FOXM1 Variant Contributes to Gefitinib Resistance via Activating Wnt/β-Catenin Signal Pathway in Patients with Non–Small Cell Lung Cancer

Non–small cell lung cancer (NSCLC), accounting for nearly 80% of all new lung cancer diagnoses, is one of the main causes of cancer-related mortality (1, 2). Fortunately, many EGFR tyrosine-kinase inhibitors (TKI) have been developed to improve the progression-free survival (PFS) of patients with EGFR-mutant NSCLC (3, 4). Although third-generation EGFR–TKIs (e.g., osimertinib) showed promising initial responses in patients with NSCLC with EGFR-activating mutation and even with T790M mutation (5), gefitinib, the first-generation EGFR–TKI, is still the standard treatment choice in China and many developing regions. Unfortunately, unpredictable resistance develops and limits its clinical application.

Several studies on gefitinib resistance mechanisms have found that EGFR T790M mutation, c-MET and HER2 amplifications, epithelial–mesenchymal transition (EMT), and cancer stemness were the main reasons for gefitinib resistance (6, 7). Strikingly, up to 30% of patients with NSCLC still suffered from progression with first-line gefitinib treatment due to unknown mechanism in our previous study (ACTIVE study, CTON1706; ref. 8). In addition, our recent study found that the concomitant mutations in EGFR were associated with lower objective response rate and shorter PFS with EGFR–TKI treatment, and the mechanism remains unclear (9). Therefore, it is important to search for new biomarkers to develop individualization of the treatment with more precision even for patients with NSCLC with EGFR mutation.

Because a large number of antitumor drugs, including gefitinib, suppressed the progression of cell cycle and the activity of DNA repair in tumor cells (10, 11), recent studies have identified that abnormal genetic alterations in cell-cycle and DNA repair pathway could be served as predictive factors for the resistance to platinum, targeted agents, and immunotherapy (12, 13). Although preclinical studies have identified several key factors of cell-cycle and DNA repair pathways in the regulation of response of gefitinib, such as ATM (11), FOXM1, and FOXO3 (14), and some previous study even found that XRCC1 399Arg/Arg, RRM1 2464GG, and ERCC1 8092CA were potential predictors for gefitinib response in general patients (15), it is still unclear whether genetic alterations in cell-cycle and DNA repair pathway affected the response of gefitinib in patients with EGFR-activating NSCLC.

In addition, our previous studies have identified that polymorphisms of pharmacokinetics factors were correlated to the PFS in a small cohort of patients with EGFR-mutant NSCLC with gefitinib treatment (16). Therefore, all these pathways and related genes possibly involved in regulation of gefitinib response, polymorphisms in all these pathways, and related genes are worthy of genotyping. In this study, 125 variants involving pharmacokinetics factors, cell-cycle and DNA repair factors in two cohorts of patients with NSCLC with EGFR mutation were analyzed to identify predictive biomarkers and develop individualization of the gefitinib treatment.

Between November 2013 and July 2018, a total of 282 patients with NSCLC were enrolled in Sun Yat-sen University Cancer Center, and were randomly assigned in a 7:3 ratio to exploratory (n = 192) and validation (n = 90) cohorts. Overall eligibility criteria were ages ≥18 years, and histologically/cytological confirmed stage IIIB/IV adenocarcinoma with a sensitive EGFR (i.e., exon 19 deletion and exon 21 L858R) mutation (determined in tumor tissue or circulating tumor cell); the absence of the resistant EGFR T790M mutation, c-MET and HER2 amplifications, etc.; all patients had at least one measurable lesion assessed by RECIST version 1.1; Eastern Cooperative Oncology Group performance status of 0 or 1, and no organ dysfunction. Patients were eligible for exclusion from the study if they were pregnant, lactating women, or likely to be pregnant; or had severe malabsorption syndrome or interstitial pneumonia or pulmonary fibrosis.

All patients received gefitinib 250 mg daily until disease progression, intolerable adverse events, or other reasons for withdrawal. A total of 2 mL peripheral blood with EDTA anticoagulation was collected and kept in −80°C refrigerator until analysis. The study was approved by Human Ethics of Sun Yat-sen University Cancer Center (B2013–038–1) and conducted in accordance with the principles of the Declaration of Helsinki and the Good Clinical Practice Guidelines of the International Conference on Harmonization. All patients provided written informed consent. This study was registered at ClinicalTrials.gov (NCT01994057).

Genomic DNA were extracted and purified with commercial kit according to the manufacturer's instruction (Tiangen). All tag SNPs were selected by Haploview 4.2 (Broad Institute; https://www.broadinstitute.org/haploview/haploview) within Chinese Beijing (CHB) in HapMap (Supplementary Table S1). The allele frequencies were lower bound 0.05 and upper bound 0.5. The threshold of linkage disequilibrium (LD) was defined as r² ≥ 0.8. The candidate polymorphisms were genotyped by using a previously published Agena MassARRAY System technique (Agena; ref. 17) and Sanger sequencing. EGFR driver mutations were detected in tumor tissue DNA by the ADx–ARMS method.

