AXL and Error-Prone DNA Replication Confer Drug Resistance and Offer Strategies to Treat EGFR-Mutant Lung Cancer

Abstract

Anticancer therapies have been limited by the emergence of mutations and other adaptations. In bacteria, antibiotics activate the SOS response, which mobilizes error-prone factors that allow for continuous replication at the cost of mutagenesis. We investigated whether the treatment of lung cancer with EGFR inhibitors (EGFRi) similarly engages hypermutators. In cycling drug-tolerant persister (DTP) cells and in EGFRi-treated patients presenting residual disease, we observed upregulation of GAS6, whereas ablation of GAS6's receptor, AXL, eradicated resistance. Reciprocally, AXL overexpression enhanced DTP survival and accelerated the emergence of T790M, an EGFR mutation typical to resistant cells. Mechanistically, AXL induces low-fidelity DNA polymerases and activates their organizer, RAD18, by promoting neddylation. Metabolomics uncovered another hypermutator, AXL-driven activation of MYC, and increased purine synthesis that is unbalanced by pyrimidines. Aligning anti-AXL combination treatments with the transition from DTPs to resistant cells cured patient-derived xenografts. Hence, similar to bacteria, tumors tolerate therapy by engaging pharmacologically targetable endogenous mutators.

Significance: EGFR-mutant lung cancers treated with kinase inhibitors often evolve resistance due to secondary mutations. We report that in similarity to the bacterial SOS response stimulated by antibiotics, endogenous mutators are activated in drug-treated cells, and this heralds tolerance. Blocking the process prevented resistance in xenograft models, which offers new treatment strategies. This article is highlighted in the In This Issue feature, p. 2483.

Figure 1.

GAS6 expression characterizes cycling persisters; upregulation of the cognate receptor AXL is essential for resistance. A, Single-cell PC9 subclones were established in the absence of drugs. Following treatment with erlotinib, we obtained RNA-seq data from each subclone. The volcano plot presents genes that are differentially expressed in the 8 highest versus the 8 lowest persistence subclones. Red dots mark genes exceeding two cutoffs: P value and change in expression. B, Shown are GAS6 expression levels in Watermelon-transduced PC9 cells treated with osimertinib for the indicated time intervals. Day 14 persisters were sorted into 3 groups according to their cycling status. C, Single-cell GAS6 expression levels in human non–small cell lung cancer tumors clinically characterized as treatment naïve (TN), residual disease (RD), and progressive disease (PD; 457, 557, and 1,088 cells per group, respectively). Data from NCT03433469. D, Single-cell RNA-seq data (29) from TN (21 samples), RD (16 samples), and PD (16 samples) were analyzed for AXL transcripts. Each dot represents a single tumor cell. Box-and-whisker plot: center line, median; box limits, upper (third) and lower (first) quartiles; whiskers indicate maximum and minimum values. Y-axis is a log₂ scale (one-way Kruskal–Wallis test). E and F, AXL-overexpressing (OX; E) or AXL-knockout (KO; F) PC9 cells, along with the respective control cells, were untreated or treated with TKIs (1 or 3 μmol/L). On day 9, surviving cells were stained and quantified (average ± SD, triplicates, and 3 experiments). G, PC9 cells (2 × 10⁶) were subcutaneously implanted in nude mice. When tumors became palpable, mice were randomized into groups (8 or 10 animals per group) that were treated daily with erlotinib (10 mg/kg/day) or osimertinib (5 mg/kg/day). Tumor volumes are presented (average ± SEM). Note that due to ulceration, one animal of the control arm was eliminated. H, 3–4 mice from each group shown in G were euthanized when tumors reached 1,500 mm³. Tumors were extracted and subjected to immunoblotting. Tubulin (TUB) served as the loading control. I and J, AXL-overexpressing PC9 cells (I) and AXL-knockout PC9 cells (2 × 10⁶, J), along with the respective control cells, were subcutaneously implanted in mice. When tumors became palpable, mice were randomized and treated as in G. n.s., not significant. *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001.

Figure 2.

