Abstract
Anaplastic lymphoma kinase (ALK) inhibitors are highly effective in patients with ALK fusion–positive lung cancer, but acquired resistance invariably emerges. Identification of secondary mutations has received considerable attention, but most cases cannot be explained by genetic causes alone, raising the possibility of epigenetic mechanisms in acquired drug resistance. Here, we investigated the dynamic changes in the transcriptome and enhancer landscape during development of acquired resistance to ALK inhibitors. Histone H3 lysine 27 acetylation (H3K27ac) was profoundly altered during acquisition of resistance, and enhancer remodeling induced expression changes in both miRNAs and mRNAs. Decreased H3K27ac levels and reduced miR-34a expression associated with the activation of target genes such as AXL. Panobinostat, a pan-histone deacetylase inhibitor, altered the H3K27ac profile and activated tumor-suppressor miRNAs such as miR-449, another member of the miR-34 family, and synergistically induced antiproliferative effects with ALK inhibitors on resistant cells, xenografts, and EML4-ALK transgenic mice. Paired analysis of patient samples before and after treatment with ALK inhibitors revealed that repression of miR-34a or miR-449a and activation of AXL were mutually exclusive of secondary mutations in ALK. Our findings indicate that enhancer remodeling and altered expression of miRNAs play key roles in cancer drug resistance and suggest that strategies targeting epigenetic pathways represent a potentially effective method for overcoming acquired resistance to cancer therapy.
Significance: Epigenetic deregulation drives acquired resistance to ALK inhibitors in ALK-positive lung cancer. Cancer Res; 78(12); 3350–62. ©2018 AACR.
Introduction
Echinoderm microtubule-associated protein-like 4 (EML4)–anaplastic lymphoma kinase (ALK) rearrangement is a distinct molecular subclassification of non–small cell lung cancer (NSCLC; ref. 1). Tumors with oncogenic ALK rearrangements are exquisitely sensitive to ALK blockade with tyrosine kinase inhibitors (TKI; ref. 2). Initial treatments with first- and second-generation ALK inhibitors (crizotinib, ceritinib, or alectinib) result in profound initial antitumor responses, with an objective response rate of up to 70% (3–5). However, ALK-positive tumors eventually acquire resistance to ALK inhibitors within 1 year of treatment. Therefore, identifying the molecular events that limit efficacy to ALK inhibitors is necessary for improving patient survival.
To date, two distinct mechanisms of acquired resistance to ALK inhibitors have been elucidated. The first mechanism involves ALK-dependent processes characterized by either the amplification of ALK itself or secondary mutations in the kinase domain that hinder the binding of ALK inhibitors (6). The other mechanism of resistance is independent of ALK and includes (1) the activation of bypass signaling pathways (EGFR, KRAS, KIT, MET, and IGF1R), which allow continuous cell proliferation and survival (7–9), and (2) processes associated with the epithelial-to-mesenchymal transition (EMT; ref. 10). In a recent study examining the relative predominance of ALK-dependent mechanisms, ALK kinase domain mutations (including the gatekeeper mutation L1196M) were found in 20% of patients with first-generation (crizotinib) failure and in 56% of patients whose cancers had progressed while being treated with second-generation ALK inhibitors (11). These results suggest that other unknown mechanisms of resistance affect the majority of patients who relapse during treatment with first- and/or second-generation ALK inhibitors.
The genetic mechanisms of acquired resistance to targeted therapy have been widely investigated, but research remains limited concerning the effects of changes to both the noncoding genome and the epigenetic landscape of drug resistance. Previous studies have shown that tumors exhibit both genetic and epigenetic heterogeneity within cell populations, and that epigenetic changes, such as DNA methylation and/or histone modifications, accumulate in the genome throughout oncogenesis (12–15). This dynamic landscape may permit the introduction of adaptive ALK-independent mechanisms of drug resistance within targeted cells. Notably, Sharma and colleagues reported that drug-tolerant cells that emerged after treatment of tumors with an EGFR inhibitor were eradicated by treatment with a histone deacetylase (HDAC) inhibitor (16). HDAC inhibitors may profoundly alter the activity of enhancers, the key DNA elements that regulate cell-specific gene expression, and thereby significantly influence the transcription of genes involved in the drug response (14, 17, 18).
For these reasons, we investigated the epigenetic mechanisms that underlie acquired resistance to ALK inhibitors. Our findings provide epigenome-based insights into novel therapeutic strategies that may overcome acquired resistance to ALK inhibitors and enhance the survival of patients with ALK rearrangements.
Materials and Methods
Generation of drug-resistant cells
H3122 cells were kindly provided by Dr. Okamato at Kyushu University (Fukuoka, Japan). H2228 cells were purchased from the ATCC. Cells were cultured in RPMI-1640 (Sigma) supplemented with 10% FBS and 1% penicillin/streptomycin at 37°C in 5% CO2, and media were exchanged every 2 to 3 days. H3122 and H2228 cells were cultured with increasing concentrations of ceritinib, starting with the IC30, to create the ceritinib (LDK378)-resistant model. The dose was increased in a stepwise manner when normal cell proliferation patterns resumed. Fresh drug was added every 72 to 96 hours. Resistant cells (LR; bulk of resistant cells, #1–12; independently derived resistant clones in colonies expanded from LR) were derived after approximately 6 months of culture in the continuous presence of 1 μmol/L ceritinib. LR cells were used in all experiments.
Generation of the in vivo–resistant model
Human H3122 cells were injected subcutaneously into the flanks of CB17 SCID mice. Tumor-bearing mice (tumor size 200–500 mm3) were treated with ceritinib (50, 75, 87.5, or 100 mg/kg) daily. Tumors initially showed a dose-dependent decrease in volume following ceritinib treatment, and when tumors showed >25% regrowth from the maximal reduction, the mice were considered resistant to ceritinib. For the microarray analysis related to Fig 2C, the mice were sacrificed to harvest and store tumor tissues. For assessments of drug efficacy, some resistant mice were randomized into four treatment groups [vehicle, ceritinib (50 mg/kg), panobinostat (5 mg/kg), and a combination of ceritinib (50 mg/kg) and panobinostat (5 mg/kg); n = 7 mice per group], as described in Supplementary Fig. S12A. Tumor volumes were measured twice a week to evaluate the tumor growth rate, which was calculated using the formula: 0.532 × length × width2. All animal experiments were performed in compliance with worldwide standard animal care conditions via the Institutional Animal Care and Use Committee (IACUC). The research proposal was approved by the IACUC of Yonsei University.
Tumor biopsy samples
Paired biopsy samples were obtained from patients with ALK rearrangements before and after treatment with ALK TKIs at Yonsei Cancer Center (Republic of Korea), the National University Cancer Institute (Singapore), or the National Taiwan University Hospital (Taiwan). All patient tumor biopsy samples were obtained using protocols approved by the Institutional Review Board. All biopsy samples were obtained after written informed consent from the patients, and all procedures were conducted in accordance with the Declaration of Helsinki.
