HCC827 cells were obtained from American Type Culture Collection and were grown in the recommended medium at 37°C with 5% CO₂ in a humidified incubator. The gefitinib-resistant HCC827/GR cells were established by gradient dose treatment for parental cells for more than 6 months.

After extraction, quantification, and purification, 10 μg of total RNA, in which ribosomal RNA was depleted, was fragmented into small pieces using divalent cations under elevated temperature. Following this, RNA fragments were reverse-transcribed to establish the final cDNA library according to the manufacturer’s instructions for the mRNA-seq sample preparation kit (Illumina). Then, paired-end sequencing was performed following the vendor’s recommended protocol on an Illumina Hiseq 4000 (LC-bio).

First, the reads that contained adapter contamination, low-quality bases, and undetermined bases were removed by Cutadapt, and the sequence quality was validated using FastQC (http://www.bioinformatics.babraham.ac.uk/projects/fastqc/). Then, Bowtie2 (Langmead and Salzberg, 2012) and Tophat2 (Kim et al., 2013) were used to map reads to the human genome, and then they were assembled by StringTie (Kim and Salzberg, 2011). All transcriptomes were combined to rebuild a comprehensive transcriptome using Perl scripts. Finally, StringTie and Ballgown (Zhang et al., 2016) were applied to estimate the expression levels of all transcripts.

The expression values of mRNAs were calculated as fragments per kilobase per million (FPKM) using StringTie (Pertea et al., 2015). The differentially expressed mRNAs (DE mRNAs) were chosen with a fold change (FC)≥2 or ≤0.5 and p ≤ 0.05 using the R package-Ballgown (Frazee et al., 2015).

To understand the biological functions of DE mRNAs, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed by using the “clusterProfiler” package in R software. GO analysis includes cellular components, molecular functions, and biological processes, which explains the biological function of genes from different perspectives. The KEGG pathway enrichment analysis was used for analyzing the degree of enrichment of DE mRNAs in pathway terms.

Gene Set Enrichment Analysis (GSEA) was conducted to explore the characteristics of gene hallmarks in gefitinib-sensitive and gefitinib-resistant cells by using OmicStudio tools at https://www.omicstudio.cn/tool. In addition, h.all.v7.2.symbols.gmt was used as the reference gene set. Gene set was considered to be significantly enriched when a normal p value was <0.05 and a false discovery rate (FDR) was <0.25.

HCC827/GR cells were seeded in 6-well plates at a density of 1 × 10⁵ cells/well and replaced with a serum-free medium when the confluence reached about 60%–80%. 10 μl TROY siRNA or negative control siRNA (GenePharma) and 10 μl Lipofectamine RNAiMAX (ThermoFisher Scientific) were diluted in 150 μl Opti-MEM (ThermoFisher Scientific). Then, the diluted siRNA was mixed with the diluted RNAiMAX reagent and incubated for 5 minutes at room temperature. The siRNA–lipid complex was added to HCC827/GR cells and incubated for at least 24 h before the subsequent experiments.

Parental and resistant HCC827 cells plated in triplicate were incubated with gefitinib for 72 h at distinct concentrations. Cell viability was measured with the cell counting kit-8 (CCK-8) assay (Dojindo Molecular Technologies) following the manufacturer’s protocol. At least three replicates were tested for each cell line on different days. The IC₅₀ values were calculated by CalcuSyn software.

The cells were lysed in a lysis buffer containing protease and phosphatase inhibitors, and cell lysate protein concentrations were determined using the BCA Protein Assay Kit (Thermo Fisher Scientific). Then, equivalent quantities (30 μg) of protein were separated with 8%–15% sodium dodecylsulfate-polyvinylidene gel electrophoresis and transferred to PVDF membranes (Merck Millipore). Then, they were probed with antibodies against TROY or Ki67 from Abcam and E-cadherin, N-cadherin, vimentin, ZEB1, ALDH1, ZO-1, or GAPDH from Cell Signaling Technology. After being incubated with secondary antibodies, the proteins were detected using an ECL kit (Beyotime Biotechnology).