RandomForestSRC (RSFSRC), an efficient machine-learning algorithm to analyze survival data, was used to select predictive feature variables for gefitinib response (18, 19). In brief, (i) codominant genetic models were analyzed in R with survival package and SNPs with P > 0.1 were removed; (ii) bootstrap samples (average 67%) were built from the exploratory cohort, and the excluded samples (33%) were named out-of-bag data (OBB data); (iii) a survival tree was built for each bootstrap data. At each node of the survival tree, randomly selected covariates and all SNPs from the step 1 with a log-rank splitting rule that maximized the survival difference between daughter nodes; (iv) the prediction error was calculated by using the OBB samples; (v) the most explanatory factors were selected as feature variables through overlap analysis by using minimal depth and VIMP, which were two popular methods for identification of feature variables; (vi) the top explanatory factors were presented into COX regression model with hazard ratio (HR) as the weight for each predictor (HR, –1.49 = 1; HR, 1.5–2.49 = 2; HR, 2.5–3.49 = 3; ref. 18); (vii) each patient was scored with this system and regrouped by their scores. Survival analysis was conducted with regrouped patients in the exploratory and validation cohorts.

HCC827, PC9, and H293T cell lines were obtained from National Collection of Authenticated Cell Cultures of China. PC9 and HCC827 cells were maintained in RPMI1640 (GIBIO) supplied with 10% FBS (Gibco). The H293T cell line was cultured in DMEM (GIBIO) with 10% FBS. All cells were cultured with the media containing 1% penicillin (100 U/mL) and streptomycin (100 U/mL) at 37°C and 5% CO₂ in a humidified chamber. Cells were checked for Mycoplasma (Beyotime, China) upon thawing. Gefitinib, niclosamide, actinomycin D, and cycloheximide were purchased from Selleck. Puromycin was provided by Yeasen (Shanghai, China). All cell lines were tested Mycoplasma negative.

The stable knockout (KO) FOXM1 (NM_202002) was accomplished by CRISPR/Cas9 system (PX459 V2.0) with a small guided RNA (Supplementary Table S2). PX459 V2.0 was transfected into the desired cells by Lipofectamine 3000 (Thermo Fisher Scientific). Puromycin (2 μg/mL for PC9 cells, 1 μg/mL for HCC827 cells) was used to select the positive cells. For overexpression of FOXM1, FOXM1 rs3742076_G, and FOXM1 rs3742076_A were cloned into GV492 lentiviral vector (GeneChem). The virus was produced, filtered, and titrated according to the instruction, and infected HCC827 and PC9 cells with P solution (GeneChem).

Western blot analysis was performed according to standard procedure. In brief, the total protein was harvested by RIPA buffer (Beyotime) supplemented with complete protease inhibitor cocktail. The proteins were separated with SDS-PAGE gels and transferred to polyvinylidene difluoride membranes (Millipore). After blotting with related primary antibodies, the membranes were incubated with the secondary antibodies. For protein stability assay, cells were treated with 10 μg/mL cycloheximide for 0, 1, 2, 4, 6 hours following SDS-page analysis. For separation of nuclear and cytoplasmic protein of cells, a Nuclear Protein and Cytoplasmic Protein Extraction Kit was used according to the manufacturer's instruction (Beyotime). All the antibodies or protein were available in commercial companies: anti-GAPDH (cat: A1906) and anti–β-catenin (cat: A19657) in Abclonal, and anti-Histone H3 (cat: 4499) in Cell Signaling Technology for Western blot; anti-FOXM1(cat: 20459) for western blot, IHC, co-immunoprecipitation (Co-IP), and chromatin immunoprecipitation (ChIP) in Cell Signaling Technology; goat anti-Rabbit IgG H&L [horseradish peroxidase (HRP); cat: A0201] was purchased from Beyotime. Normal IgG (cat: 2729) was provided by Cell Signaling Technology.

The cells were lysed in buffer for Western and IP (Beyotime) containing complete protease inhibitor cocktail. After incubation with indicated antibodies at 4°C overnight, the lyses was incubated with protein G-magnetic beads (Cell Signaling Technology) for 20 minutes at room temperature. The complexes of lyses and magnetic beads were washed with the lysis buffer for 5 times. The protein was analyzed by western blot according to standard procedures.

The extraction of total RNAs was described in previous study. In brief, RNAs were isolated with TRizol (Thermo Fisher Scientific) according to the manufacturer's instruction. The cDNA was reversely transcripted with the PrimeScript RT Reagent Kit (RR036A, Takara). The expressions of candidate genes were determined by SYBR Green Master Mix (RR820A, Takara) in Applied Biosystems 7500 Real-Time PCR Systems (ABI). GAPDH was selected as an internal control for calculation of the levels of candidate gene expression. For RNA stability assay, cells were treated 5 μmol/L actinomycin D for 0, 1, 2, 4, and 6 hours followed real-time PCR (RT-PCR) analysis. All premiers for RT-PCR analysis were shown in Supplementary Table S2.

The migration of cancer cells was detected by wound-healing assay. The monolayer cells were seeded into 6-well plates until 90% confluence. The cells were wounded by a 10-μL pipette tip and cultured with FBS-free medium for 72 hours. The migration distances were measured with a microscope and ImageJ.

To study the invasion of tumor cells, 0.5 mL (5×10⁴ cells/mL) FBS-free cells suspension was added into 24-well invasion chamber (353097, Corning) coating with Matrigel matrix (354234, Corning) and 0.75 mL culture medium containing 10% FBS were added into the wells. After incubation for 16 hours, the migrated cells were fixed with methanol and stained with crystal violet.

For Cell Counting Kit-8 (CCK-8) assay, 5,000 cells were seeded into plates with 96 wells. Cell viability was detected with CCK8 (Invigentech) according to the manufacturer's instruction. The absorbance was obtained at 450 nm with a Microplate Reader (Thermo Fisher Scientific) after incubation at 37°C for an additional 2 hours. For proliferation assay, cells were cultured with or without gefitinib for 72 hours followed by cell counting or CCK8 analysis.