Conditional EML4-ALK transgenic mice
The Cre mouse line expressing Cre-ERT2 from the mouse SPC gene locus (SPC-Cre-ERT2 mice) was kindly provided by Dr. Harold Chapman at the University of California San Francisco. Conditional EML4-ALK transgenic mice were generated as previously described (19).
Chromatin immunoprecipitation-sequencing, RNA-seq, small RNA-seq, and microarray analysis
Chromatin immunoprecipitation-sequencing (ChIP-seq), RNA-seq, and small RNA-seq were performed in duplicate with the use of Illumina NextSeq500, as described in the Supplementary Information. The data were deposited in the NCBI Gene Expression Omnibus under accession number GSE97646 (in vitro microarray data), GSE81261 (in vivo microarray data), and GSE81487 (sequencing data).
Assessment of mRNA expression by qRT-PCR
For mRNA expression analysis, total RNA was isolated using an RNeasy Mini Kit (Qiagen), and cDNA was created using an iScriptTM cDNA Synthesis Kit (Bio-Rad). PCR reactions were prepared with iQTM SYBR Green Supermix and performed using a C1000TM Thermal Cycler (Bio-Rad). The gene encoding β-actin was amplified as a control, and relative quantifications of target mRNAs were analyzed using the comparative threshold cycle (CT) method. PCR primer sequences are listed in Supplementary Table S1.
Statistical analysis
Statistical analyses were performed using SPSS 21 and GraphPad Prism 5 software, and all data are expressed as the mean ± SD or ±SE for three or more individual experiments. Statistical significance was evaluated using the Student t test, the Mann–Whitney test, the Kruskal–Wallis with Dunn post hoc test, or ANOVA with Tukey post hoc test, as appropriate. P value of <0.05 was considered significant.
Results
Establishment of an in vitro model of ALK inhibitor–resistant lung cancer
To generate an in vitro model with which to study the mechanisms of acquired resistance to ALK inhibitor in lung cancer, we cultured two ALK fusion–positive lung adenocarcinoma cell lines (H3122 and H2228) with increasing concentrations of ceritinib (a second-generation ALK inhibitor) and established the ceritinib-resistant cell lines, H3122-LR and H2228-LR (Fig. 1A). Compared with parental cells, both H3122-LR and H2228-LR cells exhibited lower levels of phosphorylated and total ALK. Even after ceritinib treatment, ERK phosphorylation was not inhibited in H3122-LR and H2228-LR cells (Fig. 1B). To clarify the mechanisms of acquired resistance to ceritinib, we sequenced the cDNA encoding the ALK tyrosine kinase domain to determine whether mutations were present. No secondary mutations affecting the ALK tyrosine kinase domain, which spans exons 21 to 27, were detected. H3122-LR and H2228-LR cells were also resistant to the ALK inhibitors crizotinib and lorlatinib (Fig. 1C). These results suggested that the acquired resistance was not ALK dependent.
Ceritinib-resistant cells exhibit signs of the EMT in an ALK-independent manner. A, H3122, H3122-LR, H2228, and H2228-LR cells and individual resistant clones from LR cells (numbered) were treated with the indicated concentrations of ceritinib. The number of viable cells was measured at 72 hours. B, H3122, H3122-LR, H2228, and H2228-LR cells were treated with 1 μmol/L ceritinib for 6 hours. Cell extracts were assayed by Western blotting to detect the indicated proteins. C, H3122, H3122-LR, H2228, and H2228-LR cells were treated with the indicated concentrations of crizotinib or lorlatinib for 72 hours, after which the number of viable cells was measured. D, Phospho-RTK array showing increased phosphorylation of AXL in H3122-LR and H2228-LR cells compared with parental cells. E, Light microscopy images of H3122, H3122-LR, H2228, and H2228-LR cells. Scale bars, 100 μm.
Ceritinib-resistant cells exhibit signs of the EMT in an ALK-independent manner. A, H3122, H3122-LR, H2228, and H2228-LR cells and individual resistant clones from LR cells (numbered) were treated with the indicated concentrations of ceritinib. The number of viable cells was measured at 72 hours. B, H3122, H3122-LR, H2228, and H2228-LR cells were treated with 1 μmol/L ceritinib for 6 hours. Cell extracts were assayed by Western blotting to detect the indicated proteins. C, H3122, H3122-LR, H2228, and H2228-LR cells were treated with the indicated concentrations of crizotinib or lorlatinib for 72 hours, after which the number of viable cells was measured. D, Phospho-RTK array showing increased phosphorylation of AXL in H3122-LR and H2228-LR cells compared with parental cells. E, Light microscopy images of H3122, H3122-LR, H2228, and H2228-LR cells. Scale bars, 100 μm.
Next, we examined the extent of phosphorylation of receptor tyrosine kinases (RTK) in both LR cell lines to evaluate bypass pathway activation. Phospho-RTK arrays revealed increased phosphorylation of EGFR, HER2, and HER3 in H3122-LRcells, but not in H2228-LR cells (Fig. 1D). Treatment with gefitinib or afatinib did not reverse ceritinib resistance in both LR cells (Supplementary Fig. S1). We found that phosphorylation of AXL was remarkably increased in both H3122-LR and H2228-LR. Consistent with previous reports indicating that AXL was associated with the EMT (10, 20–22), both LR cells showed EMT-associated morphologic changes; specifically, the shape of the cells changed from round to fibroblast-like following the development of treatment resistance (Fig. 1E). Microarray analyses of H3122-LR cells were performed, and the gene set enrichment analysis of differentially expressed genes indicated the significant enrichment of EMT-related gene signatures (Supplementary Fig. S2A; Supplementary Table S2). To confirm this finding, we performed Western blotting of H3122, H3122-LR, H2228, and H2228-LR cells. Compared with parental cells, LR cells exhibited protein changes characteristic of the EMT, such as a reduced level of the epithelial marker E-cadherin (CDH1) and increased levels of N-cadherin (CDH2) and vimentin (VIM). Interestingly, levels of both phosphorylated and total AXL were potently increased in LR cells compared with parental cells (Supplementary Fig. S2B). We next examined whether upregulation of AXL was necessary for the EMT-related changes in LR cells. When LR cells were treated with the AXL inhibitor foretinib, AXL phosphorylation was inhibited, accompanied by the upregulation of CDH1 and downregulation of CDH2 and VIM, suggesting the AXL-dependent acquisition of the EMT phenotype (Supplementary Fig. S3A). Colony formation assays revealed a significant reduction in the relative colony formation rate when H3122-LR and H2228-LR cells were treated with foretinib (Supplementary Fig. S3B), but showed no synergistic effect in combination with ceritinib.
