Targeting Monounsaturated Fatty Acid Metabolism for Radiosensitization of KRAS Mutant 3D Lung Cancer Models

Abstract

Mutations in the KRAS oncogene can mediate resistance to radiation. KRAS mutation (mut) driven tumors have been reported to express cancer stem cell (CSC)-like features and may harbor metabolic liabilities through which CSC-associated radioresistance can be overcome. We established a radiation/drug screening approach that relies on the growth of 3D spheres under anchorage-independent and lipid-limiting culture conditions, which promote stemness and lipogenesis. In this format, we screened 32 KRASmut-enriched lung cancer models. As predicted from published data, CB-839, a glutaminase inhibitor, displayed the highest degree of radiosensitization in KRASmut models with LKB1 co-mutations. Radiosensitization by inhibition of stearoyl-CoA desaturase-1, SCD1, displayed a similar genotype preference though the data also implicated KEAP1 co-mutation and SCD1 expression as potential predictors of radiosensitization. In an isogenic model, KRASmut cells were characterized by increased SCD1 expression and a higher ratio of monounsaturated fatty acids (MUFA) to saturated fatty acids. Accordingly, pharmacological inhibition or depletion of SCD1 radiosensitized isogenic KRASmut but not wild-type cells. The radiosensitizing effect was notably small, especially compared to several DNA repair inhibitors. As an alternative strategy to targeting MUFA metabolism, adding polyunsaturated FAs (PUFA) phenocopied some aspects of SCD1 inhibition, suppressed tumor growth in vivo, and opposed the CSC-like phenotype of KRASmut cells. In conclusion, we report a 3D screening approach that recapitulates clinically relevant features of KRASmut tumors and can be leveraged for therapeutic targeting of metabolic vulnerabilities. Our data highlight pronounced inter-tumoral heterogeneity in radiation/drug responses and the complexity of underlying genomic dependencies.

Introduction

Preclinical and clinical development of molecular targeted agents that have radiosensitizing properties, with or without concurrent administration of chemotherapy, remains of critical importance to the advancement of radiation oncology (1,2). Such drugs may have the greatest clinical utility when deployed against tumors that harbor resistance to radiation or concurrent chemoradiation. In fact, it has become clear that even within a given type of cancer the sensitivity of tumors to radiation and chemoradiation generally varies considerably between patients (3). Several known reasons exist for a tumor exhibiting radioresistance, many driven by tumor genotype (4).

KRAS encodes a GTPase that relays signals from the cell membrane to the nucleus (5). When mutated, KRAS becomes constitutively active and acquires oncogenic properties. KRAS mutations are common in various cancer types, including approximately 30% of non-small cell lung carcinomas (NSCLC). We and others have identified mutant KRAS as a tumor biomarker of radioresistance in NSCLC and other cancers (6–10). Different mechanisms of radioresistance have been reported and may be cancer type dependent (10–12). Recent data indicate that mutations in LKB1 (STK11) and KEAP1, which are common in NSCLC and often co-occur with mutant KRAS, can also promote radioresistance (13–15). Targeting glutaminase (GLS) may be one approach through which the radioresistance of LKB1 or KEAP1 mutant tumors can be overcome (14,15).

The microenvironment of cancers tends to be deficient in nutrients, including exogenous lipids (16,17). Yet lipids are needed for multiple cellular processes and functions including membranes, signaling, energy production, proliferation, and survival. As a result, cancer cells can obtain lipids by de novo lipogenesis, which is typically restricted to hepatocytes and adipocytes. Enhanced lipid metabolism may be essential for the survival, growth, and oncogenicity of cancer stem cells (CSC) (18–22). Increased de novo lipid biosynthesis, lipid storage, and enhanced lipid oxidation are features of CSCs (23). At the same time, CSCs need to be protected against high levels of reactive oxygen species (ROS) that could result, for example, from lipid peroxidation (18,24). Lipid desaturation pathways can protect cancer cells from lipotoxicity and generate unsaturated lipids for growth and proliferation (18). SCD1 is the key rate-limiting enzyme for lipid desaturation and a promising target for cancer therapy (25–30). It is a delta-9 fatty acid desaturase located in the ER membrane, catalyzing saturated fatty acids (SFA) to mono-unsaturated fatty acids (MUFA), which can then be converted to other lipids. SCD1 may be upregulated in CSCs where it is a therapeutic target (18,22,27,28,31–33). SCD1 has also been reported to protect certain cancers from lipid peroxidation and ferroptosis (26,34).

There are several reports of alterations in the fatty acid metabolism of KRAS mutant NSCLC (35–38); however, how these changes could be exploited to overcome tumor radioresistance remains to be established. Furthermore, tumors with mutant KRAS appear particularly prone to expressing stemness (10,39–41). Our prior work suggested that mutant KRAS promotes a CD133-high, CSC-like phenotype of NSCLC cells, which protects genomic DNA from ionizing radiation damage through a chromatin-mediated pathway (10).

