Identification of a gefitinib resistance-associated signature for predicting prognosis and therapeutic response in lung adenocarcinoma via integrated multi-omics analysis and machine learning

Identification of a gefitinib resistance-associated signature for predicting prognosis and therapeutic response in lung adenocarcinoma via integrated multi-omics analysis and machine learning | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Identification of a gefitinib resistance-associated signature for predicting prognosis and therapeutic response in lung adenocarcinoma via integrated multi-omics analysis and machine learning Dong Zhou, Zhi Zheng, Yan-Qi Li, Quan-Xing Liu, Xu-Feng Deng, and 6 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4573455/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Gefitinib resistance (GR) is widespread; therefore, alternative treatments for lung adenocarcinoma (LUAD) are needed. The study of gefitinib-resistance gene sets may lead to a better understanding of the mechanism underlying GR, methods for predicting and preventing GR, and alternative therapies. GR gene sets, single-cell data, and transcriptome data were obtained from public databases. Univariate and multivariate regression analyses and machine learning techniques were used to screen genes and construct a signature, respectively. Survival analysis and time-dependent receiver operating characteristic (ROC) curves were used to assess signature performance in internal and external data sets. Enrichment and tumor immune-microenvironment analyses were used to explore the mechanism of the signature genes in GR. Novel immunological and non-immunological therapies were explored. A signature consisting of 22 genes was successfully constructed in LUAD cohort, which performed well in both internal and external validation. The signature was closely related to chromosomal processes, DNA replication, important immune-cell infiltration, and multiple immune scores in enrichment and tumor microenvironment analyses. Further, the signature predicted immunotherapy efficacy in patients with LUAD to a certain extent, and we identified various agents other than gefitinib that may have better treatment effects in high-risk and low-risk groups, providing treatment guidance for gefitinib-resistant patients. The 22-gene signature can predict the prognosis of gefitinib-resistant patients with LUAD and immunotherapy efficacy, and provides new guidance for non-immunotherapy. LUAD gefitinib resistance single-cell RNA-sequencing signature immune microenvironment treatment Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Introduction Lung cancer remains one of the most common and deadliest malignant tumors worldwide, with a 5-year survival rate of approximately 20% (Siegel et al. 2024). Chemotherapy is the traditional treatment for advanced lung cancer, but it has limited efficacy, and its significant side effects greatly impact patients’ quality of life (Qu et al. 2024). With the advent of targeted therapies, the most common being epidermal growth factor receptor (EGFR) tyrosine kinase inhibitors (TKIs) (Noronha et al. 2024), the survival of patients with advanced driver gene-positive lung adenocarcinoma (LUAD) has greatly improved (Choi, Chang 2023). Gefitinib, a first-generation EGFR-TKI, is widely used in the treatment of advanced EGFR mutation-positive patients with LUAD because of its improved efficacy and relatively low side effects (Yang et al. 2016). However, inherent and acquired resistance to gefitinib substantially impair its clinical efficacy (Johnson et al. 2022). Drug resistance and consequent tumor metastasis are considered factors that directly affect patient prognosis. Therefore, there is an urgent need for effective and tolerable regimens to address gefitinib resistance (GR). The mechanism of GR is complex and involves multiple genes, layers, and dimensions, including genetic mutations in the EGFR pathway, abnormal bypass pathway activation, and changes in phenotypic characteristics (Reita et al. 2021). For example, the EGFR T790M mutation, is a primary acquired GR mechanism in LUAD (Ge et al. 2023). In tackling the resistance issue, the importance of exploring alternative treatments cannot be overstated. Osimertinib, a third-generation EGFR mutant-selective TKI, stands out as a promising option as it has demonstrated high effectiveness in patients with non-small cell lung cancer (NSCLC) carrying the T790M mutation (Ge et al. 2023). However, the relatively small population with acquired resistance due to T790M and emergence of secondary resistance with the use of osimertinib strongly limit its clinical application (Lu et al. 2024;Haratake et