To introduce rs3742076 into HCC827 (FOXM1 S628) cells, a 180 nt single-stranded oligodeoxynucleotides was synthesized by Sangon with the desired A > G substitution (Supplementary Table S2). The sgRNA sequences were shown in Supplementary Table S2. After selection of single cell–derived colonies and Sanger sequencing, up to 300 clones were picked and screened for successful homologous recombination (Supplementary Table S2).

PC9 wild-type, FOXM1 rs3742076_A, and FOXM1 rs3742076_G cells (1×10⁶ cells) were injected subcutaneously into the right flanks of 4-week-old Balb/c (nu/nu) mice (Animal center of Sun Yat- sen University, Guangzhou, China). The mice were randomly treated with vehicle (5% tween 80) and gefitinib (20 mg/kg) when the tumor volumes reached about 100 mm³. The tumor volumes were measured by using digital calipers and calculated by using a standard formula: tumor volume = width² × length/2. The tumor was harvested and fixed with 4% formalin. This study was approved by the Institutional Animal Care and Use Committee of Sun yat-Sen University (SYSU-IACUC-2021-B1288). All animal procedures were compliant with all relevant ethical regulations regarding animal research.

The tumor samples were dehydrated with ethanol and embedded in paraffin after being fixed with 4% formalin for more than two days. The tumor sections were cut and stained with hematoxylin and eosin. For IHC assay, the tumor section was deparaffinized and rehydrated with an alcohol series. The antigen retrieval for the tumor was through sodium citrate buffer. The tumor slides were incubated with appropriate primary antibodies at 4°C overnight followed by incubation with secondary antibodies. HRP conjugates were used to stain the IHC assay. Nuclei were counterstained with 4′,6-diamidino-2-phenylindole. Images were taken with Olympus and analyzed by Image Plus 6.0.

The total RNA was purified by Dynabeads Oligo (dT; Thermo Fisher Scientific) followed by Fragmentation through Magnesium RNA Fragmentation Kit (NEB). The cDNA was transcript by Invitrogen SuperScript II Reverse Transcriptase and digested by Uracil-DNA Glycosylase (NEB). The cDNA library size was 300–500bp and sequencing on an Illumina Novaseq 6000 (LC-Bio Technology CO., Ltd.) following the vendor's recommended protocol. The deferential transcripts and genes were analyzed by R package edgeR (https://bioconductor.org/packages/release/bioc/html/edgeR.html) or DESeq2 (http://www.bioconductor.org/packages/release/bioc/html/DESeq2.html) with fold change >2 or fold change <0.5 and P < 0.05. The differentially expressed mRNAs were enriched by gene ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG) and gene set enrichment analyses (GSEA).

The promotor sequences of CTNNB1 (NM_001904-promoter) were sub-cloned into dual-luciferase reporter vector (GV238, Genechem). The Dual Luciferase Reporter Assay System (Promega) was used to assay the activity of luciferase in overexpression FOXM1 rs3742076_A or rs3742076_G cells. The expression efficiency was calculated by the ratio of Firefly luciferase activity value and Renilla luciferase activity value.

The chromatin DNA was prepared by truChIP Chromatin Shearing Kit (Covaris) according to the manufacturer's instruction. In brief, the cells were collected and re-suspended in fixed buffer followed crosslinking protein–chromatin with freshly preparing 1% formaldehyde solution (Cell Signaling Technology). The cross-linked cells were lysed with lysis buffer B containing protease inhibitor cocktail. The nuclei were isolated and collected by centrifugation at 1,700 × g for 5 minutes at 4°C. The chromatins were sheared by an AFA Focused-ultrasonicator (Covaris) with appropriate 1-mL tubes. The ChIP reactions were used to enrichment of specific DNA with appropriate antibody according to the instruction (Active Motif). The ChIP DNA were de-cross-linked, purified, and analyzed by RT-PCR.

For sphere forming assay, single-cell suspension was prepared at the density of 2.5×10³ cells/mL and seeded into 6-well ultra-low attachment plates (Corning). The culture medium was a serum-free DMEM-F12 (Gibco) containing 20 ng/mL EGF (Gibco), 20 ng/mL bFGF (Gibco), insulin (1X, Gibco), and 0.4% BSA (Sigma) and replaced every 3 days. The size and number of the spheres were recorded after 7 days.

For clonogenic growth assay, cells were planted in 6-wells plates at a density of 2,000 cells/well. Gefitinib was added after an additional 24 hours, with media change and fresh drug addition every 3 days. Cells were fixed with methanol for 20 minutes and stained with 0.1% crystal violet after cultured with or without gefitinib for 10 to 14 days.

R version 3.6.0 (R Development Core Team, Austria) and GraphPad Prism version 9.0 (https://www.graphpad.com/scientific-software/prism/) were used for all the analyses. RandomForestSRC (https://cran.r-project.org/web/packages/randomForestSRC/index.html), ggRandomForests (https://CRAN.R-project.org/package=ggRandomForests), survival (https://cran.r-project.org/web/packages/survival/index.html), and survminer (https://cran.r-project.org/web/packages/survminer/index.html) packages was used to analyze and present the survival data. Cox regression were conducted by glmnet package (https://cran.r-project.org/web/packages/glmnet/index.html) and presented by ggplot2 (https://cran.r-project.org/web/packages/ggplot2/index.html). For bioinformatics analysis, The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) were used in this study. All statistical tests were carried out at a two-sided nominal 0.05 significance level.

