Radiation Resistance in KRAS-Mutated Lung Cancer Is Enabled by Stem-like Properties Mediated by an Osteopontin–EGFR Pathway

The KRAS gene encodes a GTPase involved in relaying signals from the cell membrane to the nucleus. Upon the introduction of point mutations (mut), most commonly at codons 12 and 13, the KRAS protein becomes constitutively active and acquires oncogenic properties. Activating mutations in KRAS are common, with the most frequent allelic mutations in codons 12 and 13 affecting almost 140,000 new cases of non–small cell lung carcinoma (NSCLC), colorectal cancer, and pancreatic cancer in the United States per year (1). KRASmut cancers often have a poor prognosis and exhibit treatment resistance, particularly to agents targeting the EGFR (2–5). For many years, it has been known that mutations in KRAS may enhance cellular resistance to ionizing radiation (IR), although this connection has not been without controversy (6–8). In particular, only sparse clinical data have come out to suggest that the radiation resistance associated with the KRASmut genotype that is observed in the laboratory can also be seen in patients (9, 10). A variety of underlying mechanisms have been investigated to explain the radiation resistance of KRASmut cancer cells, but no consistent model has emerged (8, 11–14).

Cancer stem cells (CSC), also referred to as tumor-initiating or tumor-propagating cells, may promote metastases development and recurrence after therapy (15, 16). In radiation biology, “clonogenic” tumor cells have been defined as cells that have the capacity to produce an expanding family of daughter cells and form colonies after irradiation in an in vitro assay or give rise to a recurrent tumor in in vivo models. Whether or to which degree clonogenic cells represent CSCs is unclear. Because the goal of radiotherapy in curative intent is to eradicate the last surviving clonogen or CSC, as a single clonogenic cell remaining after completion of treatment may give rise to a local tumor recurrence, the relationship between radiation resistance and CSC-like tumor cells is of considerable interest (15, 17). Generally, cancer cells with CSC-like phenotypes or markers have been found to be radiation resistant (18–20). Enhanced DNA damage response and repair pathways have been discussed as one underlying mechanism (21, 22). Interestingly, KRAS has been recently linked to CSC-like phenotypes (1, 23–26). For example, in one report, a KRAS-dependent pathway promoted tumor initiation, anchorage independence, and self-renewal (23).

EGFR together with Aurora B kinase and protein kinase C alpha (PKCα) maintains mitosis-like condensed chromatin (MLCC) in a subpopulation of KRASmut cancer cells in interphase (27). The condensed chromatin is characterized by a mitosis-typical colocalization of phosphorylated serine 10 and trimethylated lysine 9 on histone 3 (H3S10p and H3K9me3). This modification may be involved in developmentally regulated gene regulation (28), but the primary biological function of MLCC in cancer cells has remained unclear. Furthermore, the data suggest that MLCC physically protects genomic DNA against the induction of lethal DNA double-strand breaks (DSB) by IR (27).

We set out to elucidate the mechanisms of radiation resistance associated with KRAS mutations in NSCLC. We found evidence for a previously unknown link between the radiation-resistant and CSC-like phenotypes in KRASmut NSCLC cells and tumors. Unexpectedly, MLCC promoted not only radiation resistance but also CSC-like properties. Our data lead to a better understanding of the heterogeneity of KRASmut NSCLCs, where a subset of cancers contains a radiation-resistant fraction of MLCC-expressing cells that represents a novel therapeutic target.

Annotated NSCLC cell lines were originally obtained from the MGH/Sanger cancer cell line collection http://www.cancerrxgene.org/translation/CellLine in the time period 2009 to 2010, with the exception of A549 cells, which were purchased from ATCC, as described in detail in previous publications (27, 29). Cell cultures were passaged for <3 months after thawing an individual frozen vial. The identity of the cell lines had been tested as described using a set of 16 short tandem repeats (AmpFLSTR Identifier Kit, ABI). In addition, SNP profiles based on a panel of 63 SNPs assayed using the Sequenom Genetic Analyzer was used for in-house identity checking whenever a cell line was propagated and confirmed uniqueness of cell lines for the ones without available STR (27, 29). On some cell lines, additional authentication was performed by Bio-Synthesis, Inc. No cell line tested positive for mycoplasma (MycoAlert, Lonza). Three-dimensional (3D) tumor spheres and isogenic KRASwt NCI-H1703 NSCLC cells with or without stable expression of a KRASmut transgene were cultured and treated as described previously (27). For generation of A549 IR survivor cell lines, A549 cells were irradiated with 2 or 8 Gy single dose concurrently with 2 μmol/L erlotinib and seeded for colony formation. After 2 weeks, individual colonies were expanded for subsequent experiments.