Single-cell suspensions were plated at a density of 3000 cells in 6-well ultra-low-attachment plates (Corning) and cultured with DMEM/F12 (Gibco) supplemented with 20 ng/ml bFGF (PeproTech), 20 ng/ml EGF (PeproTech), 5 ng/ml heparin sodium (PeproTech), and 10% serum replacement (ThermoFisher Scientific) at 37°C in a humidified 5% CO₂ incubator for 7–14 days. The number of spheres in microscopic field was counted using Image-Pro Plus 6.0 software. The median number from three replicates was used for statistical analysis.

The expression of TROY, E-cadherin, and vimentin in lung cancer tissues was detected by IHC staining from nine NSCLC patients who obtained resistance after first-generation TKI therapy. The samples were used with the approval of the Ethics Committee of the First Affiliated Hospital of Zhejiang University. After prepared into 4-μm-thick-sections, the tissue samples were deparaffinized, rehydrated, and unmasked antigens. The samples were incubated with primary antibodies (TROY 1:200 dilution, Abcam; E-cadherin and vimentin 1:500 dilution, Cell Signal Technology) at 37°C for 1 h. After washing in PBS, the sections were stained with DAB and counterstained with hematoxylin. Then, the IHC images were captured and scored by the distribution and immunoreactivity. The final results were determined by two observers independently. The percentage of positive cells was recorded as 0 if no cell was stained, 1 for less than 10% of stained cells, 2 for 10%–50% of stained cells, and 3 for more than 50% of stained cells. The staining intensity was recorded as 0 if no staining was observed, 1 for weak staining, 2 for moderate staining, and 3 for strong staining. Then, these values were multiplied together to provide a staining index for each case. Using this method of assessment, we evaluated the IHC score of TROY, vimentin, and E-cadherin in NSCLC by staining index (scored as 0, 1, 2, 3, 4, 6, or 9).

The data in vitro were presented with mean values ±SD from three or more independent repetitions. Two-tailed unpaired Student’s t-test was conducted to determine p value, and p < 0.05 was defined as statistical significant. Clinical data were analyzed using the chi-squared test or the two-sided Student’s t-test by SPSS Statistics 22 software. The Kaplan–Meier method was applied for survival analysis, and the log-rank test was performed to detect the significance of the difference between survival curves. Multivariate analysis was used to assess the prognostic value.

To investigate the reasons for acquired resistance to EGFR-TKI, we established gefitinib-resistant NSCLC cell (HCC827/GR) by increasing the concentration of gefitinib in the culture medium gradually. The sensitivity of parental and resistant cells for gefitinib was detected by CCK-8 assay, and the IC₅₀ of resistant cells was steeply increased compared with that of parental cells (Figures 1A,B). In addition, T790M mutation in exon 20 of EGFR, c-MET amplification, and HER2 amplification had not been detected in HCC827/GR cells (Figure 1C).

Then, RNA-sequencing was conducted to explore the expression level of mRNA in gefitinib-resistant cell and their parent cell line. The differentially expressed mRNAs were selected with a fold change ≥2.0 or≤0.5 and with statistical significance (p value < 0.05). The volcano plot showed that there were 343 upregulated mRNAs and 320 downregulated mRNAs (Figure 2A), and TNFRSF19 (TROY) mRNA was strikingly overexpressed in gefitinib-resistant cells. The heat map analysis displayed the expression of the top 40 dysregulated mRNAs (Figure 2B), and the mRNAs were listed in order of ascending p value (Table 1).

In addition, GO and KEGG analyses were carried out to examine the underlying functions of DE mRNAs. The GO analysis exhibited that dysregulation of mRNAs was associated with the plasma membrane, extracellular space, extracellular region, extracellular exosome, cell adhesion, and extracellular matrix organization (Figures 2C,D). Correspondingly, KEGG analysis demonstrated that the most significantly associated pathways were PI3K-Akt signaling pathway, focal adhesion, phospholipase D signaling pathway, and ECM-receptor interaction, among others (Figures 2E,F).

