Targeting mutant KRAS signaling pathways continues to attract attention as a therapeutic strategy for KRAS-driven tumors. In this study, we exploited the power of the CRISPR-Cas9 system to identify genes affecting the tumor xenograft growth of human mutant KRAS (KRASMUT) colorectal cancers. Using pooled lentiviral single-guide RNA libraries, we conducted a genome-wide loss-of-function genetic screen in an isogenic pair of human colorectal cancer cell lines harboring mutant or wild-type KRAS. The screen identified novel and established synthetic enhancers or synthetic lethals for KRASMUT colorectal cancer, including targetable metabolic genes. Notably, genetic disruption or pharmacologic inhibition of the metabolic enzymes NAD kinase or ketohexokinase was growth inhibitory in vivo. In addition, the chromatin remodeling protein INO80C was identified as a novel tumor suppressor in KRASMUT colorectal and pancreatic tumor xenografts. Our findings define a novel targetable set of therapeutic targets for KRASMUT tumors. Cancer Res; 77(22); 6330–9. ©2017 AACR.
The RAS family of oncogenes (KRAS, NRAS, and HRAS) is the target of intense research for two reasons: the crucial role of mutant RAS proteins in tumorigenesis and the continued unmet need for therapeutic options for RAS-mutated human cancers. RAS is the most frequently mutated oncogenic driver of human cancer, and KRAS is mutated in about 20% of all human cancers, including some of the most lethal. Because of the historic difficulty in targeting RAS, genomic screening using synthetic lethal/enhancing approaches has become an attractive method to identify more tractable therapeutic targets and to further understand RAS biology. The approach identifies genetic targets whose disruption does not normally affect survival but becomes lethal/enhancing in the presence of an activating RAS mutation (1, 2).
Functional genomic screens have been revolutionized by application of the gene-specific editing technology of clustered regularly interspaced short palindrome repeats (CRISPR) and CRISPR-associated protein 9 (Cas9) system, which allows efficient and specific genome engineering in mammalian cells (3–5). By targeting the Cas9 nuclease gene using specific single-guide RNAs (sgRNA) and inducing targeted double-strand DNA breaks that are repaired by error-prone nonhomologous end joining, insertion or deletion mutations can be introduced into 5′ exons of coding genes, resulting in loss-of-function mutants. A number of groups have used pooled oligonucleotide array synthesis to generate and validate large sgRNA libraries for genome-wide loss-of-function (knockout) studies (6–11).
In cell culture models, genome-wide CRISPR-Cas9 knockout screens have revealed that KRAS wild-type (KRASWT) and mutant (KRASMUT) tumor cells show differential dependencies on downstream MAPK signaling members and adapters, and that KRASMUT tumors are highly dependent on mitochondrial oxidative phosphorylation (7, 10). To extend our knowledge of the genetic vulnerabilities of KRASMUT tumors to the in vivo setting, we performed a pooled CRISPR-Cas9 knockout screen of tumor xenografts using an isogenic pair of human colorectal cancer cell lines carrying wild-type or mutant KRAS. We identified gene knockouts that conferred selective beneficial or detrimental effects in the context of KRAS activation, including multiple metabolic vulnerabilities, highlighting the therapeutic potential of targeting cancer metabolism in KRASMUT tumors. Using a secondary validation sgRNA library, we additionally identified a novel KRAS-dependent tumor-suppressor gene in KRASWT/MUT colorectal cancer and pancreatic adenocarcinoma (PDAC) isogenic xenografts.
Materials and Methods
Cell lines and culture
The paired isogenic colorectal cancer cell lines HCT116 KRASWT/−, HCT116 KRASWT/G13D, DLD-1 KRASWT/−, and DLD-1 KRASWT/G13D (abbreviated to WT and MUT) were kind gifts from Dr. Bert Vogelstein (Johns Hopkins University, Baltimore, MD). Capan-2, SW1990, H348, and H441 cells were from the ATCC. All cell lines were cultured in DMEM (Invitrogen) with 10% FBS (Invitrogen) and 1% penicillin/streptomycin (Invitrogen). HCT116 cell lines were validated by PCR amplification and Sanger sequencing to confirm the mutation at the genomic level. Strict bio-banking procedures were followed, and cells were tested for contamination, including mycoplasma. HCT116-dCas9, DLD-1-dCas9, H358-dCas9, and Capan-2-dCas9 cell lines were generated by lentiviral transduction with lentiCas9-Blast (Addgene # 52962) and selection with 10 μg/mL of blasticidin.