The transcriptome experiment is available via the NCBI GEO repository under the identifier code GSE201020. Individual participant data are not publicly available because this requirement was not anticipated in the study protocol considering that this trial started patient enrollment in 2013. Please email the corresponding authors with requests for this data.

Patients’ characteristics and features are shown in Table 1. A total of 282 patients with NSCLC were enrolled and randomly assigned in a 7:3 ratio to exploratory (n = 192) and validation (n = 90) cohorts. The median age was 57.5 (range, 28–88) and 57.0 (range, 37–77) years in exploratory and validation cohorts, respectively. The majority of patients were in stage IV (exploratory vs. validation cohorts: 94.27% vs. 93.33%), were non-smokers (exploratory vs. validation cohorts: 83.85% vs. 78.78%), carried EGFR 19 del (exploratory vs. validation cohorts: 57.29% vs. 57.78%), and received gefitinib as first-line therapy (exploratory vs. validation cohorts: 85.42% vs. 82.22%). No significance was found in the baseline characteristics between exploratory cohort and validation cohort (Table 1).

One hundred twenty-five SNPs in 47 candidate genes were successfully genotyped in this study. Fifteen SNPs were not included in last analysis due to limit minimum allele frequency (<0.05) in enrolled patients and call rate (<95%) during genotyping (Supplementary Table S1). We found that 18 SNPs (P < 0.1) were significantly associated with PFS in patients with NSCLC (Fig. 1A; Supplementary Table S1), including XRCC1 rs3213263 (P = 0.0001), FOXM1 rs2302257 (P = 0.002), ERCC1 rs10408989 (P = 0.0085), RAD51 rs11633367 (P = 0.013), RAD51 rs34285050 (P = 0.0169), FOXO3 rs75544369 (P = 0.0172), FOXO3 rs13207511 (P = 0.018), ERCC1 rs8101974 (P = 0.0187), ERCC1 rs10417603 (P = 0.0289), FOXM1 rs10848718 (P = 0.0475), SLC22A8 rs4149179 (P = 0.0536), STK11 rs34072429 (P = 0.0644), CDK4 rs2270777 (P = 0.0702), FOXO3 rs3813498 (P = 0.0787), FOXO3 rs13220810 (P = 0.0848), STAT6 rs3024975 (P = 0.0916), IL16 rs4778889 (P = 0.0933), and CDK6 rs10225965 (P = 0.0947).

Five thousand survival trees were established on the basis of the 18 SNPs and clinical confounder factors through Randomforest Survival analysis, and the OBB error rate was 0.356 (Supplementary Fig. S1A). Top predictive variables were chosen by the intersection of the ranks between minimal depth and VIMP, including rs3213263, rs2302257, rs13207511, rs75544369, rs11633367, rs10408989, rs2270777, and rs3428505 (Supplementary Fig. S1B). To determine how PFS depends on each top variable, partial dependence plots were generated by integrating out the effects of variables besides the covariate of interesting 1, 3, and 5 years (Supplementary Fig. S1C–S1J). Taken together, rs3213263 CT, rs2302257 CC, rs75544369 GA, rs10408989 GG, rs227077 CC, rs13207511 GG, rs11633367 GG, and rs3428505 AA genotypes were risk factors for patients with NSCLC with gefitinib treatment.

To validate and construct the Gefitinib Prognosis Score for EGFR mutant patients with NSCLC (GPSE), we introduced the top predictors into a Cox multivariate model. The result showed that XRCC1 rs3213263 [CT vs. CC; HR, 1.904; 95% confidence interval (CI), 1.230–2.948; P = 0.004], FOXM1 rs2302257 (CC vs. GG; HR, 2.987, 95% CI, 1.534–5.814; P = 0.001), FOXO3 rs75544369 (GA vs. GG; HR, 1.787; 95% CI, 1.056–3.015; P = 0.031), ERCC1 rs10408989 (TT vs. GT; HR, 1.689; 95% CI, 1.127–2.531; P = 0.011; GG vs. GT; HR, 4.316; 95% CI, 1.967–9.471; P < 0.0001) were included in the model (Fig. 1B).

The GPSE and its corresponding validation set were listed in Supplementary Table S3. According to the scores of each patient, we divided the patients in four groups. In the exploratory cohort (Fig. 1C), the median PFS were 26.23 (95% CI, 14.23–35.37), 16.80 (95% CI, 12.87–23.60), 11.13 (95%CI, 8.87–16.70) and 6.25 (95% CI, 4.43–11.67) months for 0, 2, 3 and 4, and ≥5 groups (P < 0.0001), respectively. In the validating cohort (Fig. 1D), the median PFS were 18.80 (95% CI, 11.800–64.53), 13.60 (95% CI, 9.17–27.33), 13.47 (95% CI, 9.83–18.40), and 7.40 (95% CI, 3.33–NA) months for 0, 2, 3, and 4, and ≥5 groups (P = 0.0017), respectively. Collectively, rs3213263, rs2302257, rs75544369, and rs10408989 presented meaningful associations with PFS in patients with NSCLC with gefitinib in each of the datasets.

