brief communication published online: 15 June 2014 | doi: 10.1038/nchembio.1550
DNA sequencing and CRISPR-Cas9 gene editing for target validation in mammalian cells Transfection of HCT116 with Cas9 constructs that targeted HPRT1 (Fig. 1a and Supplementary Table 3 and Supplementary Fig. 3) markedly increased resistance of cells to 6-TG (Fig. 1b and Supplementary Table 4). Resequencing the targeted HPRT1 loci confirmed that the Cas9 nuclease introduces small, mostly frameshift indels and that these indels are enriched after selection with 6-TG (Fig. 1c and Supplementary Table 5 and Supplementary Fig. 4). Remarkably, almost no wild-type (WT) allele was left after drug selection, demonstrating that HPRT1 loss of function causes 6-TG resistance. The experiment confirmed the well-documented importance of HPRT1 for 6-TG–induced cell death 7 and allowed us to establish the methodology. We then applied the similar resistant screen strategy to a second example, the natural product triptolide (Fig. 2a and Supplementary Fig. 5). Triptolide is a potent antiproliferative agent, and its derivative has a therapeutic window in mouse models of pancreatic cancer9. However, triptolide’s mechanism of action is not clear. Biochemical and fractionation approaches identified several proteins that could bind triptolide in vitro, among them ERCC3 (also known as XPB), a protein necessary for RNA polymerase II function10, as well as dCTP pyrophosphatase11, polycystin 2 (ref. 12) and
a
D138fs C66fs Q144*
HPRT1 1
P38 L79 CRISPR cutting sites
c
b
50
0
0
31 100 3.1 10 [6-thioguanine] (µM)
Nontargeting Cas9 HPRT1-P38
218
50
0 Cas9 gRNA 6-TG
+ – –
+ + –
No indel HPRT1-L79
L79
P38 100
100 % reads
Identification and validation of drug-resistant mutations can provide important insights into the mechanism of action of a compound. Here we demonstrate the feasibility of such an approach in mammalian cells using next-generation sequencing of drug-resistant clones and CRISPR-Cas9–mediated gene editing on two drug-target pairs, 6-thioguanine–HPRT1 and triptolide-ERCC3. We showed that disrupting functional HPRT1 allele or introducing ERCC3 point mutations by gene editing can confer drug resistance in cells. Finding biological targets of cellularly active small molecules is important for elucidating previously uncharacterized biology and nominating new targets for drug discovery. Once a small molecule’s target has been identified, orthogonal validation becomes a necessary next step. Among existing target identification and validation approaches, identification of compound-resistant mutations has been long held as the ‘gold standard’. This genetic approach has been extensively used in lower organisms such as bacteria and yeast, resulting in discoveries of important drug-target pairs such as rapamycin-TOR1 (ref. 1). With recent advancements in next-generation sequencing technology, similar approaches have also been successfully implemented in mammalian cells2. One drawback of applying a forward genetic screen to mammalian cells, rather than model systems such as yeast, is the lack of highly efficient molecular biology tools to validate the observed mutations that confer resistance to the drug. In this work, we applied the recently discovered CRISPR-Cas9 genome editing approach3–6 to validate the results of forward genetic screens by either knocking out a gene or knocking in drug-resistant alleles and looking for rescue of drug sensitivity. We found this method to be highly efficient, and, more notably, it can be applied to validate both dominant and recessive drug-resistant alleles. As the initial proof of concept, we carried out a forward genetic screen in the KBM7 cell line to isolate mutations that could confer resistance to 6-thioguanine (6-TG; Supplementary Results, Supplementary Figs. 1 and 2). KBM7 is a nearly haploid chronic myelogenous leukemia cell line that has only one copy of all chromosomes except chromosome 8. Hence, it can be used to identify not only dominant but also recessive gain- or loss-of-function mutations in a genetic screen7. 6-TG is an anticancer drug that is converted into a toxic form, 6-thioguanosine-monophosphate, by hypoxanthine phosphoribosyltransferase (HPRT1) 8. Exome and amplicon sequencing demonstrated that all of the isolated 6-TG– resistant clones carried nonsense mutations in the single HPRT1 allele (Fig. 1a and Supplementary Tables 1 and 2). To validate these nonsense mutations, we used CRISPR constructs that cut at two different sites of HPRT1 (Fig. 1a) to inactivate the gene. Cas9 is a nuclease that cuts genomic DNA at a locus determined by a sequence of small guide RNA co-expressed with the enzyme 3–6. Repair of the cutting site tends to be dominated by the nonhomologous end-joining (NHEJ) mechanism, which is error prone and introduces frameshift indels, resulting in a ‘knockout’ inactivation of the gene3–6.