Pooled library amplification and viral production
We followed previously published protocols of screens using the GeCKO library, as developed by the Zhang lab, with slight modifications. The human GeCKO v2 A library pooled plasmid (lentiCRISPR v2) was obtained from Addgene (cat # 1000000048) and amplified according to the recommended protocols by electroporation in Lucigen Endura electrocompetent cells. For the secondary focused validation sgRNA library, oligonucleotide pools (CustomArray) were obtained with variable 20 bp sgRNA sequences flanked by universal PCR primers (74 bp total synthesized sequence, see Supplementary Experimental Procedures for full sequences). Full-length oligonucleotides were amplified by PCR using Q5 HiFi polymerase (New England Biolabs) and size-selected on 2% agarose gels. PCR inserts were cloned into BsmBI-digested lentiGuide-Puro (Addgene # 52963) using Gibson assembly (New England Biolabs), transformed into Endura electrocompetent cells, and amplified similarly to the GeCKO library. 293FT cells were resuspended in DMEM and cotransfected with 12 μg of the hGeCKO plasmid library or secondary validation mini-pool, 9 μg psPAX2 vector, and 6 μg pMD2.G vector in 10 cm plates using 48 μL of PLUS reagent (Invitrogen 11514-015) and 60 μL of Lipofectamine in Opti-MEM medium. The medium was aspirated after 6 hours and replaced with fresh DMEM/10% FBS. The supernatant was harvested after 60 hours, centrifuged at 3,000 rpm at 4°C for 10 minutes, filtered through a 0.45 μm low protein-binding membrane (Millipore), and concentrated using an Amicon Ultra-15 Centrifugal Filter (Millipore) for 40 minutes at 4°C at 4,000 rpm. The virus was then aliquoted and frozen at −80°C.
Pooled library transduction
A functional virus titer was obtained by measuring puromycin resistance after transduction via spinfection, as previous published (8). A titer resulting in 20% to 40% of cells surviving puromycin selection was calculated to correspond to an MOI of 0.2–0.5 and a single infection percentage of 77% to 89%. For the genome-wide screen, duplicate transductions with 1.6 × 108 cells each were infected at an MOI of 0.2 for coverage of approximately 500×. For the secondary validation mini-pools, triplicate transductions with 1.2 × 107 cells each were infected at an MOI of approximately 0.2 for coverage of approximately 1,000×. Transduced cells were selected with 2 μg/mL puromycin for 7 days. Cells were passaged with a seeding density of 3 × 107 cells for the genome-wide screen and 1 × 107 cells for the secondary mini-pool screen at each passage to maintain library representation.
All animal studies were approved by the UC San Diego Institutional Animal Care and Use Committee. HCT116 cells transduced with the GeCKO library or HCT116-dCas9 cells and Capan-2-dCas9 transduced with the secondary mini-library or individual sgRNA constructs were washed with PBS, resuspended in 300 μL of PBS and Matrigel (1:1 ratio), and injected subcutaneously into the flanks of nude mice (NU/J; Jackson Labs). For the primary GeCKO screen and secondary mini-library screens, 3 × 107 cells were injected per mouse and groups of 3 mice were injected per transduction replicate per cell line. After 14 days, mice were euthanized and the tumors excised and stored in RNAlater (Qiagen) solution. For individual sgRNA experiments, 1.5 × 106 HCT116, DLD-1, H358, or Capan-2 tumor cells were resuspended in PBS:Matrigel (1:1) and injected into the flanks of nude mice (3–4 mice per sgRNA, 6–8 mice per gene targeted). Tumor dimensions were measured with Vernier calipers for 24 days, and volumes (mm3) were calculated as (0.5 × width) × (height2).
