KRAS is the most frequently mutated oncogene in human cancer, and its activating mutations represent long-sought therapeutic targets. Programmable nucleases, particularly the CRISPR-Cas9 system, provide an attractive tool for genetically targeting KRAS mutations in cancer cells. Here, we show that cleavage of a panel of KRAS driver mutations suppresses growth in various human cancer cell lines, revealing their dependence on mutant KRAS. However, analysis of the remaining cell population after long-term Cas9 expression unmasked the occurence of oncogenic KRAS escape variants that were resistant to Cas9-cleavage. In contrast, the use of an adenine base editor to correct oncogenic KRAS mutations progressively depleted the targeted cells without the appearance of escape variants and allowed efficient and simultaneous correction of a cancer-associated TP53 mutation. Oncogenic KRAS and TP53 base editing was possible in patient-derived cancer organoids, suggesting that base editor approaches to correct oncogenic mutations could be developed for functional interrogation of vulnerabilities in a personalized manner for future precision oncology applications.
Repairing KRAS mutations with base editors can be used for providing a better understanding of RAS biology and may lay the foundation for improved treatments for KRAS-mutant cancers.
The RAS gene family member KRAS is the most frequently mutated oncogene in human cancer, with approximately 14% of patients harboring KRAS mutations, equivalent to approximately 2.5 million new cases per year worldwide (1, 2). Activating mutations in KRAS lead to a constitutively active RAS/RAF signaling pathway capable of inducing cell transformation by forcing cells into continuous proliferation, making the mutation an important therapeutic target (3). However, despite more than three decades of intense research efforts, targeting KRAS has proven challenging owing to its small size (∼21 kDa, 189 amino acids) and a “smooth” surface that impedes binding by small molecules (2). Moreover, KRAS exhibits exceptionally high affinity to its endogenous target, GTP, leading to an extremely strong binding that prevents direct targeting of the nucleotide-binding pocket (4). Furthermore, the high intracellular GTP concentrations (>500 μmol/L) hinders competition (5). Therefore, targeting strategies have shifted away from a pan-KRAS inhibitor to an approach supporting mutation-selective approaches. Renewed hope has been brought about by the recent development of KRASG12C-specific inhibitors (6), which has resulted in the approval of the first KRASG12C-inhibitor for clinical use (7). However, other KRAS mutations, which are more common, such as KRASG12D or KRASG13D remain “undruggable” to date.
Functional genomic studies examining the expression of gene products in cancer cell lines have revealed that KRAS-mutant cancer cells depend on KRAS function for growth and survival (8). Moreover, siRNAs that selectively inhibit mutant KRAS mRNAs (9) led to impeded growth of cancer cells both in vitro and in vivo (10). However, continuous expression/delivery of the silencing trigger is required to maintain target RNA suppression, making this approach difficult to apply in a clinical setting. Programmable nucleases have recently been developed to selectively cleave oncogenic mutations in cancer cells. They offer the advantage of working at the genomic level and achieving permanent gene disruption (11). However, detailed long-term studies following cleavage of oncogenic mutations have not yet been reported. Furthermore, this approach is not applicable to mutations occurring in tumor suppressor genes. The recently developed CRISPR adenine base editing system is independent of DNA double-strand break repair and has recently been applied to repair a cancer-associated mutation in the TERT promoter (12). However, to our knowledge, it has not been tested for the correction of an activated oncogene or a mutated tumor suppressor gene to date. To examine efficiency, specificity, and durability of these approaches, we harnessed CRISPR-Cas9 systems to selectively cleave or correct different oncogenic KRAS variants and TP53 hotspot mutations in a panel of cancer cell lines and patient-derived organoids (PDO) and analyzed the cells in culture over an extended period of time.
Materials and Methods
Since we implement a mutation-selective approach targeting point mutations, sgRNAs were manually modified to match the mutant sequence in question and manually assessed (e.g., using IDT CRISPR-Cas9 guide RNA design checker https://eu.idtdna.com/site/order/designtool/index/CRISPR_SEQUENCE). The designed sgRNAs differ in only one nucleotide from the wild-type allele sequence, making the wild-type allele the most likely off-target site. Therefore, for each experiment, mutation-selective sgRNAs were delivered to cells that do not carry the mutation to control for potential off-target toxicities.
