AXL, a TAM (TYRO3, AXL, and MERTK) family receptor tyrosine kinase, is increasingly being recognized as a key determinant of resistance to targeted therapies, as well as chemotherapy and radiation in non–small cell lung cancer (NSCLC) and other cancers. We further show here that high levels of AXL and epithelial-to-mesenchymal transition were frequently expressed in subsets of both treatment-naïve and treatment-relapsed NSCLC. Previously, we and others have demonstrated a role for AXL in mediating DNA damage response (DDR), as well as resistance to inhibition of WEE1, a replication stress response kinase. Here, we show that BGB324 (bemcentinib), a selective small-molecule AXL inhibitor, caused DNA damage and induced replication stress, indicated by ATR/CHK1 phosphorylation, more significantly in TP53-deficient NSCLC cell lines. Similar effects were also observed in large-cell neuroendocrine carcinoma (LCNEC) cell lines. High AXL protein levels were also associated with resistance to ATR inhibition. Combined inhibition of AXL and ATR significantly decreased cell proliferation of NSCLC and LCNEC cell lines. Mechanistically, combined inhibition of AXL and ATR significantly increased RPA32 hyperphosphorylation and DNA double-strand breaks and induced markers of mitotic catastrophe. Notably, NSCLC cell lines with low levels of SLFN11, a known predictive biomarker for platinum and PARP inhibitor sensitivity, were more sensitive to AXL/ATR cotargeting. These findings demonstrate a novel and unexpected role for AXL in replication stress tolerance, with potential therapeutic implications.
These findings demonstrate that the combination of AXL and ATR inhibitors could be a promising therapeutic combination for NSCLC, LCNEC, and other cancers.
AXL, a TAM (TYRO3, AXL, and MERTK) family receptor tyrosine kinase, has emerged as a key determinant of therapeutic resistance in multiple cancers, including non–small cell lung cancer (NSCLC; ref. 1). AXL overexpression has been shown by our group, as well as others, to confer acquired resistance to targeted therapies, such EGFR, ALK, and BRAF inhibitors (2–6). AXL promotes cell proliferation and survival via the RAS/MEK/ERK and the PI3K/AKT/mTOR pathways. AXL signaling has also been shown to induce invasion and migration (1), is strongly associated with a mesenchymal phenotype, and has been shown to drive epithelial-to-mesenchymal transition (EMT), which is also associated with therapeutic resistance (3, 7, 8). Several small-molecule AXL inhibitors and anti-AXL biologics are currently being investigated as monotherapies and in combination with targeted agents and chemotherapy in clinical trials for NSCLC (e.g., NCT02922777, NCT02729298, NCT02988817, and NCT03425279; refs. 1, 9–11). AXL inhibitors have also demonstrated antiviral effects and are being investigated in clinical trials for SARS-CoV-2 infection.
Replication stress, which often manifests by the slowing or stalling of replication fork progression, is considered an important driver of genomic instability, a hallmark of cancer (12). As a result of aberrant oncogenic stimulation and/or loss of cell-cycle checkpoints, cancer cells undergo increased replication of unrepaired DNA, which results in elevated replication stress. Several oncogenic alterations prevalent in NSCLC tumors, such as KRAS mutations, STK11/LKB1 loss, and MYC amplifications, have been shown to induce replication stress (13, 14). Specifically, about 40% of lung adenocarcinomas and 10% of squamous cell lung carcinomas, the two predominant histologic subtypes of NSCLC, have alterations in one or more of these genes (15–17). To alleviate replication stress, cancer cells rely on an ATR/CHK1-mediated replication stress response, which prevents collapse of stalled replication forks and premature restart of aberrant replication, both of which result in mitotic catastrophe and cell death. Given that more than 50% of NSCLC tumors have alterations in TP53 or a DNA damage response (DDR) gene (e.g., ATM, PRKDC, FANCM, POLE, and BRCA; ref. 17), NSCLC tumors are highly reliant on the ATR–CHK1 axis. Similarly, another lung cancer subtype, large-cell neuroendocrine carcinoma (LCNEC), also has a high prevalence of TP53 inactivation (92%; ref. 18), and frequent cooccurring alterations in genes associated with replication stress, such as KRAS, NRAS, STK11, or RB1 (18), suggesting that LCNEC tumors may also be dependent on the ATR–CHK1 pathway. Elevating intrinsic replication stress levels, as well as targeting the replication stress response pathways, have emerged as potential therapeutic approaches for specific subsets of NSCLC. Replication stress response inhibitors, such as ATR and CHK1 inhibitors, in combination with radiotherapy and chemotherapeutic agents or other DNA-damaging agents are in early clinical investigations in NSCLC (e.g., NCT02589522, NCT02264678, NCT01139775, and NCT02797964). Standard cancer treatments, such as chemotherapy and radiotherapy, have also been shown to induce replication stress by causing DNA damage (12, 19).
Increased AXL expression has been observed in response to both chemotherapy and radiotherapy, and is associated with resistance to these treatments (20–22). Conversely, AXL inhibition promotes sensitivity to chemotherapy and radiation (22). On the basis of these findings, we hypothesize that elevated AXL expression may play a role in tolerance of replication stress inherent to lung cancer cells or in response to treatment. Previously, our group and others have demonstrated an unexpected role for AXL in DDR (8, 23), where AXL knockdown and inhibition induced DNA damage and impaired the efficiency of homologous recombination–mediated DNA repair (8). AXL inhibition in combination with a PARP inhibitor (olaparib) and a CHK1/2 inhibitor (AZD7762) has also been reported to result in synergistic cell death (8, 24). However, the effects of AXL inhibition on the replication stress response pathway are not well understood.