We, therefore, set out to interrogate a panel of KRAS mutant and wild-type NSCLC cell lines grown under anchorage-independent and lipid-limiting 3D conditions, recapitulating some aspects of the physiologic dependence of CSC-like cells on de novo lipogenesis. In particular, we were interested in targeting MUFA metabolism for radiosensitization and understanding response heterogeneity and tumor genotype dependencies associated with this approach.

Methods

Cell lines and culture

All NSCLC cell lines were selected from the cell line collection in the MGH Center for Molecular Therapeutics (https://www.cancerrxgene.org/celllines) or purchased from ATCC. Cell line RRIDs are listed in Tab. S1A. NCI-H1703 cells (RRID:CVCL_1490) harboring wild-type KRAS with or without exogenous expression of KRAS-G12V were described previously (12). Kras-G12D Tp53 mutant murine sarcoma cells (clone 1345) were obtained from Tyler Jacks and David Kirsch (42). Cell cultures were passaged for less than 3 months after thawing an individual frozen vial, and all cell lines were reauthenticated using short tandem repeat analysis (Genetica Labcorp, Burlington, NC). No cell line tested positive for mycoplasma (MycoAlert^® Mycoplasma Detection Kit, Lonza Basel, Switzerland). Cells were cultured in 2D monolayer or as 3D spheres as described (10,12). Under 3D conditions, no lipid-containing serum was added.

RNA interference

SCD1 siRNA was purchased from Ambion Life Technologies (Cat# AM16704, siRNA ID:15069). Sense sequence (5’->3’): GCAAGU AGAUUGUCUCGAGtt. Antisense: CUCGAGACAAUCUACUUGCtc. NCI-H1703 cells were transfected with siSCD1 or a scrambled control using X-tremeGENE^™ siRNA Transfection Reagent (Roche) following the manufacturer’s guidelines.

Treatments

Stearoyl-CoA Desaturase 1 (SCD1) inhibitor (MF438, Cat #569406) and CAY10566 (Cat #HY-15823) were purchased from EMD Millipore, Darmstadt, Germany and MedChemExpress, Monmouth Junction, NJ, respectively (see Tab. S2 for all chemical structure references). Telaglenastat (CB-839, Cat #S7655) was purchased from SelleckChem, Houston, TX. Auranofin (Cat #A6733), Cis-5,8,11,14,17-Eicosapentaenoic acid (EPA, Cat #44864), Cis-4,7,10,13,16,19-Docosahexaenoic acid (DHA, Cat #D2534), and Arachidonic acid (AA, Cat #A3611) were purchased from Sigma Aldrich (RRID:SCR_008988). Drugs were aliquoted and stored according to the manufacturer’s instructions. NSCLC cell lines were screened with radiation/drug combinations in 384- well black/clear round bottom ultra-low attachment spheroid microplates (Corning #3830) as described (2). Drug concentrations were curated from the institutional database in the Center for Molecular Therapeutics following our published guidelines (2). MF438, CAY10655, CB-839, and Auranofin were used at four drug concentrations over a range of 1.5 logs with the highest concentration (D1) being 0.1 or 3.3 , 1, 1, and 2 μM, respectively. For comparison, DNA repair inhibitors were seeded in the same format: M3814 (against DNA-PK, #HY-101570, MedChemExpress, D1=0.2 μM), AZD1390 (ATM, Cat #S8680, SelleckChem, D1=3.3 nM), olaparib (PARP1/2, Cat# O-9201, LC Laboratories, Woburn, MA, D1=10 μM), Berzosertib (VE-822, M6620) (ATR, Cat #S7102, SelleckChem, D1=0.2 μM), Prexasertib (LY2606368) (CHK1, Cat #HY-18174, MedChemExpress, D1=10 nM), Adavosertib (MK-1775) (WEE1, Cat #S1525, SelleckChem, D1=1 μM). Dimethyl sulfoxide (DMSO) was used as a control for all drug studies. In other experiments, cells were seeded into 96-well ultra-low adherence microplates and treated with drugs at the indicated concentrations (20 μM for DHA, EPA, AA, and 0.1–1 μM for MF438) or/and radiation. X-ray treatments with 2, 4, or 6 Gy, or sham treatments, were carried out 24 hours after drugging of 3D tumorspheres, as described previously (2).