Small-molecule inhibitor treatment
Nude mice were injected with HCT116 tumor cells as above. After xenografts were palpable (∼7 days after injection), animals were injected intraperitoneally with 100 μL of vehicle, thionicotinamide (Spectrum; 100 mg/kg body weight) in 1% DMSO, or KHK inhibitor (Calbiochem 420640 or synthesized according to published methods; 25 mg/kg; ref. 12) in PBS. Injections were repeated every other day for 14 days (7 doses total). Tumor dimensions were measured with Vernier calipers, and volumes (mm3) were calculated as (0.5 × width) × (height2).
sgRNA library quantification by deep sequencing
Genomic DNA was extracted from tumors or cells using the previously published salt-precipitation protocol (11). The sgRNA library representation was determined using a two-step PCR process in which PCR1 amplifies the lentiviral sequence containing the 20 bp sgRNA cassette and PCR2 attaches Illumina sequencing adapters and barcodes. Primer sequences were obtained from the Zhang lab online resource (http://genome-engineering.org/gecko/) using v2Adaptor_F and v2Adapter_R for PCR1 and primers F01-F06 and R01-R02 for PCR2. All PCR reactions were performed using Herculase II Fusion DNA Polymerase (Agilent). Sufficient PCR1 reactions were performed to maintain library coverage. For the genome-wide screen, a total of 200 μg of genomic DNA template was used per sample with 10 μg of gDNA per 100 μL PCR1 reaction, ≥20 PCR1 reactions occurred per biologic sample. For the focused secondary screen, a total of 18 μg of genomic DNA template was used per sample with 6 μg of gDNA per 100 μL PCR1 reaction; ≥3 PCR1 reactions occurred per biologic sample. After pooling the PCR1 product, 10 μL was used for each PCR2 reaction. PCR2 was performed with one reaction per 104 constructs (7 PCR2 reactions per sample). PCR products were purified and quantified with Qubit and/or Bioanalyzer, and diluted libraries were sequenced on an Illumina NextSeq (TSRI core).
Data processing and analysis
Illumina NextSeq sequencing reads were demultiplexed and the adapters trimmed using cutadapt to leave only the 20 bp sgRNA spacer sequences. The sgRNA spacer sequences were then mapped to the reference human GeCKO Library A using Bowtie, allowing a maximum of one mismatch and allowing only uniquely aligning reads. Only sgRNA spacers with multiple reads were analyzed (sgRNA spacers with only a single read were filtered out). Normalized read counts were obtained by normalizing to total read count per sample (normalized reads per sgRNA = reads per sgRNA/total reads for all sgRNAs in the sample × 106 + 1). CRISPR guide scores were generated by calculating the log fold change of normalized sgRNA read counts between xenograft samples and the baseline T0 samples. The log fold change was then normalized to the median of the nontargeting control sgRNAs for each sample. RNAi Gene Enrichment Ranking (RIGER) analysis was performed using GENE-E software (Broad Institute) using the weighted sum method to convert sgRNAs to genes and 1 × 107 number of permutations. STARS analysis was performed using the STARS software v 1.2 (Broad Institute) with a threshold of 25 and 500 to 1,000 iterations.
Genome-wide CRISPR-Cas9 screen of isogenic human colorectal cancer cells harboring wild-type or mutant KRAS
To conduct a CRISPR-Cas9 knockout screen in an in vivo setting, we used a well-characterized isogenic pair of human colorectal cancer lines that differ only in the presence (HCT116MUT) or absence (HCT116WT) of an activating G13D KRAS mutation. These cell lines have previously been used by our lab to conduct a high-throughput screen of oncogenic KRAS synthetic lethal miRNAs (13). We transduced HCT116WT and HCT116MUT cells with the human GeCKO v2 library, which contains 65,383 sgRNA constructs targeting 19,050 human coding genes (3 sgRNAs per gene) and 1,864 miRNAs (4 sgRNAs per gene; ref. 8). Duplicate transductions were performed at a low MOI (0.2) to ensure transduction with only one sgRNA per cell, thereby effectively barcoding individual cells. Sufficient cell numbers (1.6 × 108 cells) were transduced to allow 500× coverage of each sgRNA within the library, and this coverage was maintained at each cell passage (>3 × 107 cells seeded per passage). Cells were selected for stable viral integration with puromycin for 7 days and then subcutaneously injected into the flanks of nude mice (3 × 107 cells per mouse, 3 mice per duplicate cell line: total n = 12 mice). The mice were euthanized 14 days later, the tumors were excised, and genomic DNA was extracted. Lentiviral sgRNA constructs were then PCR-amplified and quantified by deep sequencing (Fig. 1A). As a surrogate for cell proliferation, we analyzed enrichment or depletion of sgRNA abundance in the tumor xenografts compared with levels in the cell lines before injection. It was expected that sgRNAs targeting genes that are essential for tumor growth would be less abundant in the xenografts compared with the preinjection cells, whereas sgRNAs targeting genes that normally control growth (i.e., tumor suppressors) would be enriched in the xenografts compared with preinjection cells.