We therefore hypothesized that these variants predicted the prognosis of gefitinib in NSCLC due to altering the expression or function of these genes. To determine whether levels of these genes are correlated to the prognosis in lung cancer, we first analyzed the association between expression levels and survival in patients with lung adenocarcinoma (LUAD), we found that only FOXM1 expression level was associated with the progression-free interval and overall survival (OS) in patients with LUAD according to TCGA database (Fig. 2A and B). To further validate whether FOXM1 expression level is correlated to survival in patients with LUAD, we then performed bioinformatics analysis in GEO datasets. The results showed that the FOXM1 high expression was a risk factor for patients with LUAD in multiple datasets (Supplementary Fig. S2A–E). Meanwhile, the expression level of FOXM1 in tumor was significantly higher compared with those in normal samples (Supplementary Fig. S2F), indicating that FOXM1 was characterized as a potential oncogenic role in lung cancer cells. Notably, the expression of FOXM1, not XRCC1, FOXO3, and ERCC1, was found to be correlated to prognosis in patients with EGFR-activating mutation according to the TCGA dataset (Fig. 2C), which was similar to the reports from Kakiuchi and colleagues (20). Furthermore, FOXM1 expression level was significantly higher in gefitinib-resistant cells compared with that in sensitizing cells according to GSE123066 (Fig. 2D), which was similar to previous studies (14, 21). Collectively, these data suggested that FOXM1 expression levels might be a predictor for response of gefitinib in vivo and in vitro.

Nowadays, in pharmacogenomics studies, the candidate SNPs are usually selected through softwares of Haploview 4.2, LDpair Tool, and SNPPicker, in which SNPs are separated into different sets based on r² (the value of linkage disequilibrium), and in each set just one randomly selected SNPs (named as tagSNP) need to be detected in laboratory experiments for the sake of efficiency and feasibility. In our study, rs2302257, and rs3742076 are in the same set and rs2302257 is the tag SNP. In addition, rs2302257 and rs3742076 are in the same set due to the high r² of 0.97 (Supplementary Fig. S2G) in the Haploview 4.2. And in our investigation, the value of LD r² between these two SNPs is of 0.98 (Supplementary Fig. S2H). Rs2302257 was found to be significantly correlated with the poor PFS of the patients with NSCLC with EGFR mutation. Because rs2302257 is an intron polymorphism, according to previous reports (22, 23), the intron polymorphism may influence the protein function either by linking with some other coding regions’ SNPs or by influencing the truncate/expression of the gene. Through bioinformatics analysis by SNPSplicer (24) and Human ACtive Enhancers to interpret Regulatory (HACER; ref. 25), rs2302257 is not possible a splicing or enhancer SNP and not probable to cause a new isoform or regulate expression of FOXM1. Then we analyze the other SNPs in the same set with rs3742076.

Furthermore, we investigated whether rs3742076 or rs2302257 had a functional effect on FOXM1 by SNPinfo, which was an integrating tool for the prediction of functional SNPs in genetic association studies (26) The results showed that only rs3742076 had a potential regulating effect for FOXM1 (Supplementary Table S4), implying the association between rs2302257 and PFS of patients with NSCLC in clinical observation was mainly mediated by rs3742076; therefore, in the following study, we mainly focus on the effect of rs3742076, rather than rs2302257, on FOXM1.

We next sought to determine whether rs3742076 was associated with PFS in patients with NSCLC. Only 188 patients in exploratory cohort and 86 patients in validation cohort were genotyped successfully by Sanger sequencing. We found that rs3742076 significantly stratified patient survival in response to gefitinib in exploratory cohort [median PFS, GG vs. AA&AG: 9.2 (95% CI, 4.43–13.97) vs. 13.37 (95% CI, 12.17–19.40; P = 0.00039) HR = 2.399], which was validated in an independent cohort [median PFS, GG vs. AA&AG: 8.13 (95% CI, 7.4–NA) vs. 13.80 (95%CI, 10.97–17.90; P = 0.048) HR, 2.628; Fig. 2E and F]. Taken together, these findings strengthened our hypothesis that the FOXM1 rs3742076 determined the response to gefitinib in patients with NSCLC.

To characterize the phenotypic and the genotypic consequences of FOXM1 variation, we established stable FOXM1-KO models in PC9 and HCC827 cells by CRISPR/Cas9 system (Fig. 3A and B). To further determine whether rs3742076 promote cells resistance to gefitinib in vitro, we generated stable overexpressed FOXM1 rs3742076_A (expressed wild-type FOXM1, FOXM1^Ser628Ser) and FOXM1 rs3742076_G (expressed mutant FOXM1, FOXM1^Pro628Pro) in the FOXM1-KO models (Fig. 3C). We found that introduction FOXM1 rs3742076_G in cells significantly increased the IC₅₀ value of gefitinib (Fig. 3D and E) and enhanced colony formation with or without gefitinib treatment (Fig. 3F).

Previous studies have revealed that FOXM1 is the one of pivotal regulators in tumor migration, invasion, and proliferation (27). To determine whether the variant affect migration, invasion, and proliferation in NSCLC cells, wound-healing, Transwell, and cell counting assays were performed. Enhancement of NSCLC invasion (Fig. 3G and H) and proliferation (Fig. 3I–L) were found in NSCLC cells overexpressing mutant FOXM1, other than in cells overexpressing wild-type FOXM1, regardless of culturing with gefitinib or not. Meanwhile, although both expressers led to increased cell migration, no significant difference was found between cells overexpressing wild-type and mutant FOXM1, regardless of gefitinib incubation or not (Supplementary Fig. S3A–S3D). Collectively, these results indicated that overexpression of FOXM1 rs3742076_G enhanced the invasion and proliferation, and promoted an aggressive phenotype in NSCLC cells thus contributing to NSCLC cell resistance to gefitinib and resulting in NSCLC progression.