Cell viability assays

Cell viability was assessed 5 days after irradiation, or 5–6 days after drugging, depending on the experiment. 3D CellTiter-Glo^® (CTG) luminescence assay reagent (Cat #G9683, Promega, Fitchburg, WI) was added to the wells at a ratio of 1:2 (15 μM + 30 μM media for a 384-well plate) or 1:5 (30 μM + 150 μM for a 96-well plate) and luminescence was read with a Spect SpectraMax M5 Microplate Reader (Molecular Devices, San Jose, CA) for 96-well plates or MultiLabel reader, 2104 Envision (Perkin Elmer, Waltham, MA, USA) for 384-well plates. This format was previously validated for detecting radiosensitizing drug effects and benchmarked against the gold standard colony formation assay under 2D monolayer conditions (2,43). Short-term radiosensitization factors (SRF) for 2 Gy or 6 Gy were derived as described (2,43). SRF is the ratio of the fraction of cells (FC) as determined by CTG for IR alone divided by the FC for combined drug+IR effect and corrected for the FC observed for drug alone.

Primary sphere formation assay

Cells were seeded at limited dilution in 96-well clear ultralow adherence round bottom plates (Thermo Scientific) following established protocol (44). Cells were grown for 24 hours, drugged with DMSO control, 0.1 μM or 1 μM MF438, and irradiated with 4 Gy or sham treated. Plates were analyzed 7 days later. Wells were scored for the presence of a viable sphere in a blinded fashion. For each condition, 20 cells/wells were seeded.

Lipid peroxidation

Exponentially growing cells were treated in 6-well plates with DMSO control, positive control (RSL3, 5 μM, MedChem Cat #HY-100218A), or MF438 (1 μM). After 3 hours, cells were stained with 1.5 μM of C11-BODIPY (581/591) (Invitrogen, Cat #D3861) and 1 μg/mL Hoechst in PBS (Thermo Scientific, Cat #62249) using the manufacturer’s protocol. Images were taken on Nikon Eclipse Ti2, and analysis was performed using NIS-Elements AR. Hoechst stain was used as a binary to locate the cells. The average green and red fluorescent signals were exported for each binary. The oxidation ratio was calculated by the average green fluorescent signal (oxidized) divided by the average red fluorescent signal (non-oxidized).

Immunofluorescence microscopy

Staining and visualization of γ-H2AX foci was performed 24 hours after irradiation with 6 Gy with anti-phospho-histone H2A.X Ser139 antibody (RRID:AB_309864) (clone JBW301, Cat #05–636, Sigma Aldrich) and goat anti-mouse IgG (H+L) highly cross-adsorbed secondary Ab, Alexa Fluor^™ 488 (Cat #A-11029, Thermo Fisher Scientific Inc., Waltham, MA) using standard protocol as described (45).

Fatty acid profile analysis

Fatty acid profiles of NCI-H1703 cells (RRID:CVCL_1490) grown under 2D or 3D conditions were analyzed by gas chromatography-mass spectrometry (GC-MS) as described (22,46). Briefly, cells treated for 24 hours with inhibitor or control were subjected to fatty acid methylation by mixing with 1.5 mL hexane and 1.5 mL 14% boron trifluoride/methanol at 100°C for 1 hour. Fatty acid methyl esters were analyzed using a fully automated 6890N network GC system equipped with a flame ionization detector (Agilent Technologies, Palo Alto, California). Individual fatty acid peaks were identified by their relative retention times compared to reference standard (Nu-Chek Prep, Elysian, Minnesota), and the area percentage of each peak was analyzed using Agilent Chemstation.

Gene and protein expression analysis

RNA-seq read counts for NCI-H460 RRID:CVCL_0459, NCI-H2122 RRID:CVCL_1531, NCI-H1944 RRID:CVCL_1508 (sensitive) and NCI-H23 RRID:CVCL_1547, HCC44 RRID:CVCL_2060, A549 RRID:CVCL_A549 (resistant) were obtained from the DepMap Cancer Cell Line Encyclopedia (https://depmap.org/portal/ccle/). Differential expression analysis was performed using the DEseq2 R package (RRID:SCR_015687) from Bioconductor (47). This package performs the Wald test which tests individual coefficients, or contrasts of coefficients, without the need to fit a reduced model. The Wald test P values were adjusted for multiple testing using Benjamini and Hochberg’s procedure. Differentially expressed genes were pre-ranked from most significant upregulated genes to most significant downregulated genes using the equation sign(log2FoldChange)*(-log10(padj)). A fast pre-ranked gene set enrichment analysis was performed using the FGSEA R package from Bioconductor (RRID:SCR_020938). Hallmark gene sets were obtained from the Molecular Signatures database (MSigDB). Reverse-phase protein array data for SCD1 were also obtained through the DepMap Cancer Cell Line Encyclopedia (48).

Western blotting

Whole-cell lysates were prepared using standard methods. Specific antibodies against SCD1 (RRID:AB_2183099) (1:500 dilution, Cell Signaling rabbit mAb #2794), beta-actin (RRID:AB_1903890), and tubulin (RRID:AB_10836184) (Cell Signaling) were used. Immunocomplexes were visualized by enhanced chemiluminescence detection using the ChemiDoc Imaging System (Bio-Rad).