al. 2024). Therefore, early identification of the GR risk in patients with LUAD and early intervention are particularly important, enabling physicians to devise a treatment strategy based on disease progression prediction. The proliferation of public genome data has led to the rise of meta-analysis and computational modeling as pivotal methods for circumventing the constraints of insufficient statistical power in isolated studies (Wang et al. 2023). In the past few years, regularized regression classifiers, such as the least absolute shrinkage and selection operator and elastic net, have become increasingly recognized as efficient tools for feature selection and prediction in high-dimensional datasets (Yang et al. 2024;Fan et al. 2024). In the present study, we analyzed GR-related genes using integrated bulk and single-cell RNA-sequencing (scRNA-seq) data to a better understanding of the mechanism underlying GR, methods for predicting and preventing GR, and alternative therapies. First, we analyzed the expression characteristics of GR genes in various cell subgroups based on scRNA-seq data. Next, multiple supervised machine learning methods were used to identify GR signature genes, and a clinical prediction model was built based on the results. Finally, following a thorough evaluation of the biological mechanisms of the signature genes identified, their involvement in LUAD was validated based on database information and cell experiments. Materials And Methods Data collection and processing Whole-gene mRNA expression matrices, clinical data (phenotypes and overall survival), and single-nucleotide polymorphism mutation data of Cancer Genome Atlas (TCGA)-LUAD samples were downloaded from the UCSC website (https://xenabrowser.net/). In total, 513 tumor samples and 59 normal samples were obtained, and samples with missing information were excluded. The GSE117570 dataset, which comprises 10× scRNA-seq data of 11,485 cells, was obtained from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/), and data on P1, P3, and P4 were selected for analysis. Expression data of signature genes in particular cells were downloaded from the Tumor Immune Single-cell Hub database (http://tisch.comp-genomics.org/home/). Two GEO datasets (GSE31210 and GSE13213) with complete mRNA expression and clinical information were obtained from the GEO database to construct the validation cohort. Genes upregulated in GR were obtained from the Gene Set Enrichment Analysis (GSEA) database (https://www.gsea-msigdb.org/gsea/index.jsp, “COLDREN_GEFITINIB_RESISTANCE_UP” dataset). scRNA-seq data collection and analysis The R package Seurat was used to convert raw scRNA-seq data into Seurat objects, followed by quality control and the exclusion of unqualified cells. The gene expression in core cells was standardized using a linear regression algorithm, and the top 2,000 highly variable genes were identified using analysis of variance. Uniform manifold approximation and projection was used to identify cell subpopulations. The FindAllMarkers function in the R package Seurat and the COSG R package were employed to identify marker genes for all cell clusters and differentially expressed genes among the screened marker genes. The R package clusterProfiler was used for Kyoto Encyclopedia of Genes and Genomes (KEGG) and Reactome enrichment analyses of top 100 marker genes in each cell cluster. A single-sample GSEA score was calculated for each cell cluster based on the expression levels of GR-related genes, and all cell samples were divided into high- and low-score groups based on the median score. Data of 50 hallmark gene sets were downloaded from the GSEA database and the R package GSVA was employed to calculate hallmark scores for each cell cluster. Identification and validation of a prognostic signature Univariate Cox proportional risk regression analysis was conducted on GR-related genes in the TCGA-LUAD training cohort, using p < 0.05 as the screening criterion. Ten classical algorithms (single or combined), including random forest, least absolute shrinkage and selection operator, gradient boosting machine, survival support vector machine, supervised principal components, ridge regression, partial least squares regression for Cox, CoxBoost, Stepwise Cox, and elastic network, were used to construct a prognostic signature based on the screened genes, and the C-index was calculated for all signatures to select the most suitable one. The R packages survival, pROC, and ggplot2 were used for survival analysis and time-dependent receiver operating characteristic (ROC) analysis using the TCGA-LUAD training cohort, GSE13213 and GSE31210 validation cohorts, and their combination. Prognostic