Tumor xenografts were grown and treated as described previously (27, 30). For tumor initiation and growth studies, 2 × 10⁵ cells were injected into the flanks of nude mice (nu/nu, Charles River Laboratories). Additional experiments were performed using 7- to 14- week-old male and female NMRI (nu/nu) mice, as described previously (31). KRASmut or wild-type (wt) xenografts were either untreated and excised at a volume of approximately 180 cm³, received 50 mg/m² of erlotinib for 2 days and were harvested 4.5 hours after last application of erlotinib, were irradiated with 1 Gy and excised 30 minutes after fraction, or irradiated with 4 Gy and excised 24 hours after irradiation. Some tumors also received a combined treatment of irradiation and erlotinib. All irradiations were given using 200 kV X-rays (0.5 mm Cu), at a dose rate of approximately 1 Gy/minute. Up to 5 animals were irradiated simultaneously in specially designed plastic tubes fixed on a lucite plate, and the tumor-bearing leg was positioned in the irradiation field by a foot holder distal to the tumor. Snap-frozen samples and single-cell suspension generated from the xenografts were subjected to γ-H2AX staining or CD133 flow cytometry, respectively. In some experiments, tumor tissue samples from KRASmut and wt NCI-H1703 xenografts were maintained in complete medium, subjected to ex vivo treatment with erlotinib, and snap frozen for subsequent immunofluorescence microscopy analysis.

X-ray treatments were performed as described previously (30, 31). Erlotinib, saracatinib (LC Laboratories), and ABT-263 (Selleckchem) were dissolved in DMSO, and depending on the experiment were added 1 to 24 hours prior to irradiation at 2, 1, and 1 μmol/L final concentration, respectively. The drugs were maintained for the duration of the respective experiment. Human osteopontin (BD Biosciences) was dissolved in deionized water. Drugs were aliquoted and stored according to the manufacturers’ guidelines.

Clonogenic survival assays were performed as published previously (29). Determination of cell numbers at 3 or 5 days after irradiation was performed by using a fluorescent nucleic acid stain Syto60 or CellTiter-Glo luminescence assay (Promega), respectively, as described previously (29).

Transfections were carried out with 200 nmol/L validated SPP1 siRNA (Invitrogen) or a scrambled control siRNA (Ambion) using the X-tremeGENE Transfection Kit (Roche) as described previously (27). Western blotting and subsequent experiments were carried out 48 hours after transfection.

Cells were incubated with CD133/1 (AC133) - PE antibody (Miltenyi Biotec) with the dilution of 1:11 for up to 10⁷ cells/100 μL of buffer at 4°C for 2 hours. High and low CD133-expressing cells were subjected to sterile sorting by flow cytometry with 4 Laser BD FACSAria II Cell Sorter at MGH Pathology Flow Cytometry Core.

The presence of MLCC and its characteristics have been described previously (27). Cells with MLCC were detected using a specific antibody against colocalized H3S10p and H3K9me3, and cells displaying the typical granular interphase pattern were scored as MLCC positive as described. Staining and visualization of γ-H2AX and 53BP1 foci were also performed as described previously (27). For visualization of BIM expression (#2933, Cell Signaling Technology), the same protocol was used.

Whole-cell lysates were prepared using standard methods. Specific antibodies against osteopontin, H3K9me3, total-H3, and CRMP1 (ab8448, ab8898, ab10799 and ab62558, respectively, Abcam), BIM (#2933, Cell Signaling Technology), and β-actin (A2228, Sigma-Aldrich) were used with 1:100 dilution. Immunocomplexes were visualized by enhanced chemiluminescence detection (Amersham Corp.) using goat anti-mouse or goat anti-rabbit IgG coupled to horseradish peroxidase as a secondary antibody (Santa Cruz Biotechnology).