Normalized viability (%)
© 2014 Nature America, Inc. All rights reserved.
Yegor Smurnyy, Mi Cai, Hua Wu, Elizabeth McWhinnie, John A Tallarico, Yi Yang & Yan Feng*
+ – –
+ + +
+ + –
+ + +
Frameshift indel
Nonframeshift indel
Figure 1 | HPRT1 loss-of-function confers resistance to 6-thioguanine, as demonstrated by a forward genetic screen and CRISPR-mediated knockout. (a) Location of identified HPRT1 nonsense mutations and the designed Cas9 cutting sites. (b) Cellular viability of control and HPRT1-knockout cells after treatment with 6-thioguanine. Average of two replicates, the error bars are the s.d. (c) Abundance of different alleles in control sample and CRISPR-targeted samples before and after 6-TG selection based on the next generation amplicon sequencing.
Developmental and Molecular Pathways, Novartis Institutes for Biomedical Research, Cambridge, Massachusetts, USA. *e-mail: [email protected] nature CHEMICAL BIOLOGY | Advance online publication | www.nature.com/naturechemicalbiology
1
brief communication a
b
R337L L440P
H
GTF2H4 1 135 ERCC3 binding
Nonframeshift indel
304 381 462 ERCC3 binding
% reads
Frameshift indel 50
Missense only Missense + silent
IC50 shift 5- to 50-fold IC50 shift >50-fold
0 Cas9 gRNA Repair template Triptolide
e
Selection concentration
Other + – – –
+ + + –
+ + + +
+ – – –
+ + + –
+ + + +
800
WT HCT116
100
600
F482S pool 50
IC50 (nM)
D54H pool
400
Genetic background of the clone
WT/WT
D54H/E48fs
T 82 S L4 83 * F4
NT
W
5* E5
0 Overexpressed ERCC3 allele
1,000
T 4H
10 100 [Triptolide] (nM)
D5
1
W
0
NT
200
NT W T D5 4H F4 82 S
© 2014 Nature America, Inc. All rights reserved.
Normalized viability (%)
d
WT
DNA binding
208 327 488 542 702 782 1 44 GTF2H4 binding Helicase Helicase
OH
O O
F482S
D54H 100
ERCC3
O O
c
S40W S162I E294D D475E F482S D54H Y163C I156F S179I K487T ATPase
O
Nature chemical biology doi: 10.1038/nchembio.1550
F482S/V478fs
Figure 2 | Identification and validation of recessive triptolide-resistant ERCC3 alleles using CRISPR-mediated knock-in strategy. (a) Chemical structure of triptolide. (b) Location of resistant mutations in the ERCC3 and GTF2H4 sequences. Variants conferring more than 50-fold resistance are marked in red. Domains necessary for ERCC3 and GTF2H4 interaction were previously identified by co-immunoprecipitation15. ERCC3 helicase domains and residues necessary for ATPase activity and DNA binding are annotated on the basis of the information from the UniProt database. (c) Abundance of different alleles in control sample and CRISPR-targeted samples before and after drug selection based on next-generation amplicon sequencing. (d) Triptolide sensitivity of WT HCT116 and D54 or F482 edited pools. Data show the average of two replicates, and error bars show the s.d. (e) IC50 measurements in either WT HCT116 or single-cell clones with knock-in of mutant alleles. A plasmid carrying mCherry-ERCC3 (either WT or mutants, as shown in the figure) was transfected into cells, and resistance to triptolide was measured. NT, not transfected. Data show the average of two replicates, and error bars show the s.d.
