Genome‐wide screen for anticancer drug resistance in haploid human embryonic stem cells

Abstract Anticancer drugs are at the frontline of cancer therapy. However, innate resistance to these drugs occurs in one‐third to one‐half of patients, exposing them to the side effects of these drugs with no meaningful benefit. To identify the genes and pathways that confer resistance to such therapies, we performed a genome‐wide screen in haploid human embryonic stem cells (hESCs). These cells possess the advantage of having only one copy of each gene, harbour a normal karyotype, and lack any underlying point mutations. We initially show a close correlation between the potency of anticancer drugs in cancer cell lines to those in hESCs. We then exposed a genome‐wide loss‐of‐function library of mutations in all protein‐coding genes to 10 selected anticancer drugs, which represent five different mechanisms of drug therapies. The genetic screening enabled us to identify genes and pathways which can confer resistance to these drugs, demonstrating several common pathways. We validated a few of the resistance‐conferring genes, demonstrating a significant shift in the effective drug concentrations to indicate a drug‐specific effect to these genes. Strikingly, the p53 signalling pathway seems to induce resistance to a large array of anticancer drugs. The data shows dramatic effects of loss of p53 on resistance to many but not all drugs, calling for clinical evaluation of mutations in this gene prior to anticancer therapy.


| INTRODUCTION
Chemotherapies and other anticancer drugs stand at the forefront of cancer therapy, eradicating tumours and assisting in prolonging patient survival. [1][2][3] Nevertheless, tumours frequently show resistance to treatment, either innate or acquired following initial treatment. Resistance to anticancer drugs is a leading cause of ineffectual cancer treatment and tumour relapse, leading to poor survival rates. 4,5 The identification of genes and pathways that confer drug resistance is therefore of prime interest in the field of cancer research. The application of drug treatments to a wide panel of human cancer cell lines (hCCLs), followed by examining their response and gene expression, have yielded valuable insights regarding drug potency and mechanism. 6 However, a direct interpretation linking a specific gene to drug resistance is not trivial. A forward genetic screening approach was suggested and applied as a powerful tool to identify genes and pathways responsible for or affecting a phenotype, requiring an efficient gene interference platform. Emanuel Segal, Jonathan Nissenbaum and Mordecai Peretz contributed equally to this study. Although gene knockdown screens by using RNA interference libraries (RNAi or shRNA) provided valuable data, 7-9 they suffer from inherent limitations such as inconsistent RNA knockdown and a relatively high degree of off-targets. 10 With the recent advancement of CRISPR/Cas9 technology, [11][12][13][14] the ability to carry out comprehensive loss-of-function (LoF) screens not only sheds light on fundamental biological principles but also offered an attractive tool for drug target identification. 15 The versatile nature of the CRISPR/Cas9 tool was applied to a large number of hCCLs for the identification of cancer-essential genes and their correlation to anticancer drug response. 16 However, a well-known characteristic and limitation of hCCLs is their complex genomic profile. Aneuploidy/polyploidy, large genomic rearrangements, and a high rate of mutations, could easily mask gene/pathway relevance of essential genes. 17,18 KBM7 and subsequent derivative HAP1 19 are near-haploid (i.e., possessing a single copy of nearly all chromosomes) cancer-derived cell lines that serve a valuable role in overcoming the polyploid genetic background, [20][21][22] however they still show an abnormal genetic background and chromosomal instability 23 due to their cancerous origin.
A potential way of bypassing undesirable confounding interactions from other gene mutations is use of normal, genetically stable, human pluripotent stem cells (hPSCs) as a cancer-related model.
Both cancer cells and pluripotent cells share similarities in pivotal cellular features, making hPSCs a useful tool for cancer research. [24][25][26] In the past decade, tremendous progress in mammalian forward and reverse genetics has been made by the introduction of mouse haploid embryonic stem cells, 27,28 followed by the generation of rat, 29 monkey, 30 and recently, haploid human embryonic stem cells (hESCs 31,32 ). The generation of the normal human cells with a single allele of each gene has opened exciting avenues for basic as well as applied research in human genetics. 18,[33][34][35][36][37][38] Here, we aim to exploit haploid hESC advantages as a tool for identifying anticancer drug resistant genes. The main advantages of a haploid hESC platform are efficacy and the normal genomic background, as the efficacy of genetic targeting depends on the number of target loci that need to be manipulated. If a cell line possesses extra copies of a given gene-a common occurrence in cancers-by probability, gene knockout will be more effective in haploid cells with a single copy. Furthermore, haploid hESCs have a normal genome with no effective point-mutations, 31,39 and thus confounding mutations common in cancer cell lines can be avoided. To achieve a comprehensive perspective on mechanisms that might drive or regulate resistance to anticancer drugs, we performed genome-wide LoF screens using a wide range of drugs. We selected 10 different anticancer drugs which target fundamental cellular mechanisms and are used for a broad range of cancer indications.
Furthermore, our broad view of resistance-related genes also illuminates an interesting and relevant aspect for the role that the TP53 gene may play in chemoresistance. Collectively, our results emphasize the advantage of using haploid cells for wide LoF screens and the informative nature of using normal pluripotent cells for cancer research.