Animal experiments

All animal procedures were carried out with approval from the Institutional Animal Care and Use Committee of Massachusetts General Hospital. Female 6- to 8-week old wild-type C57Bl/6 mice (RRID:CVCL_1490) and fat-1 littermates (RRID:IMSR_JAX:020097) on a C57BL/6 background were used for tumor transplantation and growth studies as described previously (49). Transgenic mice were engineered to carry the C. elegans fat-1 gene, which converts ω-6 to ω-3 fatty acids, resulting in a ratio of ω-6/ω-3 close to 1:1 in all mouse tissues (50). 1345 tumor cells were injected at varying numbers into the flanks of mice and tumor initiation and growth were monitored using an established protocol (49).

Data access statement

The datasets used in this report are included in the supplemental file and available from the corresponding author upon reasonable request.

Results

Radiosensitizing effects of metabolic inhibitors across genomically diverse 3D NSCLC models

We examined the radiosensitizing properties of the specific SCD1 inhibitor MF438 in a previously benchmarked plate format (2). Human NSCLC cell lines were seeded into ultralow adherence wells and grown as 3D spheres under lipid-limiting conditions (Fig. 1A). Plates were drugged with MF438, irradiated or sham treated 24 hours later, and SRF values were calculated as described (2). We observed considerable variation of radiosensitization effects across 32 sphere models (Fig. 1B, Tab. S1A). Interestingly, for models with KRAS mutations, there was a stronger radiosensitization signal (median SRF=1.082) than for wild-type models (median SRF=1.032, p=0.0003) (Fig. 1C). Furthermore, KRAS/LKB1 (KL) double mutants were more radiosensitized (median SRF=1.114) than KRAS mutant and LKB1 wild-type models (median SRF=1, p=0.0005).

Figure 1.

Screening of NSCLC 3D models with radiation and metabolic inhibitors. A, Illustration of combined radiation/drug treatments using duplicate ultralow adherence 384-well plates for irradiation with 2 Gy X-rays or sham treatment (tx). B, Screening results for three metabolic inhibitors across a panel of 32 NSCLC models grown as 3D spheres under lipid-limiting conditions. Y-axis shows short-term radiosensitization factors (SRF). Individual data points represent SRF values obtained for four different drug concentrations, with four technical replicates per concentration, based on two biological repeat experiments. Data points associated with <0.5 drug viability were excluded. For each model, median SRF, interquartile range, and range of all replicates are shown. C, Comparison of pooled SRF technical replicates from Panel B as a function of tumor genotype. Horizontal lines represent median values. Statistical comparison with Mann-Whitney test, two-sided. *p≤0.05, **p≤0.01, ***p≤0.001, ****p≤0.0001; ns, not significant. mut or m, mutant; wt, wild-type.

For comparison, we also treated the 3D models with the GLS inhibitor CB-839, an agent in the NCI CTEP portfolio. CB-839 was previously shown to preferentially radiosensitize LKB1-mutant NSCLC, while the potential impact of KRAS co-mutations was not established (15). Again, we observed inter-tumoral heterogeneity of radiosensitization effects though the pattern across the models was somewhat different from MF438 (Fig. 1B). Similar to MF438, the most pronounced radiosensitization signal was seen in KL mutants (Fig. 1C). However, the effect was not uniform across these mutants, and conversely, individual models with other genotypes were also susceptible to CB-839 (Fig. 1B). In contrast, the pro-oxidant drug auranofin had generally weaker effects with only a statistical trend toward stronger radiosensitization of KRAS-mutant models (Fig. 1C). Taken together, the data highlight the utility of large cell line panels for a better understanding of heterogeneous drug effects on radiosensitization and potentially linking them to tumor genotype.

Confirmation of SCD1 inhibitor effects and further correlation with tumor genotype

To corroborate the observed radiosensitization by MF438 and gain further insight into potential genotype associations, we first sought to reproduce drug effects in a limited number of models using a primary sphere formation assay (Fig. S1A). We show evidence of radiosensitization for two of three different genotypes albeit for KRAS mutant A427 cells the effect was small (Fig. 2A). Next, we compared the effects of MF438 with another SCD1 inhibitor, CAY10566, in these cell lines and found them to be virtually identical (Fig. 2B). Lastly, the effects of CAY10566 in an expanded panel of NSCLC cell lines resembled MF438 responses, with the strongest radiosensitization signal seen in models with KRAS and LKB1 co-mutations (Fig. 2C).

Figure 2.