Cells were seeded at 10⁶ per 10-cm plate, and cross-linking was carried out using 1.5% formaldehyde. Procedures were performed using standard protocols. Primers used in the chromatin immunoprecipitation (ChIP) assays are as follows: CRMP1: forward: 5′-GAAATCTGCTTGGGAGCAGC-3′, reverse: 5′-GCCTAAAATCCCTGTCAATG-3′; BIM3: forward: 5′-CCTCCCGAGGCTTCACACCG-3′, reverse: 5′-CACCTACCCCAACCCGGCGG-3′; BIM7: forward: 5′-GCAGTGATTGGGCGTAGGAG-3′, reverse: 5′-TCAAGTGGTGGTAGTGGCGG-3′; GAPDH: forward: 5′-GTATTCCCCCAGGTTTACAT-3′, reverse: 5′- TTCTGTCTTCCACTCACTCC-3′,

The tumor sphere invasion assay was performed using the 96-Well 3D Spheroid BME Cell Invasion Assay Kit (# 3500-096-K, Trevigen), and the invasiveness of tumor spheres was analyzed following the manufacturer's protocols.

Microarray analysis was performed on RNA collected from 2 to 3 biological replicates of NCI-H1703 clones. RNA was purified using the miRNeasy Kit (Qiagen) following the manufacturer's instructions, including treatment with DNAase. Total RNA (1 μg) was used to create double-stranded cDNA using the cDNA Synthesis System (Roche). Double-strand cDNA was labeled with Cy3 using the Roche One Color DNA Labeling Kit following the manufacturer's protocol. Labeled DNA (6 μg) was hybridized to 12×135K Human Expression arrays (Roche/NimbleGen) for 20 hours following the manufacturer's protocol. Arrays were washed using a NimbleGen Wash Buffer Kit and scanned on an ms200 scanner. Images were aligned and features were extracted using NimbleScan 2.6 software.

We performed raw data quality control by principle component analysis and subsequent clustering by visualizing the distribution of expression levels to identify experimental outliers. After outlier removal, raw data files for each sample were normalized, background-corrected, and saved to logarithmic scale using a robust multiarray analysis. Normalized data were analyzed and presented using R project. The normalized data from the experiments were fit into linear models using the Bioconductor package limma. Pair-wise comparisons of interest were performed using moderated t-statistics to test for significant differential expression. FDR at the 0.05 level for each probe was computed. Probes with P < 0.05 and showing a fold change >1 were considered differentially expressed. All probes were annotated with gene names.

To better visualize the effect of differentially expressed genes, pathway analysis was performed on those transcripts that had a significance of P < 0.05. Enriched networks, pathways, cellular processes, etc., were generated through the use of IPA (Ingenuity Systems) in all three contrasts. The Database for Annotation, Visualization and Integrated Discovery (DAVID) was used to annotate genes with chromosomal location to cluster highly repressed genes in mutant samples.

Microarray expression data generated for the current study have been made publicly available through the Gene Expression Omnibus (GSE79486). Microarray expression data were also obtained from the publicly available Cancer Cell Line Encyclopedia (CCLE).

For additional genomic analyses, data from patients with lung adenocarcinoma with wt or mut KRAS were retrieved from The Cancer Genome Atlas through the cBioPortal for Cancer Genomics site (32) or the Oncomine Cancer Microarray database (33).

It has remained unclear to which extent KRAS mutations represent a biomarker of radiation resistance and which mechanisms may contribute to the radiation-resistant cellular phenotype. As a hypothesis-generating exercise, we observed inferior treatment outcomes for KRASmut compared with wt NSCLCs in a small cohort of patients with locally advanced disease treated with radiation, prompting us to investigate the heterogeneity of radiation resistance in NSCLC further (Supplementary Fig. S1A). In a panel of 18 NSCLC cell lines, KRAS codon 12/13 mutations were associated with increased survival of 62% after single dose irradiation (Fig. 1A, left). Expression of a KRASmut transgene at physiologic levels in KRASwt NCI-H1703 cells (Supplementary Fig. S1B; ref. 27) produced a comparable survival increase of 55% (Fig. 1A, right). Consistent with these data, lower numbers of IR-induced DSB were observed in KRASmut versus wt cells (Fig. 1B), which was confirmed in xenografts (Fig. 1C). Unexpectedly, the difference in radiation resistance between KRASmut and wt cells was more pronounced in tumor spheres compared with monolayer cultures (87% compared with 36% survival increase, respectively; Fig. 1D).