The majority of the preinjection sgRNA library was recovered in the tumor samples, with 93% of the sgRNA library constructs (92% of nontargeting control sgRNAs) containing multiply aligning reads in all samples and little apparent random loss during cell injection (Supplementary Fig. S1A). Replicate mice from the duplicate transductions showed good correlation at the sgRNA level (Supplementary Fig. S1B) and gene level (Supplementary Table S1). Screen performance was analyzed using previously published essential and nonessential reference gene sets (7, 14) and the 1,000 nontargeting control sgRNAs in the GeCKO library. We then compared the change in sgRNA representation in HCT116WT versus HCT116MUT xenografts using the CRISPR-Cas9 guide score (15), which is the median-corrected log fold change in abundance of each sgRNA in the xenograft compared with the preinjection cell population. Nontargeting control sgRNAs and nonessential genes clustered around the center of the plots, indicating that they were neither enriched nor depleted in either xenograft (Fig. 1B). As expected, sgRNAs targeting most (92%) of the reference essential genes were depleted in both HCT116WT and HCT116MUT xenografts compared with the preinjection cells (Fig. 1B). However, consistent with previous studies, not every sgRNA in the GeCKO library was equally efficacious, and some sgRNAs targeting essential genes were not depleted (Fig. 1B).
Pathway analysis of the in vivo GeCKO CRISPR-Cas9 screen identifies multiple metabolic vulnerabilities in KRAS mutant tumors
We next performed pathway analysis of sgRNAs that were commonly or selectively depleted in HCT116MUT versus HCT116MUT xenografts. Gene set enrichment analysis (GSEA) of KEGG pathways of commonly depleted sgRNAs (i.e., lethal or essential genes) identified ribosome and spliceosome components as the two most significant hits, which is consistent with the results of earlier in vitro screens (Fig. 1C and D; ref. 7). In contrast, pathway analysis of sgRNAs depleted in HCT116MUT xenografts but not in HCT116WT xenografts (KRAS synthetic lethal) revealed enrichment of genes associated with the MAPK signaling pathway and multiple metabolic pathways, including oxidative phosphorylation (Fig. 1C and D).
For further analysis, we examined genes for which at least 2 sgRNAs were selectively depleted in HCT116MUT xenografts (Supplementary Table S2). We applied a cutoff value of >0.5 log2 fold decrease in abundance HCT116MUT compared with HCT116WT xenografts, which effectively excluded >90% of nontargeting control sgRNAs. The gene hits defined by these criteria included previously validated targets of the MAPK signaling pathway (MAPK1/ERK2) and metabolic pathways (GFPT1; refs. 7, 16). Among the novel candidate KRAS synthetic lethal genes identified were metabolic genes in the tricarboxylic acid cycle (succinate-CoA ligase ADP-forming beta subunit, SUCLA2), pentose phosphate pathway (nicotinamide adenine dinucleotide kinase, NADK), and fructose metabolism (ketohexokinase, KHK; Fig. 2A and Supplementary Table S2). Notably, these genes did not display synthetic lethality in a parallel genome-wide screen of 14 day-cultured HCT116WT and HCT116MUT cells (Fig. 2A) or in other in vitro screens, including of HCT116 cells (7), substantiating the importance of recapitulating the in vivo tumor microenvironment in order to identify relevant therapeutic targets.
To validate the identified metabolic genes, we generated two independent sgRNAs per gene and transduced them into stable Cas9-expressing HCT116WT and HCT116MUT cell lines. The knockout cell lines were then deep sequenced to determine the CRISPR mutagenesis frequency. On-target insertions, deletions, and substitutions, most of them small, were observed for each of the sgRNAs, with similar genomic effects in HCT116WT and HCT116MUT cells (Supplementary Fig. S2A). Protein knockdown in the cells was also confirmed by immunohistochemistry (Supplementary Fig. S2B), and the cells were then injected into nude mice (n = 3–4 mice per sgRNA). Knockout of SUCLA2, NADK, and KHK significantly reduced the growth of HCT116MUT xenografts but not of HCT116WT xenografts (Fig. 2B and C). As positive controls, cells were also transduced with individual KRAS- or MAPK1-targeting sgRNAs and, as expected, they selectively decreased the growth of HCT116MUT tumors (Supplementary Fig. S2C).