For expression of Cas9 and sgRNAs from lentiviral vectors, we optimized the sequence of pL-CRISPR.EFS.GFP (a gift from Benjamin Ebert Addgene plasmid #57818; http://n2t.net/addgene:57818; RRID:Addgene_57818; ref. 13), where the tracer sequence was modified to increase sgRNA stability and enhance its assembly with Cas9 protein (14). Streptococcus pyogenes Cas9 and GFP were linked via P2A and were expressed from the EFS promoter, whereas the sgRNAs were expressed from the human U6 pol III promoter. Protospacers targeting mutations were cloned into pL-CRISPR.EFS.GFP by cloning complementary oligonucleotides into the vector. Unless a guanine was the first base in the protospacer, a guanine was added to the 5′ end of the protospacer before cloning to boost the expression of the gRNA from the human U6 promoter (15). Likewise, for ABE8e timecourses, protospacers were cloned into the LRT2B vector expressing TdTomato (a gift from Lukas Dow Addgene plasmid #110854; http://n2t.net/addgene:110854; RRID:Addgene_110854; ref. 16), using BsmBI/BbsI sites following the standard protocol (Supplementary Table S1; ref. 15). Briefly, oligos for gRNA were phosphorylated and annealed in a 10 μL reaction containing 100 pmol of each gRNA oligo (top and bottom oligos), 1 μL 10× T4 ligation buffer and 0.5 μL T4 polynucleotide kinase. The reaction was run in thermocycler at 37°C for 30 minutes, 95°C for 5 minutes, and then ramped down to 25°C at rate of 0.1°C/min. The reaction was then diluted 250 times in water and 2 μL was used for the restriction/ligation golden gate protocol. In this reaction, 50 ng backbone plasmid was added to the diluted annealed oligos together with 1 μL Tango buffer (Thermo Fisher Scientific), 0.5 μL 10 mmol/L DTT, 1 μL 10 mmol/L ATP, 0.5 μL BsmBI/BbsI, and 0.25 μL T4 DNA ligase in 10 μL final reaction. To digest the vector and ligate oligos into it, the following temperature profile was repeated six times in a thermocycler: 37°C for 5 minutes, 23°C for 5 minutes followed by incubation at 37°C for 15 minutes, and finally at 80°C for 5 minutes. Next, 2 μL of this final reaction was used to transform DH5α Escherichia coli cells. A single colony was picked, grown in liquid LB-antibiotic media before plasmid DNA was purified (Thermo Fisher GeneJet DNA Miniprep Kit). Plasmids were sequenced with a forward primer to the U6 promoter: 5′-GAGGGCCTATTTCCCATGATTCC-3′. For the base editor cloning, we used the plasmid backbone pLenti-FNLS-P2A-GFP-PGK-Puro (a gift from Lukas Dow Addgene plasmid #110869; http://n2t.net/addgene:110869; RRID:Addgene_110869; ref. 16), which is a lentiviral vector bearing a codon optimized CBE (BE3) in combination with GFP and the puromycin resistance gene. The CBE in the backbone was replaced with NG-ABE8e (a gift from David Liu; Addgene plasmid #138491; http://n2t.net/addgene:138491; RRID:Addgene_138491) through intermediary Acc65I and BstZ17I restriction sites. The ligation product was transformed into bacteria for overnight incubation. Single clones were picked, grown, and miniprepped followed by sequencing employing several primers aligning the full sequence of the base editor in addition to diagnostic test digests confirming the correct integration into the backbone. Finally, Plasmid DNA were transformed into E. coli DH5α and cells were grown overnight at 37°C with constant shaking. Plasmid DNA was purified using QIAGEN-tip 20 Maxiprep Kit (Qiagen 10023), according to the manufacturer's protocol. The DNA was resuspended at 1 μg/μL, validated by Sanger sequencing and then used for lentivirus production.
PANC-1 (CRL-1469), RKO (CRL-2577), HCT116 (CCL-247), A549 (CCL-185), and HEK293T (CRL-11268) cells were purchased from ATCC and maintained in DMEM (high glucose, GlutaMAX, pyruvate; Gibco) supplemented with 10% (v/v) FBS (Gibco) and antibiotics (100 U/mL penicillin, 100 μg/mL streptomycin; Gibco) and kept at 37°C, 5% CO2. Cell lines were routinely tested and confirmed to be Mycoplasma-free (latest on June 1, 2022). For all cell lines used, cells were allowed to recover after thawing for two passages before performing experiments.
TA cloning was performed using the TA Cloning Kit with pCR2.1 vector (Thermo Fisher Scientific). The amplicons were separated using agarose gel electrophoresis and the target bands were extracted and purified using the Gel extraction Purification Kit (Bioline). The DNA concentration was measured with a Nanodrop spectrophotometer. For TA cloning ligation, 0.1 to 0.3 pmol of DNA fragments were mixed with pCR2.1 vector according to manufacturer's protocol, incubated at 16°C for 1 hour. The ligation product was transformed into competent E. coli DH5α employing blue/white screening. White colonies were picked the next day and plasmid DNA was extracted via Mini-prep. Finally, PCR reactions using M13 forward and reverse primers were performed, followed by Sanger sequencing.
Lentivirus production and transduction
Lentiviral particle production and infection were performed as described previously (17). Briefly, seven million HEK293T cells were seeded in 10 cm dishes and transfected on the next day at ∼80% confluency with 2 μg pMD2.G (a gift from Didier Trono Addgene plasmid #12259; http://n2t.net/addgene:12259; RRID:Addgene_12259), 6 μg psPAX2 (a gift from Didier Trono Addgene plasmid #12260; http://n2t.net/addgene:12260; RRID:Addgene_12260) and 10 μg of the transfer vector (e.g., pL-CRISPR.EFS.GFP.gRNA) using 45 μg PEI (1 mg/mL) per dish. Plasmid maxiprep kits (Qiagen) were used to provide transfection-level DNA. After ∼20 hours of transfection, the medium was changed to complete DMEM and 72 hours after transfection the viral supernatant was collected, filtered through a 0.45 μmol/L filter and centrifuged for 2 hours at 100,000 × g at 4°C. The supernatant was decanted and the viral pellets were resuspended in PBS overnight at 4°C on a shaker. For long-term storage, the virus particles were kept in cryovials at −80°C. Where indicated, the viral supernatant was concentrated using Amicon Ultra-15 Centrifugal Filter Devices (Merck) according to the manufacturer's instructions. When a new cell line was used for the first time, the amount of virus needed to infect 50% of the cells was determined by titration. Transductions were typically performed in 96-well plates in the presence of protamine sulfate (final concentration 5 μg/mL; Sigma-Aldrich) and spin-infected for 1 hour at 1,000 × g at 37°C.
RKO, PANC-1, A549, and HCT116 were typically transduced in 96-well plates and the percentage of infected cells was analyzed measuring GFP/TdTomato expression using a MACSQuant VYB Analyzer (Miltenyi Biotec). At 72 hours after transduction, cells were trypsinized and collected for flow cytometry analysis. Viable single cells were gated using the forward and side scatter followed by doublets exclusion. The BFP fluorescence signal was measured using a violet 405 nm laser, the GFP fluorescence signal was measured using a blue 488 nm laser, whereas the TdTomato was assayed using a 561 nm yellow laser and the log area of the signal was collected. Cell sorting was carried out using a BD FACSAria III Cell Sorter (BD Biosciences). For time course experiments, adherent cells were processed every 48 to 72 hours by flow cytometry in technical triplicates. Briefly, for each time point, medium was aspirated completely using a multichannel pipette. Then, cells were carefully washed with 200 μL sterile PBS and treated with 30 μL/well trypsin, enough to cover the adherent layer of cells. Cells were incubated for 5 minutes at 37°C. Next, 170 μL complete medium was added directly to the cells. The cell suspension was homogenized by pipetting up and down vigorously. Then, 30 μL was transferred to a new 96-well plate for later acquisitions, filled up to 200 μL with prewarmed complete medium. Of the remaining 170 μL cell solution, 50 μL was used for flow cytometry acquisition and the rest was used for gDNA isolation.