We noticed that FOXM1 protein level in rs3742076_G cells was higher compared with that in rs3742076_A cells (Fig. 3C), implying that the differences in RNA or protein stability might be caused by the variant. Next, we sought to determine the RNA stability in cells by incubating with actinomycin D. Unfortunately, no significant difference was found between cells expressing FOXM1 rs3742076_G and rs3742076_A (Supplementary Fig. S3E and S3E1), indicating that the variant had limited effect on mRNA stability of FOXM1. Rather, the variant increased the protein stability of FOXM1 as evaluated in the CHX chase assay (Supplementary Fig. S3F and S3F1), which was similar to our results from the in silico analysis (Supplementary Fig. S3G) by I-Mutant v2.0 (28). Taken together, these results indicated that the variant upregulated FOXM1 through increasing protein stability.

To further explore the functional role of rs3742076 in NSCLC, differentially expressed genes were identified from transcriptome analysis between FOXM1 rs3742076_A and rs3742076_G cells. We found that FOXM1 rs3742076_G regulated dissimilar signaling pathways in rs3742076_G cells compared with those in rs3742076_A cells (Fig. 4A) through KEGG enrichment analysis, specifically the upregulation of the key regulator CTNNB1 (encoding β-catenin) in rs3742076_G cells for wnt signaling (Fig. 4B). GSEA revealed enrichment of pathway in wnt signaling (Fig. 4C) and signaling pathways regulating pluripotency of stem cells (Fig. 4D). Consistent with RNA-sequencing analysis, RT-PCR and western blot (Fig. 4E and F) validated that β-catenin was upregulated in rs3742076_G cells with a significant activation in wnt/β-catenin signaling compared with that in rs3742076_A cells (Fig. 4G).

We next sought to determine the clinical significance of β-catenin in patients with lung cancer by analyzing the transcriptomic dataset from GEO (GSE29013, GSE30219, and GSE31210). We found that upregulation of CTNNB1 was significantly associated with poorer survival rates in patients with LUAD (Fig. 4H). Collectively, these results indicated that β-catenin was identified as the major target of FOXM1 rs3742076_G in NSCLC cells.

Next, we noticed that wnt/β-catenin signaling pathway has been identified as a key regulator in cancer stem cells (29) and EMT (30), resulting in drug resistance and tumor progression. By sphere forming assay, we found that rs3742076_G enhanced mammosphere formation compared with rs3742076_A or empty vector (Fig. 4I). In addition, we found that FOXM1 rs3742076_G induced expression of N-cadherin, whereas suppressed expression of E-cadherin compared with rs3742076_A (Fig. 4J), indicating that FOXM1 rs3742076_G promoted EMT-related properties in NSCLC cells. Taken together, these results demonstrated that the FOXM1 variant-enhanced properties were associated with EMT and stemness in NSCLC cells.

To further validate that the effects of FOXM1 rs3742076_G on gefitinib resistance are mainly mediated by the activation of β-catenin and wnt signaling, we examined whether inhibition of β-catenin could attenuate the aggressive phenotype. A β-catenin inhibitor, niclosamide (31), markedly suppressed the wnt activities (Fig. 5A and B). Depletion of β-catenin significantly inhibited proliferation, invasion, and colony formation abilities in rs3742076_G cells (Fig. 5C–E). Collectively, our results suggested that inhibition of β-catenin alleviated FOXM1 variant-induced gefitinib resistance in NSCLC.

We next questioned the detailed molecular mechanisms of FOXM1 rs3742076_G in upregulation of β-catenin in NSCLC cells. The previous study has identified TTTGTTTGTTTT in CTNNB1 as the transactivated binding site of FOXM1 in endothelial cell (32). The correlation analysis of the dataset (33) revealed a significant positive correlation between FOXM1 and CTNNB1 in patients with NSCLC with EGFR-activating mutation (Fig. 5F), indicating that β-catenin might be a transcript target for FOXM1. To next assess whether the variant had an impact on the transactivation of β-catenin, a luciferase reporter assay was performed with constructs driven by a human FOXM1 containing rs3742076_A or rs3742076_G. We found that FOXM1 rs3742046_G drove greater luciferase activity on β-catenin promoter compared with rs3742046_A (Fig. 5G), indicating that the variant enhanced the transcription activity of FOXM1. To further examine whether the variant enhanced the binding to β-catenin promotor, we next identified the potential FOXM1-binding site in the promotor region of β-catenin by hTFtarget (hust.edu.cn; Supplementary Table S5) followed by a ChIP-qPCR validation. The results showed that FOXM1-bound β-catenin promoter DNA resulted in a markedly increase in rs3742046_G cells compared with those in rs3742046_A cells (Fig. 5H). Taken together, our results demonstrated that the variant rs3742076_G in transactivation domain (TAD) of FOXM1 could be characterized as a gain-of-unction variation by increasing activation of β-catenin through direct binding to the promotor of β-catenin.

Recent studies identified that FOXM1 activated wnt/β-catenin signaling by directly binding to β-catenin through protein–protein interaction (PPI) in leukemia stem cell (34). Therefore, we hypothesized that the variant enhanced PPI of FOXM1 and β-catenin in NSCLC cells. By co-IP assay, we found that the PPI of FOXM1 and β-catenin was enhanced in rs3742076_G cells compared with that in rs3742076_A cells (Fig. 5I). To determine the intracellular localization of PPI of FOXM1 and β-catenin, nuclear, and cytoplasmic proteins were separated by a nuclear and cytoplasmic protein extraction kit. We first sought to identify the intracellular localization of FOXM1 and β-catenin, the results showed that FOXM1 was typically localized to nuclear in rs3742076_A cells (Fig. 5J), which was consistent with previous study (35). Interestingly, FOXM1 and β-catenin were largely translocated to cytoplasm in rs3742076_G cells compared with those in rs3742076_A cells (Fig. 5J). By co-IP assays, β-catenin was predominantly co-localized with FOXM1 in the cytoplasm in rs3742076_G cells, indicating that the PPI of FOXM1 rs3742076_G and β-catenin was mainly localized to the cytoplasm (Fig. 5K). Taken together, FOXM1 rs3742076_G enhanced wnt/β-catenin activity by promoting co-localization of FOXM1 and β-catenin in the cytoplasm.