In this study, we found that AXL, which is associated with resistance to therapy, is expressed at higher levels in subsets of treatment-naïve NSCLC and LCNEC tumors, as well as treatment-resistant, relapsed NSCLCs. We also examined the effects of AXL inhibition with the selective small-molecule AXL inhibitor, BGB324 (bemcentinib), and AXL knockdown on DNA damage and replication stress in NSCLC and LCNEC cells. BGB324 induced DNA damage and activated the ATR/CHK1 axis, an effect that was more pronounced in a TP53-deficient background. Furthermore, high AXL levels predicted resistance to ATR inhibitors, while the combination of BGB324 and ATR inhibitor was synergistic. Together, these findings demonstrate a novel effect of BGB324 on the ATR/CHK1 axis and indicate the combination of AXL and ATR inhibitors for treatment of TP53-deficient NSCLC and LCNEC.
Materials and Methods
Antibodies for Western blotting were purchased from Cell Signaling Technology: phospho-CHK1 (S345) (2348, 1:1,000), CHK1 (2360, 1:1,000), γH2Ax (9718, 1:1,000), AXL (8661, 1:1,000), Cyclin B1 (4138), phospho-cdc2 (Y15) (4539), phospho-Histone H3 (S10) (53348), phospho-AKT (S473) (4060), AKT (9272), phospho-S6 (S240/244) (2215), S6 (2217), and p21 (2946); Bethyl Laboratories: RPA32 (A300-244, 1:1,000), phospho-RPA32 (S4/S8) (A300-245, 1:1,000), phospho-RPA32 (S33) (A300-246, 1:1,000), and phospho-KAP1 (S824) (A300-246, 1:1,000); and Santa Cruz Biotechnology: β-actin (sc-47778, 1:2,000) and GAPDH (sc-20357, 1:2,000). BGB324 was generously provided by BerGenBio and AZD6738 by AstraZeneca. VX-970 was purchased from Selleck Chemicals. Vinculin (V9131), propidium iodide, and hydroxyurea were purchased from Sigma-Aldrich.
Human NSCLC, LCNEC, and mesothelioma cell lines were obtained from The University of Texas MD Anderson Lung Cancer Moon Shot Program. Genetically engineered mouse model (GEMM)-derived NSCLC cell lines, 344SQ, 344SQpLKO.1shControl, 344SQshAxl-10.1, and 344SQshAxl-12.1, were provided by D.L. Gibbons. Cell lines were cultured in RPMI1640 supplemented with 10% FBS, 100 IU/mL penicillin, and 100 μg/mL streptomycin at 37 °C in a humidified atmosphere of 5% CO2. All cell lines were in early passages and maintained in culture for less than 2 months, human cell lines were authenticated by short tandem repeat profiling and tested regularly for Mycoplasma contamination using MycoAlert Plus (Lonza).
Single-agent and combination viability assays
NSCLC and LCNEC cell lines (2,000 cells/well) were seeded in 96-well white bottom microtiter plates. After overnight attachment, cells were treated with BGB324, ATR inhibitors, or DMSO control at indicated concentrations in triplicate for 120 hours. Cell viability was measured by using CellTiter-Glo (Promega) and luminescence was read on a Synergy HT Microplate Reader (BioTek). For single-drug treatments, dose–response curves were modeled using nonlinear curve fitting and the IC50 drug concentration was estimated using our previously published method (drexplorer software; ref. 25). Replicate reproducibility was determined by concordance correlation coefficient and goodness of fit by residual SE. For drug combination experiments, AUC for the observed effect of the combination was compared with the AUC for the additive effect predicted by the BLISS model. The difference between the two AUCs, denoted by ΔAUC, was computed. ΔAUC ≤ −0.1 was considered to be a greater than additive and −0.1 < ΔAUC < 0.1, an additive effect of the drug combination, on the basis of an estimated 10% margin of experimental variability (26). Combination index (CI) at 50% fraction affected was computed using the Chou–Talalay model (27).
Reverse phase proteomic array
Reverse phase proteomic array (RPPA) was performed as described previously (26).
Cells were seeded in 10-cm dishes and treated as indicated. Cells were washed with ice-cold PBS and lysed with RPPA lysis buffer supplemented with protease and phosphatase inhibitor cocktail. The lysate was centrifuged at 14,000 rpm for 10 minutes to remove cell debris. Total protein concentration of the supernatant was measured using DC Protein Assay Reagent (Bio-Rad). Cell lysate (30 μg) was boiled for 5 minutes at 100°C with 2 × Laemmli buffer, resolved on a 15% polyacrylamide gel, and electroblotted on a nitrocellulose membrane. Membranes were blocked in 1 × Casein Blocking Solution (Bio-Rad) for 1 hour at room temperature and incubated overnight with primary antibodies at specified dilutions at 4°C. Membranes were washed with TBS with 0.1% Tween-20 and incubated with appropriate horseradish peroxidase–linked secondary antibodies for 1 hour at room temperature. The immunoblots were visualized using the SuperSignal West Pico Plus Chemiluminescence Substrate (Thermo Fisher Scientific) on a Bio-Rad ChemiDoc Touch Imaging System. Relative band intensities were quantified using ImageJ software and normalized to loading control.