Massive enhancer remodeling accompanies acquired resistance to ALK inhibitors
To evaluate how the enhancer landscape is altered during ceritinib resistance, we first tested for global changes in histone modifications associated with enhancers (H3K4me1 and H3K27ac) and three other histone modifications with well-understood functions by Western blotting (23). Notably, compared with parental cells, H3122-LR cells displayed a decrease in global H3K27ac levels. Among the repressive signals (H3K27me3 and H3K9me3), the H3K27me3 level was slightly increased. The levels of H3K36me3, which is associated with transcribed regions in gene bodies, were similar between samples (Fig. 2A). H3K27 acetylation and methylation are mutually exclusive, and the acetyltransferase CBP/p300 complex and methyltransferase Polycomb repressive complex (PRC2) act in opposition to one another (24, 25). To elucidate the reasons for the global decrease in H3K27ac, we analyzed the expression of CBP/p300 and PRC2 components in H3122-LR cells using qRT-PCR (Fig. 2B).The mRNA expressions of CREBBP and EP300 were decreased in H3122-LR cells, suggesting that the global decrease in H3K27ac levels was at least partially due to the decreased expression of H3K27 acetyltransferase. To explore the expression changes in other epigenetic regulators during acquired resistance, we analyzed the expression levels of 720 genes encoding epigenetic regulators including 101 chromatin remodelers, 158 histone modification writers, 90 histone modification readers, and 66 histone modification erasers (26) in both H3122-LR cell– and H3122 cell–derived ceritinib-resistant xenograft tumors using RNA-seq and expression microarray, respectively. The expression of 124 genes was commonly increased, and the expression of 155 genes was commonly decreased in both in vitro and in vivo ceritinib-resistant models (Fig. 2C; Supplementary Fig. S4). Chromatin remodeler genes, such as CHD2, CHD6, GADD45A, and GADD45G, were upregulated, whereas CHD4, CHD5, and CHD7 were downregulated in resistant cells and xenograft tumors. Consistent with in vitro findings, EP300 and CREBBP were also decreased in resistant xenograft tumors. These data suggest that massive epigenetic remodeling may have occurred during ceritinib resistance in vitro and in vivo.
Dynamic changes in the H3K27ac signal in H3122-derived ceritinib-resistant cells. A, Decreased global levels of H3K27ac in LR cells compared with parental cells. Western blotting using the indicated antibodies, with total histone H3 as a loading control. B, Relative fold changes in mRNA levels of H3K27 acetyltransferase components (CREBBP and EP300) and H3K27 methyltransferase components (EZH2, EED, and SUZ12) in LR cells. Ct values were normalized to β-actin. The mean ± SD (n = 3) from two independent experiments is shown, and the P value was determined using the Student t test. *, P < 0.05; **, P < 0.01. C, Heatmap showing the expression profiles of the 124 commonly increased and 155 commonly decreased epigenetic regulator genes in both in vitro and in vivo ceritinib-resistant models. Generation of the in vivo–resistant model is described in Materials and Methods. The list of 720 genes encoding epigenetic regulators was downloaded from a database of human epigenetic factors and complexes, EpiFactors (http://epifactors.autosome.ru/; ref. 26). The expression profiles of H3122 and LR cells were obtained from RNA-seq data. The expression profiles of H3122 xenograft tumor (Con) and ceritinib-resistant xenograft tumor were obtained from the microarray data. Colors represent Z scores. D, Plot of the average level of H3K27ac at all peaks of H3122 or LR. E, Heatmaps of H3K27ac in H3122 and LR cells. The regions in which the H3K27ac level decreased or increased in LR cells were identified in the promoter or enhancer and ranked by the difference observed between H3122 and LR cells. Each row represents one peak centered at the midpoint between two 5-kb flanking sequences. F, Left plot, Kyoto Encyclopedia of Genes and Genomes pathway enrichment analysis of the nearest genes of H3K27ac loss or gain regions in the enhancer. Right plot, transcriptome analysis showing the percentage of genes that were upregulated (red) or downregulated (blue) ≥ 1.5-fold in LR cells.
Dynamic changes in the H3K27ac signal in H3122-derived ceritinib-resistant cells. A, Decreased global levels of H3K27ac in LR cells compared with parental cells. Western blotting using the indicated antibodies, with total histone H3 as a loading control. B, Relative fold changes in mRNA levels of H3K27 acetyltransferase components (CREBBP and EP300) and H3K27 methyltransferase components (EZH2, EED, and SUZ12) in LR cells. Ct values were normalized to β-actin. The mean ± SD (n = 3) from two independent experiments is shown, and the P value was determined using the Student t test. *, P < 0.05; **, P < 0.01. C, Heatmap showing the expression profiles of the 124 commonly increased and 155 commonly decreased epigenetic regulator genes in both in vitro and in vivo ceritinib-resistant models. Generation of the in vivo–resistant model is described in Materials and Methods. The list of 720 genes encoding epigenetic regulators was downloaded from a database of human epigenetic factors and complexes, EpiFactors (http://epifactors.autosome.ru/; ref. 26). The expression profiles of H3122 and LR cells were obtained from RNA-seq data. The expression profiles of H3122 xenograft tumor (Con) and ceritinib-resistant xenograft tumor were obtained from the microarray data. Colors represent Z scores. D, Plot of the average level of H3K27ac at all peaks of H3122 or LR. E, Heatmaps of H3K27ac in H3122 and LR cells. The regions in which the H3K27ac level decreased or increased in LR cells were identified in the promoter or enhancer and ranked by the difference observed between H3122 and LR cells. Each row represents one peak centered at the midpoint between two 5-kb flanking sequences. F, Left plot, Kyoto Encyclopedia of Genes and Genomes pathway enrichment analysis of the nearest genes of H3K27ac loss or gain regions in the enhancer. Right plot, transcriptome analysis showing the percentage of genes that were upregulated (red) or downregulated (blue) ≥ 1.5-fold in LR cells.
We next performed ChIP-seq for H3K27ac and H3K4me1 to explore whether H3122-LR cells display alterations in enhancer landscapes. The average level of H3K27ac showed an up to 22% decrease at the center of H3122-LR cells (Fig. 2D). A total of 78,353 H3K27ac peaks were observed in H3122 cells, and we identified 26,576 regions with a decreased H3K27ac signal and 8,888 regions with an increased H3K27ac signal in H3122-LR cells (Supplementary Fig. S5A and S5B). The average H3K4me1 signal was decreased in the regions with a lower H3K27ac signal and increased in regions with a higher H3K27ac signal (Supplementary Fig. S5C and S5D). Most of the regions with a loss or gain of signal were located in introns or intergenic regions (Supplementary Fig. S5E).