| Cell lines and culture
The following cell lines were used in this study: Haploid hESCs 31 and h-pES10-based mutant library recently established by us. 18 Diploid hESC-TP53 LoF mutation with GFP-tagged tubulin-TUB::GFP.; female 293T cells, obtained from R. Weinberg (Whitehead Institute).
The library of mutated hESC were cultured at 37 C and 5% CO 2 on matrigel-coated plates (Corning) in feeder-free mTeSR1 (STEMCELL Technologies) medium supplemented with 10 μM ROCK inhibitor Y-27632 (Stemgent) for 1 day after splitting. Before reaching confluency, cells were passaged by a quick trypsinization using TrypLE Select (Thermo Fisher Scientific), plated in feeder-free conditions. WA09 hESCs were cultured on feeder layer growth-arrested mouse embryonic fibroblasts (MEFs) in standard hESC growth medium, com-

| Anticancer drug selection and calibration
All 10 anticancer drugs were selected based on their selective function and importance as drugs that are frequently used in the clinic.

| Cell competition assay
In order to assess the resistance or insensitivity of TP53 LoF against 3 | RESULTS

| Assessment of hESCs for cancer resistance screen
To assess the relevance of hESCs for cancer resistance, we tested the response of hESCs to anticancer drugs from the AOD. hESCs were exposed to a range of AOD compounds and cell viability was previously measured. 42,43 We compared this viability data to the NCI-60 Growth Inhibition Data. A total of 88 shared compounds were ranked by cell response from most to least potent and we then correlated our data with the NCI hCCLs' average response. Comparing hESCs to hCCLs (mean) provided a significant positive correlation (R = 0.78, p = 0.0012; Figure 1A), indicating the similarities in responses between hESCs and hCCLs to a given anticancer drug. Augmented by the knowledge about the molecular and cellular similarities between pluripotent cells and cancer cells (e.g., tumour formation capacity, telomerase activity, oncogene upregulation and high proliferation rate), the correlation suggests that human pluripotent cells may offer a platform for anticancer drug screening while avoiding the genomic complexity characterizing cancer cell lines.

| Genome-wide CRISPR LoF screen for anticancer drugs
The field of anticancer drugs encompasses hundreds of compounds, widely varied in their mechanisms of action. We selected 10 different anticancer drugs for our screening ( Figure 1B). Selection was based on drug mechanism (e.g., protein kinases, DNA damaging agents, etc.), their diverse cancer indications, and the presence of a drug in clinical use. Following selection of suitable drugs, we calibrated a dose response curve for each in order to select an optimal concentration for our genome-wide LoF screens. The haploid hESC-CRISPR/Cas9 LoF library 18 was exposed to the selected anticancer drugs in at least two different concentrations to allow both strong and moderate selection, capturing a wider snapshot of drug-resistant genetic repertoire. We evaluated cell survival daily and once cells showed recovery, they were harvested for DNA extraction and replating at a ratio of 3:1, respectively. The replated cells were allowed to proliferate and then re-exposed to the drug for further selection. By the end of the experiment, we sequenced 4-6 library samples alongside their respective controls ( Figure 1C). Next, we analysed the changes in sgRNA abundance between the control and treatments via our CRISPR analysis pipeline. 18

