domain is required. As described above, these founders enabled us to determine that the tyrosinase domain of TYRP1 is critical for protein function and the first cytoplasmic domain of S45A2 is dispensable (Figure 1H, 2F; Table S2). The in-frame mutation in Fgf5 founder #3728 (c.283_300del; p.I95_Q100del) demonstrated the functional importance of mutating the well-conserved FGF (fibroblast growth factor) domain (residues 85-214), involved in binding its receptor (Figure 4H, S9; Table S2). The in-frame mutation in Tgm3 founder #4724 (c.737_739del; p.D246A,P247del) revealed that mutation of an uncharacterised region downstream of the first cluster of Ca²⁺ binding sites and upstream of the third transglutaminase domain impacts TGM3 protein function (Figure 4I, S9; Table S4). This mutation may specifically disrupt an unknown functional domain or it may impact TGM3 protein level, e.g. by disrupting splicing, post-translational processing or protein stability. The in-frame mutations in Bmp5 founders #4727 (c.600_611del; p.I200M,Y201_K204del) and #4734: c.595_596 (insCTCATCCTT; p.Y199SinsHPY) revealed that mutation of the propeptide can affect BMP5 protein function, perhaps by affecting its cleavage (Figure 4J, S9; Table S4).

Crispants can form an allelic series that shed light on protein domain function

We next asked if we could use this method to obtain non-null alleles that would shed light on protein domain function. For this, we aimed to produce truncated and in-frame alleles by targeting Adamts20 (A disintegrin-like and metallopeptidase (reprolysin type) with thrombospondin type 1 motif, 20), associated with the “belted” phenotype. This gene is composed of 41 exons, alternatively spliced to form six isoforms. The most common isoform contains exons 3-41 (Figure 5A). ADAMTS20 protein consists of N-terminal Reprolysin and Disintegrin domains and 15 C-terminal thrombospondin type 1 (TSP) repeats. Adamts20^bt is a missense mutation (c.1598C>T; p.L533P) in the Disintegrin domain⁴³. Several other missense and nonsense mutations in Adamts20 phenocopy Adamts20^bt(Figure 5A). Of these, the best characterised truncated allele is Adamts20^bt-Bei1 (c.2860C>T; p.R954X), which lies downstream of the disintegrin domain in the third TSP repeat⁴³. We designed an sgRNA nearby R954X in exon 22 and produced 11 founders (Table 1, S1). Seven (63.6%) displayed the full white belt phenotype, three (27.3%) had a small patch of white fur and one (9.1%) appeared “wild-type” (Figure 5B, S4).

Figure 5:

Three of the seven founders with the full phenotype were compound heterozygotes. Two of these (#2984 and #2986) each carried a truncating indel and a large “unseen” deletion. Given that the trait is recessive, we can conclude that the truncated alleles are sufficient to cause the belted phenotype (Figure 5C, S4; Table S5), consistent with the Adamts20^bt-Bei1 allele that we mimicked. We sequenced the “unseen” deletions by ELF-CLAMP. #2986 carried a 936 bp deletion, which was the largest deletion we observed in this study (Figure 5D; Table S6). #2984 carried a 626 bp deletion, with an additional 16 bp deletion 633 bp upstream (Figure 5D; Table S5). The third compound heterozygote (#3745) carried a long in-frame deletion (Δ45 bp): c.2782_2826del; p.M928_D942del and a frameshift mutation (Figure 5B-C, S10; Table S5), suggesting that mutating the TSP3 repeat is sufficient to abolish ADAMTS20 protein function.

The other four founders with the full phenotype were mosaic. One (#2983) carried a 17 bp deletion within the genotyping amplicon and several large deletions (sequenced using ELF-CLAMP). We predict that the large deletions are truncating/null since they span one/both ends of exon 22, so this exon will be excluded in Adamts20 transcripts. This founder therefore carries one truncating allele and several null alleles (Figure 5C, S4; Table S5). The founder with the wild-type appearance (#3748) also carried one short mutation within the genotyping amplicon and several large deletions. This mutation was an in-frame (-6 bp) insertion-deletion (c.2795T>A, 2798_2807delinsTTGC; p.V932_336delinsDLA) (Figure 5C, S10; Table S6).