We defined promoters as ±2.0-kb regions around the transcription start site and enhancers as regions with H3K27ac and H3K4me1 signals that did not overlap with promoters. In enhancers, 18,044 regions showed H3K27ac loss and 8,509 regions showed H3K27ac gain in LR cells (Fig. 2E). The nearest genes of enhancer H3K27ac loss regions were associated with Jak-STAT signaling (e.g., CSF2, IFNGR1, and PIK3R1), Hippo signaling (e.g., CCND1, SMAD3, and DLG1), and miRNAs. Among the 5,606 genes in these regions, 10.6% (594) genes displayed a ≥1.5-fold downregulation in LR cells (Fig. 2F). The nearest genes to enhancer H3K27ac gain regions were associated with focal adhesion (e.g., GSK3B, VAV2, and LAMA3), proteoglycans (e.g., WNT5A, PXN, and FZD2), and the ErbB signaling pathway (e.g., EGFR, PTK2, and MAP2K2). Among the 3,776 genes in these regions, 18.7% (707) genes were upregulated ≥1.5-fold in LR cells (Fig. 2F). In promoters, 8,532 regions showed H3K27ac loss and 379 regions showed H3K27ac gain in LR cells (Fig. 2E). The genes displaying H3K27ac loss in their promoters were associated with the spliceosome (e.g., TRA2B, SNRPB2, and DDX23), RNA transport (e.g., NUP160, PNN, and SUMO2), and cell cycle (e.g., CCND1, CDK2, and PKMYT1). Among the 6,031 genes in these regions, 10.5% (632) genes were downregulated ≥1.5-fold in LR cells (Supplementary Fig. S6A). The genes displaying an H3K27ac gain in their promoters were associated with focal adhesion (e.g., SPP1, ACTN1, and CAV2), Ras signaling (e.g., RAP, RASAL2, and PDGFB), and glutathione metabolism (e.g., GPX2, MGST3, and GCLM). Although only 293 genes were present in this category, 41.9% (123) genes were upregulated ≥1.5-fold in LR cells (Supplementary Fig. S6B). Supplementary Table S3 shows gene expression levels in H3122 and LR cells and the promoter or enhancer H3K27ac gain or loss category for each gene. Together, these results indicate that massive enhancer remodeling during ceritinib resistance is associated with changes in the expression of mRNAs.
miR-34a and miR-449a reduce ceritinib-resistant cell proliferation in vitro and in vivo
We found the differentially enriched regions of H3K27ac were associated with miRNAs (Fig. 2F). So we next examined changes in miRNA expression during the acquisition of ceritinib resistance. We performed small RNA-seq and identified 265 upregulated and 188 downregulated miRNAs (Supplementary Table S4). The levels of miR-221 and miR-222, which promote the EMT and tumorigenicity by targeting PTEN and TIMP (27), were increased in LR cells (Fig. 3A). Among the downregulated miRNAs, we noted miR-34a, which represses AXL expression and inhibits the EMT (Fig. 3A; ref. 28). In addition, decreased H3K27ac levels in miR-34a loci were noted in LR cells. The expression of miR-449, another member of the miR-34 family, was also downregulated in LR cells. In addition to AXL, miR-34/-449 targets such as FGFR1 were also upregulated in LR cells (Supplementary Fig. S7A). miR-34a and miR-449a/b contain the same seed sequence, which is complementary to the AXL 3′UTR (Supplementary Fig. S7B). The relative levels of miR-34a and miR-449a were significantly reduced in both H3122-LR and H2228-LR cells compared with the parental cells (Fig. 3B). Data mining of The Cancer Genome Atlas also suggested an inverse correlation between AXL and miR-34a/449a expression in tissue samples of patients with lung adenocarcinoma (Supplementary Fig. S7C). We next examined whether forced expression or repression of miR-34a and miR-449a regulated cell proliferation and AXL-dependent EMT-related gene expression. The efficiency of mimics and inhibitors of miR-34a and miR-449a are described in Supplementary Fig. S8A and S8B. Forced expression of miR-34a and miR-449a in H3122-LR and H2228-LR cells significantly reduced cell viability (Fig. 3C) and yielded fewer colonies in colony formation assays (Fig. 3D). In addition, LR cells treated with miR-34a and miR-449a mimics exhibited decreased levels of p-AXL, total AXL, CDH2, and VIM and increased levels of CDH1(Fig. 3E). Based on these data, miR-34a and miR-449a negatively regulate AXL-dependent EMT gene expression during the acquisition of ceritinib resistance.
miR-34a and miR-449a disrupt the viability of ceritinib-resistant cells and regulate AXL expression. A, Scatter plot of miRNA expression in H3122-LR cells compared with H3122 cells. Red dots, >2-fold increase; blue dots, >2-fold decrease. Green, positions of the indicated miRNAs. Representative ChIP-seq profiles of regions with decreased H3K27ac signals at miR-34a loci. B, Relative expression of miR-34a and miR-449a in H3122-LR and H2228-LR cells compared with H3122 and H2228 cells, respectively. C, Cell viability assay using H3122-LR and H2228-LR transfected with miR-34a and miR-449a mimics. The mean ± SD are shown (n = 3), and the P value was determined using the Student t test. *, P < 0.05; **, P < 0.01. D, Colony formation assay. The mean ± SD is shown (n = 3), and the P value was determined using the Student t test. **, P < 0.01; ***, P < 0.001. E, Mimics of miR-34a and miR-449a were transfected into LR cells, and cell lysates were subjected to Western blotting to detect the indicated proteins. F, Comparison of the volumes of the xenograft tumors from miR-34a mimic–transfected H3122-LR cells and miR-449a mimic–transfected H3122-LR cells. Data are presented as the mean ± SD (the one-tailed Mann–Whitney U test: P < 0.05 compared with miR-C in each group, n = 3).
miR-34a and miR-449a disrupt the viability of ceritinib-resistant cells and regulate AXL expression. A, Scatter plot of miRNA expression in H3122-LR cells compared with H3122 cells. Red dots, >2-fold increase; blue dots, >2-fold decrease. Green, positions of the indicated miRNAs. Representative ChIP-seq profiles of regions with decreased H3K27ac signals at miR-34a loci. B, Relative expression of miR-34a and miR-449a in H3122-LR and H2228-LR cells compared with H3122 and H2228 cells, respectively. C, Cell viability assay using H3122-LR and H2228-LR transfected with miR-34a and miR-449a mimics. The mean ± SD are shown (n = 3), and the P value was determined using the Student t test. *, P < 0.05; **, P < 0.01. D, Colony formation assay. The mean ± SD is shown (n = 3), and the P value was determined using the Student t test. **, P < 0.01; ***, P < 0.001. E, Mimics of miR-34a and miR-449a were transfected into LR cells, and cell lysates were subjected to Western blotting to detect the indicated proteins. F, Comparison of the volumes of the xenograft tumors from miR-34a mimic–transfected H3122-LR cells and miR-449a mimic–transfected H3122-LR cells. Data are presented as the mean ± SD (the one-tailed Mann–Whitney U test: P < 0.05 compared with miR-C in each group, n = 3).
To further investigate the antiproliferative effects of miR-34a and miR-449a in vivo, miR-34a or miR-449a mimic–transfected LR cells were injected into nude mice to produce xenografts. Three weeks after injection, the miR-34a and miR-449a groups formed substantially smaller tumors compared with the controls (Fig. 3F). We also tested the effects of miR-34a and miR-449a inhibitors on H3122 parental cells. These inhibitors significantly increased cell viability (Supplementary Fig. S9A), increased the colony formation rate (Supplementary Fig. S9B), and increased tumor volumes in xenograft models (Supplementary Fig. S9C).