| Validation of candidate genes
Our screens yielded a substantial number of plausible candidate genes. However, beyond the large screening analysis, individual validations are informative in specifically linking a gene's LoF mutation and resistance to a given drug. Our selection of candidate genes was based on their CS ranking, their molecular relevance to the drug's mechanism of action, and/or on the relevance of the candidate gene in cancer. We selected several candidate genes in which LoF mutations may confer resistance to anticancer drugs with diverse mechanisms of action. All selected candidate genes were confirmed as expressed in hESC as described. 18 Azacytidine is given as a treatment for myelodysplastic syndromes and acute myeloid leukaemia. It has a known mechanism of action via inactivation of the de-novo DNA methylase 44 and indeed Tet Methylcytosine Dioxygenase 3 (TET3) was highly enriched in our screen (CS = 1.9, FDR = 0.0002; Figure 2A). Therefore, we selected the TET3 gene for further validation. sgRNAs against TET3 were designed and via the Lenti-CAS9 vector system, TET3 KO hESCs were generated. The mutated cells exhibited a dramatic decrease in sensitivity to azacytidine treatment (IC50 = 0.2 μm for WT and 5 μM for mutant; Figure 3A). Sunitinib acts as a cell signalling inhibitor by targeting tyrosine kinase receptors, primarily via the VEGF signalling pathway, which plays a role in both tumour angiogenesis and tumour cell proliferation. 45   To further support the relevance of the findings in genome-wide screening in hESCs to actual tumours, we analysed cancer patient data

| TP53 LoF effect on anticancer drug resistance
The importance of TP53 in cancer with its pivotal and diverse roles in cell maintenance, growth, and survival is well documented. 53 LoF mutations in TP53 have been identified as such that allow cell survival and provide growth advantage in culture. Nevertheless, this growthadvantage characteristic by itself is masking an effect that mutations in the TP53 gene have on drug resistance. When we compared the TP53 sgRNA reads (normalized as percent of all guides in the culture), we found dramatic differences among the 10 drugs tested ( Figure 5A).
While the controls did not exhibit a massive enrichment of TP53 sgRNAs, this did occur in certain drug screens more than others (e.g., methotrexate and olaparib; Figure 5A). Interestingly, the enrichment of the TP53 sgRNAs in the azacytidine assay was far less than for most of the other drugs. In order to address this interesting phenomenon and to distinguish between culture-growth-advantage per se and drug-resistance effects, we generated a cell competition assay.
As an elegant, easy-to-monitor system, we generated an hESC line with a LoF mutation in TP53 and a tubulin-tagged GFP fluorophore.
Mixing the mutant cells with WT cells allowed for reliable detection in cell population dynamics over time and treatment. We tested four dif-

| System efficacy
In this project, we established a system for using hESCs to identify mutations that may confer drug resistance. This platform is favourable to alternative approaches as it was established in haploid cells with a clean genetic background, which aids in the identification of specific genetic pathway activity. Along with the aforementioned common biology between ESCs and cancer cells, our comparison of response to anticancer drugs between our haploid hESCs and the CCL data ( Figure 1) provided strong support for utilizing our hESCs in genomewide screen for drug-resistant mutations. We selected 10 drugs arrayed across mechanisms of actions and indications that are actively used in cancer treatment. Our haploid hESC LoF library was exposed to these drugs and subsequent analysis enabled identification of potential resistance-conferring mutated genes and the pathways involved.