Human AXL Stealth siRNAs (#HSS100897, #HSS100898, and #HSS183343) and negative control siRNA (#12935112) were purchased from Thermo Fisher Scientific. Cells were transfected using Lipofectamine 2000 (Thermo Fisher Scientific) following the manufacturer’s instructions for 48 hours. For stable knockdown, mouse Axl short hairpin RNA (shRNA) constructs (TRCN0000023310 and TRCN0000023312) were purchased from Horizon/GE-Dharmacon. shRNAs used were expressed in the pLKO.1 puro vector with a scramble sequence as the control. Lentiviruses were generated by cotransfection of Axl shRNA constructs with psPAX2/pMD2.G into 293T cells using Lipofectamine LTX (Thermo Fisher Scientific). Following transduction into 344SQ cells and puromycin selection, Axl knockdown was confirmed by quantitative reverse transcriptase PCR and Western blotting.
Clonogenic survival assay
Calu-1 and H1299 cells were seeded in a 6-well plate at 250 cells per well. After overnight attachment, cells were treated with DMSO, BGB324 (1 μmol/L), VX-970 (1 μmol/L), or their combination for 48 hours. Colonies were allowed to grow in drug-free media for 2 weeks and stained with 0.25% crystal violet.
Calu-1 and H1299 cells (200 cells/well) in 24-well plates were treated with DMSO, BGB324 (1 μmol/L), VX-970 (1 μmol/L), or the combination for 24 hours. Cells were fixed with 4% formaldehyde and permeabilized with 0.03% triton X-100. After blocking with 5% normal goat serum, cells were incubated overnight with γH2Ax Antibody (Cell Signaling Technology, #9718, 1:500) at 4°C, followed by addition of the secondary goat anti-rabbit AlexaFluor-546–conjugated antibody (4 μg/mL; A11010, Thermo Fisher Scientific). Nuclei were stained with DAPI (1 μg/mL, Sigma). Fluorescence microscopy was performed with an Olympus IX73 Microscope System.
A total of 0.25 × 106 cells (Calu-1 and H1299) were plated in a 10-cm dish and after overnight attachment, treated with DMSO, BGB324 (1 μmol/L), VX-970 (1 μmol/L), or the combination for 72 hours. For mitotic release assay, cells were treated with Nocodazole (200 ng/mL; Sigma) and the inhibitors for 24 hours and released into fresh media. Cells were harvested at indicated timepoints, fixed in 70% ethanol overnight at 4°C, and stained with 50 μg/mL propidium iodide and 250 μg/mL RNAase A for 1 hour at 37°C. Cells analyzed were on a LSR Fortessa Flow Cytometer (BD Biosciences) and data were analyzed using FlowJo Software (Treestar).
Intracellular flow cytometry staining of phospho histone H3
Calu-1 cells were plated and treated as described above. Cells were harvested, fixed overnight with Fixation Buffer (BioLegend), and permeabilized with intracellular staining Perm Wash Buffer (BioLegend). Cells were then stained using Alexa Fluor 488 anti-phospho Histone H3 (Ser10) antibody (BioLegend, #613408) and propidium iodide (50 μg/mL) and analyzed on an LSR Fortessa flow cytometer.
H1299 cells were treated with indicated concentrations of BGB324, VX-970, or their combination for 48 hours. Apoptosis was measured using Annexin V-FITC Apoptosis Detection Kit (BD Biosciences), per the manufacturer’s instructions.
Data statistics and bioinformatics analyses were performed using R (version 3.3.0, https://www.r-project.org/) and Bioconductor packages (https://www.bioconductor.org/). Mutation and frequency analysis were done using Fisher exact or χ2 test, as indicated. ANOVA followed by Tukey post hoc test was used to compare across treatment groups. To identify proteins most highly correlated with drug sensitivity or EMT score, we used Spearman rank correlation test. For mRNA and RPPA expression data analyses, Benjamini–Hochberg method was used to control FDR (28).
AXL is overexpressed in subsets of treatment-naïve and relapsed NSCLC tumors
EMT has been implicated as a mechanism of resistance to multiple therapies (29). To determine whether EMT was increased following treatment in NSCLC clinical samples, we examined EMT scores of patient tumors using our previously established 77-gene Pan-cancer EMT signature (30). Using the EMT signature, tumors expressing epithelial genes at higher levels relative to mesenchymal genes in the EMT signature have an EMT score < 0 and exhibit an epithelial phenotype. Conversely, tumors with a mesenchymal phenotype, have an EMT score > 0 (3, 30). In the two treatment-naïve NSCLC clinical cohorts, The Cancer Genome Atlas (TCGA) lung adenocarcinoma (LUAD; ref. 15) and TCGA lung squamous cell carcinoma (LUSC; ref. 16; total n = 1,016 tumors), an average of 31% of the tumors had a mesenchymal phenotype (Fig. 1A). In tumors from patients with treatment-resistant advanced NSCLC, who had received prior systemic treatment, including chemotherapy, radiation, and EGFR inhibitors, but subsequently relapsed (BATTLE-1; ref. 31 and BATTLE-2; ref. 32; total n = 239 tumors), approximately 53% of tumors exhibit a mesenchymal phenotype (Fig. 1B). This statistically significant enrichment of mesenchymal tumors in the treatment refractory cohorts (P < 0.001 by χ2 test) suggests that prior therapy may induce EMT in NSCLC.
As we have previously shown AXL to be both a marker of a mesenchymal phenotype and a driver of EMT (3, 8), we next directly examined AXL expression in the NSCLC tumor cohorts. In the treatment-naïve NSCLC patient cohorts (LUAD and LUSC), we observed a subset of tumors that expressed higher levels of AXL mRNA (Fig. 1A) and AXL protein (Supplementary Fig. S1A and S1B). Furthermore, 48% of LUAD tumors and 66% of LUSC tumors expressing high levels of AXL were mesenchymal (EMT score > 0; Fig. 1C). This suggests that early cotargeting of AXL along with standard therapies could be useful in preventing resistance and improving treatment outcomes in patients with treatment-naïve NSCLC. Similarly, consistent with AXL’s role in therapeutic resistance, a larger subset of relapsed NSCLC tumors from the BATTLE-1 and BATTLE-2 cohorts also expressed high AXL (Fig. 1B). Together, these findings underscore the therapeutic utility of targeting AXL, particularly in treatment-resistant NSCLC.