To further investigate whether rs3742076 is directly involved in the regulation of FOXM1 protein expression and wnt/β-catenin signaling activity, CRISPR/Cas9-mediated genome editing was applied to introduce rs3742076 G/G to HCC827 cells (Fig. 6A). We found that cells with rs3742076 G/G genotype had a higher FOXM1 protein level compared with the parental cells carrying rs3742076 A/A genotype, upregulating the expression of β-catenin and wnt activity (Fig. 6B–D). In addition, we also found that NSCLC invasion and proliferation was significantly enhanced by introducing rs3742076 G/G to HCC827 cells regardless of culturing with gefitinib or not (Fig. 6E and F). Collectively, introduction of rs3742076 G/G resulted in upregulation of β-catenin leading to gefitinib resistance and tumor aggressiveness (Fig. 6G).

To verify the role of the variant of FOXM1 in mediating gefitinib resistance in NSCLC progression, we performed a cell-derived xenograft mouse model with PC9 cells engineered with overexpression of FOXM1 rs3742076_A, rs3742076_G or empty virus. Consistent with our previous in vitro results, overexpression of FOXM1 significantly abolished the inhibitory effects of gefitinib treatment in vivo and FOXM1 rs3742076_G tumor were more resistance to gefitinib compared with rs37420746_A tumor, as demonstrated by the increased tumor size and tumor weight (Fig. 7A–C). In addition, Ki67 and β-catenin expression levels in FOXM1 rs3742076_G tumor were higher than those in rs3742076_A tumor with or without gefitinib treatment (Fig. 7D–H). Collectively, rs3742076_G promoted an aggressive phenotype of NSCLC cells and conferred to gefitinib resistance in vivo.

In this study, we established a predictive system for prognosis of EGFR-mutant NSCLC with gefitinib treatment and validated it in an independent cohort. By bioinformatics analysis, we found that FOXM1 upregulation was associated with poor prognosis in patients with lung cancer, especially in EGFR-mutant lung cancer. We identified that rs3742076_G, located in the TAD of FOXM1, upregulated the expression of FOXM1 through increasing protein stability, which resulted in tumor progression. By in vitro and in vivo study, we also elucidated an oncogenic role of FOXM1 rs3742076_G in facilitating EGFR-mutant NSCLC proliferation, invasion, and resistance to gefitinib through the activation of the wnt/β-catenin signaling pathway. Mechanistically, FOXM1 missense variant-induced gefitinib resistance by activating the wnt/β-catenin signaling pathway through directly binding to β-catenin in cytoplasm and promoting transcription of β-catenin in nucleus. Finally, niclosamide, a market-available inhibitor of β-catenin, markedly reversed rs3742076_G-induced gefitinib resistance and aggressive phenotype, which may be a particularly suitable combination for patients with NSCLC carrying rs3742076_G.

In the majority of previous studies, genes or other genomic features from cell lines (36) or TCGA (37) were used to identify predictive indicators for patients with gefitinib therapy besides EGFR mutation, and several scores system were established, including gefitinib response score with expressions levels of 12 genes (20), EGFR structure–function-based approach (38), and EGFR scores containing the activity of the EGFR pathway (37). Before the current study, integrative prognostic models dedicated to EGFR-mutant NSCLC with gefitinib treatment were lacking based on SNPs signature. The GPSE risk score may serve as a prognostic tool for EGFR-mutant NSCLC patient stratification, including the variants in XRCC1, FOXM1, FOXO3, and ERCC1.

FOXM1, located in the chromosomal band 12p13, is an evolutionarily conserved transcriptional regulator. It contains forkhead box or winged helix domain, N-terminal repressor domain, and C-terminal TAD (39). FOXM1 is considered as a major actor in the mitotic (40) and DNA repair programs (41), which participates in maintenance of the quiescence and self-renewal capacity of stem cells (42). FOXM1 was upregulated in EGFR–TKI resistant cells compared with parental cells by promoting EMT and proliferation properties, such as gefitinib, erlotinib, and osimertinib (14, 43), which implied that FOXM1 level might be the one of key factors for the sensitivity of EGFR–TKIs. Indeed, FOXM1 overexpression is associated with a poor clinical prognosis in different tumors and drives cancer progression and recurrence (27). Moreover, FOXM1 inhibition synergized with EGFR-targeting agents in preclinical models, which have been recently identified as therapeutic targets in EGFR TKI-resistant NSCLC (43). These reports were complementary to our observations, and our findings confirmed FOXM1 overexpression involved in invasion, proliferation, and drug resistance, which were consistent with these results. Taken together, FOXM1 has been contemplated as an oncogenic role, whose signaling network is involved in cancer pathophysiology and drug resistance.