To verify the CRISPR knockdown results, we examined the effects on tumor growth of small-molecule inhibitors of NADK and KHK (no inhibitors of SUCLA2 were available). NADK has previously been proposed as a possible therapeutic target in cancer, because NADPH is essential to support enhanced biosynthesis and to control redox status (Fig. 2D; ref. 17). Consistent with this, administration of thionicotinamide (thioNa), a small-molecule inhibitor of NADK (18), showed significant inhibition of the growth of HCT116MUT xenografts but not of HCT116WT xenografts (Fig. 2E and F). KHK has also recently been described as a possible therapeutic target for hepatocellular carcinoma (19, 20). KHK exists as two isoforms, KHK-A and KHK-C, which are generated by alternative splicing of exons 3A or 3C, respectively. KHK-C has the greater affinity for fructose and is expressed specifically in hepatocytes. KHK-A is more widely expressed and has been shown to phosphorylate phosphoribosyl pyrophosphate synthetase 1 (PRPS1), which is involved in de novo purine synthesis (Fig. 2D). We found that KHK-A is the predominant isoform in HCT116 cells and other KRAS mutant cancer cell lines of different lineages [non–small cell lung cancer (NSCLC) and PDAC; Supplementary Fig. S2D]. We also confirmed that the 2 validation sgRNAs targeted exons 3A or 4, which would knockout KHK-A expression, and that the exon 3C-targeting sgRNAs had no effect on the growth of HCT116MUT xenografts (Supplementary Fig. S2E), consistent with the predominant expression of KHK-A in these cells. To determine the effect of KHK inhibition on xenograft growth, mice were administered a commercially available small-molecule KHK inhibitor (Fig. 2D and Supplementary Fig. S2F; ref. 12), which binds to the conserved ATP-binding domain present in both KHK isoforms (Supplementary Fig. S2G). This compound significantly inhibited the growth of HCT116MUT xenografts, but not HCT116WT xenografts (Fig. 2G and H), confirming the CRISPR knockout data and establishing the importance of this enzyme for the growth of KRAS-mutant tumors.
A secondary in vivo sgRNA screen identifies candidate KRAS-dependent tumor-suppressor and synthetic lethal genes
We next performed a smaller focused screen with higher depth (more sgRNAs per gene) and coverage per construct (1,000× coverage) to further validate the genome-wide screen and to detect hits with greater power. Recent studies subsampling large genome-wide CRISPR screens support this strategy in which a broad genome-wide screen is first performed with a limited number (3–4) of sgRNAs per gene and then relaxing the FDR threshold (75%) to generate candidates for a secondary screen of a limited number of genes with greater depth (∼6 sgRNAs per gene; ref. 6).
To generate a candidate gene list for our secondary screen library, we used the RIGER algorithm to rank significantly depleted sgRNAs in the genome-wide screen. This method considers all sgRNAs for a gene similarly to a gene set in GSEA and takes into account the sum effect of all sgRNAs against a particular gene (21). To find candidate KRAS-enhancing and lethal genes (i.e., targeted by sgRNAs enriched or depleted, respectively, in HCT116MUT xenografts), we identified genes overlapping in multiple RIGER comparisons of the isogenic paired screen (see Supplementary Experimental Procedures and Supplementary Fig. S3A and S3B), and finally obtained 152 candidate KRAS lethal genes and 160 candidate KRAS-enhancing genes. Of note, the candidate gene lists did not completely overlap with the pathway analysis list, and the secondary library did not include the validated hits from the pathway analysis.
We then created a custom-pooled sgRNA library consisting of approximately 2,500 sgRNAs targeting 250 genes (9 sgRNAs per gene) selected from the KRAS-lethal (∼150 genes) and KRAS-enhancing (∼70 genes) candidates, sgRNAs targeting 20 essential genes, and 230 nontargeting control sgRNAs (Supplementary Table S3). Analysis of the CRISPR-Cas9 guide scores from the genome-wide screen confirmed that, on average, the sgRNAs targeting the candidate KRAS-lethal and -enhancing genes were selectively depleted or enriched, respectively, in HCT116MUT xenografts (Fig. 3A).