Genotype correlation of SCD1 inhibitor effects. A, Primary sphere formation assay using limited dilution in four representative cell lines from Fig. 1B. Cells seeded at the density shown were irradiated with 4 Gy and exposed to inhibitor at “high” (1 μM) and “low” (0.1 μM) concentrations. Shown is a representative replicate from up to three biological repeats. B, Radiosensitization factors (SRF) for the same four cell lines as in Panel A using two SCD1 inhibitors in parallel (MF438 over a dose range of 100 nM - 3.3 μM and CAY10655 over 30 nM - 1 μM). For each model, median SRF, interquartile range, and range of all replicates from two biological repeats are shown. C, Comparison of pooled SRF technical repeats for 27 NSCLC cell lines grown in 3D, analogous to Fig. 1C. D, Cell viability after treatment with MF438 (D4, D3, D2, and D1 being 3, 10, 33, and 100 nM, respectively). Data points represent median viability for 8 technical replicates per drug concentration based on two biological repeats. Statistical comparison by Fisher Exact, two-sided. *p≤0.05, **p≤0.01, ***p≤0.001, ****p≤0.0001; ns, not significant.

We considered that treatment relevant co-mutations are common in lung adenocarcinoma and that both LKB1 and KEAP1 can impact SCD1 function (34). Indeed, out of 13 KEAP1 mutations present in our primary NSCLC panel, 9 co-occurred with LKB1 and 6 cell lines were triple mutants for KRAS, LKB1, and KEAP1 (Fig. 1B). Accordingly, the presence of KEAP1 mutations correlated with SCD1 inhibitor mediated radiosensitization (Fig. 2C, Fig. S1B). At the same time, the contribution of mutant LKB1 to radiosensitization was numerically small in isogenic cell pairs and not statistically significant (Fig. S1C). Notably, only 3 of the 6 triple mutants were clearly radiosensitized and none of the LKB1/KEAP1 double mutant and KRAS wild-type cell lines (Fig. 1B), demonstrating the complexity of any underlying genotype association and highlighting the need for additional biomarkers to predict drug effects across genomically diverse tumor models.

Lastly, we examined whether the radiosensitizing effects of MF438 correlated with drug alone toxicity across our 3D models (Fig. 2D, Tab. S1B). MF438 conferred relatively little loss of cell viability (< 0.5) in the majority of models. Only 7/32 (22%) models displayed hypersensitivity at higher drug concentrations, with no apparent association with KRAS or LKB1 genotype. Further, there was no significant correlation between drug alone effects on cell viability and radiosensitizing drug effects (Fig. S2A), suggesting distinct roles for SCD1 in response to radiation vs promotion of 3D cell growth without external stressors.

Expression and activity of regulators of fatty acid metabolism in KRAS-mutant NSCLC

As MF438-mediated radiosensitization was more pronounced in non-isogenic KRAS-mutant cell lines, we next sought to further investigate this association in an isogenic KRAS model. We selected a cell line, NCI-H1703, that was not radiosensitized at baseline but showed sensitivity to the drug alone (Fig. 1B, 2D). NCI-H1703 spheres harboring endogenous wild-type KRAS exhibited increased MF438 sensitivity upon expressing KRAS-G12V compared to an empty vector transfected clone (10,12) (Fig. 3A). As SDC1 is a regulator of ferroptosis, we measured lipid peroxidation levels indicative of ferroptosis using C11-BODIPY staining (Fig. 3B). This showed increased susceptibility of KRAS mutant cells to SCD1 inhibition suggesting a greater reliance on SCD1 activity in mutant compared to wild-type cells. Accordingly, we observed the highest SCD1 protein expression in the presence of mutant KRAS (Fig. 3C, Fig. S2B). KRAS mutant cells were particularly susceptible to SCD1 inhibition when grown in 3D lipid-limiting conditions compared to 2D (Fig. S2C).

Figure 3.

SCD1 activity and expression in KRAS mutant and wild-type NSCLC models. A, Cell viability after MF438 treatment of isogenic NCI-H1703 cells with endogenous wild-type (wt) KRAS or with exogenous expression of mutant (mut) KRAS in 3D. Data points represent mean +/− SE of 8 replicates. B, Measurement of lipid peroxidation with C11 BODIPY 581/591 in NCI-H1703 cells following treatment with negative control (Ctrl, DMSO), ferroptosis activator (RSL3, 5 μM), or SCD1 inhibitor MF438 (1 μM). Left, representative images. Right, bars represent mean + SE normalized oxidation ratio for two to three biological repeats. C, Expression levels of SCD1 in 2D versus 3D cultures of isogenic NCI-H1703 cells with/without mutant KRAS. D, Left, Representative images showing individual fatty acid peaks as determined by GC-MS. Right, Quantification of fatty acid peaks plotting the ratio of SFA to MUFAs for different genotypes and culture conditions of isogenic NCI-H1703 cells. Bars represent mean + SE based on three measurements. E, SCD1 expression levels comparing the most strongly radiosensitized (S) models in Fig. 1B (NCI-H460, NCI-H1944, NCI-H2122, NCI-H441, A427, LU65 NCI-H2023, NCI-H1651, NCI-H2126) to the remaining models (R). Data points correspond to individual cell lines with KL double mutants highlighted. F, Fast pre-ranked gene set enrichment analysis comparing radiosensitized (S) versus resistant (R) models with KL mutations. Normalized enrichment scores for Hallmark pathways with adjusted p≤0.05 are shown. See also Fig. S3. All statistical comparisons by t-test or Mann-Whitney test, two-sided; *p≤0.05, **p≤0.01, ***p≤0.001, ****p≤0.0001.