In agreement with this difference in radiation resistance, the fraction of KRASmut cells with the recently described radioresistance marker MLCC (27) was substantially increased when cells were grown under sphere versus monolayer conditions (Fig. 1E and F). This did not reflect any differences in cell-cycle distribution (Supplementary Fig. S1C). As we reported previously for monolayer cultures (27), MLCC expression in spheres was sensitive to erlotinib (Fig. 1F and G), implicating EGFR in promoting the radiation-resistant phenotype of KRASmut cells. Strikingly, the radiosensitizing effect of erlotinib was lost upon fractionated irradiation of KRASmut spheres (Fig. 1H; Supplementary Fig. S1D), which suggested enrichment with a radioresistant subpopulation. Because tumor spheres are known to promote the growth of CSC-like cells (24, 34), these observations encouraged us to further pursue potential links between MLCC and stemness in KRASmut NSCLC cells.

In keeping with the enrichment of tumor spheres for CSC-like cells, we observed an increase in the fraction of cells with high expression of the CSC marker CD133 (CD133^high; Fig. 2A; Supplementary Fig. S2A) and greater vimentin expression (Supplementary Fig. S2B) in spheres compared with monolayer cultures. Furthermore, KRASmut spheres contained a higher number of CD133^high expressers (Fig. 2B) and were larger in size than wt spheres (Supplementary Fig. S2C). CD133^high but not CD133^low cells were enriched for erlotinib-sensitive MLCC (Fig. 2C). We next subjected KRASmut cells to irradiation and clonally expanded surviving cells (Supplementary Fig. S2D and S2E). CD133^high cells from these IR survivors were further enriched for MLCC-containing cells compared with the cell population preirradiation (Fig. 2D). Consistent with this finding, fractionated but not single-dose irradiation as shown in Fig. 1H also increased the percentage of CD133^high cells compared with unirradiated controls (Supplementary Fig. S2F). Together, these data establish a link between the radiation-resistant MLCC cell fraction and CD133^high expression.

To further investigate the CSC-like phenotype of our KRASmut model, we performed experiments where KRASmut cell suspensions with a high fraction of CD133^high expressers proved to be substantially more tumorigenic than KRASwt cells when injected into nude mice (Fig. 2E). The increased fraction of CD133^high–expressing cells was preserved in these KRASmut xenografts (Fig. 2F). Furthermore, KRASmut CD133^high cells demonstrated the most pronounced invasion phenotype (Fig. 2G). Finally, EGFR inhibition downregulated not only MLCC in KRASmut cells (Fig. 1F and G), but also suppressed invasion (Fig. 2G), sphere growth, and CD133 expression (Fig. 2H). In summary, these data suggest that EGFR is important for maintaining the radiation-resistant CSC-like phenotype.

To identify novel factors that may promote the radiation-resistant phenotype of KRASmut NSCLC cells, we surveyed differentially expressed genes and pathways between isogenic KRASmut and wt NCI-H1703 cells (Fig. 3A; Supplementary Fig. S3). One of the most significantly upregulated genes in KRASmut cells, compared with either endogenous wt or exogenous overexpression of wt KRAS, was SPP1. The protein product of SPP1 is osteopontin. Osteopontin is a member of the SIBLING (small integrin–binding ligand and N-linked glycoprotein) family, and it has a secreted and an intracellular form, which have been reported to be elevated in many cancer types (35–37). We found osteopontin levels to be elevated in both whole-cell lysates and in media of KRASmut NCI-H1703 cultures compared with wt controls (Fig. 3B). Osteopontin expression was also notably increased in other KRASmut NSCLC cell lines (Fig. 3C). To determine the importance of SPP1/osteopontin for the radiation resistance of KRASmut cells, we depleted osteopontin by RNAi in a total of 8 NSCLC cell lines (Fig. 3D; Supplementary Fig. S4A). We first studied our isogenic NCI-H1703 model where osteopontin depletion led to radiosensitization of KRASmut but not wt cells (Fig. 3E; Supplementary Fig. S4B). Conversely, addition of recombinant osteopontin to the media enhanced the radiation resistance of KRASwt but not mut cells (Fig. 3F). Depletion of osteopontin also increased the number of IR-induced DSB in KRASmut cells but failed to do so in wt cells (Fig. 3G).