Human organoids culture
Human PDOs were derived from resection material of pancreatic ductal adenocarcinoma (PDAC), colorectal carcinoma, and patients with gastric cancer who underwent surgery at the Department of Visceral, Thoracic and Vascular Surgery at the University Hospital Carl Gustav Carus of the TU Dresden. The study was approved by the Ethical Committee of the TU Dresden (#EK451122014). PDAC, colorectal cancer, and gastric cancer PDOs were cultured as described earlier (18, 19) and mutations were verified by Sanger sequencing. For passaging, PDOs from three to four wells from a 48-well plate were pooled in a 15 mL reaction tube and mechanically dissociated by repeated pipetting up and down through a fire-polished glass pipette. After centrifuging 5 minutes at 300 × g, the supernatant was aspirated and the pellet resuspended in Matrigel before plating as 20 μL drops in a new 48-well plate. Subsequently, Matrigel drop polymerization was induced by incubating for 10 minutes at 37°C and 250 μL organoid medium added to each well. For long-term storage, organoids were disrupted as for passaging using the Pasteur pipette. Fragments were dispensed in cold Recovery Cell Culture Freezing Medium (500 μL/well) and placed in 1 mL cryotubes. PDOs were frozen overnight in −80°C freezer in a cryo-freezing container (Mr. Frosty, Nalgene). On the next day, cells were transferred to liquid nitrogen. For thawing, cryotubes were warmed at 37°C and cells were resuspended in 10 mL advanced DMEM/F12 and then centrifuged for 5 minutes at 300 × g. Pelleted cells were resuspended in 20 μL Matrigel and placed in a well of a pre-warmed 48 well plate and placed at 37°C for 10 minutes. Finally, the Matrigel drop was overlaid with 250 μL of prewarmed medium containing all growth factors including RHOKi.
Genotyping of base edited cells
Genomic DNA was isolated using the QIAamp DNA Blood Mini Kit according to manufacturer's instructions. Targeted KRAS exon2 or TP53 exon7 PCR amplification was performed using high-fidelity Phusion polymerase using the following protocol:
Nuclease-free water up to 50 μL
10 μL HF buffer
1 μL dNTP mix (10 mmol/L each)
1,25 μL 20 μmol/L (FwdPrimer)
1,25 μL 20 μmol/L (RevPrimer)
250ng gDNA or 0.5 μL cell lysate
0.5 μL Phusion DNA polymerase
Reagents were mixed, briefly spun down at room temperature and run in a thermocycler with the following cycling conditions:
1 cycle: 30 seconds 98°C (initial denaturation)
30 cycles: 10 seconds 98°C (denaturation)
20 seconds 65°C (annealing, specific to primer pairs)
30 seconds 72°C (extension)
1 cycle: 5 minutes 72°C (final extension)
1 cycle: ∞ 8°C (hold).
A 5-μL aliquot of each PCR reaction was run on agarose gel, confirming correctly-sized bands and purity of procedure through a blank no-template water control. PCR products were purified using ISOLATE II PCR and Gel Kit (Bioline) according to manufacturer's instructions and DNA concentrations were quantified using a Nanodrop spectrophotometer. The appropriate amount of DNA together with the respective sequencing primers (forward or reverse primer were used) were submitted for Sanger sequencing, following the vendor's protocol.
EditR to quantify base editing efficiency
EditR is a free online tool (20) to quantify sequencing reads from raw ab1 files. For quantification of KRAS base editing, we amplified a 544 bp PCR product spanning KRAS residues 12 and 13 of the gRNA-treated cells. Cells transduced with the nontargeting gRNA were used as control. The obtained ab1 files of the potentially edited region was uploaded to EditR together with the gRNA protospacer sequence (∼20 bp). EditR generated plots displaying editing efficiencies at each base within the protospacer.
PANC-1 cells expressing ABE8e-GFP were seeded in a 24-well plate and transduced with the gRNA-tdTomato construct as described above. The cell-cycle assay was performed 7 days after infection with the Click-iT EdU Alexa Fluor 647 Flow Cytometry Assay Kit (Invitrogen; Thermo Fisher Scientific), according to the manufacturer's instructions. To stain for the total DNA content, fxCycle Violet stain (Invitrogen; Thermo Fisher Scientific) was used at a 1:1,000 dilution. The flow cytometry was carried out on a MACSQuant X (Miltenyi Biotec), followed by the analysis of the acquired data with the FlowJo software (FlowJo, RRID:SCR_008520).
PANC-1-ABE8e-GFP sorted cells expressing the base editor were infected with pLenti.sgG12D-1-TdTomato, at MOI = 1, resulting in ∼50% infection rate to create an internal competition between cells bearing the base editing gRNA and cells expressing only the base editor. Four days after infection, cells were seeded into a μ‐Slide 8 well‐chambered coverslip slide (Ibidi) containing 300 μL of FluoroBrite DMEM (Gibco) supplemented with 10% FBS, 4 mmol/L GlutaMAX, and 1% Pen/Strep. Time‐lapse microscopy‐based imaging was performed over 5 days using the Deltavision Elite deconvolution microscope. Images were acquired on FITC, TRITC, and Cy5 channels every 30 minutes using a 20×/1.42 plan‐Apochromat objective, at 37°C with 5% CO2. Subsequently, images were deconvolved and z‐projected using softWoRx analysis software (SoftWoRx, RRID:SCR_019157).