The levels of MUFAs are generally markedly increased in cells cultured in lipid-depleted conditions (51). SCD1 synthesizes 16:1 (palmitoleate) and 18:1 (oleate) MUFAs from the 16:0 (palmitate) and 18:0 (stearate) SFAs, respectively (see Fig. S2D for illustration). Accordingly, we found that the 16:0/16:1 and 18:0/18:1 ratios, which are surrogates of SCD1 activity, were significantly altered in KRAS mutant NCI-H1703 cells with the greatest effect seen under 3D conditions (Fig. 3D). Therefore, SCD1 expression and activity are enhanced in the presence of mutant KRAS in this tumor model.

However, this observation could not be scaled to a larger panel of non-isogenic cell lines where SCD1 expression did not correlate with KRAS status (Fig. S2E), suggesting that KRAS mutation itself is an insufficient predictor across different genomic backgrounds. Indeed, NSCLC cell lines most radiosensitized by MF438 (Fig. 1B) displayed the highest levels of SCD1 gene (p=0.09) and protein expression (p=0.04) independent of KRAS status (Fig. 3E). Even among the six KL double-mutant cell lines, SCD1 protein expression trended with radiosensitization (p=0.1) (Fig. 3E). Furthermore, gene set enrichment analysis revealed significant upregulation of components of fatty acid metabolism in these radiosensitized cell lines, including fatty acid synthase (FASN) which is upstream of SCD1 (Fig. 3F, Fig. S3A–D). Consistent with this observation, we did see higher FAS protein expression in our KRAS mutant NCI-H1703 model (Fig. S4A). In human lung adenocarcinomas, FAS and SCD1 were among the 20 highest differentially upregulated proteins when comparing tumors with KRAS or/and LKB1 mutations to wild-type tumors (Fig. S4B).

Taken together, the available data support the notion that SDC1 inhibition radiosensitizes a subset of non-isogenic NSCLC models that are characterized by a complex interaction of SCD1 expression and tumor genotype, underlining the need for developing combination biomarkers.

Radiosensitization of KRAS-mutant cells in an isogenic 3D model

Given that KRAS-mutant NCI-H1703 cells demonstrated increased SCD1 expression and activity, we next tested whether SCD1 inhibition indeed radiosensitized these cells. Isogenic cells were treated with and without MF438 before irradiation, which revealed a small difference between the KRAS genotypes (Fig. 4A). Radiosensitization by MF438 was also observed when using residual radiation-induced DNA double-strand breaks as a readout (Fig. 4B), consistent with data that accumulating SFAs cause ER stress and impair homologous recombination repair (52). A similar result was observed when depleting SCD1 with validated siRNA (Fig. 4C). Combined inhibition and depletion of SCD1 had no further effects over depletion alone suggesting that sufficient disruption of SCD1 function was achieved (Fig. S4C).

Figure 4.

Targeting SCD1 for radiosensitization of isogenic NCI-H1703 spheres with mutant (mut) or wild-type (wt) KRAS. A, Radiosensitization by MF438 (0.1 μM) using 6 Gy irradiation. Bars represent mean SRF +/− SEM based on five biological repeats. B, Left, representative images of γ-H2AX foci 24 hours following irradiation with 6 Gy with and without MF438 treatment. Right, Fractions of cells with at least 20 foci per nucleus. Bars represent mean + SE based on three biological repeats. C, Left, Western blot with an anti-SCD1 antibody for cells treated with scrambled (scr) or validated siRNA. Relative SCD1 band intensities are displayed. Right, Radiosensitization by siRNA against SCD1 using 6 Gy irradiation. Bars represent mean SRF +/− SE based on two repeat experiments. D, Radiosensitization by small molecule inhibitors of the targets indicated. Data represent median, interquartile range, and range for 32 pooled replicates across four different drug concentrations and two biological repeats. MF438 data are included for reference. Statistical comparisons by t-test or Mann-Whitney test, two-sided; *p≤0.05, **p≤0.01, ***p≤0.001.

Lastly, we wanted to compare the relatively small degree of radiosensitization by MF438 in this model (SRF=1.08, Fig. 4A) to other established radiosensitizing agents. Notably, several DNA repair inhibitors, as well as CB-839, appeared to have more potent radiosensitizing effects than MF438 (Fig. 4D). The only other KRAS-mutant specific radiosensitizer identified was an inhibitor of CHK1 kinase, which we previously implicated in KRAS mutation mediated radioresistance in this model (53). These data suggest that more impactful perturbations of MUFA metabolism might be needed for clinical translation.