The pooled lentiviral sgRNA libraries were generated by oligonucleotide array synthesis, transduced into HCT116WT and HCT116MUT cells, and screened in xenografts versus preinjection cells in a similar manner to the genome-wide screen (Fig. 3B). However, the focused screen was performed at a higher coverage of 800 to 1,000× in triplicate transductions (3 mice per cell line, 3 transductions; total of 18 mice). Replicate mice from the triplicate transductions again showed good correlation at the sgRNA level (Supplementary Fig. S3C) and gene level (Supplementary Table S4). Also similar to the findings with the genome-wide screen were (i) the majority of nontargeting control sgRNAs were neither enriched nor depleted in the xenografts, (ii) many of the sgRNAs targeting essential genes were depleted in both xenografts, and (iii) multiple MAPK1-targeting sgRNAs were selectively depleted in HCT116MUT xenografts (Fig. 3C and D). A new observation was that multiple KRAS-targeting sgRNAs were selectively depleted in HCT116MUT xenografts, which was likely due to the increased coverage and depth of the secondary library (Fig. 3C and D). Using the STARS algorithm (6) to rank the genes at an FDR of <5%, we found that KRAS, MAPK1, SNRPC (small nuclear ribonucleoprotein polypeptide C), and predicted miR-4663 were depleted to a greater extent in HCT116MUT xenografts compared with HCT116WT xenografts (Fig. 3E and Supplementary Table S4). Relaxing the FDR cutoff to 25%, POP5 (POP5 homolog subunit of ribonuclease P/MRP), SF3B2 (splicing factor 3b subunit 2), and LENG9 (leukocyte receptor cluster member 9) emerged as potential KRAS synthetic lethal candidates (Fig. 3E and F). SNRPC, POP5, and SF3B2 also scored as essential hits in the Toronto Knockout HCT116MUTin vitro screen (7). Thus, our secondary screen identified previously unknown candidate KRAS synthetic lethal genes for future validation studies.
In addition to the depleted sgRNAs, the secondary screen identified several sgRNAs that were highly enriched in the tumor xenografts compared with preinjection cells, suggesting that they confer a proliferation advantage in vivo and thus may be novel tumor suppressors in the context of mutant KRAS. In support of this, we found that the most highly enriched sgRNAs targeted the known tumor suppressors NF2 and RALGAPB, with the HCT116MUT xenografts showing enrichment of multiple sgRNAs (Supplementary Fig. S4A). NF2 encodes the protein Merlin and is a regulator of Hippo signaling, which exerts tumor suppressive effects through the Wnt/β-catenin (22, 23) and LIN28B–let-7 (24) pathways. RALGAPB encodes the regulatory subunit of the GTPase-activating protein for the Ral GTPase, which is involved in signaling downstream of RAS. Loss of RalGAPβ has been shown to increase mTOR activity, and its knockdown increases PDAC invasion (25).