Panobinostat alters the H3K27ac profile and miRNA expression
We examined whether treatment with the HDAC inhibitor, panobinostat (also known as LBH589), could reverse the H3K27ac profile of H3122-LR cells. We conducted H3K27ac ChIP-seq, RNA-seq, and small RNA-seq after treatment with panobinostat (30 nmol/L) for 24 hours. In promoters, 8,718 regions showed H3K27ac gain and 1,207 regions showed H3K27ac loss in panobinostat-treated H3122-LR cells. In enhancers, 14,092 regions showed H3K27ac gain and 17,904 regions showed H3K27ac loss in response to panobinostat treatment (Fig. 4A). The genes with a gain of H3K27ac in their promoters were associated with insulin signaling, metabolic pathways, and ubiquitin-mediated proteolysis. Among the 6,754 genes in these regions, 5.0% (343) genes were upregulated ≥1.5-fold following panobinostat treatment (Supplementary Fig. S10A). The genes with a loss of H3K27ac in their promoters were associated with p53 signaling, NFκB signaling, and the cell cycle. Among the 947 genes in these regions, 6.3% (60) were downregulated ≥1.5-fold in response to panobinostat treatment (Supplementary Fig. S10B). The nearest genes of enhancer H3K27ac gain regions were associated with focal adhesion, calcium signaling, and hippo signaling. Among the 7,030 genes located in these regions, 4.5% (319) genes were upregulated ≥1.5-fold (Fig. 4B). The genes nearest to the enhancer H3K27ac loss regions were associated with NFκB signaling, hippo signaling, and PI3K-Akt signaling. Among the 5,693 genes located in these regions, 3.1% (178) were downregulated ≥1.5-fold (Fig. 4B). Supplementary Table S3 shows gene expression levels in panobinostat-treated H3122-LR cells and the promoter or enhancer H3K27ac gain or loss category for each gene. Taken together, the panobinostat treatment altered the H3K27ac signal in both promoters and enhancers, as well as the expression levels of genes involved in cancer-related pathways in H3122-LR cells.
Panobinostat induces changes in the prevalence of H3K27ac and miRNA expression. A, Heatmaps for H3K27ac levels in H3122-LR cells following panobinostat treatment (30 nmol/L for 24 hours). The regions in which the H3K27ac level was increased or decreased in the panobinostat-treated cells were identified and ranked by the difference between the panobinostat-treated and -untreated LR cells. Each row represents one peak centered at the midpoint, with a 5-kb flanking sequence. B, Left plot, Kyoto Encyclopedia of Genes and Genomes pathway enrichment analysis of the nearest genes of H3K27ac loss or gain regions in the enhancer. Right plot, transcriptome analysis showing the percentages of genes that were upregulated (red) or downregulated (blue) ≥1.5-fold in panobinostat-treated LR cells. C, Heatmap showing the relative levels of miRNAs that were downregulated in cells with acquired resistance to ceritinib or upregulated upon panobinostat treatment. Colors represent Z scores. D, Left plot, genome browser view of H3K27ac ChIP-seq data for miR-34a levels in three cell lines: H3122, LR, and panobinostat-treated LR cells. Right plot, qRT-PCR analysis of miR-34a levels. The mean ± SD is shown (n = 3), and the P value was determined using the Student t test. **, P < 0.01; NS, not significant. E, Left plot, genome browser view of H3K27ac ChIP-seq data for miR-449a/b/c levels. Right plot, qRT-PCR analysis of miR-449a levels. The mean ± SD (n = 3) from two independent experiments is shown, and the P value was determined using the Student t test. **, P < 0.01; ***, P < 0.001.
Panobinostat induces changes in the prevalence of H3K27ac and miRNA expression. A, Heatmaps for H3K27ac levels in H3122-LR cells following panobinostat treatment (30 nmol/L for 24 hours). The regions in which the H3K27ac level was increased or decreased in the panobinostat-treated cells were identified and ranked by the difference between the panobinostat-treated and -untreated LR cells. Each row represents one peak centered at the midpoint, with a 5-kb flanking sequence. B, Left plot, Kyoto Encyclopedia of Genes and Genomes pathway enrichment analysis of the nearest genes of H3K27ac loss or gain regions in the enhancer. Right plot, transcriptome analysis showing the percentages of genes that were upregulated (red) or downregulated (blue) ≥1.5-fold in panobinostat-treated LR cells. C, Heatmap showing the relative levels of miRNAs that were downregulated in cells with acquired resistance to ceritinib or upregulated upon panobinostat treatment. Colors represent Z scores. D, Left plot, genome browser view of H3K27ac ChIP-seq data for miR-34a levels in three cell lines: H3122, LR, and panobinostat-treated LR cells. Right plot, qRT-PCR analysis of miR-34a levels. The mean ± SD is shown (n = 3), and the P value was determined using the Student t test. **, P < 0.01; NS, not significant. E, Left plot, genome browser view of H3K27ac ChIP-seq data for miR-449a/b/c levels. Right plot, qRT-PCR analysis of miR-449a levels. The mean ± SD (n = 3) from two independent experiments is shown, and the P value was determined using the Student t test. **, P < 0.01; ***, P < 0.001.
We next examined whether panobinostat altered the miRNA expression profiles of H3122-LR cells. Among the downregulated miRNAs in LR cells, 70% maintained low expression levels (e.g., miR-34a, miR-375, and miR-29b-2), whereas 30% showed increased expression levels (e.g., miR-449, miR-192, and miR-132) after panobinostat treatment (Fig. 4C). We also observed an increase in the expression of 45 miRNAs after panobinostat treatment (e.g., miR-1271, miR-144, and miR-184), although the expression levels of these miRNAs were not initially decreased in H3122-LR cells (Fig. 4C). Consistent with the low expression level of miR-34a after panobinostat treatment, the miR-34a locus maintained the reduced H3K27ac signal (Fig. 4D). In contrast, miR-449a showed an increased H3K27ac signal and a significant increase in expression upon treatment with panobinostat (Fig. 4E). Based on these data, panobinostat is involved in altering the expression profiles of both mRNAs and miRNAs by inducing changing in the H3K27ac landscape in H3122-LR cells.
Panobinostat induces ablation of acquired resistant cells in vitro and in vivo
We next examined whether panobinostat could restore sensitivity to ceritinib in H3122-LR and H2228-LR cells using colony formation assays (Fig. 5A). Although panobinostat alone decreased tumor colony formation in a dose-dependent manner, combination of low-dose panobinostat (10 nmol/L) and ceritinib (0.5 μmol/L) produced significant reduction of colony formation, compared with panobinostat alone. Western blotting showed that treatment with panobinostat in combination with ceritinib resulted in a pronounced decrease in phosphorylated ERK levels (Fig. 5B) and reversed the EMT phenotype by upregulating CDH1 and downregulating p-AXL, total AXL, CDH2, and VIM (Fig. 5C). We also observed a significant increase in the number of apoptotic H3122-LR and H2228-LR cells following treatment with panobinostat in combination with ceritinib (Supplementary Fig. S11A). An analysis of the cell cycle revealed that panobinostat or ceritinib plus panobinostat induced cell-cycle arrest in the G1 phase and reduced the number of cells in the S phase transition in LR cells (Supplementary Fig. S11B). When markers related to apoptosis and cell-cycle arrest were examined by Western blotting, the panobinostat treatment increased the phosphorylation of cleaved PARP, cleaved caspase 3, and p21, but decreased pCDK2 expression (Supplementary Fig. S11C).