Targeting MUFA metabolism with ω-3 PUFA

MUFAs are considered an abundant fatty acid species under the selected 3D sphere growth conditions, and because they are relatively less susceptible to lipid peroxidation their abundance should restrict the detrimental effects of ROS and support cell proliferation and viability (23). We reasoned that supplementation with polyunsaturated fatty acids (PUFA), which are more prone to lipid peroxidation, may at least partially phenocopy the effects of SCD1 inhibition (Fig. S5A). Supplementation with PUFAs mildly suppressed SCD1 protein expression, and ω-3 PUFA potentially increased SFA levels in KRAS mutant spheres (Fig. S5B,C). Correspondingly, a clear effect on cell viability was observed for KRAS mutant compared to wild-type tumor cells (Fig. 5A). Adding SCD1 inhibition did not enhance this effect, suggesting already maximum perturbation of MUFA metabolism. Different PUFAs elicited distinct responses for some endpoints, with ω-3 PUFA but not ω-6 affecting SCD1 activity and cell viability (Fig. 5A, Fig. S5C,D). Interestingly, ω-3 selectively increased cellular ROS levels but not lipid peroxidation (Fig. S5E,F).

Figure 5.

Targeting MUFA metabolism with PUFAs in KRAS mutant (mut) and wild-type (wt) NSCLC models. A, Upper, illustration of effects of different treatments (ctrl, control; EPA, eicosapentaenoic acid; DHA, docosahexaenoic acid at 20 μM each) on the growth of NCI-H1703 spheres. Lower, fraction of cells measured by CTG assay after 5-day treatment of spheres with the drugs indicated relative to untreated control. B, Growth of Kras/Tp53 mutant 1345 sarcoma allografts in fat-1 transgenic mice versus wt littermates. Data points represent mean +/− SE based on five animals in each group. C, Radiosensitization of isogenic NCI-H1703 spheres by EPA, DHA or AA (arachidonic acid) at 20 μM each. Bars represent mean + SE based on three to four biological repeats. D, Fraction of NCI-H1703 cells with 20 or more foci 24 hours after 6 Gy irradiation with or without EPA treatment. E, Radiosensitization of NSCLC cell lines grown as 3D spheres by PUFA treatments as indicated. Bars represent mean + SE based on three biological repeats. Statistical comparisons by t-test, two-sided; *p≤0.05, **p≤0.01, ***p≤0.001, ****p≤0.0001.

As MUFAs can support stemness (23), we also asked whether ω-3 supplementation would impact the reported CSC-like phenotype of KRAS mutant models (10). Indeed, we observed a marked reduction of CD133 expression (Fig. S6A) and suppression of a previously described chromatin stem cell marker similar to loss of SCD1 function (Fig. S6B,C). Lastly, in a transgenic mouse model made to endogenously produce ω-3 (fat-1 mouse (50)), there was significantly reduced growth of Kras mutant tumors by day 35 compared to a reduced presence of ω-3 (Fig. 5B, Fig. S7A). Notably, ω-3 did not suppress the tumor take rate compared to controls without ω-3 (Fig. S7B).

Next, ω-3 but not ω-6 PUFA supplementation radiosensitized KRAS-mutant spheres more than wild-type spheres in the NCI-H1703 model (Fig. 5C). However, this was not observed when treating 2D monolayer cultures (Fig. S7C). Preferential radiosensitization of KRAS mutant spheres by ω-3 correlated with increased levels of unrepaired DSB (Fig. 5D). Interestingly, treatment with both ω-3 and MF438 was more radiosensitizing than either treatment alone but we did not pursue this further (Fig. S7D). When scaling the observed radiosensitization by ω-3 to additional non-isogenic NSCLC models, the association with mutant KRAS was less robust (p=0.07) (Fig. 5E). Again, these data point toward the heterogeneity of treatment responses and the need for biomarkers that go beyond just KRAS status.

Discussion

It is not established how many cancer cell line models are required to characterize the in vitro radiosensitizing properties of a given targeted agent. A minimum number of two cell lines has been recommended, which may be appropriate for studying DNA repair inhibitors (1,54,55). However, concerning small molecule inhibitors targeting cancer metabolism, we found considerable variation in radiosensitizing effects across an unbiased panel of 32 NSCLC models when targeting SCD1, GLS, or the cellular redox system (Fig. 1B). Interestingly, the majority of models were not or only weakly radiosensitized in this screening format. For both SCD1 and GLS inhibitors, the strongest radiosensitization signal was seen among models with triple co-mutations in KRAS, LKB1, and KEAP1, although not every model was radiosensitized. An additional predictor of radiosensitization was SCD1 expression in both isogenic and non-isogenic comparisons (Fig. 3C,E). These observations highlight clinically relevant heterogeneity in drug responses and the potential complexity of underlying genomic dependencies. The data also point towards the utility of larger cell line panels to adequately capture this heterogeneity and correlate responses with combination genomic biomarkers.