analysis of risk groups and clinical phenotypes The proportions of clinical phenotype (sex, T stage, N stage, M stage, stage, recurrence, metastasis, and overall survival) subcategories in the high- and low-risk groups were calculated, and resistance scores were systematically compared among the subcategories. Univariate and multivariate Cox regression analyses were used to assess the influence of the prognostic signature and other common clinical phenotypic features on the prognosis of patients with LUAD. Functional enrichment analysis Correlation coefficients between the resistance score and all genes were calculated, and the 100 genes with the highest positive and negative correlations were screened out. The R package clusterProfiler was used for Gene Ontology (GO), KEGG, and Reactome functional enrichment analyses of the screened genes. After calculating the score for each sample based on the constructed signature, all samples were divided into high- and low-risk groups, and GSEA was applied to the differentially expressed genes between the two groups. The cancer-related hallmark gene set file (h.all.v2023.2.Hs.symbols.gmt) was downloaded from the Molecular Signatures Database (https://www.gsea-msigdb.org/gsea/msigdb). The R package GSEA (v1.38.2) was used to calculate the normalized enrichment score and false discovery rate, and the R package ggplot2 (v3.3.6) was used to display the GSEA results. Analysis on tumor immune microenvironment (TIME) and tumor stemness LUAD samples were ordered according to stemness indices (mRNAsi and mDNAsi) and their overall survival and distribution of high- and low-risk groups were shown. Pearson correlation coefficients between stemness indices and resistance scores were calculated. Immune-cell infiltration was analyzed using the MCPCOUNTER, EPIC, XCELL, CIBERSORT, IPS, QUANTISEQ, ESTIMATE, and TIMER algorithms in the TIMER database (https://cistrome.shinyapps.io/timer/), and the infiltration scores of all types of immune cells were compared between the high- and low-risk groups using the R package IOBR. Multiple immune microenvironment-related indices, including the microenvironment, ESTIMATE, immune, and stroma scores, were calculated and compared between the high- and low-risk groups. The expression levels of various chemokines and their receptors were compared between the high- and low-risk groups, and Pearson correlation coefficients between the expression levels of significantly differentially expressed chemokines and the resistance scores were calculated. Analysis of drug efficacy prediction in gefitinib-resistant populations Data of patients who received immunotherapy was downloaded from the Tumor Immune Dysfunction and Exclusion (TIDE) database (http://tide.dfci.harvard.edu), and true or false response to immunotherapy and TIDE and exclusion scores were compared between the high- and low-risk groups. GSE135222, an anti-programmed cell death protein 1 (PD-1)/PD-1 ligand (PD-L1) immunotherapy dataset from GEO, was used to evaluate the relationship between the GR signature and LUAD immunotherapy response. Treatment data of several non-immune chemotherapy drugs were downloaded from the Genomics of Drug Sensitivity in Cancer database (https://www.cancerrxgene.org/), and the R package oncoPredict was employed to evaluate the drug half-maximal inhibitory concentration (IC50) values for each sample. Cell culture and reagents PC-9 normal NSCLC cells were obtained from the Cell Bank of the Chinese Academy of Science (Shanghai, China). PC-9 gefitinib-resistant cells (strain CTCC-ZHYC-0141) were purchased from Meisen Cell Technology (Zhejiang, China). All cells were maintained at 37 °C in a 5% CO 2 humidified atmosphere. PC-9 normal and gefitinib-resistant cells were cultured in RPMI-1640 medium supplemented with 10% fetal bovine serum and penicillin-streptomycin (100 U/ml and 100 mg/ml, respectively) (all from Thermo Fisher Scientific China, Shanghai, China). Before the start of this study, the resistance of the PC-9 resistant strain was confirmed using cell viability assays. The PC-9 gefitinib-resistant strain was maintained in medium supplemented with gefitinib at concentrations of 2.5 mM, 5 mM, and 10 mM for the first to third generations, respectively, and 10 mM from the fourth generation onward. Gefitinib (184475-35-2), AZD 7762 (1246094-78-9), and BI.2536 (755038-02-9) were purchased from Sigma-Aldrich China (Shanghai, China) and dissolved in dimethylsulfoxide at a stock concentration of 50 mM. All drugs were stored at –20 °C in 5-mL aliquots. RNA extraction and reverse transcription quantitative (RT-q)PCR Resistance gene expression in PC9 normal and gefitinib-resistant cells was