Data were analyzed using GraphPad Prism version 6 (GraphPad Prism, RRID:SCR_002798). Unless otherwise stated, time points in time-course experiments are presented as the SDs (presented as error bars) of three independent experiments, performed in biological triplicates. For base editing timecourses, the raw FACS points were processed using functions implemented in RStudio Version 1.2.1335 (RStudio, RRID:SCR_000432) and the statistical difference between the mean percentage at end-point of experimental gRNA and that of nontargeting gRNA/no gRNA was determined using unpaired two-tailed Student t test. A P < 0.05 was considered to be statistically significant.
Deep sequencing of potential off-target editing
Pooled sequencing reads were demultiplexed using the cutadapt software package. Fastq files generated were analyzed using CRISPResso2 (21) to calculate editing frequencies on both ON and OFF target sites. To calculate A to G editing efficiencies, we added the frequencies of all alleles that has at least one A converted to G in the quantification window, which is defined as −10 and +10 from the center of the gRNA. For quantification of DNA edits via nanopore sequencing, PCR products from target regions were sequenced with the native barcoding approach using SQK-LSK109 and EXP-NBD104 on a r9.4.1 flowcell (Oxford Nanopore Technologies). Sample preparation was done according to manufacturer's protocol (Version: NBA_9093_v109_revD_12Nov2019) and sequencing was performed for 24 hours. Basecalling and demultiplexing was performed with guppy v5.0.7 (Oxford Nanopore Technologies) using the dna_r9.4.1_450bps_sup model. The reads were filtered via filtlong v0.2.0 (https://github.com/rrwick/Filtlong) for a minimum mean phred score of 17 and minimum length 400 bp. DNA sequences were then mapped on the reference sequence from the PCR fragment (chr12:25,245,072–25,245,615, GRCh38/hg38) with minimap2 (version 2.17-r974-dirty; ref. 22). The edited region around Glycine 12 and Glycine 13 was then extracted from the alignments in R (version 4.1.2) via the stackStringsFromBam function from the R-package GenomicAlignments (version 1.30.0; ref. 23). The extracts were then converted to amino acid sequence with the bioseq package (version 0.1.3) and further counted and visualized using tidyverse packages (version 1.3.1).
Paired-end RNA-seq data were mapped to human genome hg38 from genecode https://www.gencodegenes.org/human/ with STAR_2.6.1d aligner (STAR, RRID:SCR_004463; ref. 24). First, STAR was used to build a genome index from the human genome fasta file. We then mapped reads of each sample in paired end mode. We analyzed two biological replicates for each of the three samples (RKO, RKO-ABE8e, and RKO-ABE8e-sgG12D-1). The output bam files were later indexed with the samtools software package (SAMTOOLS, RRID:SCR_002105).
To calculate the percentage of A-to-I editing in each sample, we used REDItools v1.3 (REDItools, RRID:SCR_012133; ref. 25) to quantify percentage of A to I out of total adenosines in each RNA-seq dataset. The REDItoolDenovo.py script was used to output the empirical distribution of all possible substitutions in each RNA-seq dataset. REDItools v1.3 was run on the bam files generated in the first step according to the following parameters; min mapping quality 30, min per base quality 25, min read coverage 10. Finally, we calculated the percentage of A to I editing by dividing the number of A to G substitutions (AG) over the total number of adenosines in the dataset (AA). These percentages were used to calculate the mean, SD and perform unpaired t test of two biological replicates in each sample using the GraphPad online calculator https://www.graphpad.com/quickcalcs/ttest1.cfm (GraphPad Prism, RRID:SCR_002798).
The datasets generated during and/or analyzed during this study are available from the corresponding author F. Buchholz on request. Researchers interested in accessing the data can contact F. Buchholz at email@example.com. It can take a few months to negotiate data use agreements and gain access to the data with Data Use Agreements with the Technical University of Dresden, Germany.
CRISPR/Cas9-mediated cleavage of KRAS mutations promotes the appearance of oncogenic escape variants
To examine the precision and versatility of CRISPR/Cas9 nuclease to target different KRAS mutations in a panel of cancer cell lines, we tested some of the most prevalent KRAS mutations G12D (PANC-1), G13D (HCT116), and G12S (A549), employing a cellular fitness competition assay (Fig. 1A). In this assay, reduction of the percentage of GFP-positive cells over time indicates that the infected cells expressing the nuclease–sgRNA complex have a growth disadvantage in comparison with the noninfected cells. To correct for possible off-target cleavage of the KRAS wild-type (WT) locus, we included a KRAS-WT cell line as a negative control. Interestingly, although the employed sgRNAs did not show appreciable effects in the KRAS-WT control cell line (Supplementary Fig. S1A), the percentage of GFP-positive cells progressively declined over time in all three cell lines bearing the mutation-targeting sgRNA, suggesting that disruption of mutant KRAS impaired their further growth (Fig. 1B–D). Hence, CRISPR/Cas9-mediated cleavage of oncogenic mutations can be used to probe the dependence of the tumor cells on the mutation. However, although the noninfected cells outcompeted the targeted cells in the pool, we noticed that the depletion curves flattened at a low percentage of GFP-positive cells after prolonged culturing, an observation seen across all three tested cell lines. Of note, these persistent cells appeared healthy and proliferated in the culture for several weeks despite carrying the mutation-targeting sgRNA and Cas9. To investigate the KRAS locus of these cells, we sorted the residual HCT116 GFP-positive cells followed by sequencing of the PCR product of the KRAS locus (Fig. 1E). Interestingly, about half of all the reads matched the wild-type KRAS allele sequence. This was expected, because HCT116 cells are heterozygous for the G13D mutation with the second allele being KRAS wild type. The fact that about 50% of the sequence reads reflected KRAS wild type suggests the wild-type sequence is not cleaved employing this sgRNA, likely owing to the employed proximal SpCas9 NGG PAM. Analyzing the mutant allele, we observed that 13% of cells still carried the original G13D mutation, not offering a clear-cut explanation why these cells survived the continuous targeting. In contrast, 8% of the clones showed other well-known cancer driver mutations of KRAS, including G13V, G12S, and K16E (Fig. 1E). Deep sequencing of the amplified DNA fragment confirmed the emergence of these alternative KRAS cancer driver mutations and identified additional ones (e.g., G13Y; Supplementary Fig. SS1B). These mutants were probably induced as a consequence of G13D mutation targeting, likely through double-strand DNA break (DSB) formation, followed by error-prone cell intrinsic DNA break repair in the HCT116 cells. Importantly, these mutant alleles are resistant to targeting by the initially designed sgRNA and therefore escaped the inactivation of the G13D driver mutation. We conclude that while the approach of using Cas9 nucleases to inactivate cancer mutations is efficient and informative, emerging oncogenic variants at the cleavage site will likely hamper this approach for many future applications, including potential clinical usage. Furthermore, nucleases are not useful for mutations occurring in tumor suppressor genes, because cleavage of these mutations is unlikely to reconstitute their wild-type activity.