Antiproliferative effects of panobinostat on ceritinib-resistant cells. A, Colony-formation assays using H3122-LR and H2228-LR cells show the efficacy of ceritinib and panobinostat. The mean ± SEM are shown (ANOVA with Tukey post hoc test: **, P < 0.01 and ***, P < 0.001 compared with DMSO treatment alone; ##, P < 0.01 compared with the value for the indicated comparison in each cell line; n = 4). B, Western blots of phosphorylated AKT, t-AKT, phosphorylated ERK, and t-ERK in LR cells treated with ceritinib and increasing doses of panobinostat. C, Western blots for factors encoded by downstream effector genes of miR-34a and miR-449a, including phosphorylated AXL, t-AXL, CDH1, CDH2, and VIM. D, Ceritinib-resistant tumors derived from H3122 cells were treated with vehicle (n = 5), ceritinib 75 mg/kg (n = 7), panobinostat 5 mg/kg (n = 7), and both ceritinib 75 mg/kg and panobinostat 5 mg/kg (n = 7), as indicated. Photographs of excised tumors from each group are shown. The mean ± SD is shown (Kruskal–Wallis with Dunn post hoc test: NS, not significant; *, P < 0.05 and ***, P < 0.001 compared with the vehicle group; #, P < 0.05 compared with the value for the indicated comparison; n = 4). E, SPC-Cre-ERT2/EML4-ALK transgenic mice were treated with tamoxifen to induce EML4-ALK–mediated lung cancer. The mice were treated with ceritinib (75 mg/kg orally), and an MRI was performed to monitor the tumor response. The mice showed a complete response to ceritinib after 2 weeks of therapy, but continued treatment with ceritinib revealed disease progression after 16 weeks. The mice were further assigned to three treatment groups: (i) 75 mg/kg ceritinib (C), (ii) 5 mg/kg panobinostat (P), or (iii) 75 mg/kg ceritinib and 5 mg/kg panobinostat (C + P). These mice were examined to determine whether the treatment regimen overcame acquired resistance. CR, complete response; PD, progressive disease. Red arrows, lung tumor nodules.
Antiproliferative effects of panobinostat on ceritinib-resistant cells. A, Colony-formation assays using H3122-LR and H2228-LR cells show the efficacy of ceritinib and panobinostat. The mean ± SEM are shown (ANOVA with Tukey post hoc test: **, P < 0.01 and ***, P < 0.001 compared with DMSO treatment alone; ##, P < 0.01 compared with the value for the indicated comparison in each cell line; n = 4). B, Western blots of phosphorylated AKT, t-AKT, phosphorylated ERK, and t-ERK in LR cells treated with ceritinib and increasing doses of panobinostat. C, Western blots for factors encoded by downstream effector genes of miR-34a and miR-449a, including phosphorylated AXL, t-AXL, CDH1, CDH2, and VIM. D, Ceritinib-resistant tumors derived from H3122 cells were treated with vehicle (n = 5), ceritinib 75 mg/kg (n = 7), panobinostat 5 mg/kg (n = 7), and both ceritinib 75 mg/kg and panobinostat 5 mg/kg (n = 7), as indicated. Photographs of excised tumors from each group are shown. The mean ± SD is shown (Kruskal–Wallis with Dunn post hoc test: NS, not significant; *, P < 0.05 and ***, P < 0.001 compared with the vehicle group; #, P < 0.05 compared with the value for the indicated comparison; n = 4). E, SPC-Cre-ERT2/EML4-ALK transgenic mice were treated with tamoxifen to induce EML4-ALK–mediated lung cancer. The mice were treated with ceritinib (75 mg/kg orally), and an MRI was performed to monitor the tumor response. The mice showed a complete response to ceritinib after 2 weeks of therapy, but continued treatment with ceritinib revealed disease progression after 16 weeks. The mice were further assigned to three treatment groups: (i) 75 mg/kg ceritinib (C), (ii) 5 mg/kg panobinostat (P), or (iii) 75 mg/kg ceritinib and 5 mg/kg panobinostat (C + P). These mice were examined to determine whether the treatment regimen overcame acquired resistance. CR, complete response; PD, progressive disease. Red arrows, lung tumor nodules.
We also established H3122 xenograft tumors in nude mice, followed by treatment with ceritinib (50 mg/kg) to generate ceritinib-resistant tumors (defined as >25% regrowth in tumor volume; Supplementary Fig. S12A). Relative expression levels of AXL, CDH1, CDH2, and VIM were compared between baseline and resistant tumors. Resistant tumors showed significant upregulation of AXL and VIM, whereas CDH1 was significantly downregulated (Supplementary Fig. S12B). Mice bearing resistant tumors were then randomized to four treatment groups of 7 mice each: control, ceritinib (50 mg/kg), panobinostat (5 mg/kg), and a combination of ceritinib (50 mg/kg) and panobinostat (5 mg/kg). Consistent with the in vitro findings, treatment with the combination of ceritinib and panobinostat displayed superior antitumor efficacy compared with ceritinib or panobinostat alone (Fig. 5D). Tumors obtained from sacrificed mice showed a decrease in AXL expression after panobinostat treatment. In addition, we observed increased expression of miR-34a and miR-449 after treatment with ceritinib and panobinostat (Supplementary Fig. S12C).
We next examined the efficacy of ceritinib and panobinostat in an EML4-ALK transgenic mouse model (19). The mice showed a complete response to ceritinib after 2 weeks of therapy, although continued treatment for up to 16 weeks led to disease progression (Fig. 5E). To investigate whether the combination of ALK inhibitor and panobinostat could overcome acquired resistance, mice were further assigned to 3 treatment groups of 6 mice each: (1) ceritinib 75 mg/kg, (2) panobinostat 5 mg/kg, and (3) ceritinib 75 mg/kg and panobinostat 5 mg/kg. Indeed, combined treatment with ceritinib and panobinostat led to pronounced tumor shrinkage and complete remission after 2 weeks, whereas panobinostat alone led to minimal tumor shrinkage, and continued treatment with ceritinib alone led to further growth of the lung nodules (Fig. 5E).