assessed using RT-qPCR. The gene-specific qPCR primers are listed in Table S1. Total RNA was extracted from the cells using TRIzol reagent (Invitrogen Life Technologies) and reverse-transcribed into cDNA using a Takara kit (Takara Biotechnology). qPCR analysis was performed using SYBR Green master mix (Thermo Scientific). Cell proliferation assay CCK8 kits purchased from Sigma-Aldrich China (Shanghai, China) were used to detect the inhibitory effects of gefitinib, AZD7762, and BI.2536 on the viability of PC-9 normal and gefitinib-resistant cells. The cells were seeded in 96-well culture plates at 6 × 10 3 cells/well, cultured for 24 h, and exposed to increasing concentrations of gefitinib, AZD7762, and BI.2536 for an additional 24 h. Ten microliters of CCK-8 solution was added to each well, and the plates were incubated for 2 h. The absorbance at 450 nm was measured using a microplate reader (Thermo Scientific). Statistical analysis Data are presented as mean ± SD or SEM and were analyzed using SPSS 26.0. Differences were considered significant at * p < 0.05, ** p < 0.01, and *** p < 0.001. The R packages used were mentioned above. Plots were generated using R Studio 4.3.3 or GraphPad Prism 8.0. Data collection and processing Whole-gene mRNA expression matrices, clinical data (phenotypes and overall survival), and single-nucleotide polymorphism mutation data of Cancer Genome Atlas (TCGA)-LUAD samples were downloaded from the UCSC website (https://xenabrowser.net/). In total, 513 tumor samples and 59 normal samples were obtained, and samples with missing information were excluded. The GSE117570 dataset, which comprises 10× scRNA-seq data of 11,485 cells, was obtained from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/), and data on P1, P3, and P4 were selected for analysis. Expression data of signature genes in particular cells were downloaded from the Tumor Immune Single-cell Hub database (http://tisch.comp-genomics.org/home/). Two GEO datasets (GSE31210 and GSE13213) with complete mRNA expression and clinical information were obtained from the GEO database to construct the validation cohort. Genes upregulated in GR were obtained from the Gene Set Enrichment Analysis (GSEA) database (https://www.gsea-msigdb.org/gsea/index.jsp, “COLDREN_GEFITINIB_RESISTANCE_UP” dataset). scRNA-seq data collection and analysis The R package Seurat was used to convert raw scRNA-seq data into Seurat objects, followed by quality control and the exclusion of unqualified cells. The gene expression in core cells was standardized using a linear regression algorithm, and the top 2,000 highly variable genes were identified using analysis of variance. Uniform manifold approximation and projection was used to identify cell subpopulations. The FindAllMarkers function in the R package Seurat and the COSG R package were employed to identify marker genes for all cell clusters and differentially expressed genes among the screened marker genes. The R package clusterProfiler was used for Kyoto Encyclopedia of Genes and Genomes (KEGG) and Reactome enrichment analyses of top 100 marker genes in each cell cluster. A single-sample GSEA score was calculated for each cell cluster based on the expression levels of GR-related genes, and all cell samples were divided into high- and low-score groups based on the median score. Data of 50 hallmark gene sets were downloaded from the GSEA database and the R package GSVA was employed to calculate hallmark scores for each cell cluster. Identification and validation of a prognostic signature Univariate Cox proportional risk regression analysis was conducted on GR-related genes in the TCGA-LUAD training cohort, using p < 0.05 as the screening criterion. Ten classical algorithms (single or combined), including random forest, least absolute shrinkage and selection operator, gradient boosting machine, survival support vector machine, supervised principal components, ridge regression, partial least squares regression for Cox, CoxBoost, Stepwise Cox, and elastic network, were used to construct a prognostic signature based on the screened genes, and the C-index was calculated for all signatures to select the most suitable one. The R packages survival, pROC, and ggplot2 were used for survival analysis and time-dependent receiver operating characteristic (ROC) analysis using the TCGA-LUAD training cohort, GSE13213 and GSE31210 validation cohorts, and their combination. Prognostic analysis of risk groups and clinical phenotypes The proportions of clinical phenotype (sex, T stage, N stage, M stage, stage, recurrence, metastasis, and overall survival) subcategories in the high- and low-risk groups were calculated, and resistance scores