| Specificity of genes conferring resistance
In identifying the genes and pathways that allow resistance to the different anticancer drugs, various aspects of the selected genes align with previously known cellular effects of the drug, though it should be noted that our candidate genes have not been previously directly or specifically linked to resistance to these drugs. Among the enriched mutations in the azacytidine screen, and considering its known activity as a demethylating agent, we identified TET3, a gene encoding for a known demethylating enzyme. Sunitinib has known targets in VEGF receptors, and indeed we could identify FLT4/VEGFR3 mutation as conferring resistance to the drug. In the case of vemurafenib, previous work has shown that PMAIP1 is downregulated following inhibition of the BRAF pathway, the target of vemurafenib. 54 We were able to show that lack of PMAIP1, a proapoptotic protein, grants a specific resistance to treatment with the drug. We could indeed validate the differential capacity of mutations in TP53 to confer resistance to the drug ( Figure 5B). These results support the power and sensitivity of our screen, as it detected a further level of gene-drug response in addition to the well documented growth advantage of TP53 mutants. As a final piece of supporting evidence, we were able to find TCGA data for a handful of bleomycin- Another pathway, not previously identified in the context of drug resistance, which was enriched across several screens was the aminoacyl-tRNA biosynthesis pathway, responsible for attaching tRNA to its respective amino acid preceding translation during gene expression ( Figure 2C). While these genes were enriched in several drug screens, the strongest effect (and greatest pathway representation) was seen in the azacytidine screen. While a mechanistic interpretation of these findings is outside the scope of the current work, we suggest that perhaps as azacytidine causes demethylation and increased gene transcription, damage to tRNA production helps curb the perturbation as the cell's resources are misspent, conversely assisting survival in cells with a mutated pathway. has not been previously reported as granting drug resistance, and its effect was validated in four drugs (three for resistance and one sensitivity; Figure 3). Moreover, the genes validated in our study

| Trends across multiple screens
have not been previously confirmed as resistance-granting to anticancer treatment.
It should also be acknowledged that current drug treatments for cancer are overwhelmingly administered in combinations of multiple drugs designed for maximum efficacy, a fact not addressed by our single-agent screens. While initial drug selection attempted to account for this (i.e., selection of drugs also used as monotherapies), clinical practices of combination therapy will require further elaboration of screens in our system to address this point.

| Genome-wide screens
As mentioned above, work examining drug response in cancer cell lines has been previously performed. 6 Databases similar to the TCGA have been established covering gene-drug interactions among numerous other biological properties of several hundreds of cancer cell lines, though as stated the genetic stability of these cells is suboptimal. 55 Genome-wide CRISPR screens in hundreds of hCCLs have been performed with the purpose of optimizing drug treatment through treatment protocol prioritization. 56 However, CRISPR-Cas9 has been shown to be affected by copy number aberrations, which are very prevalent in cancers, skewing and even invalidating results. 57 Research utilizing a genome-wide CRISPR approach on the nearlyhaploid CCLs HAP1 and KBM7 58 would bypass this interference, however such work is still plagued by the aforementioned chromosomal instability, 23 in addition to point mutations, 18  (B) WT/TP53 À/À cell competition assay. Mixture of WT and TP53 À/À hESC were used to test TP53 effect on drug resistance. Plates were seeded with 1:50 ratio of TP53 À/À :WT cells, then treated with either azacytidine, methotrexate, or olaparib. Changes in TP53 À/À -GFP hESC were measured by FACS. Used as designation of statistical significance, p-value * < 0.05, ** < 0.01, *** < 0.001.
approach, synergizing the utility of healthy cells with a clean genetic background and true haploidy, enabling the full benefit of this method to be brought to bear with greater capacity for detection, indicating multiple pathways and genes not previously identified for the drugs examined here. For example, the effects of the p53 pathway shown here would likely be masked in hCCLs without these advantages. Collectively, this work has shown strong evidence for the power, sensitivity and advantage of our resistance-screening system, and should complement existing knowledge in the furtherance of efficient treatment of cancer.

CONFLICT OF INTEREST STATEMENT
Oded Kopper and Nissim Benvenisty are V.P. R&D and CSO of NewStem Ltd, respectively.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding authors upon reasonable request.