To further assess the spectrum of EMT and AXL expression across multiple cancers, we examined EMT scores and TCGA gene expression data from 32 different thoracic and extrathoracic malignancies (Supplementary Fig. S2). Among other cancers, some of the highest AXL mRNA expression was seen in mesothelioma, which was also one of the most mesenchymal tumors (Fig. 1A; Supplementary Fig. S2). In the recently published cohort of treatment-naïve LCNEC tumors (18), 29% of tumors had a mesenchymal EMT score, along with high levels of AXL expression in many of these tumors (Fig 1A; Supplementary Fig. S1C). LCNECs are rare pulmonary tumors, accounting for about 3% of lung cancer diagnoses, with genomic similarities to NSCLC (STK11 and NRAS mutations), as well as neuroendocrine features like small-cell lung cancer (SCLC; ref. 18). LCNECs, also similar to SCLC, are primarily treated with platinum-based chemotherapy and lack any specific targeted therapy options (33). Overall, these observations identify subpopulations of patients with higher AXL expression in both treatment-naïve and relapsed NSCLC tumors, and across multiple cancer types, that may benefit from cotreatment with AXL inhibitors.
AXL inhibition by selective small-molecule inhibitor, BGB324, results in DNA damage and ATR/CHK1 activation
BGB324 (bemcentinib, R428) is a selective AXL inhibitor that is currently in phase I and II clinical trials in combination with EGFR inhibition, chemotherapy, and immunotherapy in previously treated advanced NSCLC (NCT02424617, NCT02922777, and NCT03184571; refs. 9, 34). To assess the effect of BGB324 on growth inhibition, we screened a panel of 23 NSCLC cell lines with varying AXL levels, as determined by RPPA, and genetic background in 5-day cell proliferation assays. Both NSCLC adenocarcinoma and squamous cell carcinoma cell lines showed a range of sensitivities to BGB324 (IC50 values range from 0.67 to >9.61 μmol/L; median, 2 μmol/L; Fig. 2A). In addition, we tested BGB324 in a panel of additional lung cancer types, including LCNEC and mesothelioma. Several LCNEC cell lines also showed a similar sensitivity to BGB324 (median IC50, 2.3 μmol/L; Fig. 2A). Next, as oncogenic drivers of NSCLC have differential sensitivities to treatments (35), we looked for associations between these genes and sensitivity to BGB324. However, in our panel of cell lines (Supplementary Table S1), no association between these oncogenes and BGB324 sensitivity was observed.
Next, to test our hypothesis that AXL promotes tolerance of replication stress and DNA damage, we examined markers of replication stress and DNA damage following AXL inhibition by BGB324. In a panel of NSCLC and LCNEC cell lines treated for 24 hours with BGB324, changes in protein expression were analyzed by Western blotting. In TP53-deficient (mutant or homozygous deletion) NSCLC cell lines (H1651 and Calu-1) and the NRAS-mutant/TP53-deficient LCNEC cell line, H1299, treatment with BGB324 caused a dose-dependent accumulation of γH2Ax, a marker of double-stranded DNA breaks (Fig. 2B). We next examined the effects in a mesenchymal murine lung cancer cell line (344SQ) derived from a KrasLA1/+Trp53R172HΔG GEMM (36) and observed a similar increase in γH2Ax accumulation following BGB324 treatment. As with AXL inhibition by BGB324, siRNA-mediated depletion of AXL also resulted in DNA damage (Fig. 2C). This is consistent with our previous findings (8) and further supports a role for AXL in DDR.
Beyond DNA damage, we also observed that BGB324 induced a strong increase in CHK1 phosphorylation at serine 345 in these TP53-deficient cell lines. ATR/CHK1 activation commonly occurs in response to exposed ssDNA, such as during replication stress. Furthermore, hyperphosphorylation of the 32 kDa ssDNA binding replication protein A 2 (RPA2 or RPA32) at serine 4 and 8 (S4/8) was also detected at higher concentrations (Fig. 2B). AXL knockdown also similarly activated CHK1 phosphorylation, although the effect appeared to be variable (Fig. 2C).
To assess the temporal dynamics of early-DNA damage response to AXL inhibition by BGB324, we examined the time course of activation of key DDR pathway proteins in the TP53-deficient Calu-1 and H1299 cells following BGB324 treatment. CHK1 phosphorylation was detected as early as 0.5 hours posttreatment, signaling the onset of replication stress, and was sustained up to 24 hours (Fig. 2D). Phosphorylation of RPA32 (S4/8), however, was observed only at 8 hours following BGB324 treatment and was coincident with increasing γH2Ax accumulation, suggesting possible collapse of stalled replication forks and formation of double-stranded DNA breaks (Fig. 2D; ref. 37). No significant increases in phosphorylation of KAP1, an ATM substrate, were observed (Fig. 2D). AXL phosphorylation, in basal condition, as well as upon ligand (GAS6) stimulation, measured by immunoprecipitation, was also confirmed to be inhibited both at 1 and 24 hours after BGB324 treatment (Fig. 2E; Supplementary Fig. S3A and S3B). As described previously (38), we also noted a dose-dependent increase in total AXL levels following 24-hour treatment with BGB324 (Supplementary Fig. S3C).