were systematically compared among the subcategories. Univariate and multivariate Cox regression analyses were used to assess the influence of the prognostic signature and other common clinical phenotypic features on the prognosis of patients with LUAD. Functional enrichment analysis Correlation coefficients between the resistance score and all genes were calculated, and the 100 genes with the highest positive and negative correlations were screened out. The R package clusterProfiler was used for Gene Ontology (GO), KEGG, and Reactome functional enrichment analyses of the screened genes. After calculating the score for each sample based on the constructed signature, all samples were divided into high- and low-risk groups, and GSEA was applied to the differentially expressed genes between the two groups. The cancer-related hallmark gene set file (h.all.v2023.2.Hs.symbols.gmt) was downloaded from the Molecular Signatures Database (https://www.gsea-msigdb.org/gsea/msigdb). The R package GSEA (v1.38.2) was used to calculate the normalized enrichment score and false discovery rate, and the R package ggplot2 (v3.3.6) was used to display the GSEA results. Analysis on tumor immune microenvironment (TIME) and tumor stemness LUAD samples were ordered according to stemness indices (mRNAsi and mDNAsi) and their overall survival and distribution of high- and low-risk groups were shown. Pearson correlation coefficients between stemness indices and resistance scores were calculated. Immune-cell infiltration was analyzed using the MCPCOUNTER, EPIC, XCELL, CIBERSORT, IPS, QUANTISEQ, ESTIMATE, and TIMER algorithms in the TIMER database (https://cistrome.shinyapps.io/timer/), and the infiltration scores of all types of immune cells were compared between the high- and low-risk groups using the R package IOBR. Multiple immune microenvironment-related indices, including the microenvironment, ESTIMATE, immune, and stroma scores, were calculated and compared between the high- and low-risk groups. The expression levels of various chemokines and their receptors were compared between the high- and low-risk groups, and Pearson correlation coefficients between the expression levels of significantly differentially expressed chemokines and the resistance scores were calculated. Analysis of drug efficacy prediction in gefitinib-resistant populations Data of patients who received immunotherapy was downloaded from the Tumor Immune Dysfunction and Exclusion (TIDE) database (http://tide.dfci.harvard.edu), and true or false response to immunotherapy and TIDE and exclusion scores were compared between the high- and low-risk groups. GSE135222, an anti-programmed cell death protein 1 (PD-1)/PD-1 ligand (PD-L1) immunotherapy dataset from GEO, was used to evaluate the relationship between the GR signature and LUAD immunotherapy response. Treatment data of several non-immune chemotherapy drugs were downloaded from the Genomics of Drug Sensitivity in Cancer database (https://www.cancerrxgene.org/), and the R package oncoPredict was employed to evaluate the drug half-maximal inhibitory concentration (IC50) values for each sample. Cell culture and reagents PC-9 normal NSCLC cells were obtained from the Cell Bank of the Chinese Academy of Science (Shanghai, China). PC-9 gefitinib-resistant cells (strain CTCC-ZHYC-0141) were purchased from Meisen Cell Technology (Zhejiang, China). All cells were maintained at 37 °C in a 5% CO 2 humidified atmosphere. PC-9 normal and gefitinib-resistant cells were cultured in RPMI-1640 medium supplemented with 10% fetal bovine serum and penicillin-streptomycin (100 U/ml and 100 mg/ml, respectively) (all from Thermo Fisher Scientific China, Shanghai, China). Before the start of this study, the resistance of the PC-9 resistant strain was confirmed using cell viability assays. The PC-9 gefitinib-resistant strain was maintained in medium supplemented with gefitinib at concentrations of 2.5 mM, 5 mM, and 10 mM for the first to third generations, respectively, and 10 mM from the fourth generation onward. Gefitinib (184475-35-2), AZD 7762 (1246094-78-9), and BI.2536 (755038-02-9) were purchased from Sigma-Aldrich China (Shanghai, China) and dissolved in dimethylsulfoxide at a stock concentration of 50 mM. All drugs were stored at –20 °C in 5-mL aliquots. RNA extraction and reverse transcription quantitative (RT-q)PCR Resistance gene expression in PC9 normal and gefitinib-resistant cells was assessed using RT-qPCR. The gene-specific qPCR primers are listed in Table S1. Total RNA was extracted from the cells using TRIzol reagent (Invitrogen Life Technologies) and reverse-transcribed into cDNA using a Takara kit (Takara Biotechnology). qPCR analysis was