TKIs induce DNA breaks but AXL restrains DNA fragmentation. Control (EV) and AXL-overexpressing (OX) PC9 cells were used. A, Cells were incubated for 72 hours with TKIs (10 nmol/L), and viability was determined (means ± SD, triplicates). B, Cells were treated for 48 hours with TKIs (1 μmol/L), and the fractions undergoing early/late apoptosis were determined (means ± SD, triplicates). C, Cells were treated as in B, and cell extracts were probed for the indicated proteins using immunoblotting and quantification (underneath numbers). D, Cells grown on coverslips were treated as in A. Fixed cells were stained for γH2A.X, actin, and DAPI, and images were captured using a confocal microscope (images are not shown). For the quantification of γH2A.X foci, we randomly selected 30 cells from each sample. The dot plot depicts average counts (lines) of foci per nucleus. Each dot corresponds to a single cell. The results represent means (± SD) from three experiments. E, The indicated cells were incubated with TKIs (1 μmol/L) for 24 hours. DNA damage was estimated according to the means of 40 comets (see Supplementary Fig. S3B) scored per condition. Results represent means (±SD) from 3 experiments. F, Cells were treated for 48 hours as in A and ROS levels were assayed. Data are means (±SD) from three experiments. G, OX or control PC9 cells (10⁶) were incubated for 48 hours with osimertinib (1 μmol/L), and DNA libraries of DSBs were prepared using BLISS. This step was followed by nucleotide sequencing. DSB counts were normalized across the whole genome, and relative abundance of breaks at gene bodies (top) or promoters (bottom) were measured. Data are means ± SD. H, Cells stably expressing pDRGFP were transfected with pCBASce-I and then treated as in A, prior to flow cytometry. Mock transfection was used for normalization. Quantification of relative HR capacities is shown (means ± SD, triplicates). n.s., not significant. *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001.

Figure 3.

AXL overexpression augments TKI-induced upregulation of both RAD18 and TLS polymerases and enhances the emergence of the T790M mutation. A, PC9 tumors were extracted at the marked timing (see Fig. 1I) and immunoblotted as indicated. B, PC9 cells were treated with osimertinib (1 μmol/L), or DMSO, for 0–9 days. Red arrows indicate TKI replenishments. Proteins and RNA were subjected to immunoblotting and real-time PCR, respectively. C, Pairs of tumor slices were obtained from 7 patients, before and after treatment with osimertinib. The corresponding slides were incubated with primary antibodies specific to AXL, RAD18, and DNA polymerase eta. Slices from the same patient are indicated with a distinct color. Immunoreactivities were evaluated, and scores were generated (two-way ANOVA with the Sidak multiple comparisons test). D, Control and RAD18 KO cells were treated for 9 days with TKIs. Fractions of surviving DTPs from 3 experiments are shown. E, Naïve and RAD18-knockout PC9 cells were probed using anti-AXL or anti-RAD18 antibodies. Thereafter, the cells were processed for proximity ligation analysis (red dots). Representative images and PLA quantification are shown. Scale bars, 20 μm. F, Cells were treated for 2–9 days with erlotinib (1 μmol/L) or H₂O₂ (40 μmol/L). Extracts were subjected to immunoprecipitation with an anti-RAD18 (or control) antibody and immunoprecipitates were immunoblotted (IB) as indicated. G, Naïve and AXL-overexpressing cells, along with PC9ER cells (erlotinib resistant), were treated with erlotinib (1 μmol/L). Genomic DNA was isolated after 9 days and the T790M point mutation was assayed using digital PCR. Shown are fractions of T790M in cellular DNA. Data are means ± SEM (4 experiments). H, PC9 cells were treated with erlotinib (1 μmol/L) and an anti-AXL monoclonal antibody (20 μg/mL), as indicated. Genomic DNA was isolated after 9 days and the T790M point mutation was assayed using digital PCR. Shown are fractions of T790M in cellular DNA. Data are means ± SEM (4 experiments). *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001.

Figure 4.