To elucidate the mechanisms by which osteopontin knockdown may cause radiosensitization, we cotreated cells with erlotinib. Pharmacologic inhibition of EGFR enhanced the number of induced DSB, as reported previously (27). However, in an osteopontin-depleted background, EGFR inhibition had no effect (Fig. 3G), suggesting that osteopontin and EGFR regulate DSB induction in an epistatic manner. Furthermore, both osteopontin and EGFR promoted MLCC (Supplementary Fig. S4C), and upon disruption of MLCC through pharmacologic inhibition of histone methyltransferases, both osteopontin and EGFR lost their ability to regulate DSB induction (Fig. 3H; Supplementary Fig. S4D). In complete agreement with these data, osteopontin depletion led to radiosensitization only if MLCC was not disrupted (Fig. 3I). Finally, we did not find that inhibition of DNA-PKcs, PI3K, or MEK resulted in KRASmut-specific radiosensitization (Supplementary Fig. S4E and S4F). Together, these data indicate that the radiation resistance of KRASmut NSCLC cells is promoted by SPP1/osteopontin through EGFR-dependent MLCC, which represses DSB induction.

To better understand the importance of SPP1 across different genomic contexts, we compared SPP1 expression in a panel of NSCLC cell lines (Fig. 4A) as well as human lung cancers (Supplementary Fig. S5A). Unexpectedly, we did not observe a significant difference in SPP1 expression between KRASmut and wt genotypes. We thus hypothesized that functional differences may exist between high SPP1 expression in KRASmut versus KRASwt cells. Indeed, osteopontin depletion by siRNA radiosensitized KRASmut but not wt cells in a chromatin- and EGFR-dependent manner using DSB induction and cell survival as endpoints (Supplementary Fig. S5B and S5C). Together, these data, summarized in Fig. 4B, suggest that osteopontin enhances radiation resistance most strongly in high SPP1-expressing KRASmut cells but not in high SPP1-expressing KRASwt cells.

Next, we asked which factors may support radiation resistance in low SPP1-expressing KRASmut cells. In A549 cells with high SPP1 expression (Fig. 4A), siSPP1 transfection caused statistically significant radiosensitization compared with cells transfected with control siRNA, but this was not seen in SW1573 cells, which have low SPP1 (Fig. 4B and C). Depletion of osteopontin abolished the radiosensitizing effect of erlotinib in A549 cells (Fig. 4C), consistent with the data shown in Fig. 3I. In contrast, erlotinib retained its radiosensitizing effect in osteopontin-depleted SW1573 cells.

Osteopontin has been reported to signal to EGFR through Src family kinases (SFK; ref. 38). We thus hypothesized that SW1573 cells may depend on SFK signaling even though osteopontin did not play a role for radiation resistance. Treatment with a potent inhibitor of SFK, particularly c-Src, indeed achieved radiosensitization similar to erlotinib (Fig. 4C). Furthermore, although osteopontin depletion only suppressed sphere growth and CD133 expression in A549 but not SW1573 cells, SFK inhibition affected these endpoints in both the cell lines (Fig. 4D and E) and disrupted the formation of MLCC (Supplementary Fig. S5D). Taken together, these data suggest that the osteopontin/SFK pathway is epistatic with EGFR signaling in terms of promoting stem-like features and radiation resistance. Therefore, osteopontin overexpression may promote MLCC-dependent radioresistance in some but not all KRASmut NSCLCs.

To better understand the functional link between MLCC and CSC-like phenotype, we surveyed genes repressed in KRASmut compared with wt NCI-H1703 cells and noted a downregulated gene cluster comprised of CRMP1, JAKMIP1, and STK32B (Fig. 5A; Supplementary Fig. S6A). Interestingly, CRMP1, collapsin response mediator protein-1, belongs to the CRMP family and has been implicated in antagonizing tumor invasion and metastasis (39). CRMP1 was significantly repressed in KRASmut versus wt NSCLCs (Fig. 5B). Using ChIP, we demonstrated that H3K9me3 is enriched at the CRMP1 locus in KRASmut compared with wt cells and sensitive to erlotinib treatment, which correlated well with CRMP1 protein levels (Fig. 5C and D). These observations support the idea that MLCC-mediated downregulation of genes such as CRMP1 promotes the KRASmut phenotype.