TAB1 (ref. 13). At the same time, none of these proteins has been conclusively demonstrated to be the primary cytotoxicity target in the cellular context. Among the total of 26 triptolide-resistant KBM7 clones we isolated, amplicon sequencing identified ERCC3 mutations in 15 (58%) of the clones (Supplementary Tables 6–8). In addition, six (23%) resistant clones were WT ERCC3 but carried either one of two mutations in GTF2H4, a binding partner of ERCC3 that stimulates its ATPase activity14 (Supplementary Tables 6 and 9). Most of the mutations were located either in domains of ERCC3 and GTF2H4 responsible for the formation of the binary complex15 or in the DNA-binding domain of ERCC3 (ref. 14) (Fig. 2b). Among all of the surveyed genes, ERCC3 had the highest number of mutations, and no mutations were identified in other previously nominated triptolide targets, such as DCTPP1 (ref. 11), PC2 (ref. 12), TAB1 (ref. 13) and ADAM10 (ref. 16) (Supplementary Table 9, Supplementary Fig. 6 and Supplementary Data Set 1). Although identifying multiple mutations in the same gene or protein complex is a promising initial result, the ultimate proof of the importance of a variant is the independent validation experiment in which the desired difference in phenotypes was demonstrated in a WT-mutant pair of otherwise isogenic cell lines. To our surprise, overexpression of six potent resistant ERCC3 alleles in 293FT induced less than a 2-fold shift in half-maximum inhibitory concentration (IC50) (Supplementary Fig. 7) compared to the 10- to 20-fold increases seen in the initial screen (Supplementary Table 6). We reasoned that these ERCC3 mutants could be recessive because the use of near-haploid KBM7 cells allowed us to isolate both dominant and recessive mutations. 2
Thus, gene editing–mediated ‘knock-in’ strategies become an attractive validation approach because they allow creation of both hetero- and homozygous mutant models expressing the variants from the native locus17,18. For the proof-of-concept knock-in experiment, we selected D54H and F482S mutant alleles of ERCC3 because they resulted in very high IC50 shifts and spanned different domains of the protein (Fig. 2b and Supplementary Table 6). We selected two separate Cas9 guide RNAs (gRNAs) that targeted GN19NGG motifs around D54H and F482S single nucleotide polymorphisms (SNPs). Cells were transfected with a plasmid expressing Cas9 together with a gRNA (Supplementary Table 10) and a synthetic 120-nucleotide (nt) oligonucleotide (Supplementary Table 11 and Supplementary Figs. 8 and 9), carrying the respective missense mutation and an additional silent mutation that served (i) as a marker for homologous recombination products as opposed to de novo missense mutations and (ii) as an additional mismatch between the gRNA sequence and the knock-in product, thereby preventing recutting of the site by Cas9. After Cas9 cuts, cells can use a combination of NHEJ or homologous recombination (HR) events to repair the Cas9-induced cut, resulting in various knockout or knock-in products (Supplementary Fig. 8). To enrich for knock-in products, we used a short-term selection with an IC90 dose of triptolide to kill sensitive cells. The relative number of different alleles was initially determined by amplicon sequencing of pooled samples (Fig. 2c and Supplementary Table 12 and Supplementary Fig. 4). In addition, we used fluorescence-activated cell sorting (FACS) to sort drug-selected pools into single-cell-derived clones and genotyped these clones using TA cloning and Sanger sequencing (Supplementary Table 13 and Supplementary Fig. 8).
nature chemical biology | Advance online publication | www.nature.com/naturechemicalbiology
© 2014 Nature America, Inc. All rights reserved.
Nature chemical biology doi: 10.1038/nchembio.1550 Analysis of amplicon sequencing showed that before drug selection, the desired point mutations were present in 5.2% of alleles in the D54H locus and 0.3% of alleles in the F482 locus. Killing sensitive cells with the short drug treatment greatly enriched both mutant alleles to more than 35% in the resistant pools (Fig. 2c). These enriched pools showed large IC50 shifts similar to those of the original isolated mutant clones (Fig. 2d). Silent co-mutations in the repair oligonucleotides were introduced into >99% of D54H mutant alleles and ~3% of the F482S allele (Fig. 2c). Carryover of these silent SNPs indicates that the alleles are the product of HR and not de novo mutagenesis. The lower rate of coappearance of silent SNPs in F482S is presumably due to the larger distance between the two SNPs in the oligonucleotide and is rather common to see with single-stranded oligonucleotide donors19. There were no spontaneous mutations in the rest of the ERCC3 coding sequence, further confirming that the resistance is resulting from the knock-in mutant alleles. Upon analysis of the TA cloning results, several additional findings became apparent. First, most of the resistant clones harbored one knock-in allele from HR and one knockout allele with frameshift mutation resulting from NHEJ. In addition, we identified that, in the case of D54 locus, some of the drug-resistant clones in fact carried small non-frameshift deletions, resulting in the removal of 6–12 amino acids around Asp54 (Supplementary Fig. 8 and Supplementary Table 13). Notably, no WT allele was found in the genotyped clones, in agreement with the very low WT allele frequency (