Although more enriched in HCT116MUT xenografts, NF2 and RALGAPB also score as the most highly enriched genes in HCT116WT xenografts (Fig. 4A). Using the STARS algorithm, we identified 6 genes for which sgRNAs were significantly enriched in HCT116MUT xenografts at an FDR of <5% (Fig. 4B and Supplementary Table S4). One of these, INO80C, which encodes a subunit of the conserved ATP-dependent chromatin-remodeling complex INO80, scored consistently highly in HCT116MUT xenografts (FDR < 8% in all 3 replicates) but not in HCT116WT xenografts (bottom 75% of enriched sgRNAs). Moreover, many INO80C-targeting sgRNAs were enriched only in HCT116MUT xenografts and ranked highly based on CRISPR guide scores (Fig. 4A). These findings suggested that INO80C is a novel potential tumor suppressor in KRASMUT tumors. The potential clinical relevance of this candidate novel tumor suppressor was confirmed by interrogating PDAC datasets from The Cancer Genome Atlas (TCGA), which revealed frequent (4%–18%) deep deletions of INO80C (Supplementary Fig. S4B). Moreover, analysis of additional TGCA datasets identified a significant association between INO80C deletion and worse prognosis in patients with KRASMUT colorectal cancer, NSCLC, and PDAC (Supplementary Fig. S4C). We tested the pooled validation library in xenografts from a KRAS-mutant PDAC cell line (Capan-2) in replicate (n = 3 mice per replicate) and confirmed that INO80C also enriched as a top hit (Fig. 4C). We then directly tested the effect of INO80C knockout on tumor xenograft growth by transducing stable Cas9-expressing HCT116MUT and HCT116WT cells with two independent INO80C-targeting sgRNAs. Indeed, INO80C knockout significantly and selectively enhanced the growth of HCT116MUT xenografts (Fig. 4D and Supplementary Fig. S4D). This effect was confirmed with a second pair of isogenic colorectal cancer cell lines, DLD-1-KRASWT and DLD-1-KRASMUT, although the increase in tumor growth was not as pronounced as for HCT116 xenografts (Fig. 4D and Supplementary Fig. S4D). Finally, we performed CRISPR knockout of INO80C in a KRASMUT PDAC cell line, Capan-2, and a KRASMUT NSCLC cell line, H348, and found significant enhancement of tumor growth (Fig. 4E and Supplementary Fig. S4E). Collectively, these data not only identify INO80C as a novel potential tumor-suppressor gene but also establish its relevance in human tumors harboring mutant KRAS.
KRAS oncogenic mutations occur in many of the most lethal types of cancers. Despite years of study, KRASMUT cancers remain among the most difficult to treat, due in large part to the lack of targeted agents. There is thus a significant unmet need to develop therapies that selectively inhibit activated KRAS and/or its downstream effector pathways. The advent of CRISPR-Cas9–based genetic screens has dramatically improved our ability to interrogate the genome in an unbiased manner, and such screens hold great promise for extending our knowledge of genetic vulnerabilities in different oncogenic contexts. Oncogenes exert their effects through a myriad of pathways that influence not only intracellular processes but also the interaction of tumor cells with their microenvironment to overcome external checkpoints on tumorigenesis. One of the hallmarks of cancer is aberrant cell metabolism, as reflected by the Warburg effect (26). Constitutively active oncogenic KRAS is known to rewire the metabolic program of cancer cells to support the energetic and biosynthetic demands of continued proliferation (16). KRASMUT cells show increased uptake of glucose, which can be utilized to support elevated nucleotide biosynthesis (16, 27). Recent studies highlight the fact that in vivo conditions such as tumor xenografts more accurately recapitulate the microenvironment associated with key metabolic phenotypes than do cell cultures, thereby facilitating the search for therapeutic targets (27).
Here, we conducted a genome-wide CRISPR-Cas9 screen of tumor xenografts formed by isogenic KRASMUT and KRASWT colorectal cancer cells to identify candidate genes that selectively enhance or inhibit tumor growth in vivo. Pathway analysis identified genetic vulnerabilities in multiple metabolic pathways, such as nucleotide synthesis and redox balance in which NADK and KHK play important roles. Genetic knockout and small-molecule inhibition of these targets more potently reduced the growth of KRASMUT xenografts than of KRASWT xenografts, suggesting they may be novel therapeutic targets in KRAS-mutant cancers.
In support of NADK as a cancer therapeutic target, a recent large-scale functional screen of low-frequency mutations in PDAC identified activating mutations in NADK that behaved as oncogenic drivers (28). NADK inhibition also inhibited the growth of a number of tumor types in vitro and in vivo (17). Activated KRAS maintains low intracellular levels of reactive oxygen species (ROS) while vigorously promoting metabolic activity. Thus, perturbation of these pathways to tip the redox balance may be a particularly effective treatment for KRASMUT tumors, as supported by the recent discovery that high doses of vitamin C selectively kill KRASMUT xenografts by increasing ROS that inhibit glyceraldehyde 3-phosphate dehydrogenase (GAPDH), resulting in an energetic crisis and cell death (29). Decreases in NADP/NADPH ratios activate GAPDH and enhance oxidative pentose phosphate pathway flux, which mitigates oxidative stress by increasing NADPH levels. Thus, by inhibiting the conversion of NADP to NADPH, NADK inhibition may decrease the ability of KRASMUT tumors to cope with oxidative stress.