Compared with the TP53-deficient cell lines, CHK1 phosphorylation in response to BGB324 was less pronounced in cell lines with intact TP53 (Fig. 2F), with no detectable increase in γH2Ax. In a TP53 wild-type background, treatment with BGB324, instead, resulted in the accumulation of p21, a cyclin-dependent kinase inhibitor, indicating the onset of senescence (Supplementary Fig. S3D). Overall, these findings suggest that TP53-deficient lung cancer cells could be more susceptible to the replication stress and DNA damage induced by BGB324.
AXL inhibition sensitizes cancer cells to ATR inhibitors
As BGB324 treatment results in DNA damage and replication stress (and thus, activating the ATR-CHK1 checkpoint), we hypothesized that AXL inhibition–induced replication stress would sensitize cancer cells to ATR inhibition. To test this hypothesis, we first determined the single-agent activities of two selective ATR inhibitors currently in clinical trials, VX-970/M6620 (berzosertib) and AZD6738 (ceralasertib), in a panel of 25 NSCLC and LCNEC cell lines. Both VX-970 and AZD6738 showed potent cytotoxicity in a subset of cell lines (Supplementary Fig. S4A and S4B). However, despite potent inhibition of ATR-mediated CHK1 phosphorylation, several cell lines showed inherent resistance to the ATR inhibitors. Interestingly, AXL expression was higher in cell lines that were resistant to the ATR inhibitors, VX-970 [fold change (FC) between sensitive vs. resistant = −1.9; P = 0.04; Fig. 3A] and AZD6738 (FC = −1.4; P = 0.3).
To determine whether AXL inhibition could enhance sensitivity of lung cancer cells to ATR inhibitors, NSCLC and LCNEC cells were treated with fixed ratio concentrations of BGB324 and VX-970 in 5-day cell proliferation assays. In the ATR inhibitor–resistant cell lines, Calu-1 and H2250, the combination of BGB324 and VX-970 had a greater than predicted additive effect on decreasing cell viability (i.e., ΔAUC ≤ −0.1; Fig. 3B). Chou–Talalay CIs of 0.24, 0.24, and 0.58, in Calu-1, H2250, and H1299 cell lines, respectively, calculated at 50% fraction affected for the drug combination, also suggested a synergistic interaction. Even at a nongrowth inhibitory concentration, BGB324 potentiated the cytotoxic effect of VX-970 (Supplementary Fig. S4C). This synergistic effect was further confirmed in a clonogenic assay, wherein the clonogenic survival of Calu-1 and H1299 cells treated with the drug combination was more effectively decreased as compared with either single agent (Fig. 3C). To confirm whether the observed effects were AXL dependent, we knocked down Axl in the Kras/Trp53-mutant GEMM-derived 344SQ cells (Supplementary Fig. S4D) and assessed its effect on sensitivity to ATR inhibition. While the parental 344SQ cells were inherently highly sensitive to VX-970 (Supplementary Fig. S4A), Axl knockdown further increased the sensitivity of 344SQ cells to ATR inhibition (Supplementary Fig. S4E). Together, these data show that AXL inhibition by BGB324 strongly sensitizes inherently resistant AXL-high lung cancer cells to ATR inhibitors.
To test whether the observed effects of the BGB324/VX-970 combination were specific to ATR inhibition or were more broadly applicable to DDR inhibitors, we next tested the combination of BGB324 with other DDR inhibitors. In addition to ATR inhibitors, BGB324 also modestly sensitized cancer cells to CHK1 inhibition (LY2606368/prexasertib; Fig. 3D; Supplementary Fig. S5A). In contrast, combinations with other DDR inhibitors, including an ATM inhibitor (AZD0156), a DNAPKC inhibitor (NU7441), or a WEE1 inhibitor (AZD1775) showed no effect (Fig. 3D; Supplementary Fig. S5A). These results further support the idea that the sensitization of NSCLC and LCNEC cells to VX-970 was due to the specific increase in replication stress and activation of ATR/CHK1 axis by BGB324. Next, we observed by RPPA analysis that phosphorylation of mTOR and its downstream mediators, such as S6 and MDM2, was significantly inhibited following treatment of Calu-1 cells with BGB324/VX-970 combination (Fig. 3E). AXL exerts its oncogenic pro-survival effects through PI3K/AKT/mTOR signaling (1). Furthermore, mTOR has also been shown to mediate DDR and replication stress (39, 40). To determine whether the observed synergistic effect of BGB324/VX-970 combination was mediated by the mTOR pathway, we tested the combination of VX-970 with AZD2014, a potent and selective inhibitor of mTOR complexes, mTORC1 and mTORC2. AZD2014 partially recapitulated the effects seen with BGB324, suggesting observed synergistic interaction maybe mediated, in part, by inhibition of AXL/mTOR signaling (Supplementary Fig. S5B).
Combination of AXL and ATR inhibitors caused significant DNA damage
Having seen increased γH2Ax accumulation (DNA damage), CHK1 phosphorylation (replication stress), and RPA32 hyperphosphorylation (Fig. 2B) in TP53-deficient cell lines (Calu-1, H2250, H1299, and 344SQ) following 24-hour treatment with BGB324, we were interested to understand the effect of AXL/ATR cotargeting on these markers. BGB324-induced increases in pCHK1 were abrogated by the ATR inhibitor, VX-970 (Fig. 4A). Greater induction of phospho-RPA32 (S4/8) was detected as early as 4 hours in H1299 cells (Fig. 4B) and was pronounced at 24 hours in most cell lines (Fig. 4A) when both drugs were combined as compared with treatment with the single agents, suggesting fork collapse and progression to double-strand breaks. Consequently, γH2Ax levels (DNA damage) were significantly increased upon combined AXL/ATR inhibition (Fig. 4A). Appreciable γH2Ax levels in H2250 could not be detected by Western blotting. Induction of pKAP1 at 24 hours was also evident in H1299 and 344SQ cells, when both drugs were combined. Prolonged treatment with hydroxyurea, a ribonucleotide reductase inhibitor that induces replication fork stalling and replication stress, also showed a similar increase in RPA32 phosphorylation and γH2Ax (Fig. 4B). The effects of combined AXL/ATR inhibition were further confirmed by using another ATR inhibitor, AZD6738, which likewise increased γH2Ax and phospho-RPA32 levels (Fig. 4C).