We next investigated differential gene expression in KRASmut and wt NSCLCs harboring putatively activating alterations of MLCC pathway genes (Supplementary Fig. S6B). Among the most downregulated proteins in KRASmut tumors with an altered MLCC pathway was BIM (BCL2-like 11), a proapoptotic member of the B-cell CLL/lymphoma 2 family of proteins (Fig. 5E; Supplementary Fig. S7A). We confirmed that BIM repression was dependent on intact EGFR signaling both in cells and tumors (Fig. 5F and G; Supplementary Fig. S7B and S7C). Intriguingly, the BIM locus was also enriched with H3K9me3 in KRASmut cells, similar to CRMP1, but not in wt cells (Fig. 5H). To reactivate BIM, we treated KRASmut cells with the BH3 mimetic ABT-263, which releases BIM from complexes with BCL-2 and BCL-xL (40). This treatment preferentially radiosensitized KRASmut spheres relative to wt cells, which already have high BIM levels (Fig. 5I; Supplementary Fig. S7D and S7E).

KRASmut NSCLCs with a putatively activated MLCC pathway and low BIM were significantly enriched for co-occurring genomic alterations in the TP53 and CDKN2A tumor suppressors (Fig. 5J), and furthermore, TP53 mutation trended with increased SPP1 expression (Supplementary Fig. S7F). Finally, the MLCC-positive subset of KRASmut tumors was associated with a worse survival than the remaining KRASmut tumors in a univariate analysis (Fig. 5K). Together, these findings indicate the presence of a KRASmut tumor subset with a stem-like phenotype that confers not only cellular radiation resistance but also worse prognosis.

Our efforts to uncover mechanisms of radiation resistance in KRASmut NSCLC have revealed novel insight into a stem-like phenotype of this aggressive cancer type. Key to our findings is a previously identified, but not well understood, small subpopulation of KRASmut cancer cells that is characterized by condensed chromatin (termed MLCC; ref. 27). Dense chromatin can protect genomic DNA against lethal DSB upon exposure to IR, thereby conferring radiation resistance, but the primary biological function of MLCC is poorly understood (27, 28, 41, 42). Our data support a novel, dual model of MLCC that not only suppresses DSB induction but also serves to repress putative negative regulators of stem-like properties, such as invasion and metastasis (CRMP1, BIM; Fig. 5L). We have connected this CSC-like phenotype to EGFR and osteopontin pathways, which cooperate in a previously unrecognized manner to regulate MLCC and DSB induction.

There exists controversy concerning the concept of CSC (43, 44). We have defined CSC as tumor-initiating cells, demonstrated by the enhanced ability of sphere-derived KRASmut cells to grow as xenografts in nude mice and heterogeneously express CD133 in these tumors (Fig. 2E and F). The MLCC-expressing radiation-resistant subpopulation, which is enriched within the CD133^high fraction of cells (Fig. 2C and D), has shown notable plasticity. For example, single-cell–derived KRASmut populations have the ability to reacquire MLCC expression in a subset of cells (Fig. 2D; Supplementary Fig. S2D and S2E). These observations suggest a dynamic stem-like epigenetic state and argue against the presence of a clonally derived CSC population. Other data lend support to the emerging picture that KRASmut tumors are more stem-like than their wt counterparts and that anchorage-independent growth conditions may reveal KRAS-dependent phenotypes (1, 23–25, 45). Interestingly, the PKC phosphorylation site S181 on KRAS is important for the oncogenic functions of mutated KRAS, and PKC inhibition represses tumorigenesis in KRASmut cancers. In addition, PKCα is required for MLCC expression in KRASmut cells (27). Together, these data strongly suggest that PKCα promotes the stem-like phenotype of KRASmut cancers (Fig. 5L). Osteopontin and SFK have not been previously implicated in promoting the radiation resistance of KRASmut versus wt tumor cells (Fig. 3,4). However, several recent reports highlight their importance for stem-like properties, while the role of EGFR in this regard is less clear (46–50). Osteopontin is an upstream activator of Src and EGFR signaling (38, 47, 51). Our data are thus entirely consistent with the hypothesis that this pathway promotes a stem-like phenotype in KRASmut NSCLC.