In addition, GSEA unveiled that EMT, interferon alpha response, G2M checkpoint, E2F targets, hypoxia, and UV response were the top six relevant pathways (Figure 3), and EMT was the overriding cause of gefitinib resistance.

To investigate the role of TROY in gefitinib-resistant NSCLC cells, we first examined the expression of TROY in gefitinib-sensitive and -resistant NSCLC cell lines. Unsurprisingly, the mRNA and protein expression were both increased in gefitinib-resistant cells when compared to parental cells (Figures 4A,B). Then, we knockdown TROY by siRNA, and the results showed that TROY siRNA can effectively reduce TROY protein expression (Figure 4B), and cell growth was significantly reduced in TROY knockdown HCC827/GR cells when treated with gefitinib (Figure 4C), indicating that TROY is one of the reasons for gefitinib acquired resistance in NSCLC.

Phenotypic change was observed in acquiring resistant cells. Microscopically, HCC827/GR cells exhibited spindle cell–like morphology which differs from parental HCC827 (Figure 5A). It was foreseeable that HCC827/GR cells displayed EMT features with the decreased expression of E-cadherin and ZO-1 and the increased expression of N-cadherin and vimentin. In addition, ZEB1, an EMT transcriptional regulator, was upregulated too (Figure 5B). Since EMT has been identified as an indispensable reason for drug resistance in tumors, we detected the EMT features in TROY knockdown HCC827/GR cells. Interestingly, EMT‐related proteins N-cadherin, vimentin, and ZEB1 were significantly reduced after transfecting with TROY siRNA, and the expression of E‐cadherin and ZO-1 was elevated (Figure 5B). It suggested that TROY may play a crucial role in the EMT process in acquired resistance of EGFR-TKI.

Increasing evidence manifests that EMT is related to the obtainment of CSC features, in consequence, conferring resistance to cancer therapy. Sphere-formation assay is commonly used to detect cancer stem cell characteristics; thus, we compared the abilities of sphere formation between parental and resistant HCC827 cells. The results showed that gefitinib-resistant HCC827/GR cells displayed higher abilities of tumor sphere formation, while knockdown of TROY weakened the sphere-formation abilities in HCC827/GR cells (Figures 6A,B). Consistently, pluripotency markers, such as ALDH1, CD133, and SOX2, were elevated to varying degrees in HCC827/GR cells. In addition, Ki67, a marker indicating the degree of cell proliferation, was observed to be downregulated in HCC827/GR cells. Likewise, downregulated TROY gave rise to downregulated expression of stemness markers in gefitinib-resistant cells (Figure 6C). These data indicate that TROY may play a key role in the CSC regulation through EMT in NSCLC.

The expression of TROY, E-cadherin, and vimentin in tumor samples obtained from nine NSCLC patients who were resistant to the first-generation EGFR-TKI and without T790M mutation were determined by immunohistochemistry staining (Figures 7A,B). The expression of E-cadherin was found in all tumor tissues before treatment and decreased in patients after EGFR-TKI resistance with the expression of vimentin elevated. Concurrently, the expression of TROY was found to increase in patients after drug resistance (Table 2).

Survival analysis of TROY was conducted by the online tool Kaplan–Meier Plotter (https://kmplot.com/) (Győrffy et al., 2013). The first progression (FP) was 32.9 and 20.1 months in TROY-low expression and TROY-high expression LUAD patients, respectively. Meanwhile, overall survival (OS) was 88.7 and 63 months in TROY-low expression and TROY-high expression patients, respectively. In addition, multivariate analysis confirmed that TROY expression (HR = 1.29, 95% CI 1.09–1.52) was an independent prognostic factor in LUAD (Figures 7C,D). Either FP or OS was closely related with the expression of TROY in LUAD patients, which means TROY could be used as a prognosis biomarker for NSCLC.