DNA sequencing and CRISPR-Cas9 gene editing for target validation in mammalian cells Transfection of HCT116 with Cas9 constructs that targeted HPRT1 (Fig. 1a and Supplementary Table 3 and Supplementary Fig. 3) markedly increased resistance of cells to 6-TG (Fig. 1b and Supplementary Table 4). Resequencing the targeted HPRT1 loci confirmed that the Cas9 nuclease introduces small, mostly frameshift indels and that these indels are enriched after selection with 6-TG (Fig. 1c and Supplementary Table 5 and Supplementary Fig. 4). Remarkably, almost no wild-type (WT) allele was left after drug selection, demonstrating that HPRT1 loss of function causes 6-TG resistance. The experiment confirmed the well-documented importance of HPRT1 for 6-TG–induced cell death 7 and allowed us to establish the methodology. We then applied the similar resistant screen strategy to a second example, the natural product triptolide (Fig. 2a and Supplementary Fig. 5). Triptolide is a potent antiproliferative agent, and its derivative has a therapeutic window in mouse models of pancreatic cancer9. However, triptolide’s mechanism of action is not clear. Biochemical and fractionation approaches identified several proteins that could bind triptolide in vitro, among them ERCC3 (also known as XPB), a protein necessary for RNA polymerase II function10, as well as dCTP pyrophosphatase11, polycystin 2 (ref. 12) and
a
D138fs C66fs Q144*
HPRT1 1
P38 L79 CRISPR cutting sites
c
b
50
0
0
31 100 3.1 10 [6-thioguanine] (µM)
Nontargeting Cas9 HPRT1-P38
218
50
0 Cas9 gRNA 6-TG
+ – –
+ + –
No indel HPRT1-L79
L79
P38 100
100 % reads
Identification and validation of drug-resistant mutations can provide important insights into the mechanism of action of a compound. Here we demonstrate the feasibility of such an approach in mammalian cells using next-generation sequencing of drug-resistant clones and CRISPR-Cas9–mediated gene editing on two drug-target pairs, 6-thioguanine–HPRT1 and triptolide-ERCC3. We showed that disrupting functional HPRT1 allele or introducing ERCC3 point mutations by gene editing can confer drug resistance in cells. Finding biological targets of cellularly active small molecules is important for elucidating previously uncharacterized biology and nominating new targets for drug discovery. Once a small molecule’s target has been identified, orthogonal validation becomes a necessary next step. Among existing target identification and validation approaches, identification of compound-resistant mutations has been long held as the ‘gold standard’. This genetic approach has been extensively used in lower organisms such as bacteria and yeast, resulting in discoveries of important drug-target pairs such as rapamycin-TOR1 (ref. 1). With recent advancements in next-generation sequencing technology, similar approaches have also been successfully implemented in mammalian cells2. One drawback of applying a forward genetic screen to mammalian cells, rather than model systems such as yeast, is the lack of highly efficient molecular biology tools to validate the observed mutations that confer resistance to the drug. In this work, we applied the recently discovered CRISPR-Cas9 genome editing approach3–6 to validate the results of forward genetic screens by either knocking out a gene or knocking in drug-resistant alleles and looking for rescue of drug sensitivity. We found this method to be highly efficient, and, more notably, it can be applied to validate both dominant and recessive drug-resistant alleles. As the initial proof of concept, we carried out a forward genetic screen in the KBM7 cell line to isolate mutations that could confer resistance to 6-thioguanine (6-TG; Supplementary Results, Supplementary Figs. 1 and 2). KBM7 is a nearly haploid chronic myelogenous leukemia cell line that has only one copy of all chromosomes except chromosome 8. Hence, it can be used to identify not only dominant but also recessive gain- or loss-of-function mutations in a genetic screen7. 6-TG is an anticancer drug that is converted into a toxic form, 6-thioguanosine-monophosphate, by hypoxanthine phosphoribosyltransferase (HPRT1) 8. Exome and amplicon sequencing demonstrated that all of the isolated 6-TG– resistant clones carried nonsense mutations in the single HPRT1 allele (Fig. 1a and Supplementary Tables 1 and 2). To validate these nonsense mutations, we used CRISPR constructs that cut at two different sites of HPRT1 (Fig. 1a) to inactivate the gene. Cas9 is a nuclease that cuts genomic DNA at a locus determined by a sequence of small guide RNA co-expressed with the enzyme 3–6. Repair of the cutting site tends to be dominated by the nonhomologous end-joining (NHEJ) mechanism, which is error prone and introduces frameshift indels, resulting in a ‘knockout’ inactivation of the gene3–6.