AXL induces purine metabolism and increases PuMB. A, RNA from PC9 cells and AXL-KO cells was sequenced, and differentially expressed genes were subjected to KEGG Pathway Enrichment analysis. B, A waterfall plot depicting fold changes in metabolite abundance in AXL-KO relative to PC9-WT (control) cells. Metabolites were rank-ordered, and specific compounds are indicated (assays performed in triplicates). C, Fractional labeling of IMP in PC9-KO and PC9-WT (control) cells using [amide-15N]glutamine and 24 hours of incubation. D, Fractional labeling of purines determined in AXL-KO and PC9-WT cells incubated for 24 hours with [U-13C]glucose (means ± SD; 3 experiments). E, An AXL expression vector and MYC promoter reporter plasmid were transfected into HEK293 cells. Renilla was used as a control. Luminescence reading was taken 48 hours later. F,PPAT and PAICS promoter reporter plasmids were transfected into HEK293 cells, along with AXL and MYC vectors. Luminescence readings (means ± SD; 3 experiments) were normalized to a GAPDH reporter. G, PC9 cells were treated with osimertinib (1 μmol/L) or DMSO for the indicated time intervals. RNA was isolated and subjected to real-time PCR using primers corresponding to the indicated transcripts. H, PC9 cells were either untreated or treated for 9 days with TKIs at the indicated concentrations. Control cells were treated with DMSO. The indicated proteins were detected using immunoblotting. Tubulin served as the gel loading control. I, PuMB was analyzed in the TCGA lung adenocarcinoma data set (n = 506 patients) and presented versus EGFR's mutational status. J, A cohort of 10 treatment-naïve patients, for whom both tumoral single-cell RNA-seq data and tumor whole-exome sequencing data were available, was analyzed for AXL expression and PuMB. Dots show individual data points and the diagonal represents a regression line. K, The status of AXL expression level and presence of EGFR or RAS mutations in the TCGA data set are shown. L, A core MYC gene-expression signature was analyzed against the level of AXL expression in the cohort of 506 patients. ***, P ≤ 0.001.

Figure 5.

Combining an anti-AXL antibody and EGFR blockers prevents tumor relapses. A–C, PC9 cells (3 × 10⁶) were implanted in CD1-nu/nu mice. When tumors became palpable, mice were randomized into groups of 8 animals that were treated (hatched area) with the indicated antibodies (total dose: 0.2 mg/mouse/injection) once every three days, or daily with osimertinib (5 mg/kg/day). Data are means ± SEM (A: single agent, B: dual treatments, C: triple drug combinations). D, H1975 cells (3 × 10⁶) were subcutaneously implanted in CD1-nu/nu mice. When tumors became palpable, mice were randomized into groups of 10 animals each, which were treated for 30 days (hatched area) with the indicated antibodies (total dose: 0.2 mg/mouse/injection), once per 3 days, or with osimertinib (5 mg/kg/day). Mice were euthanized when tumor size reached 1,500 mm³. E, The PDX tumor model TM00193 was implanted in NSG mice. When tumors became palpable, mice were randomized into groups (7 mice per group) that were treated (hatched area) with the indicated antibodies (see A) or with osimertinib (10 mg/kg/day). Tumor volumes (averages ± SD) are indicated. CTX, cetuximab; Osi, osimertinib. *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001.

Figure 6.

The triple treatment (anti-AXL antibody + cetuximab + osimertinib) inhibits tumors that have already acquired resistance to osimertinib. A, PC9 cells (3 × 10⁶) were subcutaneously implanted in CD1-nu/nu mice. When tumors became palpable, mice were treated with osimertinib (5 mg/kg/day) and after tumors regressed and later relapsed (300 mm³), animals were randomized into groups (n = 7) that were treated for 30 days (hatched area) with the indicated drugs. Thereafter, all treatments were stopped, but we continued monitoring the animals. B–G, Tumor volumes per individual mice that were initially treated with osimertinib (5 mg/kg/day) and later switched to the triple drug combination (3×). Note that all treatments were terminated once tumors disappeared, but we kept monitoring the animals. CTX, cetuximab, Osi, osimertinib. **, P ≤ 0.01; ***, P ≤ 0.001.