Unlike TYRP1 and S45A2 where loss-of-function phenotypes are cell-autonomous, ADAMTS20 protein is secreted⁴⁴ and therefore it is not necessary for all cells to express functional ADAMTS20 for an animal to have a wild-type appearance. We therefore conclude that this short in-frame insertion-deletion in TSP3 does not disrupt ADAMTS20 protein function.

A partial phenotype could result from lower levels of secreted functional ADAMTS20 protein or from the presence of a hypomorphic variant. Two founders (#3734 and #3743) displayed small white patches, and carried multiple variant alleles including short in-frame indels (Figure 5B; Table S6). A third partially belted founder (#3744) carried a 42 bp deletion (c.2792_2833del; p.S931C, V932_C945del) and several large deletions (Figure 5B-C, S10; Table S6). We propose that the Δ42 bp allele is hypomorphic due to its size and the overlap with the non-functional 45 bp deletion in founder #3745 (Figure 5C). In support of this hypothesis, several published Adamts20 alleles cause partial belted phenotypes, one of which lies in exon 22 and encodes a missense mutation in the TSP3 repeat (Adamts20^m5Btlr: c.2779T>G; p.C927G; Figure 5A,C).

Using one sgRNA, we have produced eleven founders that include a non-null allelic series comprising three truncated alleles and three in-frame alleles with varying impact on coat colour that allowed us to classify them as loss-of-function, functional or hypomorphic.

Discussion

Phenotypic assessment of crispant founders is replacing historical methods for genetic screening in zebrafish and Xenopus studies. Here, we demonstrate that crispant mouse founders display recessive traits and so could be adopted for initial in vivo assessment of mutant phenotypes. The major concern in the mouse community is that genetic mosaicism arising from mutagenesis occurring after cell division in zygotes will impact the phenotype^24,45,46. Incomplete editing and the presence of short in-frame variants that retain function are of particular concern. By using recessive visible traits, we could determine whether individual founders displayed the expected phenotype. Incomplete editing was very rare: we achieved 100% editing in all but two animals in this study. Genetic mosaicism was common, but did not impact the phenotype when all alleles present disrupted function, including short in-frame indels.

The standard approach for generating null alleles is targeting a critical exon with an sgRNA¹². Repair by NHEJ/MMEJ can generate frameshift mutations, which lead to the presence of premature termination codons (PTC) that target transcripts for nonsense mediated decay (NMD). This approach does not consider in-frame indels (which theoretically represent a third of repair products) to be sufficient to disrupt genes. Our results demonstrate that the best strategy to generate a null allele is to ensure that an sgRNA abolishes protein function by considering the protein domain structure and targeting an essential functional domain. We also show that targeting a domain of unknown function can reveal its functional importance in crispants. Short in-frame indels induced by the sgRNAs in this study disrupted protein function in Tyrp1, Fgf, Tgm3 and Bmp5, but not in Slc45a2 and Adamts20.

It is recommended to avoid targeting too close to the N-terminus of a gene because transcripts with a PTC <150 nt downstream of the start codon may also escape NMD, due to translation from a downstream in-frame start codon⁴⁷. Our sgRNA targeting Slc45a2 was within this range and there are three in-frame start codons just downstream. However, the presence of an ATG downstream of a PTC is not always sufficient to rescue gene function, as demonstrated by the null phenotype of the published mutation that we mimicked (Slc45a2^uw-10Btlr c.39T>A; p.Y13X). Slc45a2 founder #4062 revealed an independent mechanism by which targeting too close to the N-terminus could fail to disrupt a gene: mutagenesis can generate an alternative start codon, resulting in a variant where only the first few amino acids are lost.

On-target genome damage resulting from Cas9-mediated breaks is increasingly reported^32,33. This includes large deletions, insertions and rearrangements. These events are often missed by routine genotyping assays: most commonly PCR amplification and sequencing of <1 kb regions around the target site. We detected the presence of large deletions by qPCR and sequenced a subset of these using ELF-CLAMP. When aiming to generate null alleles, genome damage is not an issue as long as it does not extend beyond the targeted gene. In Bmp5 founder #3725, we detected deletions that extended to the Myo5a locus (700 kb away). We speculate, however, that extremely long deletions cannot occur at a high frequency as the loss of flanking genes will likely result in lethality - as is known for large deletions spanning Myo5a and Bmp5⁴⁸.