Base editing corrects the G12S KRAS mutation in A549 cells
Base editing has emerged as a technology that allows precise genome editing without the introduction of DSBs (26). Relevant to this work, the described G12D, G13D, and G12S point mutations in the cell lines are potentially all addressable by an adenine base editor (ABE; Supplementary Fig. S2). We therefore decided to test whether inactivation of KRAS mutations using adenine base editing was feasible. Owing to the lack of a well-positioned SpCas9 NGG PAM, we decided to use a less restrictive Cas9 variant that utilizes an NG PAM (27). To enable efficient base editing, we combined this nCas9 variant with the recently described enhanced base editor ABE8e (28), hereafter referred to as NG-ABE. To allow stable expression, the KRAS mutant cell lines were first infected with a lentivirus expressing NG-ABE, in combination with a puromycin resistance cassette and GFP (Fig. 2A). The NG-ABE expressing cells were then subjected to a cell competition assay, analogous to the prior Cas9-nuclease experiments. Half of the cells were infected with a lentivirus expressing the mutant KRAS-specific sgRNA together with TdTomato to monitor the percentage of sgRNA-positive cells over time (Fig. 2A). We first tested this approach with four different sgRNAs in the human lung carcinoma cell line A549, harboring a homozygous KRAS G12S mutation (Fig. 2B). Interestingly, although the percentage of TdTomato-positive cells did not change over time in cells transduced with a no gRNA-TdTomato control, all four targeting sgRNAs led to a progressive loss, although with different kinetics, of TdTomato-positive cells over time (Fig. 2C), suggesting that correction of mutant KRAS back to the wild-type sequence impaired cell proliferation. Twelve days after infection, we isolated gDNA from the pool of cells infected with sgG12S-2-TdTomato. After PCR-amplification of KRAS codon 12 and Sanger sequencing, analysis revealed substantial editing at the targeted adenine (“A”), indicating conversion of the mutation back to the wild-type sequence (Fig. 2D; Supplementary Fig. S3). Importantly, RKO-ABE8e cells (KRAS WT) were not affected by infection with any of the four sgRNAs (Supplementary Fig. S4B), suggesting that targeting mutant KRAS with ABE8e-sgG12S did not affect proliferation of KRAS WT cells.
Moderate KRAS G13D base editing and prominent bystander edits in HCT116
Next, we tested six sgRNAs in the human colon cancer cell line HCT116, harboring a heterozygous KRAS G13D mutation (Fig. 3A). Only one of the six tested sgRNAs (sgG13D-6) showed a continuous depletion of TdTomato-positive cells over time (Fig. 3B), suggesting that many sgRNAs do not function efficiently to correct the G13D mutation. Twelve days after infection, we isolated gDNA from the pool of cells infected with sgG13D-6-TdTomato, showing a depletion down to 25% at the end of the experiment. After PCR-amplification, which included KRAS codon 13, and Sanger sequencing, analysis revealed moderate editing at the targeted “A” (Fig. 3C). In addition to the targeted adenine, two adjacent “A”s were also noticeably changed as bystander edits (Fig. 3C). Although the altered “A” at V14 results in a silent mutation, conversion of the “A” at K16 leads to a nonsynonymous mutation, potentially creating an oncogenic KRAS isoform.
Efficient KRAS G12D base editing in PANC-1 cells
Finally, we tried to address the most frequent KRAS mutation—G12D and to this goal, we tested six sgRNAs in a human carcinoma cell line of the exocrine pancreas, PANC-1, harboring a heterozygous KRAS G12D mutation (Fig. 4A). In four of six sgRNAs, a rapid depletion of TdTomato-positive cells was observed, reaching depletion levels down to 1% with two different sgRNAs, suggesting that the infected cells are largely outcompeted by noninfected cells (Fig. 4B). PCR amplification of KRAS codon 12 in cells infected with sgG12D-1-TdTomato at early time points after infection and subsequent Sanger sequencing revealed a high ratio of A>G conversions, signifying efficient and rapid correction back to wild-type KRAS sequence (Fig. 4C; Supplementary Fig. S5A). Importantly, potential bystander edits of adjacent “A”s would not change the protein sequence (Fig. 4A and C). Similar to our previous findings, all the tested sgRNAs showed no growth retardation in RKO-ABE8e cells (KRAS WT), ruling out the possibility that the effects seen are due to off-target effects (Supplementary Fig. S4A). To investigate a potential emergence of other oncogenic KRAS mutations, as observed in the nuclease treated cells, we sorted the remaining TdTomato-positive cells 60 days after infection, followed by gDNA isolation and sequencing of the KRAS locus. Remarkably, all of the sequenced clones harbored the KRAS wild-type sequence (Supplementary Fig. S5B), demonstrating the high accuracy of the base editor to correct the oncogenic mutation. However, this result also suggests that some rare PANC-1 cells can survive the correction of the KRAS mutation and continue to proliferate in culture.