KHK has also emerged as a potential therapeutic target in cancer. A recent study identified a switch from KHK-C to KHK-A isoform expression in hepatocellular carcinoma mediated by c-Myc and heterogeneous nuclear ribonucleoprotein H1/2 (19, 20). This switch has two key effects: a decrease in fructose metabolism through KHK-C, which reduces ROS levels, and an increase in PRPS1 activity, which enhances nucleotide production. Both of these effects are predicted to benefit KRAS-mutated cancers, and fructose has also been shown to promote the growth of pancreatic cancer cells (30).
Our secondary focused screen revealed known and novel genes as potential synthetic lethal partners with KRASMUT. Although the identification of KRAS and MAPK1 was expected, the novel genes include SNRPC, POP5, SF3B2, LENG9, and predicted miR-4663. These targets should be validated in future studies. SNRPC (also known as U1C) encodes a protein component of the U1 small nuclear ribonucleoprotein involved in spliceosome formation (31) and interacts with the RNA-binding protein EWS (32, 33). EWS is involved in chromosomal translocation in human cancers such as Ewing sarcoma (EWS/FL-1 fusion) and desmoplastic small round cell tumor (EWS/WT1 fusion), which are thought to promote oncogenesis by acting as transcriptional activators (34). EWS/FL-1 fusion protein has been shown to constitutively activate MAPK1 transcription and thus MAPK signaling (35). POP5 encodes a protein associated with RNase MRP and RNase P complexes but has no known roles in cancer (36). SF3B2 encodes a splicing factor that has been shown to be targeted by HIV to induce cell-cycle arrest (37). LENG9 encodes an uncharacterized protein of unknown function, and miR-4663 encodes a predicted but unvalidated miRNA identified by RNA sequencing from breast cancer tissue (38).
The positive-selection side of the secondary screen, which yielded a much higher signal compared with the dropout side, identified NF2 and RALGAPB as potential tumor-suppressor genes in both KRASWT and KRASMUT xenografts. Another candidate tumor suppressor in KRASMUT xenografts was INO80C, which we also validated by demonstrating that knockout enhanced the growth of KRASMUT HCT116 and DLD-1 colorectal cancer xenografts as well as KRASMUT H358 NSCLC xenografts. Based on the frequent deep deletions noted in PDAC TCGA datasets, we additionally examined the effects of INO80C deletion in a KRASMUT PDAC cell line, Capan-2, and found that it also enhanced tumor growth in vivo. Although the SWI/SNF family of chromatin remodelers, which is frequently mutated or lost in many tumors including PDAC, has been more widely studied, the INO80 complex has recently been implicated in carcinogenesis (39, 40). INO80 is a large multi-subunit complex that maintains genome stability through nucleosome editing, such as removal of the histone variant H2A.Z (41). In yeast, the homolog of INO80C (Ies6) combines with ACTR5 (Arp5) to form a complex (42) shown to be important in demarcation of transcriptional units through inhibition of H3K79 methylation (43). Intriguingly, knockout of this subcomplex in yeast alters transcription of metabolic gene networks that downregulate glycolysis and upregulate mitochondrial energy-generating processes (42). Further analysis of the role of INO80C in KRASMUT tumors may reveal novel elements of RAS biology.
Our study demonstrates the feasibility of genome-wide pooled CRISPR-Cas9 knockout screens of tumor xenografts for uncovering genetic vulnerabilities that may be amenable to therapeutic targeting. We identified NADK and KHK as candidate metabolic gene targets that were not previously identified using in vitro screens. We also found that a smaller positive-selection screen was particularly effective in identifying candidate tumor-suppressor genes such as INO80C.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Conception and design: E.H. Yau, T.M. Rana
Development of methodology: E.H. Yau, I.R. Kummetha, T.M. Rana
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E.H. Yau, I.R. Kummetha, G. Lichinchi, T.M. Rana
Writing, review, and/or revision of the manuscript: E.H. Yau, T.M. Rana
Study supervision: T.M. Rana
Other (performed experiments): R. Tang, Y. Zhang
We thank Steve Head and the staff of the next-generation sequencing core facility at The Scripps Research Institute for help with deep sequencing, and members of the Rana lab for helpful discussion and advice.
This work was supported in part by the NIH (CA177322). E.H. Yau was supported by the NCI of the NIH under Award Number T32CA121938.
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.