Although FOXM1 plays a pivotal role in a series of cellular progress in cancer cells, there is still limited knowledge about the role of the polymorphisms of FOXM1 in drug resistance. The identified A > G rs3742076 variant, located in TAD of FOXM1, conferred resistance to gefitinib in EGFR-mutant NSCLC through stimulating an aggressive phenotype in cancer cells. A previous study revealed that the regulation of FOXM1 activity was dependent on TAD structure switching (44), implying that the structure of TAD played a vital role on transcriptional activity of FOXM1. Interestingly, the variant was located in TAD of FOXM1, which might affect transcriptional activity of FOXM1, and lead to abnormal expression of the target genes of FOXM1. By an in silico tool (28), we found a possible protein stability change of FOXM1 by substitution A to G in rs3742076. These results were reinforced by the CHX chase assay, which confirmed that the Proline substitutions enhanced FOXM1 stability, and resulted in a higher expression of FOXM1.

We then assessed whether rs3742076 variant affected the main cellular processes in which FOXM1 participates. Higher wnt activity and cytoplasmic β-catenin accumulation were detected mainly in FOXM1 rs3742076_G cells in this study. Consistently, previous studies have identified that the canonical wnt pathway protected the NSCLC cells from EGFR inhibition (45, 46). Recent studies have revealed that FOXM1 activated wnt signaling by directly binding to β-catenin in leukemia stem cells (34) and increased the transcription of β-catenin and wnt signaling activity in endothelial cell (32). Interestingly, our functional characterization results showed that FOXM1 rs3742076_G upregulated wnt activity by activating β-catenin in transcriptional progress and promoting β-catenin accumulation in cytoplasmic through ChIP and Co-IP assays, which could be explained by the location of the mutation inside the TAD of FOXM1. In addition, we introduced FOXM1 rs3742076_G into HCC827 (FOXM1 rs3742076_A) by CRISPR/Cas9, and found that both the protein levels of FOXM1 and β-catenin in rs3742076_G cells were higher compared with those in rs3742076_A cells. Finally, inhibition of β-catenin alleviated FOXM1 rs3742076_G-dependent aggressive phenotype and gefitinib resistance in NSCLC, which may have some potential therapeutic implications. Taken together, these results indicated a pathogenic role for FOXM1 missense variants in gefitinib resistance by activating wnt/β-catenin signaling pathway through directly binding to β-catenin in cytoplasm and promoting transcription of β-catenin in nucleus.

Although large studies have revealed that somatic mutations played a role in tumor relapse and drug response, there is emerging evidence, indicating that germline polymorphisms can be biomarkers to select therapeutic agents clinically, such as MET^N375S in Her2 inhibitors (47) and germline BRCA1/2 variants in PARP inhibitors (48). In the study, we found that the FOXM1 germline mutation drove cancer malignancy and resistance to gefitinib in EGFR-mutant NSCLC. These variants from germline have already existed before tumorigenesis [i.e., PC9 cells (FOXM1 rs3742076^Ser/Pro)], which may be filtered out in the analysis when compared with the normal tissue using next-generation sequence. Indeed, there is still limited knowledge about the role of these germline variants in drug response and tumor progression. However, all these data have revealed that the germline mutation may not be ignored in the future.

There are some limitations in this study. First, the rs3742076 variant identified in this study is a low-frequency one (7.6%) and the patients enrolled were only the ones treated with gefitinib treatment in Asian population, restricting its potential role as to be a widely used biomarker for pan-ErbB receptor inhibitors in EGFR-mutant NSCLC. However, FOXM1 expression level has been well characterized as a prognostic marker in multiple datasets from European populations, especially in EGFR-mutant NSCLC, suggesting its potential indicative role for patients with NSCLC with gefitinib treatment. Second, this study is a retrospective study, although these results were validated in an independent cohort, prospective studies are still invaluable in the future. Third, the combination of niclosamide and gefitinib in patients with NSCLC with rs3742076_G needs to be tested in clinical trials.

In conclusion, we identified the genetic variant in FOXM1 influencing interpatient variability in gefitinib response through survival analysis in multipronged experiments, and highlighted the variant as a predictive biomarker for gefitinib resistance and provided insights into the molecular mechanisms of gefitinib resistance in NSCLC. Our current work contributes to the emerging notion that the germline polymorphism could impact the selection of EGFR–TKIs in EGFR-mutant NSCLC.

No disclosures were reported.

S. Guan: Conceptualization, resources, data curation, methodology, writing–original draft, writing–review and editing. X. Chen: Software, visualization. Y. Chen: Formal analysis. W. Xie: Supervision. H. Liang: Methodology. X. Zhu: Methodology. Y. Yang: Resources. W. Fang: Resources. Y. Huang: Resources. H. Zhao: Resources. W. Zhuang: Resources. S. Liu: Resources. M. Huang: Supervision, funding acquisition, writing–review and editing. X. Wang: Project administration, writing–review and editing. L. Zhang: Project administration, writing–review and editing.

This work was supported by the National Natural Science Foundation of China (grant nos. 81973398, 81473283, 81730103, 81573507, and 82020108031), The National Key Research and Development Program (grant nos. 2017YFC0909300 and 2016YFC0905000), Guangdong Provincial Key Laboratory of Construction Foundation (grant no. 2017B030314030), Science and Technology Program of Guangzhou (grant no. 201607020031), National Engineering and Technology Research Center for New Drug Druggability Evaluation (Seed Program of Guangdong Province (grant no. 2017B090903004), the 111 project (grant no. B16047), China Postdoctoral Science Foundation (grant nos. 2019M66324, 2020M683140, and 2020M683139) and Natural Science Foundation of Guangdong Province (grant no. 2022A1515012549). We are grateful to Kornel Tomczyk (Faculty of Foreign Language Education, Sun Yat-sen University, Guangzhou, China) and Keyue Wang (Faculty of life science and medicine, King's College London) for their help with the review of this article. We would also like to thank the patients who gracefully donated their samples and time to this project.

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.