performed using SYBR Green master mix (Thermo Scientific). Cell proliferation assay CCK8 kits purchased from Sigma-Aldrich China (Shanghai, China) were used to detect the inhibitory effects of gefitinib, AZD7762, and BI.2536 on the viability of PC-9 normal and gefitinib-resistant cells. The cells were seeded in 96-well culture plates at 6 × 10 3 cells/well, cultured for 24 h, and exposed to increasing concentrations of gefitinib, AZD7762, and BI.2536 for an additional 24 h. Ten microliters of CCK-8 solution was added to each well, and the plates were incubated for 2 h. The absorbance at 450 nm was measured using a microplate reader (Thermo Scientific). Statistical analysis Data are presented as mean ± SD or SEM and were analyzed using SPSS 26.0. Differences were considered significant at * p < 0.05, ** p < 0.01, and *** p < 0.001. The R packages used were mentioned above. Plots were generated using R Studio 4.3.3 or GraphPad Prism 8.0. Results Analysis of LUAD scRNA-seq data Single-cell data of LUAD samples comprising 11,485 cells were obtained from GSE117570, and samples P1, P3, and P4 were selected for analysis. The R package SingleR was used for cell-cluster dimension reduction, visualization, and annotation, identifying 11 major cell subgroups, including B cells, T helper 2 (Th2) cells, CD8 + T effector cells, plasmacytoid dendritic cells, endothelial cells, malignant cells, M1 macrophages, M2 macrophages, monocytes, natural killer (NK) cells, and plasma cells (Fig. 1A). The top five highest and lowest differentially expressed genes in each cell subgroup are shown in Fig. 1B. Fig. 1C shows the top 10 expressed marker genes in each cell subgroup identified using the R package COSG. To intuitively visualize the biological processes the core cell types were involved in, we conducted KEGG and Reactome pathway functional enrichment analyses of the top 100 marker genes (Fig. 1D, E). Notably, the marker genes in B, CD8 + T effector, and NK cells were significantly enriched in chemokine signaling (Fig. 1D), whereas Th2 cells were strongly associated with PD-L1 expression and the PD-1 checkpoint pathway in cancer (Fig. 1D). Marker genes in Th2 cells, plasmacytoid dendritic cells, M1 macrophages, and monocytes were significantly enriched in interleukin signaling (Fig. 1E). The expression levels of GR-related genes in each cell cluster are shown in bubble plots in Fig. 2A. The R package GSVA was used to score each cell cluster based on GR-related gene expression, and the clusters were divided into two groups with high and low resistance scores using the median as a threshold (Fig. 2B). Fig. 2C displays the distribution of the cell clusters, resistance score values, and the distribution of cell clusters in the high- and low-score groups. The proportions of all cell clusters in the high- and low-score groups are shown in a bar chart in Fig. 2D. The proportions of all cell clusters in samples P1, P3, and P4 are shown in Supplementary Fig. 1A. Supplementary Fig. 1B shows the distributions of cell populations in the high- and low-drug resistance score groups separately, and the line chart in Supplementary Fig. 1C more intuitively displays the differences in the proportions of the clusters between the two groups. The proportions of M2 macrophages and malignant cells were significantly higher in the high-resistance score group than in the low-resistance score group, whereas the number of Th2 cells was the highest in the low-score group. These results indicated that a high resistance score may predict a tumor microenvironment more favorable for tumor growth, which is associated with poor prognosis, suggesting the feasibility of using GR-related genes for predicting the prognosis of LUAD. Identification of candidate genes and GR signature establishment We obtained 71 GR-related genes with expression profiles from merged TCGA and GEO data. Univariate Cox analysis identified 22 prognostically significant genes ( p < 0.05): SLC47A1 , UBE2M , OLFM1 , SLC7A1 , TRAF5 , TIMM50 , PPIF , TMEM158 , PRELID1 , CDH2 , IKBIP , PPP2R1B , TUB , CEBPG , UCHL1 , NFKBIZ , MDM4 , UBA2 , MAP1B , LIX1L , ALDH1B1 , and NAA10 . Of these, 15 genes were negatively correlated with prognosis, and seven were positively correlated. The hazard ratio and p -value of each gene are shown in a forest plot in Fig. 3A. The expression levels of these 22 genes were generally higher in PC-9 gefitinib-resistant LUAD cells than in normal PC-9 cells (Fig. 3B). To construct the prognostic prediction model, we employed 10 traditional algorithms for precision calculation, using TCGA data as the training set and GSE31210, GSE13213, and merged data as the validation sets. By comparing the