Phenotypic assessment of crispants has several applications. While the phenotypes assessed here were detectable in individual animals, traits with lower penetrance could be assessed in cohorts of crispants (generated from a single round of mutagenesis). In situations when CRISPR/Cas9-mediated mutagenesis results in no live pups, it can be hard to determine whether this is due to technical failure or because the targeted gene is essential during embryogenesis⁴⁹. The presence of developmental phenotypes in crispant embryos would suggest that the targeted gene is essential. Therefore, mouse lines could be generated by producing founders with incomplete editing (by reducing targeting efficiency or injecting CRISPR/Cas9 reagents into one blastomere in 2-cell embryos), which could be bred to generate heterozygotes. Similarly, when knocking out both copies of a gene causes infertility in both sexes, defects in gametogenesis could be detected in crispants, and mouse lines could instead be established using this strategy. Crispants can also be used for initial phenotypic assessment when investigating genetic interactions. IMPC mouse lines are generated on C57BL/6N and crispants could be produced on other strain backgrounds to assess for phenotypic modifiability. It is often desired to study the interaction between two genes. Unfortunately, we show that targeting both genes simultaneously dramatically reduces viability. Instead, we propose to establish/obtain a mouse line with a mutation in the first gene and use this line as donors to mutate the second gene in crispants. This strategy can be applied to screen for genetic suppressors of disease – which can reveal potential therapeutic targets^6–8.

Making genotype-phenotype associations with non-null alleles helps to dissect protein function. We produced compound heterozygous crispants carrying an in-frame mutation and a null allele in five of the six genes that we targeted for disruption. For recessive traits, the presence or absence of the null phenotype in these founders revealed whether in-frame indels were sufficient to abolish protein function. C-terminal deletions can be used to assess the importance of the domain(s) downstream of a truncating mutation. As demonstrated here for Adamts20, truncating and in-frame variants can be generated using the same sgRNA. Compound heterozygous crispants could be bred to establish mouse lines with the alleles of interest for phenotypic confirmation and mechanistic studies. Mosaic animals carrying potentially informative variants could be also bred to segregate alleles for phenotypic analysis. We demonstrate that a single round of mutagenesis can generate many alleles and that these can be identified via thorough genotyping of crispant founders.

Using phenotypic assessment of crispants as a first stage in multiple experimental pipelines will inform on the most promising candidates to prioritise for further study. This will save considerable time and resources and dramatically reduce animal numbers.

Materials and Methods

Accession numbers for genes, transcripts and proteins

Nucleotide and amino acid numbering refer to the following accession numbers for mouse genes, transcripts and proteins. Tyrp1 gene (NCBI ID: 22178,), Tyrp1 transcript (NM_031202.3) and TYRP1 protein (P07147.1). Slc45a2 gene (NCBI ID: 22293), Slc45a2 transcript (NM_053077) and S45A2 protein (P58355.1). Fgf5 gene (NCBI ID: 14176), Fgf5 transcript (NM_010203.5) and FGF5 protein (P15656.1). Tgm3 gene (NCBI ID: 21818), Tgm3 transcript (NM_009374.3) and TGM3 protein (Q08189.2). Bmp5 gene (NCBI ID: 12160), Bmp5 transcript (NM_007555.4) and BMP5 protein (NP_031581.2). Pax1 gene (NCBI ID: 18503), Pax1 transcript (NM_008780.2) and PAX1 protein (NP_032806.2). Adamts20 gene (NCBI ID: 223838), Adamts20 transcript (NM_177431.5) and ADAMTS20 protein (P59511.2).

Experimental model and subject details

Mouse lines

Animal procedures were approved by the Animal Use Committee at the Canadian Council on Animal Care (CCAC)-accredited animal facility, The Centre for Phenogenomics (TCP). All mice were housed in a specific-pathogen-free (SPF) facility. They were maintained on a 12-h light/dark cycle and given ad libitum access to food and water. Mice were fed a standard diet (HarlanTeklad 2918), consisting of 18% protein, 6% fat and 44% carbohydrates. They were housed in individually ventilated cages (IVC) with wood chippings, tissue bedding and additional environmental enrichment in groups of up to five animals.

