Aldehyde dehydrogenase 1A1 confers erlotinib resistance via facilitating the reactive oxygen species-reactive carbonyl species metabolic pathway in lung adenocarcinomas

Background: Acquired resistance to epidermal growth factor receptor (EGFR)-tyrosine kinase inhibitors (TKIs) such as erlotinib is a major challenge to achieve an overall clinical benefit of the targeted therapy. Recently, aldehyde dehydrogenase 1 (ALDH1) induction has been found to render lung adenocarcinomas resistant to EGFR-TKIs, and targeting ALDH1A1 becomes a novel strategy to overcome resistance. However, the molecular mechanism underlying such effect remains poorly understood. Methods: Comprehensive assays were performed in a panel of lung adenocarcinoma cell lines and xenografts that acquired resistance to erlotinib. Cancer phenotype was evaluated by cell viability, apoptosis, migration, and epithelial-mesenchymal transition analysis in vitro, tumorsphere formation analysis ex vivo, and tumor growth and dissemination analysis in vivo. Reactive oxygen species (ROS) and reactive carbonyl species (RCS) were detected based on fluorescent oxidation indicator and liquid chromatography coupled to mass spectrometry, respectively. Protein target was suppressed by RNA interference and pharmacological inhibition or ecto-overexpressed by lentivirus-based cloning. Gene promoter activity was measured by dual-luciferase reporting assay. Results: Knockdown or pharmacological inhibition of ALDH1A1 overcame erlotinib resistance in vitro and in vivo. ALDH1A1 overexpression was sufficient to induce erlotinib resistance. Metabolomic analysis demonstrated lower ROS-RCS levels in ALDH1A1-addicted, erlotinib-resistant cells; in line with this, key enzymes for metabolizing ROS and RCS, SOD2 and GPX4, respectively, were upregulated in these cells. Knockdown of SOD2 or GPX4 re-sensitized the resistant cells to erlotinib and the effect was abrogated by ROS-RCS scavenging and mimicked by ROS-RCS induction. The ALDH1A1 overexpressed cells, though resisted erlotinib, were more sensitive to SOD2 or GPX4 knockdown. The ALDH1A1 effect on erlotinib resistance was abrogated by ROS-RCS induction and mimicked by ROS-RCS scavenging. Detection of GPX4 and SOD2 expression and analysis of promoter activities of GPX4 and SOD2 under the condition of suppression or overexpression of ALDH1A1 demonstrated that the RCS-ROS-metabolic pathway was controlled by the ALDH1A1-GPX4-SOD2 axis. The ROS-RCS metabolic dependence mechanism in ALDH1A1-induced resistance was confirmed in vivo. Analysis of public databases showed that in patients undergoing chemotherapy, those with high co-expression of ALDH1A1, GPX4, and SOD2 had a lower probability of survival. Conclusions: ALDH1A1 confers erlotinib resistance by facilitating the ROS-RCS metabolic pathway. ALDH1A1-induced upregulation of SOD2 and GPX4, as well as ALDH1A1 itself, mitigated erlotinib-induced oxidative and carbonyl stress, and imparted the TKI resistance. The elucidation of previously unrecognized metabolic mechanism underlying erlotinib resistance provides new insight into the biology of molecular targeted therapies and help to design improved pharmacological strategies to overcome the drug resistance.

KEGG-pathway cluster analysis based on 1343 genes whose expression levels significantly changed. Untargeted metabolomics analyses were performed using LC-qTOF-MS/MS. The data was acquired and processed using MultiQuant software, version 2.0 (AB Sciex). Metabolites showing significantly different levels HCC827-ER5 cells compared with HCC827 cells were subjected to metabolic pathway-enrichment analysis using MetaboAnalyst 4.0 cell-viability assays, and this effect was abrogated by the GSH-synthesis inhibitor, BSO. The cells were exposed to erlotinib for 72 h. BSO was added 12 h before the point of the cell viability measurement. (F) DOX-induced ALDH1A1 expression increased the migration of HCC827 and PC9 cells, and these effects were reversed by a 12-h treatment with 100 μM BSO. BSO free data of each corresponding cell line as control.

Supplementary tables
Supplementary table S1. Oligonucleotides used in this study.

Oligonucleotides
Real-time RCR  The viability assay was performed by CCK8 analysis. The cells were exposed to erlotinib for 72 h.

Metastasis analysis
A total of 3×10 6 cells suspended in 100 μL PBS were subcutaneously inoculated into the left and right flanks of 5-week-old BALB/c nu/nu athymic mice. Metastasis of the subcutaneously inoculated tumors was analyzed by detecting the disseminated tumor foci in distant tissues.
Tissues were removed for pathological analysis at the end point of the test or when the mouse was to die. After formalin-fixed 4-µm-thick sections were cut for hematoxylin and eosin (H&E) stain, the number and the area of disseminated tumor foci were counted and measured by using the NanoZoomer-S210 system (Hamamatsu, Japan). In bioluminescent imaging assay for metastasis, HCC827 cells transfected with luciferase (3×10 6 ) were subcutaneously inoculated into the left and right flanks of 5-week-old BALB/c nu/nu athymic mice. At the end of the test, the brains of the mice were isolated for detection of tumor metastasis basing on bioluminescent imaging analyzed by using the IVIS system (PerkinElmer, USA).

RNA-Seq analysis
Total RNA was extracted according to the manufacturer's instructions (Takara, 9767) from the indicated cell samples. RNA was subjected to RNA-Seq analysis on a BGISEQ-500 system by the Beijing Genomics Institute (BGI). FPKM was used to calculate the gene expression level.
Differentially expressed gene screening was performed using the Noiseq method. Gene Ontology (GO) annotation was based on the GO database (http://www.geneontology.org/). The KEGG pathway database (http://www.genome.jp/kegg/) was used to perform the pathway enrichment analysis of differentially expressed genes.

Untargeted metabolomics
Sample preparation: Cells were seeded in 6-well plates (2 × 10 5 /well) in triplicate and allowed to adhere overnight. Cell samples were extracted with 1 mL (acetonitrile). The flow rate was set at 0.4 mL/min. The total elution time was 15 min and the elution condition was set as following: 0-2 min, 5% A; 2-9 min, 5%→55% A; 9-11 min, 55% A; including a 4 min equilibration time at 5% A. The metabolites were ionized using the electrospray ionization interface operating in negative ion mode and positive ion mode. IonSpray voltage was set at −4500 V (ESI-) and 5500 V (ESI+), curtain gas was kept at 35 psi, ion source temperature was 500 °C (ESI-) and 550 °C (ESI+), and nebulizing gas and drying gas were 55 psi. The collision energies were optimized with respect to the analyte between 15 eV and 45 eV to maximize the analyte response. Data was acquired and processed using MultiQuant software version 2.0 (AB Sciex).