ALK-rearranged patient samples
To explore the clinical relevance of our findings, we obtained paired tumor biopsies from 14 patients with EML4-ALK rearrangements before and/or after ALK inhibitor therapy. All 14 patients had been treated with ALK inhibitor (crizotinib or ceritinib), and both pre- and posttreatment samples were available for 9 of these patients (Table 1). We analyzed each sample for miR-34a, miR-449a, and AXL expression levels. In situ hybridization (ISH) of miR-34a and miR-449a showed that the expression levels of miR-34a or miR-449a were decreased in posttreatment tumor biopsies compared with pretreatment biopsies (n = 6/9, 67%). Immunohistochemistry for AXL showed increased expression in posttreatment tumor biopsies from 5 of 8 paired patients (63%; Fig. 6A). Of these 5 cases, three showed concomitant upregulation of AXL and downregulation of miR-34a or miR-449a. A representative image of a patient biopsy is shown in Fig. 6B. We also performed sequencing for secondary mutations in the ALK gene to interpret our findings in the context of previously identified mechanisms of resistance (Table 1). Among the eight samples available for ALK sequencing, two samples (25%) revealed the L1196M mutation, which is the most commonly known gatekeeper mutation. Interestingly, these two samples with theL1196M mutation showed relatively lower AXL expression (IHC score = 2), compared with others without secondary mutations (IHC score = 3–4). More importantly, among the other six samples without secondary mutations, two samples (33%) showed concomitant upregulation of AXL and downregulation of miR-34a or miR-449a. Coupled with the preclinical data, our findings suggest that repression of miR-34a or miR-449a and activation of AXL may be mutually exclusive mechanisms of acquired resistance to ALK TKIs in NSCLC.
Expression pattern of miR-34a, miR-449a, and AXL in pre- and post-ALK TKI paired samples (n = 14)
. | Type of ALK inhibitor . | Duration of treatment (month) . | Best response to ALK inhibitor . | miR-34a score for ISH . | miR-449a score for ISH . | AXL score for IHC . | ALK secondary mutation (posttreatment) . | |||
---|---|---|---|---|---|---|---|---|---|---|
Patient . | . | . | . | Pretreatment . | Posttreatment . | Pretreatment . | Posttreatment . | Pretreatment . | Posttreatment . | . |
1 | Ceritinib | 8 | SD | NA | 0 | NA | 0 | NA | 4 | NA |
2a | Crizotinib | 13 | PR | 4 | 1 | 2 | 4 | 3 | 3 | NA |
3a | Crizotinib | 16 | PR | 3 | 1 | 2 | 2 | 4 | 4 | None |
4a | Ceritinib | 11 | PR | 0 | 0 | 0 | 0 | 4 | 2 | L1196M |
5a | Ceritinib | 6 | SD | 0 | 0 | 0 | 0 | 1 | 3 | None |
6a,b | Ceritinib | 10 | PR | 0 | 0 | 3 | 1 | 2 | 4 | None |
7a,b | Crizotinib | 14 | PR | 0 | 0 | 4 | 2 | 1 | 4 | NA |
8 | Ceritinib | 2 | PD | NA | 1 | NA | 1 | NA | 4 | None |
9 | Ceritinib | 5 | SD | 4 | NA | 2 | NA | 2 | NA | NA |
10a | Crizotinib | 23 | SD | 2 | 2 | 2 | 2 | 2 | 3 | None |
11 | Crizotinib | 11 | PR | NA | 1 | NA | 1 | NA | 2 | L1196M |
12a,b | Crizotinib | 7 | PR | 1 | 1 | 2 | 0 | 2 | 4 | None |
13a | Ceritinib | 16 | PR | 2 | 1 | 4 | 4 | NA | NA | NA |
14 | Crizotinib | 14 | PR | NA | 3 | NA | 1 | NA | NA | NA |
. | Type of ALK inhibitor . | Duration of treatment (month) . | Best response to ALK inhibitor . | miR-34a score for ISH . | miR-449a score for ISH . | AXL score for IHC . | ALK secondary mutation (posttreatment) . | |||
---|---|---|---|---|---|---|---|---|---|---|
Patient . | . | . | . | Pretreatment . | Posttreatment . | Pretreatment . | Posttreatment . | Pretreatment . | Posttreatment . | . |
1 | Ceritinib | 8 | SD | NA | 0 | NA | 0 | NA | 4 | NA |
2a | Crizotinib | 13 | PR | 4 | 1 | 2 | 4 | 3 | 3 | NA |
3a | Crizotinib | 16 | PR | 3 | 1 | 2 | 2 | 4 | 4 | None |
4a | Ceritinib | 11 | PR | 0 | 0 | 0 | 0 | 4 | 2 | L1196M |
5a | Ceritinib | 6 | SD | 0 | 0 | 0 | 0 | 1 | 3 | None |
6a,b | Ceritinib | 10 | PR | 0 | 0 | 3 | 1 | 2 | 4 | None |
7a,b | Crizotinib | 14 | PR | 0 | 0 | 4 | 2 | 1 | 4 | NA |
8 | Ceritinib | 2 | PD | NA | 1 | NA | 1 | NA | 4 | None |
9 | Ceritinib | 5 | SD | 4 | NA | 2 | NA | 2 | NA | NA |
10a | Crizotinib | 23 | SD | 2 | 2 | 2 | 2 | 2 | 3 | None |
11 | Crizotinib | 11 | PR | NA | 1 | NA | 1 | NA | 2 | L1196M |
12a,b | Crizotinib | 7 | PR | 1 | 1 | 2 | 0 | 2 | 4 | None |
13a | Ceritinib | 16 | PR | 2 | 1 | 4 | 4 | NA | NA | NA |
14 | Crizotinib | 14 | PR | NA | 3 | NA | 1 | NA | NA | NA |
Abbreviation: SD, stable disease; PR, partial response; NA, not available tissue.
aPatients with matched pre- and postsamples available.
bPatients who showed decrease in miR-449a with concomitant increase in AXL expression.
ISH for miR-34a and miR-449a and IHC for AXL in patients with ALK rearrangements. A, Scores obtained from patient tissue samples obtained pre- and post-ALK inhibitor treatment for (i) miR-34a and miR-449a expression based on ISH assays and (ii) AXL expression based on immunohistochemical staining. The black dotted line represents paired samples that showed a decrease in miR-34a or miR-449 expression levels. The green dotted line represents a concomitant decrease in miR-34a or miR-449 and an increase in the AXL expression level. Data are presented as the mean ± SD. The P value was determined using the Mann–Whitney U test (two-tailed). B, Representative staining patterns for miR-34a, miR-449a, and AXL in patient tissues obtained pre- and post-ALK inhibitor treatment. Scale bar, 50 μm. Scramble is the negative control, and U6 is the positive control for ISH. C, Model of the mechanism of acquired resistance to ALK inhibitor due to massive enhancer remodeling. Resistant cells showed an EMT phenotype with AXL activation. During the acquisition of resistance, H3K27ac loss and repression of miR-34a led to the activation of its target genes, including AXL. Panobinostat induced H3K27ac gain and activation of miR-449 and inhibition of its target genes, such as AXL.
ISH for miR-34a and miR-449a and IHC for AXL in patients with ALK rearrangements. A, Scores obtained from patient tissue samples obtained pre- and post-ALK inhibitor treatment for (i) miR-34a and miR-449a expression based on ISH assays and (ii) AXL expression based on immunohistochemical staining. The black dotted line represents paired samples that showed a decrease in miR-34a or miR-449 expression levels. The green dotted line represents a concomitant decrease in miR-34a or miR-449 and an increase in the AXL expression level. Data are presented as the mean ± SD. The P value was determined using the Mann–Whitney U test (two-tailed). B, Representative staining patterns for miR-34a, miR-449a, and AXL in patient tissues obtained pre- and post-ALK inhibitor treatment. Scale bar, 50 μm. Scramble is the negative control, and U6 is the positive control for ISH. C, Model of the mechanism of acquired resistance to ALK inhibitor due to massive enhancer remodeling. Resistant cells showed an EMT phenotype with AXL activation. During the acquisition of resistance, H3K27ac loss and repression of miR-34a led to the activation of its target genes, including AXL. Panobinostat induced H3K27ac gain and activation of miR-449 and inhibition of its target genes, such as AXL.