Wild-type C57BL/6J (The Jackson Laboratory) and Tg(CAG-EGFP)D4Nagy (JAX stock #018979; backcrossed onto C57BL/6J for >8 generations)⁴² were used to derive embryos. Wild-type or homozygous Tg(CAG-EGFP)D4Nagy females were used as embryo donors at 3-4 weeks. Wild-type or hemizygous Tg(CAG-EGFP)D4Nagy males were used for breeding to generate embryos. CD-1 (ICR) (Charles River Laboratories) outbred albino stock females were used as pseudopregnant recipients.

Method details

Generation of “crispant” founders

Crispant founder mice were generated by the Model Production Core at The Centre for Phenogenomics. Donor females were superovulated by intraperitoneal (IP) injection of 5 IU of pregnant mare serum gonadotropin (PMSG, Prospec, HOR-272) at 10 am (room light cycle: 5 am/on, 5 pm/off) followed 48 hours later by an IP injection of 5 IU human chorionic gonadotrophin (hCG, EMD Millipore 230734) and mated overnight with proven breeder males. The next morning the females were checked for the presence of a vaginal copulation plug as evidence of successful mating. Oviducts were dissected at approximately 22 hours post hCG and cumulus oocyte complexes were released in M2 medium (Millipore MR-015D) and treated with 0.3 mg/ml hyaluronidase (Sigma H4272) as previously described⁵⁰. Fertilized embryos were selected and kept at 37°C, 6% CO₂ in microdrops of KSOM media with amino acids (KSOM^AA, Life Global medium LGGG-050 from Cooper Surgical) covered by embryo tested paraffin oil (Life Global LGPO-500 from Cooper Surgical) prior to electroporation. Embryos were briefly cultured in KSOM^AA and transferred into the oviducts of 0.5 dpc pseudopregnant CD-1(ICR) female recipients shortly after manipulations⁵⁰.

Reagents for Cas9 RNA-guided nuclease mediated mutations

To produce null alleles, single guide RNAs (sgRNAs) were designed to target exons that had previously been deleted or mutated with a truncating mutation in knock-out mouse lines (Table S1). The Adamts20 sgRNA was designed to target nearby the truncating mutation, Adamts20^{bt- Bei1} (c.2860C>T; p.R954X)⁴³. sgRNAs were designed using CRISPOR (http://crispor.tefor.net/)⁵¹ and CHOPCHOP (https://chopchop.cbu.uib.no/)⁵² with off-target PAM set to -NRG. Off target criteria: at least three mismatches to unintended targets, at least one of which was in the 11 bp “seed” region adjacent to the PAM. All sgRNA had a predicted frameshift frequency >67%, assessed using inDelphi²⁶. See Table S7 for sgRNA sequences. Recombinant sgRNAs (Synthego) and Cas9 protein (IDT) were used for electroporation. Mixes consisted of Cas9 protein (6 µM) and sgRNA (9 µM total or 4.5 µM each for 2 sgRNAs) in 1X buffer (100 mM KCl, 20 mM Hepes, pH 7.2-7.4) prior to dilution in an equal volume of OptiMEM (ThermoFisher 31985062) immediately before electroporation.

Electroporation

Electroporation was performed as described^12,14. Briefly, zygotes were treated with Acid Tyrode’s (Sigma T1788) for a few seconds to weaken the zona pellucida, rinsed through Opti-MEM media (Thermo Fisher Scientific 31985062) and placed into 5 µl 1:1 mixture of Cas9 RNP and Opti-MEM between electrodes on 1-mm gap electrode slide BEX CUY21 EDIT II connected to BioRad Gene Pulser XCell electroporator. Twelve square pulses at 30 V with 1 ms pulse duration and 100 ms interval were applied. The zygotes were retrieved from the slide, rinsed with KSOM^AA and kept at 37°C, 6% CO2 until ready for embryo transfers.