Not much is known about the factors that restrict stem-like properties. Chromatin condensation, such as MLCC, may be one way to control gene expression in this regard. This modification was recently found to control mesenchymal differentiation in noncancerous cells (28). Our data indicate that in KRASmut cells, MLCC downregulates CRMP1, a suppressor of invasiveness and metastasis in lung cancer (39, 52). However, it has not been previously linked to KRAS status. We also found an association of BIM repression with a stem-like phenotype (Fig. 5E–H). Interestingly, a recent study suggests that BIM-induced apoptosis may temper breast cancer metastasis by governing the survival of disseminating tumor cells (53). Our data suggest that low BIM levels are associated with stemness and radiation resistance in only a subset of KRASmut NSCLCs, which is enriched for TP53 mutations (Fig. 5J). Similarly, BIM expression in KRASmut NSCLC cell lines was heterogeneous, and low BIM correlated with resistance to MEK/PI3K inhibitors (54). BIM re-expression or pharmacologic release from BCL-2/xL complexes restored treatment sensitivity, similar to our data (Fig. 5I).

Finally, we report a previously unrecognized mechanism of radiation resistance in stem-like NSCLC cells, that is, suppression of DSB induction, a mechanism that stands in contrast to previous data showing enhanced repair of DNA damage in CSCs (18). We highlight two important connotations of this observation. First, our data do not rule out the possibility that other KRASmut tumor subsets may exhibit altered DNA damage response/repair as an alternate mechanism of radiation resistance. Second, our findings, particularly in Figs. 1H and 2D and Supplementary Fig. S2F, raise the interesting possibility that enrichment for CSC-like, MLCC-positive tumor cells could account for at least some of the failures of combined radiotherapy with EGFR-directed targeted agents seen in the clinic. Because multiple factors in addition to EGFR may promote the MLCC-positive phenotype and because this phenotype can arise de novo in cell populations derived from single cells that survive IR (Supplementary Fig. S2D and S2E), we hypothesize that more than one molecular target will need to be inhibited to overcome the radiation resistance of KRASmut tumors in patients. This concept is consistent with the emerging notion that several CSC-associated pathways should be blocked to reduce stem cell load and treatment resistance (50, 55). Dissecting the intertumoral heterogeneity of KRAS-mutated cancers with regard to stemness and sensitivity to DNA-damaging agents should be an important subject of future studies.

J. Whetstine is a consultant/advisory board member for Celgene and Qsonica. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M. Wang, T.S. Hong, M. Krause, M. Baumann, H. Willers

Development of methodology: M. Wang, L. Marcar, Q. Liu, X. Li, M. Krause, H. Willers

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Han, J. Black, Q. Liu, X. Li, L.V. Sequist, R.H. Mak, C.H. Benes, K. Gurtner, M. Krause, M. Baumann, J.X. Kang, H. Willers

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Wang, J. Han, L. Marcar, K. Nagulapalli, R.H. Mak, K. Gurtner, M. Krause, H. Willers

Writing, review, and/or revision of the manuscript: M. Wang, J. Han, Q. Liu, L.V. Sequist, R.H. Mak, C.H. Benes, T.S. Hong, K. Gurtner, M. Baumann, H. Willers

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.X. Kang

Study supervision: M. Wang, H. Willers

Other (design, data generation, discussion): J. Whetstine

This work was supported by Dana-Farber/Harvard Cancer Center SPORE in Lung Cancer, NCIP50 CA090578 to H. Willers, American Cancer Society123420RSG-12-224-01-DMC to H. Willers, UK Wellcome Trust102696 to C.H. Benes, Cancer Clinical Investigator Team Leadership Award awarded by the NCI through a supplement to P30CA006516 to T.S. Hong, Federal Share of program income earned by Massachusetts General Hospital on C06 CA059267, Proton Therapy Research and Treatment Center (T.S. Hong and H. Willers), R01GM097360 to J.R. Whetstine, and an American Lung Association Lung Cancer Discovery Award to J.R. Whetstine.

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.