Discussion
Although epigenetic mechanisms underlying acquired resistance to ALK TKIs remain elusive, we are the first group to show that downregulation of miR-34a or miR-449a induces the EMT using an integrated analysis of ALK-positive lung cancer models and a large clinical cohort of paired ALK-positive NSCLC specimens with acquired resistance to ALK TKIs. According to the analysis of clinical specimens, our resistance mechanism may occur mutually exclusively with secondary mutations in the ALK kinase domain. Consistent with the results from a previous study, we detected the L1196M mutation in 25% of cases (11), and concomitant upregulation of AXL and downregulation of miR-34a or miR-449a were observed in two samples without secondary ALK mutations.
Reversible transitions of epigenetic states enable tumor cells to resist cancer therapy (16, 29, 30), and an expanding body of literature supports a model in which miRNAs are involved in treatment resistance in various cancers (31–34). In the model described here, ceritinib-resistant cells exhibited an EMT phenotype with AXL activation. Genome-wide H3K27ac loss was also observed during resistance. A loss of enhancer activity led to the repression of miR-34a and activation of its target genes, including AXL (Fig. 6C). miR-34a is known to antagonize many different oncogenic processes, including apoptosis, proliferation, EMT, stemness, and Wnt signaling (35, 36). miR-449 is also a potent inducer of cell death, cell-cycle arrest, and/or cell differentiation (37). These two miRNAs belong to the same family and are structurally related. Intriguingly, the panobinostat treatment sensitized otherwise resistant mesenchymal phenotypes to ALK inhibition. From a therapeutic perspective, treatment with miRNAs is a new concept that should be investigated. miR-34a has shown antitumor efficacy in preclinical solid cancer models (38). Mechanistically, panobinostat induced an increase in H3K27ac levels, leading to the upregulation of miR-449. Although miR-34a expression did not increase following panobinostat treatment, miR-449 upregulation may compensate for the absence of miR-34a, because both miRNAs function redundantly by targeting AXL (12, 13).
The EMT is a fundamental process for tissue formation during development and in adult cell renewal. Activation of EMT signals in cancer is thought to be associated with stem-like properties, including migratory capabilities, and chemoresistance (29, 39–41). For EMT states, chromatin regulation of key transcription factors (TF) and their downstream targets is critical (29). EMT-related pathways, such as TGFβ, PI3K, MAPK, Hedgehog, and Wnt, can be controlled epigenetically via DNA methylation, histone modifications, or changes in small/noncoding RNAs (42). However, our understanding of the mechanism underlying the EMT during the acquisition of chemoresistance is still limited. Recently, epigenetic repression of transposable elements, including H3K9me3-mediated repression of LINE-1s, is known to occur via a mechanism in which a subpopulation of cancer cells transiently survive otherwise lethal drug exposures (43). In the present study, we primarily focused on regions with a decreased H3K27ac signal following acquired resistance, but regions with a higher H3K27ac signal were also observed. We found that specific TF binding motifs, such as those of MYC, TEAD4, and FOXA1, are enriched in these regions. Further studies are necessary to determine whether these TFs are important for the acquisition of drug resistance. Genetic and epigenetic mechanisms are not mutually exclusive in cancer evolution (44). Epigenetic adaptations may provide initial resistance and allow tumor cells to survive until they acquire secondary mutations that further drive progression (16, 45, 46). Furthermore, mutations in noncoding regions, such as enhancers and promoters, influence gene expression and can be new drivers for cancer progression and evolution (47–49).
We found that activation of AXL kinase occurred in the context of the EMT in ALK-positive NSCLCs. Increased expression of AXL and its ligand, growth arrest-specific protein 6 (GAS6), have been reported in EGFR-mutant NSCLC tumors obtained during resistance (22). AXL overexpression has also been implicated as a mechanism of resistance to ALK TKIs (50). AXL expression was important in the promotion of a mesenchymal phenotype, but concomitant inhibition of AXL and ALK did not restore sensitivity to ALK inhibitors. We, thus, postulate that epigenetic alterations in miRNAs targeting AXL are the key driver of resistance. Supporting this mechanism, we showed that reversing deacetylation with an HDAC inhibitor could overcome resistance in in vivo models.
Taken together, our findings highlight the importance of considering the epigenetic aspects of cancer drug resistance. Enhancer remodeling and miRNA expression profile changes should be considered as a novel acquired resistance mechanism associated with ALK inhibitors. This study provides a rationale for the development of new strategies for targeting epigenetic pathways to overcome resistance to ALK inhibitors.
Disclosure of Potential Conflicts of Interest
J.C.-H. Yang is a consultant/advisory board member for Novartis, Pfizer, Roche/Genentech, and Takeda. No potential conflicts of interest were disclosed by the other authors.
Authors' Contributions
Conception and design: M.R. Yun, S.M. Lim, Y.S. Kim, M. Kim, B.C. Cho
Development of methodology: M.R. Yun, S.M. Lim, S.-K. Kim, K.-H. Pyo, B.C. Cho
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.R. Yun, S.M. Lim, H.M. Choi, K.-H. Pyo, S.K. Kim, J.M. Lee, Y.W. Lee, H.R. Kim, K. Haam, N. Huh, H.S. Shim, R.A. Soo, J.-Y. Shih, J.C.-H. Yang, M. Kim, B.C. Cho
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.R. Yun, S.M. Lim, S.-K. Kim, S.K. Kim, J.W. Choi, J.-H. Kim, H.S. Shim, R.A. Soo, J.C.-H. Yang, M. Kim, B.C. Cho
Writing, review, and/or revision of the manuscript: M.R. Yun, S.M. Lim, H.R. Kim, M.H. Hong, H.S. Shim, R.A. Soo, J.-Y. Shih, J.C.-H. Yang, M. Kim, B.C. Cho
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.R. Yun, S.M. Lim, S.K. Kim, M. Kim
Study supervision: M. Kim, B.C. Cho
Acknowledgments
We thank all the patients who donated samples for this study. B.C. Cho is supported by a grant of the Korean Health Technology R&D Project, Ministry of Health and Welfare, Republic of Korea (HI12C1186), Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science, ICT & Future Planning (2016R1A2B3016282), and the Kim Jung Sook Julie Foundation. M. Kim is supported by the Collaborative Genome Program for Fostering New Post-Genome Industry of the NRF (2017M3C9A5028693) and the KRIBB Research Initiative grant. S.M. Lim is supported by the NRF grant funded by the Korean government (2016R1C1B1013299). M.R. Yun is supported by the NRF grant funded by the Korean government (2014R1A1A1006865).
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.