Genomic DNA extraction

For initial PCRs, DNA was extracted from tail tips by incubation in alkaline lysis buffer (25 mM NaOH, 0.2 mM EDTA) at 95°C for 30 minutes, followed by neutralisation with an equal volume of 40 mM Tris-HCl. For additional assays, DNA was purified from ∼1 cm tail after termination by digestion in 500 μl lysis buffer (50 mM Tris HCl pH 8.0, 100 mM NaCl, 50 mM EDTA, 0.5% SDS) with 0.2 mg/ml Proteinase K overnight at 55°C. Samples were then treated with 0.1 mg/ml RNase at 37°C for 1-2 hours and gDNA was extracted with 500 μl PCI (phenol:chloroform:isoamyl alcohol, Sigma). gDNA was precipitated from the aqueous phase with 1 volume cold isopropanol, and dissolved in TE. DNA was purified from snap-frozen half spleen and portion of liver tissue by homogenisation in 3 ml lysis buffer (50 mM Tris HCl pH 7.5, 100 mM NaCl, 5 mM EDTA). Proteinase K was added to a final concentration of 0.4 mg/ml and SDS to 1% (w/v). After overnight incubation at 55°C, samples were treated with 0.1 mg/ml RNase at 37°C for 1-2 hours and gDNA was extracted with 3 ml PCI (phenol:chloroform:isoamyl alcohol, Sigma). gDNA was precipitated from the aqueous phase with 2.5 volumes 100% ethanol and 0.1 volumes 3M NaOH, and dissolved in TE.

Genotyping PCRs

Tail tip genomic DNA was PCR amplified for genotyping. From Pax1 founders, gDNA was also derived from spleen and liver. From selected Bmp5 founders, gDNA was also derived from spleen. Quanta polymerase was used for all reactions and amplicon lengths were measured on a QIAxcel Advanced Instrument. Amplicons were Sanger sequenced using a forward nested primer for Tyrp1, Slc45a2, Fgf5, Tgm3, Pax1 and Adamts20 and a reverse nested primer for Bmp5 (a polyT run upstream of the target site prevents the use of a forward primer for Sanger sequencing). See Table S8 for primer sequences and annealing temperatures.

Sanger deconvolution analysis

Sanger deconvolution was performed using DECODR v3.0²⁹ and ICE (Synthego)³⁰. Long in-frame indels were defined as >21 bp (consistent with ICE). Sequence alignments between Sanger chromatograms and the wild-type reference were performed using Snapgene (Version 6.1.2).

qPCR

Quantification of DNA copy number around genotyping primer annealing sites was performed by qPCR using gDNA diluted to 2.5 ng/µl. Each region of interest (ROI) was detected by region-specific primers and normalised to the TRFC locus (see Table S8 for primer sequences). qPCR was performed on a Viia7 instrument (ABI) instrument using SYBR Green PCR Master Mix (Invitrogen). PCR amplification conditions: annealing temperature 58°C, 40 cycles. Relative abundance = E^TFRC^CT^TFRC / E^ROI^CT^ROI. Abundance was normalised to the mean of three wild-type animals.

Deep-amplicon sequencing (DAS)

Genotyping PCR amplicons were sent for Premium PCR sequencing at Plasmidsaurus. The sequencing yield was 2914 – 4201 reads per sample. FASTQ format data were analysed using the CRISPResso2 software⁵³ with the following parameters: Cas9, single end reads, 50% minimum homology. 1501 - 2672 reads per sample aligned and were included in the analysis. For quantification, the editing window was 3 bp either side of the Cas9 cut site. Indels observed in >0.5% of reads were plotted.

Enrichment of Long DNA Fragments using Capture and Linear Amplification (ELF-CLAMP)

This protocol was adapted from Chang et al.³¹ for use with genomic DNA (gDNA). This protocol consists of two selection steps for viewpoints of interest (in-vitro CRISPR/Cas9 cutting and site-specific fusion of a biotinylated T7 promoter) followed by specific enrichment using linear amplification (in-vitro transcription). The resulting RNA is subsequently analysed using direct-RNA sequencing on an Oxford Nanopore Technologies MinION device. The samples were run in three batches.

sgRNAs for in vitro CRISPR/Cas9 cutting of viewpoints in the genomic DNA were produced by in vitro transcription of a dsDNA generated by annealing a ssDNA containing a T7 promoter (TAATACGACTCACTATAGGGAG), the specific sequence of the gRNA and a part of the common gRNA sequence (GTTTTAGAGCTAGAAATA), to a reverse oligo containing only the common sequence (see Table S9 for oligo sequences). This partially annealed template is next made fully double stranded using DNA Polymerase I, Large (Klenow) Fragment (New England Biolab, M0210). In vitro transcription of the resulting template to generate full-length gRNAs was performed by using the T7 RiboMAX Express Large Scale RNA Production System (Promega, P1320) followed by DNase treatment, according to the manufacturer’s instructions. sgRNAs were then isolated by 1:1 phenol-chloroform-IAA extraction followed by isopropanol with sodium acetate precipitation.