To further confirm whether these increases were AXL dependent, Kras/Trp53-mutant 344SQ Axl-knockdown cells were treated with VX-970. In cells lacking AXL, addition of VX-970 resulted in enhanced DNA damage and RPA32 hyperphosphorylation, as compared with the parental cells (Fig. 4D). As expected, the effect of the drug combination on DNA damage was less pronounced in TP53 intact cell lines (Fig. 4E). The effect of VX-970 and BGB324/VX-970 was also unexpectedly more pronounced in TP53 wild-type cell lines harboring STK11 comutations (Supplementary Fig. S5C), suggesting that there may be other molecular contexts in which the combination could be effective. Together, these data suggest that the combined inhibition of AXL and ATR induces severe replication stress and DNA damage, resulting in cell death. In general, this effect appeared to be more pronounced in TP53-deficient lung cancer cells.
We next analyzed the nuclear distribution of γH2Ax using immunofluorescence. In asynchronous H1299 and Calu-1 cells treated with BGB324 for 24 hours, a heterogenous pattern of diffuse pan-nuclear γH2Ax staining along with a few distinct punctate foci was detected (Fig. 5A). γH2Ax foci formation is typically seen with DNA damage, indicating double-stranded DNA breaks, while uniform pan-nuclear staining has been observed during lethal replication stress (41, 42). As expected, treatment with the ATR inhibitor or hydroxyurea showed a pan-nuclear γH2Ax staining, characteristic of replication stress (Fig. 5A; Supplementary Fig. S6). Furthermore, in response to BGB324/VX-970 combination, most of the cells showed widespread γH2Ax foci formation. In a fraction of cells, intense pan-nuclear γH2Ax staining, likely a result of severe replication stress and DNA-PK hyperactivation at stalled replication forks (41), was also observed following treatment with the combination (Fig. 5A; Supplementary Fig. S6).
BGB324 and VX-970 combination induced mitotic catastrophe through premature cdc2 activation
In response to DNA damage, TP53-deficient cell lines fail to activate the G1-phase checkpoint, instead, relying on the ATR/CHK1-mediated G2–M-phase checkpoint to halt cell-cycle progression and repair DNA damage. ATR/CHK1 activation causes the inhibitory phosphorylation of the mitotic cyclin-dependent kinase, cdc2/CDK1, at Tyr 15, resulting in G2–M-phase arrest, while ATR inhibition abrogates this checkpoint, allowing cells to progress prematurely into mitosis. Because the BGB324/VX-970 combination induced significant DNA damage in TP53-deficient cell lines, we next examined its effect on cell-cycle progression. In asynchronous Calu-1 cells, treatment with the BGB324/VX-970 combination for 24 hours strongly increased the number of cells in G1-phase, as compared with either single agent alone (Fig. 5B). To test whether this increase was due to restoration of the G1-phase checkpoint, cells were treated with BGB324 and/or VX-970, followed by addition of nocodazole, a microtubule inhibitor. Cell-cycle profiles posttreatment showed that majority of the cells progressed through the cell cycle and were arrested at M-phase by nocodazole, with no significant induction of G1-phase checkpoint. These results indicate that the increase in G1-phase population was likely as a result of downregulation of cdc2 phosphorylation and abrogation of the G2-phase checkpoint, which was seen here (Fig. 5C and D). A previous study reported a similar effect of BGB324 on cdc2 dephosphorylation, as well as a synergistic combination with antimitotic agents (43). Of note, treatment with either BGB324 or the BGB324/VX-970 combination appeared to induce a small fraction of cells to arrest at the G1-phase, suggesting a partial, but incomplete G1-phase checkpoint restoration. This was further supported by a significant decrease in RB phosphorylation in cells treated with the inhibitor combination (Fig. 5C). To further examine mitotic exit following treatment with the inhibitors, Calu-1 cells were treated with DMSO, BGB324, VX-970, or the combination in the presence of nocodazole. Upon release from the mitotic arrest, the control cells returned to normal cell-cycle kinetics by 72 hours. On the other hand, cells treated with the ATR inhibitor or the drug combination failed to reenter cell cycle efficiently, with an exacerbation of the endoreduplication and polyploidy induced by nocodazole (Fig. 5B; ref. 44). Consistent with this, expression of phospho-histone H3, a mitosis marker, was increased in Calu-1 and H2250 cells (Fig. 5D). A similar increase in histone H3–positive cells was also observed by flow cytometry (Fig. 5B; Supplementary Fig. S7A). These findings, together with cdc2 dephosphorylation by BGB324/VX-970 combination (Fig. 5C and D), suggest an aberrant mitotic exit in the presence of significant DNA damage. In the LCNEC cell line, H1299, despite cdc2 reactivation, expression of phospho-histone H3 was significantly inhibited at 72 hours and the cells persisted in prolonged G2–M-phase arrest (Fig. 5D; Supplementary Fig. S7B). In this same model, an increase in proportion of cells undergoing early and late apoptosis was detected by Annexin-V/propidium iodide staining at 72 hours following treatment of H1299 cells with the drug combination (Fig. 5E). Together, these data suggest that DNA damage induced by BGB324/VX-970 combination results in cancer cell death through multiple mechanisms, including mitotic catastrophe and apoptosis.