Normalized viability (%)
© 2014 Nature America, Inc. All rights reserved.
Yegor Smurnyy, Mi Cai, Hua Wu, Elizabeth McWhinnie, John A Tallarico, Yi Yang & Yan Feng*
+ – –
+ + +
+ + –
+ + +
Frameshift indel
Nonframeshift indel
Figure 1 | HPRT1 loss-of-function confers resistance to 6-thioguanine, as demonstrated by a forward genetic screen and CRISPR-mediated knockout. (a) Location of identified HPRT1 nonsense mutations and the designed Cas9 cutting sites. (b) Cellular viability of control and HPRT1-knockout cells after treatment with 6-thioguanine. Average of two replicates, the error bars are the s.d. (c) Abundance of different alleles in control sample and CRISPR-targeted samples before and after 6-TG selection based on the next generation amplicon sequencing.
Developmental and Molecular Pathways, Novartis Institutes for Biomedical Research, Cambridge, Massachusetts, USA. *e-mail: [email protected] nature CHEMICAL BIOLOGY | Advance online publication | www.nature.com/naturechemicalbiology
1
brief communication a
b
R337L L440P
H
GTF2H4 1 135 ERCC3 binding
Nonframeshift indel
304 381 462 ERCC3 binding
% reads
Frameshift indel 50
Missense only Missense + silent
IC50 shift 5- to 50-fold IC50 shift >50-fold
0 Cas9 gRNA Repair template Triptolide
e
Selection concentration
Other + – – –
+ + + –
+ + + +
+ – – –
+ + + –
+ + + +
800
WT HCT116
100
600
F482S pool 50
IC50 (nM)
D54H pool
400
Genetic background of the clone
WT/WT
D54H/E48fs
T 82 S L4 83 * F4
NT
W
5* E5
0 Overexpressed ERCC3 allele
1,000
T 4H
10 100 [Triptolide] (nM)
D5
1
W
0
NT
200
NT W T D5 4H F4 82 S
© 2014 Nature America, Inc. All rights reserved.
Normalized viability (%)
d
WT
DNA binding
208 327 488 542 702 782 1 44 GTF2H4 binding Helicase Helicase
OH
O O
F482S
D54H 100
ERCC3
O O
c
S40W S162I E294D D475E F482S D54H Y163C I156F S179I K487T ATPase
O
Nature chemical biology doi: 10.1038/nchembio.1550
F482S/V478fs
Figure 2 | Identification and validation of recessive triptolide-resistant ERCC3 alleles using CRISPR-mediated knock-in strategy. (a) Chemical structure of triptolide. (b) Location of resistant mutations in the ERCC3 and GTF2H4 sequences. Variants conferring more than 50-fold resistance are marked in red. Domains necessary for ERCC3 and GTF2H4 interaction were previously identified by co-immunoprecipitation15. ERCC3 helicase domains and residues necessary for ATPase activity and DNA binding are annotated on the basis of the information from the UniProt database. (c) Abundance of different alleles in control sample and CRISPR-targeted samples before and after drug selection based on next-generation amplicon sequencing. (d) Triptolide sensitivity of WT HCT116 and D54 or F482 edited pools. Data show the average of two replicates, and error bars show the s.d. (e) IC50 measurements in either WT HCT116 or single-cell clones with knock-in of mutant alleles. A plasmid carrying mCherry-ERCC3 (either WT or mutants, as shown in the figure) was transfected into cells, and resistance to triptolide was measured. NT, not transfected. Data show the average of two replicates, and error bars show the s.d.