CRISPR/Cas9 cutting of viewpoints in genomic DNA was done as follows: 999 ng each gRNA was incubated with 30.5 pmol Alt-R S.p. Cas9 Nuclease V3 (Integrated DNA Technologies) in a total of 2.5 μl NEBuffer 3.1 (New England Biolabs) at room temperature for 10 minutes. 2.5 μg genomic DNA was combined with the sgRNA-Cas9 complexes (RNPs) in a total volume of 25 μl (in nuclease-free water) and incubated overnight at 37°C, followed by enzyme deactivation at 65°C for 30 minutes. Nuclear RNA and gRNAs were removed by adding 125 U RNase If (New England Biolabs, M0243) and incubation at 37°C for 45 minutes, followed by enzyme deactivation at 70°C for 20 minutes. The cut genomic DNA was purified and concentrated by adding 1 volume AMPure XP beads (30 μl) (Beckman Coulter, A63880) and eluted in 23 μl nuclease-free water. To repair single stranded damage, 20 μl of cut genomic DNA was mixed with 2.5 μl NEBNext FFPE Repair Buffer and 0.5 μl NEBNext FFPE Repair Mix (New England Biolabs, M6630), followed by incubation at 20°C for 15 minutes and addition of 3 volumes of AMPure XP beads (70 μl) for purification. DNA was eluted in 30 μl nuclease-free water.

Site-specific fusion of a biotinylated T7 promoter was done by adding probes on one side or both sides of the newly generated cut site. Probes consisted of the following components: a first biotinylated base, a short linker sequence, the recognition site for the SbfI restriction enzyme (CCTGCA^GG), the complete T7 promoter sequence (TAATACGACTCACTATAGGGAG) and a 30 bp sequence that is complementary to the sequence directly bordering the CRISPR/Cas9 cut site (see Table S9 for probe sequences). Biotinylated probes were diluted to 1 μM by mixing 2 μl of probes with nuclease-free water in a final volume of 80 μl. This was added to 28.5 μl of cut and repaired genomic DNA, 10 μl of 5X OneTaq buffer, 1 μl of 10 mM dNTPs and 1 μl of OneTaq Polymerase (New England BioLabs, M0480). To generate (partially) double stranded DNA, the reaction was incubated in a thermal cycler with the following steps: 95°C for 8 minutes, 1°C decrease per 15 sec to 65°C, 68°C for 5 minutes, rapid decrease to 4°C. Nuclease-free water was subsequently added to a final volume of 200 μl. Samples were pooled here for downstream processing: in each experimental batch, all samples with proximal probes were pooled and all samples with distal probes were pooled.

5 μl per probe of Dynabeads MyOne Streptavidin C1 beads (Thermo Fisher Scientific, 65001) were washed according to the manufacturer’s instructions, followed by resuspension in 200 μl of Binding&Wash buffer (10 mM Tris-HCl, pH 7.0, 1mM EDTA, 2M NaCl). The total volume of beads was added to the cut and T7 promoter-fused genomic DNA, followed by incubation at room temperature for 30 minutes in a HulaMixer (Thermo Fisher Scientific). Beads were washed three times in 200 μl of 1X Binding&Wash buffer according to the manufacturer’s instructions, and further washed in 43 μl nuclease-free water with 5 μl CutSmart Buffer at 37°C for 20 minutes. Bound DNA fragments were released by adding 43 μl nuclease-free water, 5 μl CutSmart Buffer, and 20 U SbfI (New England BioLabs, R3642), followed by incubation at 37°C for 20 minutes. The released DNA was purified and concentrated by adding 1 volume AMPure XP beads and eluted in 11 μl nuclease-free water.