Phenotypic characterization of KRAS G12D correction
The depletion of sgRNA-expressing cells could be caused by different scenarios, such as slowed cell growth, cell-cycle arrest, or cell death. As oncogenic KRAS is known to regulate critical components involved in cell cycle (29, 30), we first characterized the cells with corrected KRAS locus by performing cell-cycle analyses. Indeed, we observed a decrease in proliferating cells reflected by a significant increase in the fraction of cells in G0–G1 phase of the cell cycle (Fig. 4D). In line with existing work (31), this result indicates that mutant KRAS is required for G1 progression to S phase. Sequencing of infected cells at an early time point (3 days after infection) revealed that the base editing reaction is rather rapid (>50% at 3 dpi; Supplementary Fig. S5A). This rapid correction contrasted the relatively slow decline of gRNA expressing cells in the competition assay (∼80% Tomato-positive at day 3 after infection). To investigate the dynamics of corrected cells, we performed long-term time-lapse microscopy, where the GFP-sorted PANC-1 cells expressing NG-ABE were infected at a ratio of ∼50% with the sgG12D-1-TdTomato virus and cultured over 5 days. Intriguingly, we observed numerous cells carrying the base editing sgRNA-TdTomato that rounded up and bursted in the time-lapse movies (Fig. 4E; Supplementary Movie S1), indicating that some edited cells undergo cell death. Of note, a few other TdTomato-positive cells seemed to prolifate at a regular rate, indicating that these cells (even though many of them presumably carry already the KRAS wild-type sequence) were not affected, yet. Importantly, cells infected with control sgRNA-TdTomato continued to divide at a regular rate as observed in the time-lapse movies (Supplementary Movie S2). Taken together, these experiments indicate that the expression of mutant KRAS is required for continuous cell proliferation and that correction of mutant KRAS using base editing leads to G0–G1 cell-cycle arrest and sporadic cell death in corrected cells.
Off-target analysis of ABE8e targeting KRAS G12D mutation
To assess possible unintended editing at potential DNA sites with homology to the employed KRAS sgRNA, we identified most likely human off-targets using Cas-OFFinder (Fig. 5A; ref. 32). The top 9 potential off-target sites were PCR amplified from PANC-1-ABE8e expressing cells before and after infection with the top-performing sgRNA-1 for KRASG12D mutation correction, followed by deep sequencing. Analysis of more than one million sequencing reads revealed greater than 50% A>G mutational correction for the on-target (Supplementary Fig. S6A), consistent with data obtained using EditR. In contrast, no appreciable changes at any of the off-target sites were detected (Fig. 5B), echoing previous reports that ABEs are rather specific (33).
Base editors might also cause off-target editing at mRNA level (34). To investigate possible deamination of adenines in a transcriptome-wide manner, we isolated mRNA from control cells and cells stably expressing ABE8e and the top-performing sgRNA-1 for the G12D mutation. We then measured the A-to-I substitution frequency across the transcriptome for the samples by deep sequencing. Analyzing more than 30 million reads for each sample, we did not detect significant differences in A-to-I conversion in transcriptomes between WT cells, cells expressing ABE8e alone, or cells expressing ABE8e together with the sgRNA (Fig. 5C), consistent with previous data demonstrating neglectable mRNA off-target editing of latest generation ABEs (28).
Efficient TP53 R273H base editing in PANC-1 cells
On the basis of the success of base editing mutations in the activated oncogene KRAS, we thought to apply this technology to another class of genes frequently mutated in cancer cells. Tumor suppressor genes are genes that regulate cells during replication and cell division. The tumor suppressor gene TP53 is the most frequently mutated gene in cancer (35), and like KRAS, it is notoriously difficult to target with conventional pharmacologic approaches (36). What is more, targeting of mutations in tumor suppressor genes with nucleases is not useful, as cleavage of the mutation is unlikely to recover the wild-type activity of the tumor suppressor gene. We therefore decided to test, whether base editing could be applied to correct mutations in TP53 and if so, what consequences this correction might have. Of note, PANC-1 cells harbor an inactivating hotspot mutation in TP53 (p.R273H, c.818G>A). Importantly, this mutation should be addressable by ABEs (Fig. 6A). As loss of tumor suppressors function often occurs through loss of heterozygosity (35), correction of only one allele might be sufficient to revert the phenotype. Towards testing this hypothesis, we cloned one sgRNA into the TdTomato lentiviral vector to repair the R273H mutation in PANC-1 cells, harboring a hemizygous TP53 mutation. Remarkably, swift depletion of TdTomato-positive cells was observed, suggesting that the infected cells were depleted and largely outcompeted by noninfected cells (Fig. 6B). Moreover, PCR amplification of TP53 exon seven 3 days after infection with sgR273H-TdTomato and subsequent sanger sequencing revealed ∼79% of A>G conversions, signifying an efficient and rapid correction back to the wild-type TP53 sequence (Fig. 6C). To investigate whether TP53 DNA correction in the genome restored its protein function, we performed RT-qPCR for a panel of p53 target genes. Interestingly, rapid induction of expression was observed for p21, an early, canonical transcriptional target of p53, regulating cell cycle and apoptosis (Fig. 6D). Furthermore, other key p53 target genes inducing cell growth arrest such as MDM4 and DNA damage inducible gene GADD45A, as well as the apoptosis regulator PUMA, were upregulated (Fig. 6D), indicating that restoration of wild-type p53 reactivates its function as a gatekeeper for cell growth and division.