TAB1 (ref. 13). At the same time, none of these proteins has been conclusively demonstrated to be the primary cytotoxicity target in the cellular context. Among the total of 26 triptolide-resistant KBM7 clones we isolated, amplicon sequencing identified ERCC3 mutations in 15 (58%) of the clones (Supplementary Tables 6–8). In addition, six (23%) resistant clones were WT ERCC3 but carried either one of two mutations in GTF2H4, a binding partner of ERCC3 that stimulates its ATPase activity14 (Supplementary Tables 6 and 9). Most of the mutations were located either in domains of ERCC3 and GTF2H4 responsible for the formation of the binary complex15 or in the DNA-binding domain of ERCC3 (ref. 14) (Fig. 2b). Among all of the surveyed genes, ERCC3 had the highest number of mutations, and no mutations were identified in other previously nominated triptolide targets, such as DCTPP1 (ref. 11), PC2 (ref. 12), TAB1 (ref. 13) and ADAM10 (ref. 16) (Supplementary Table 9, Supplementary Fig. 6 and Supplementary Data Set 1). Although identifying multiple mutations in the same gene or protein complex is a promising initial result, the ultimate proof of the importance of a variant is the independent validation experiment in which the desired difference in phenotypes was demonstrated in a WT-mutant pair of otherwise isogenic cell lines. To our surprise, overexpression of six potent resistant ERCC3 alleles in 293FT induced less than a 2-fold shift in half-maximum inhibitory concentration (IC50) (Supplementary Fig. 7) compared to the 10- to 20-fold increases seen in the initial screen (Supplementary Table 6). We reasoned that these ERCC3 mutants could be recessive because the use of near-haploid KBM7 cells allowed us to isolate both dominant and recessive mutations. 2
Thus, gene editing–mediated ‘knock-in’ strategies become an attractive validation approach because they allow creation of both hetero- and homozygous mutant models expressing the variants from the native locus17,18. For the proof-of-concept knock-in experiment, we selected D54H and F482S mutant alleles of ERCC3 because they resulted in very high IC50 shifts and spanned different domains of the protein (Fig. 2b and Supplementary Table 6). We selected two separate Cas9 guide RNAs (gRNAs) that targeted GN19NGG motifs around D54H and F482S single nucleotide polymorphisms (SNPs). Cells were transfected with a plasmid expressing Cas9 together with a gRNA (Supplementary Table 10) and a synthetic 120-nucleotide (nt) oligonucleotide (Supplementary Table 11 and Supplementary Figs. 8 and 9), carrying the respective missense mutation and an additional silent mutation that served (i) as a marker for homologous recombination products as opposed to de novo missense mutations and (ii) as an additional mismatch between the gRNA sequence and the knock-in product, thereby preventing recutting of the site by Cas9. After Cas9 cuts, cells can use a combination of NHEJ or homologous recombination (HR) events to repair the Cas9-induced cut, resulting in various knockout or knock-in products (Supplementary Fig. 8). To enrich for knock-in products, we used a short-term selection with an IC90 dose of triptolide to kill sensitive cells. The relative number of different alleles was initially determined by amplicon sequencing of pooled samples (Fig. 2c and Supplementary Table 12 and Supplementary Fig. 4). In addition, we used fluorescence-activated cell sorting (FACS) to sort drug-selected pools into single-cell-derived clones and genotyped these clones using TA cloning and Sanger sequencing (Supplementary Table 13 and Supplementary Fig. 8).
nature chemical biology | Advance online publication | www.nature.com/naturechemicalbiology
© 2014 Nature America, Inc. All rights reserved.
Nature chemical biology doi: 10.1038/nchembio.1550 Analysis of amplicon sequencing showed that before drug selection, the desired point mutations were present in 5.2% of alleles in the D54H locus and 0.3% of alleles in the F482 locus. Killing sensitive cells with the short drug treatment greatly enriched both mutant alleles to more than 35% in the resistant pools (Fig. 2c). These enriched pools showed large IC50 shifts similar to those of the original isolated mutant clones (Fig. 2d). Silent co-mutations in the repair oligonucleotides were introduced into >99% of D54H mutant alleles and ~3% of the F482S allele (Fig. 2c). Carryover of these silent SNPs indicates that the alleles are the product of HR and not de novo mutagenesis. The lower rate of coappearance of silent SNPs in F482S is presumably due to the larger distance between the two SNPs in the oligonucleotide and is rather common to see with single-stranded oligonucleotide donors19. There were no spontaneous mutations in the rest of the ERCC3 coding sequence, further confirming that the resistance is resulting from the knock-in mutant alleles. Upon analysis of the TA cloning results, several additional findings became apparent. First, most of the resistant clones harbored one knock-in allele from HR and one knockout allele with frameshift mutation resulting from NHEJ. In addition, we identified that, in the case of D54 locus, some of the drug-resistant clones in fact carried small non-frameshift deletions, resulting in the removal of 6–12 amino acids around Asp54 (Supplementary Fig. 8 and Supplementary Table 13). Notably, no WT allele was found in the genotyped clones, in agreement with the very low WT allele frequency (