10 μl of the eluted T7 promoter-fused DNA was in vitro transcribed using the T7 RiboMAX Express Large Scale RNA Production System (Promega, P1320) according to the manufacturer’s instructions with minor modifications: 10 μl of libraries were added to 15 μl 2x T7 RiboMax buffer, 3 μl T7 Enzyme Mix, 2 μl 5M Betaine (Sigma-Aldrich, B0300) and 1 μl SUPERase RNase Inhibitor (Thermo Fisher Scientific, AM2694) and incubated at 37°C for 60 minutes. In each batch, the proximal and distal samples were now combined. The RNA was poly(A) tailed using Poly(A) Tailing Kit (Thermo Fisher Scientific, AM1350) according to the manufacturer’s instructions and incubated at 37°C for 10 minutes. The poly(A) tailed RNA was purified and concentrated by adding 100 μl of Agencourt RNAClean XP beads (Beckman Coulter, A63987) and eluted in 12 μl nuclease-free water. RNA concentration was determined using the RNA HS Assay on a Qubit Fluorometer (Thermo Fisher Scientific, Q32852).

Nanopore direct-RNA sequencing libraries were prepared using the Direct-RNA Sequencing Kit, version SQK-RNA002 (Oxford Nanopore Technologies) according to the manufacturer’s instructions with minor modifications: both adaptor ligation steps were performed for 15 minutes and the reverse-transcription step was performed at 50°C for 30 minutes with 1 μl of SUPERase RNase Inhibitor supplemented. The final Agencourt RNAClean XP beads purification step was done using 24 μl of beads for a stringent size selection while maintaining enough yield.

Direct-RNA sequencing was done for 72 hours, using FLO-MIN106 R9.4.1 flowcells on a MinION (MK 2.0) sequencing device with the MinKNOW software (Oxford Nanopore Technologies).

Direct-RNA sequencing reads (POD5 files, converted from FAST5) were basecalled with Dorado v0.6.1 using the high-accuracy (“hac”) model without read trimming. FASTQ reads were binned by viewpoint using Cutadapt, retaining only those with the expected viewpoint sequence at the 5’ end. A subsequent filtering step excluded off-target ELF-CLAMP products by requiring the presence of the corresponding downstream gene sequence. Filtered reads were then aligned to the soft-masked Mus musculus reference genome (GRCm39) using BWA-MEM. For visualisation, SAM files were converted to BAM format and loaded into the Integrative Genomics Viewer (IGV). All analysis scripts are available at https://github.com/fbm48/Tillotson, and the basecalled FASTQ files are deposited in the NCBI GEO under accession number GSE289584.

Alignments using SAM files were analysed to determine the sequence of large “unseen” deletions, and to check for the presence of genome damage in regions flanking short in-frame indels (in compound heterozygous founders). Deletions detected by ELF-CLAMP were confirmed by PCR amplification and Sanger sequencing (see Table S8 for primer sequences and annealing temperatures).

Statistical analysis

Percentage viability was compared by Fisher’s exact tests using GraphPad Prism 9. Significance thresholds: not significant (NS) P > 0.05; * P < 0.05; ** P < 0.01; *** P < 0.001.

Acknowledgements

This work was funded by a Sir Henry Wellcome Postdoctoral Fellowship (210913/Z/18/Z) to R.T. and a Canadian Institutes of Health Research Foundation Grant (FDN-15423) to M.J.J.. R.T. is currently a University of Edinburgh Chancellor’s Fellow. L.-H. C. was funded by the Blood and Transplant Research Unit in Precision Cellular Therapeutics. The authors wish to acknowledge Sandra Tondat, Amit Patel and Maribelle Cruz in the TCP Model Production Core for their contribution to the generation of the crispant founders, and members of the Justice laboratory, James Davies, Grzegorz Kudla, and Duncan Sproul for helpful discussions.

Additional information

Author contributions

Conceptualisation, R.T., M.J.J. and L.M.J.N.; methodology, R.T., M.G., L-H.C. and L.G.L.; formal analysis, R.T., F.B.M. and P.G.; investigation, R.T., J.R. L-H.C. and C.T.; writing – original draft, R.T.; writing –review & editing, all of the authors; visualisation, R.T. and J.R.; supervision, M.J.J., and L.M.J.N.; project management, J.R.; funding acquisition, R.T. and M.J.J.

Funding

Wellcome Trust

https://doi.org/10.35802/210913

Canadian Institutes of Health Research (FDN-15423)

University of Edinburgh (Chancellor's Fellowship)

Blood and Transplant Research Unit in Precision Cellular Therapeutics