Biomarkers of response to BGB324 and VX-970 combination
To identify markers that predict sensitivity to AXL/ATR cotargeting, we screened the AXL/ATR inhibitor combinations in a panel of 24 lung cancer cell lines (Fig. 6A; Supplementary Fig. S7C). In four NSCLC cell lines with primary resistance to ATR inhibition, the combination of BGB324 and VX-970 was synergistic (ΔAUC ≤ −0.1). Similarly, in five of eight LCNEC cell lines, a greater than additive combinatorial effect on inhibiting cell proliferation was observed (Fig. 6B).
Next, to identify gene expression signatures associated with sensitivity to the AXL/ATR inhibitor combination, we performed an exploratory analysis of differential gene expression, between cell lines that showed a greater than additive response (ΔAUC ≤ −0.1) and those that showed the least additive response (0 ≤ ΔAUC < 0.1) to the drug combination, using gene set enrichment analysis (GSEA). Gene sets associated with MYC-regulated genes, IFN response, as well as DNA repair were enriched in the cell lines most susceptible to AXL/ATR cotargeting (Fig. 6C). In line with this, cMYC levels were significantly decreased following treatment with BGB324/VX-970 combination (Fig. 6D). To identify proteomic biomarkers, we correlated proteomic profiles of cell lines to the observed combinatorial effect (ΔAUC). NSCLC cell lines that exhibited a greater than additive effect to BGB324/VX-970 combination (ΔAUC ≤ −0.1) expressed higher mTOR levels and pathway activation, consistent with data above that mTOR signaling was turned off by the combination (Fig. 6D). p16INK4a expression was also higher in these cell lines, which also expressed low RB protein levels (rho = −0.35; P = 0.08, data not shown). In addition to its role in promoting G1-phase arrest, p16INK4a impairs homologous DNA repair and sensitizes cells to DNA-damaging agents (45, 46). Interestingly, cell lines that showed a greater than additive response to BGB324/VX-970 combination had low SLFN11 protein levels (rho = 0.64; P < 0.001; Fig. 6D and E). In response to DNA damage and replication stress, SLFN11, a DNA/RNA helicase, is recruited to stalled replication forks by RPA1 and induces an irreversible replication arrest through chromatin unwinding (47). As a result, high SLFN11 levels have been associated with sensitivity to DNA-damaging agents and DDR inhibitors (48–50). In the absence of SLFN11, resistance to drugs that induce DNA damage or replication stress has been overcome by targeting the ATR/CHK1 axis (47, 51). Consistent with these previous findings, many cell lines with high SLFN11 expression were sensitive to single-agent BGB324 or VX-970 (without significant further enhancement of response when combined, indicated by an additive response). While, in SLFN11-low cell lines, a relatively greater in vitro effect of the combination (as compared with single agents), indicated by a greater than additive response (ΔAUC ≤ −0.1), was observed. Together, these results suggest that lung cancer cells with elevated cMYC-induced replication stress and a compromised replication stress and DDR (as indicated by lower SLFN11 expression), were more susceptible to AXL/ATR cotargeting.
AXL upregulation is emerging as a mechanism of resistance to treatments targeting replication stress and DNA damage (e.g., chemotherapy; refs. 21, 52 and radiation; refs. 20, 22), and the replication stress response (WEE1 inhibitors; ref. 53) in various cancers. Similar to our previous study (8), we show here that inhibition of AXL using BGB324 (bemcentinib) resulted in DNA damage accumulation. In addition, we found increased replication stress and ATR/CHK1 axis activation following treatment with BGB324. Together, these findings support a protective role for AXL in suppressing DNA damage and tolerance of replication stress. Although the precise mechanism of how AXL inhibition results in replication stress is unclear, we speculate that this may be mediated, in part, via RAD51 depletion. We have shown previously that AXL inhibition and knockdown causes RAD51 depletion (8), which has been linked to defective replication fork protection and ATR/CHK1 activation (8, 54). Furthermore, AXL-mediated mTOR signaling could also have an effect on replication stress, as mTOR complexes 1 and 2 are known to regulate CHK1 and promote the transcriptional abundance of DNA replication licensing factors, such as CDC6 (39, 40).
We observe elevated levels of DNA damage and replication stress induced by AXL inhibition in NSCLC cell lines with a TP53-deficient background as compared with cells with an intact TP53 axis. In the absence of a p53-dependent G1-phase checkpoint, cancer cells are likely to be more susceptible to replication stress and reliant on ATR/CHK1 signaling (55). This could explain the striking induction of γH2Ax and pCHK1 in response to BGB324 treatment in these cells. On the other hand, consistent with recent reports (56), we found that TP53 deficiency alone did not increase sensitivity of NSCLC cell lines to ATR inhibitors. Our data here show that NSCLC cell lines with high AXL levels were more resistant to ATR inhibitors. In line with this, we demonstrate greater than additive effects of AXL and ATR inhibitor combinations in many NSCLC cell lines. These findings also extend to LCNEC cell lines that frequently harbor TP53 mutations. Particularly, LCNEC cell lines respond to both BGB324 as a single agent and its combination with ATR inhibitors. Given the rarity of targetable mutations and shared characteristics, LCNEC tumors are treated similar to SCLC, predominantly with chemotherapy (18). In light of our findings that a significant subset of mesenchymal LCNEC tumors express high levels of AXL, AXL inhibition and AXL/ATR cotargeting could be a therapeutically useful strategy to treat these aggressive tumors that otherwise lack effective targeted therapies. Mechanistically, we show here that combined inhibition of AXL and ATR induced significant DNA damage and abrogated the G2–M-phase checkpoint. In contrast to the results we observed in NSCLC and LCNEC cell lines, a recent study, which examined the combination of BGB324 and a CHK1/2 inhibitor (AZD7762), found no increase in DNA damage or CHK1 phosphorylation following BGB324 treatment in melanoma cell lines (24). Differences in genetic background and tumor type likely contribute to these differential effects. Similarly, our group has previously shown in SCLC that AXL overexpression confers primary and acquired resistance to inhibition of WEE1, a DDR kinase in the replication stress response pathway (53). However, unlike AXL/ATR combination, the combination of BGB324 and WEE1 inhibitor, AZD1775, mostly showed an additive effect in the NSCLC cell lines tested. SCLC has a higher basal replication stress and frequency of DDR mutations, which could explain its greater susceptibility to AXL/WEE1 inhibitor combination. Of note, we observed that while loss of AXL, by siRNA knockdown, induced DNA damage, some residual AXL appeared to be required for CHK1 phosphorylation. Additional studies are needed to better understand AXL’s direct role in replication stress.