Finally, we thought to investigate a possible synergy by combining the correction of KRAS G12D with TP53 R273H in PANC-1 cells. Towards this goal, we cloned the TP53 sgR273H in a BFP-expressing lentiviral vector and co-infected cells with the KRAS sgG12D-TdTomato lentivirus. We then followed single- and TdTomato-BFP double positive cells over time. Importantly, RKO cells (KRAS WT, TP53 WT) did not show any growth disadvantage in response to double- or single infections with sgG12D-TdTomato and sgR273H-BFP containing viruses (Fig. 6E). In contrast, we observed a sharp decline of the double positive population that was significantly stronger than the decline of the single positive cells (Fig. 6E; Supplementary Fig. S7), suggesting that co-repair of mutant KRAS and TP53 has an additive effect in eliminating corrected cells.
Base editing in mutant KRAS and TP53 PDOs
To move cancer-driver base editing closer to translational implementation, we employed a more clinically-relevant culture model and investigated the approach in KRAS G12D-mutant PDOs. Base editing was performed in a KRAS-G12D mutant PDO (DD442) and a KRAS WT PDO control line (DD107) and the effects were analyzed utilizing the established competition assay (Fig. 7A). In infected DD107 control organoids, no change in the percentage of TdTomato-positive cells versus noninfected cells was observed with the sgRNA1-G12D, demonstrating that expression of NG-ABE in conjunction with sgG12D-1 is well tolerated in PDOs with a KRAS WT background. In contrast, we detected a progressive, albeit slow decline in TdTomato-positive cells over time in the DD442 PDO (Fig. 7B). Although the depletion was not as efficient as in the G12D mutant PANC-1 cell line, we detected >30% correctly edited A>G bases 10 days after infection (Fig.7C; Supplementary Fig. S8).
To extend the approach to a different tumor type and to a different oncogenic mutation, we targeted a hemizygous TP53 hotspot mutation (p.R175H, c.524G>A) in the colorectal carcinoma PDO line DD663 (Fig. 8A). Remarkably, a rapid decline of TdTomato-positive cells was observed, reaching less than 25% of TdTomato-positive cells 20 days after infection (Fig. 8B), suggesting that the infected cells were depleted and largely outcompeted by noninfected cells. Concomitantly, PCR amplification of TP53 exon four 3 days after infection with the sgR175H-TdTomato lentivirus and subsequent Sanger sequencing revealed >70% of A>G conversions, signifying efficient and rapid correction back to the wild-type TP53 sequence (Fig. 8C). These results demonstrate for the first time that BE-mediated correction of oncogenic mutations is possible in PDOs and that this correction impedes cancer organoid growth.
Programmable nucleases enable the targeting of specific genomic sequences including cancer driver mutations. Mutated in one quarter of cancer patients, KRAS represents the most frequently mutated oncogene in human cancers (3), whereas its pharmacologic inhibition is limited (4). We demonstrate that different KRAS mutations can be selectively targeted using Streptococcus pyogenes CRISPR-Cas9 nuclease, even though the difference from the wild-type allele is only a single nucleotide. Cleaving KRAS mutant variants in the classical 2D cell lines HCT116, A549, and PANC-1 cells rapidly depleted targeted cells up to 80%, a finding that is in line with previous studies, suggesting that many cancer cells are genetically dependent on KRAS mutations (8, 37, 38). However, long-term expression of Cas9 also selected for oncogenic escape variants that were resistant to targeting with the original sgRNA. In recent years, two studies have successfully used CRISPR-based cleavage of KRAS mutations G12V, G12D, and G13D and concluded that this approach can potentially be used for the treatment of patients with cancer (39, 40). In contrast to our study, these studies did not report the occurrence of oncogenic escape clones following Cas9 treatment. Considering these results, our work urges caution with Cas9 nuclease-based approaches to disable activating mutations in oncogenes, especially when such an approach is sought for potential clinical applications.
The development of second-generation ABE variants with increased deaminase activity (28) together with broadened compatibility with Cas homologs provided the opportunity for these genome editing tools to be tested for the correction of cancer mutations. Our work pioneers this approach for KRAS and TP53 mutations. Our results demonstrate that KRAS mutation correction was possible in all tested cell lines, although with different success rates. In A549 cells bearing the G12S variant, all four sgRNAs showed progressive depletion and editing. A weak bystander editing was also observed at a neighboring adenine, yielding a silent mutation. In contrast, in HCT116 cells, only one sgRNA showed a measurable depletion, despite similar sequences of the sgRNAs. This could be owing to the position of codon 13 and the targeted “A” within the editing window, as well as the unpredictable sequence preference of the deaminase. This result encourages for the testing of several sgRNAs per mutation, harnessing different PAMs, sgRNA lengths, and target “A” position within the editing window. A machine learning model (BE-Hive) has been recently described to predict editing efficiencies of several base editors across different target sites (41). It would be interesting to evaluate, if this algorithm can predict the efficiency and specificity of sgRNAs to correct cancer mutations in cells. Other strategies to improve editing efficiencies include the use of alternative Cas9 homologs compatible with ABE8e, which are available with different PAM preferences (28), or using an ABE where TadA replaces the Cas9 HNH-domain, allowing for an alternative editing window (42). Furthermore, base editors with optimized architectures, such as rigid linkers have recently been described (43), which might improve base editing of cancer mutations. Nevertheless, potentially harmful bystander edits always have to be considered, as we observed for the KRAS G13D correction, where editing of a bystander “A” in a subset of cells caused an amino acid change, possibly leading to an oncogenic variant. Importantly, these bystander edits can be predicted and their consequences can be evaluated beforehand. Nevertheless, versions of base editors that reduce bystander edits would further improve their utility (44).
Remarkably, in PANC-1 cells bearing the G12D variant, base edited cells showed substantial levels of editing and were depleted almost entirely, after treatment with two independent sgRNAs. Moreover, base editing depletion dynamics seemed comparable with the nuclease approach (compare Figs. 1B and 4B). However, base editing seems to be advantageous, because in contrast to the nuclease approach, we managed to deplete the ABE expressing cells almost entirely, without the occurrence of escape variants. Although this result illustrates the robustness and accuracy of the base editing approach to correct the oncogenic driver mutations, it also highlights that some rare cells still escape the consequence of oncogene mutation correction, likely representing a reservoir of resistant clones that would ultimately result in clinical relapse. It will be interesting to molecularly study these resistant cells in the future, as they could reveal mechanisms how cancer cells can bypass their KRAS dependency.