While we did not perform comprehensive comparisons between 2D and 3D cell culture conditions, our findings suggest that the impact of fatty acid metabolism on cancer cells may be best performed with cells grown as spheres under lipid-limiting conditions that mirror the limited nutrient supply in the tumor microenvironment to some extent. Exogenous lipid restriction has been shown to promote activation of the SREBP1 transcription factor, which regulates SCD1 and de novo fatty acid synthesis in cancer cells (51). Accordingly, cells grown under lipid-limiting conditions are more sensitive to pharmacologic inhibition of SCD1 (Fig. S2C) (51).

We found the highest radiosensitization signal in NSCLC models with KL mutations (Fig. 1B, Fig. 2C). Similarly, A549 cells, which are KL mutant, were radiosensitized by CB-839 in 2D culture while an A549 clone with reconstituted wild-type LKB1 was not (15). Loss of LKB1 function is associated with a wide range of metabolic perturbations (56). Interestingly, AMPK downstream of LKB1 has been reported to inhibit SREBP1 (57), suggesting that LKB1 loss could indirectly affect SCD1 function. It is, therefore, not entirely surprising that for SCD1 inhibition the strongest radiosensitization signal was also seen in KL mutant tumor models, comparable to CB-839 (Fig. 1B,C), though we did not analyze the underlying mechanisms in greater detail. However, LKB1 mutation did not always associate with radiosensitization (Fig. 1B, 2A, Fig. S1C). Furthermore, LKB1 and KRAS wild-type NCI-H1703 cells could be rendered sensitive to MF438/radiation treatment upon exogenous expression of mutant KRAS (Fig. 4A, B). It is possible that these cells harbor a baseline metabolic alteration that facilitated this phenotype. Notably, wild-type NCI-H1703 was one of the few cell lines outside the KL genotype that was susceptible to CB-839 and radiation, perhaps suggesting altered LKB1 function at baseline.

Because the in vitro radiosensitizing effects of MF438 were overall of limited magnitude, especially when compared to some DNA repair inhibitors in the same model (Fig. 4D), we sought to explore other means of targeting MUFA metabolism for radiosensitization. Omega-3 (ω-3 PUFA) is an essential fatty acid that requires an exogenous supply. It affects lipid metabolism in multiple ways including a reported inhibition of SCD1 (22,58). While we observed a mild reduction of SCD1 protein expression (Fig. S5B) we note that PUFA supplementation may impact cellular MUFA utilization and that PUFAs are more prone to lipid peroxidation (23). However, unexpectedly, SCD1 inhibition affected lipid peroxidation more strongly than PUFAs (Fig. 3B, S5F). Consistent with the described anti-cancer effects of ω-3 we observed growth suppression of Kras mutant tumors in mice engineered to express high levels of ω-3 (Fig. 5B) (50). However, we acknowledge that this anti-tumor effect of ω-3 may not be KRAS mutation specific as similar observations have been made with other tumor types (22,49). Consistent with other data (22), ω-3 suppressed CSC features in vitro though the tumor initiation rate was unaffected (Fig. S6A–C, Fig. S7B). Lastly, the difference of ω-3 versus ω-6 effects across some assay readouts was notable (Fig. 5C,E, Fig. S5D,E). It is thought that the ω-6 AA and its eicosanoid metabolites promote cell proliferation and tumor growth (59). ω-3 may suppress the production of AA-derived eicosanoids, as well as other aspects of lipogenesis in cancer cells, thereby exerting anti-proliferative and anti-stemness effects (22), but more studies will be needed to elucidate these functions in lung cancer.

Our study has several limitations. We have not been able, with the cell line models available to us, to develop robust biomarker combinations for a prediction of drug effects. However, our data highlight the underlying complexity which will inform future investigations. Shared mechanisms of radiosensitization across cell lines remain to be investigated. We also have not confirmed radiosensitization effects in vivo; however, it is unclear whether the relatively modest effect size seen in vitro would translate well in vivo, and it also remains unclear whether results seen in the isogenic cell line model are scalable. We acknowledge that our chosen xenograft model, which was arbitrarily restricted to female mice, is a suboptimal model for studying cancer metabolism but we note that the findings mirrored our in vitro observations.

We believe our data highlight opportunities for future investigations into exploiting tumor genotype-correlated metabolic vulnerabilities and their impact on radiosensitivity. Systemic disruptions of lipid metabolism or other metabolic processes can have considerable toxicity, which has limited the clinical translation of metabolic inhibitors to date (30,60,61). We speculate that combining drugs such as SCD1 inhibitors with radiation may be a viable alternative strategy because lower, less toxic drug concentrations may be sufficient for radiosensitization especially when there is increased target dependence, such as in radioresistant CSC-like cells harboring mutant KRAS. Additionally, there may exist avenues to modify radiation responses through dietary modification including lipid supplements (22,51). Future studies in this space are warranted.