Finally, we identify that cell lines with low levels of SLFN11 were sensitive to AXL/ATR cotargeting. SLFN11, a DNA/RNA helicase, induces an irreversible replication arrest in response to DNA damage and replication stress (47). Previously, we and others have reported SLFN11 as a predictive marker of sensitivity to DNA-damaging agents, such as cisplatin and PARP inhibitors, in SCLC (49, 57). High SLFN11 expression has also been associated with platinum response in LCNEC tumors (58). Conversely, SLFN11 inactivation, such as by promoter hypermethylation, conferred resistance to chemotherapy in NSCLC and other cancers (49, 59). Consistent with our observations, in the absence of SLFN11, resistance to drugs that induce DNA damage or replication stress has been overcome by targeting ATR (such as with ATR/CHK1 inhibitor combinations; refs. 48, 51). Therefore, our findings suggest the AXL and ATR inhibitor combination could be effective in platinum-resistant SLFN11-deficient NSCLC and LCNEC.
In conclusion, these findings underscore a critical role for AXL in tolerance of replication stress and DNA damage in lung cancer, which can shed light on the implication of its upregulation in response to multiple anticancer treatments. In addition, our results show that TP53-deficient NSCLC and LCNEC are specifically susceptible to the novel combination of AXL and ATR inhibitors. We further identify SLFN11, a clinically used biomarker, to be a strong predictor of response to AXL/ATR cotargeting.
J.V. Heymach reports grants and personal fees from AstraZeneca, GlaxoSmithKline, and Spectrum, personal fees from EMD Serono during the conduct of the study, Boehringer Ingelheim, Bristol-Myers Squibb, Catalyst, EMD Serono, Foundation Medicine, Hengrui Therapeutics, Genentech, Guardant Health, Eli Lilly, Merck, Novartis, Pfizer, Roche, Sanofi, Seattle Genetics, and Takeda, and personal fees and other from Spectrum outside the submitted work, as well as a patent for WO 2020/150480 issued. D.L. Gibbons reports grants from LUNGevity Foundation during the conduct of the study and Takeda, grants and personal fees from AstraZeneca, Sanofi, Janssen R&D, and Ribon Therapeutics, and personal fees from Alethia Biotherapeutics outside the submitted work. L.A. Byers reports other from AstraZeneca, GenMab, Sierra Oncology, Tolero Pharmaceuticals, PharmaMar, AbbVie, Bristol-Myers Squibb, Alethia, Merck, Pfizer, and Jazz Pharmaceuticals outside the submitted work. No disclosures were reported by the other authors.
K. Ramkumar: Conceptualization, validation, investigation, visualization, methodology, writing–original draft. C.A. Stewart: Investigation, writing–review and editing. K.R. Cargill: Investigation, visualization. C.M. Della Corte: Investigation, writing–review and editing. Q. Wang: Formal analysis, visualization. L. Shen: Formal analysis, visualization. L. Diao: Formal analysis, visualization. R.J. Cardnell: Writing–review and editing. D.H. Peng: Methodology. B.L. Rodriguez: Investigation, methodology. Y.-H. Fan: Formal analysis. J.V. Heymach: Resources. J. Wang: Formal analysis, writing–review and editing. C.M. Gay: Methodology, writing–review and editing. D.L. Gibbons: Resources, supervision, funding acquisition, writing–review and editing. L.A. Byers: Conceptualization, supervision, funding acquisition, methodology, project administration, writing–review and editing.
Support for this research was provided by the NIH/NCI CCSG P30-CA016672 grant (Bioinformatics Shared Resource), NIH/NCI T32 CA009666 (to C.M. Gay), NIH/NCI R01-CA207295 (to L.A. Byers), the University of Texas-Southwestern and MD Anderson Cancer Center Lung SPORE (5 P50 CA070907), ASCO Young Investigator Award (to C.M. Gay), the LUNGevity foundation (to D.L. Gibbons and L.A. Byers), MD Anderson Cancer Center Physician Scientist Award (to L.A. Byers), Andrew Sabin Family Fellowship (to L.A. Byers), the Rexanna Foundation for Fighting Lung Cancer (to J.V. Heymach and L.A. Byers), and through generous philanthropic contributions to The University of Texas MD Anderson Lung Cancer Moon Shot Program (to J.V. Heymach, J. Wang, and L.A. Byers). LCNEC mRNA expression data were generously provided by Dr. Julie George, University of Cologne. BGB324 was provided by BerGenBio ASA. AZD6738, AZD2014, and AZD1775 were provided by AstraZeneca.
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.