The time-lapse microscopy experiments revealed the phenotypic consequences of KRAS correction in PANC-1 cells. Although sequencing revealed that the correction within the cells happens rather quickly (>50% at 3dpi; Supplementary Fig. S5A), cells carrying the sgG12D-1 did not all die or arrest at this early time point. Instead, several dividing red cells seemed healthy during the observation period, suggesting that these cells survive the KRAS correction for an extended period of time. It is known that constitutive KRAS activation boosts several key aspects of cancer cell metabolism such as glycolysis, micropinocytosis, and autophagy to maintain high-energy levels (45, 46). Hence, cells with higher metabolites might persist longer after the correction of the KRAS mutation (47). Moreover, recent characterization of PANC-1 cells has revealed a pleomorphic phenotype with differential stem cell marker expression (48). It is therefore possible that the heterogenicity within the cells in culture accounts for the divergent behavior observed.
We extended the base editing approach to correct cancer mutations in tumor suppressors, namely the tumor suppressor TP53. Of note, many TP53 missense mutations are G>A missense mutations and are therefore addressable, in particular with Cas9 variants with modified PAM preferences. Intriguingly, the recovery of only one wildtype allele of the hemizygous TP53 mutation was sufficient to reactivate p53 function that led to a sharp depletion of corrected cells, likely triggered through p53-mediated upregulation of cell-cycle/apoptosis regulator genes, such as p21, GADD45A, MDM4, and PUMA. Of particular relevance, TP53 is the most common KRAS co-mutation at 39.4% (49), a fact that has profound implications for prognosis (50). The demonstration that co-correction of two mutations was possible and led to accelerated depletion of targeted cells offers new perspectives to identify synergistic dependencies in cancer cells. In this context, it will be interesting to investigate whether co-correction of TP53 and KRAS can prevent the emergence of resistant cells, as was observed when KRAS G12D was corrected alone.
Moving closer to the clinic, PDOs were tested in terms of their accessibility for base editing. An edited PDO with a KRASG12D mutation demonstrated a progressive depletion, albeit less efficient than in the PANC-1 cells, employing the same sgG12D-1. Importantly though, the KRAS wild-type PDO showed no signs of sensitivity towards the ABE8e-sgG12D-1 complex even after 1 month of treatment. Moreover, sequencing of sorted KRAS wild-type PDO demonstrated only KRAS WT at codon 12, without signs of any editing of “A” bases. In contrast, sequencing of treated G12D mutant PDO carrying the ABE8e-sgG12D-1 complex revealed an editing efficiency of >30%, indicating that slow regression of infected cells seems to reflect a slower editing kinetics. Why the base editing appears to be slower in pancreatic organoids is not apparent at first glance, but strategies to enhance base editing efficiency in PDOs seem feasible. For example, lower base editing efficiency in organoids could be due to the presence of agents in the organoid media influencing the base editing reaction. Fibroblast growth factor (FGF) 10 in the organoid medium may mask KRAS dependence by facilitating activation of the MAPK pathway via the PI3K–AKT axis (51). In addition, possible patient-specific mutational and transcriptional signatures (e.g., DNA repair fidelity) might influence the efficiency of the base editing reaction. Adaptation of the organoid culture conditions might allow more efficient base editing by testing base editing in growth factor reduced culture conditions. Although different mutations/sgRNAs were exploited, more efficient base editing was observed in the colorectal cancer PDO for a TP53 hotspot mutation, indicating that tissue origin of the cancer cells might also influence base editing efficiencies. However, already without further possible improvements, our study demonstrates that base editing technology can be harnessed to correct driver mutations in cancer organoids. This finding allows for testing the vulnerability of individual tumors in vitro to phenotypically profile these cells.
To our knowledge, this is the first study demonstrating efficient targeting of oncogenic recurrent mutations in activated oncogenes and tumor suppressor genes using base editing in both classical 2D cell lines and organoids. Undeniably, the development of base editors for cancer therapy is challenging because in order to be very efficient, every cancer cell in the body would have to be reached. However, only in the past two years, base editors have progressed towards clinical use for several diseases, providing proof-of-concept for the use of base editing in vivo (52). As base editors continue to advance towards clinical applications, their continued optimization to maximize their efficacy, specificity, and ability to target cancer cells in vivo remain important priorities (53, 54). Given the rapid development of base editing technology (55, 56) and delivery (57), we speculate that correction of cancer driver mutations by base editing may become a therapeutic modality in the future.
F. Buchholz reports grants from DKFZ/DKTK and EU during the conduct of the study. No disclosures were reported by the other authors.
S. Sayed: Conceptualization, validation, investigation, visualization, writing–original draft, writing–review and editing. O.A. Sidorova: Investigation, visualization, writing–review and editing. A. Hennig: Investigation. M. Augsburg: Methodology. C.P. Cortés Vesga: Investigation. M. Abohawya: Formal analysis, visualization. L.T. Schmitt: Validation. D. Sürün: Validation, writing–review and editing. D.E. Stange: Resources, writing–review and editing. J. Mircetic: Investigation, writing–review and editing. F. Buchholz: Conceptualization, resources, supervision, funding acquisition, writing–review and editing.
The project was financially supported by the European commission H2020 UPGRADE, grant no. 825825 (to S. Sayed) and Deutsches Krebsforschungszentrum (DKFZ Heidelderg): DKTK Joint Funding Program “Breaking therapy resistance in AML (RiskY-AML).” The authors would like to thank all members of the Buchholz lab for the fruitful discussions. Graphical schemes were created with www.BioRender.com.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/).