algorithms and more than 100 algorithm combinations, we found that the random forest algorithm (mean C-index, 0.758) had the best fit (Fig. 3C). Subsequently, we constructed a GR signature scoring model based on the weights and coefficients for the 22 GR genes provided by the random forest algorithm (Fig. 3D). Among the genes screened by the random forest algorithm, the expression of the top 10 most important genes in multiple cells in multiple datasets is shown in a heatmap in Supplementary Fig. 2. The genes were highly expressed in different cell types, suggesting that the GR signature may affect the prognosis of patients with LUAD via multiple pathways. Internal and external validation of the GR signature using TCGA, GEO, and merged cohorts We first validated GR signature performance in the TCGA cohort. Based on the gene expression weighting algorithm mentioned above, we calculated GR-related scores for all samples, which were then divided into high- and low-resistance score groups based on the median score. As shown in Fig. 4A, the survival time of patients in the high-score group was significantly shorter than that in the low-score group ( p < 0.0001), and the time-dependent ROC curve showed that the GR signature maintained high precision in the TCGA cohort for 7 years (Fig. 4A). Subsequently, two suitable LUAD-GSE datasets (GSE13213 and GSE31210) were selected from the GEO database for external validation of signature reliability. The GR scores of all samples were calculated using the same algorithm, and the samples were divided into high- and low-score groups. In both external cohorts, survival in the high-score group was significantly worse than that in the low-score group (Fig. 4B, C, p = 0.0036 and p = 0.0019), and the accuracy of the GR signature remained high for 5 years, as shown by the time-dependent ROC curve (Fig. 4B, C). In a merged cohort of the TCGA and two GEO cohorts, the GR signature was reliable for prognosis prediction (Fig. 4D). Thus, the GR signature was demonstrated to be reliable in its ability to predict survival in both internal and external cohorts. Next, we collected clinical phenotypic data for all samples in the TCGA cohort and explored their correlation with the GR score. The data showed that the higher the T stage (Fig. 5A), N stage (Fig. 5B), M stage (Fig. 5C), and grade (Fig. 5D), the higher the GR score (all p < 0.05). GR scores were also significantly higher in patients with recurrence or metastasis than in those without (Fig. 5E). Among the patients with complete followup data, the scores were significantly higher among deceased patients than among the survivors (Fig. 5F, p < 0.0001). Collectively, these results demonstrated the accuracy and reliability of the GR signature in predicting clinical phenotypes and patient outcomes. Interestingly, the mean GR score was higher in men than that in women (Fig. 5G, p < 0.001), suggesting that the GR signature may be related to sex hormones or other genes; however, further studies are needed to verify this. Univariate (Supplementary Fig. 3A) and multivariate (Supplementary Fig. 3B) Cox regression analyses of the resistance score and clinical characteristics, including age, sex, TNM stage, recurrence, and metastasis, revealed that the resistance score performed well in both analyses and served as an independent prognostic factor, which further confirmed the reliability of the GR signature. Functional enrichment and TIME analysis of GR signature To confirm the prognostic value of the GR signature in patients with LUAD from another perspective, we subjected the 100 genes with the highest positive (Fig. 6A) and negative (Fig. 6B) correlations with the GR signature resistance score to GO, KEGG, and Rectome functional enrichment analyses. The top 20 enriched pathways for each analysis (Fig. 6C–E) revealed that the GR signature was closely associated with cell cycle (particularly, chromosomal processes) and DNA replication, suggesting that the signature genes may affect the proliferation and survival of LUAD cells by interfering with mitotic processes, thus affecting patient prognosis. Based on epigenetic and transcriptome data, we calculated the mDNA stemness index (si) and mRNAsi for each sample and assessed their correlation with overall survival and risk (Supplementary Fig. 4A, B). Pearson correlation analysis revealed a positive correlation between the GR score and the mDNAsi (Supplementary Fig. 4A, r = 0.11, p < 0.05) and mRNAsi (Supplementary Fig. 4B, r = 0.35, p < 0.001) in the TCGA cohort. By applying GSEA to the high- and low-risk groups, we found that in many cancers, particularly LUAD, high GR scores were positively correlated with some cancer-promoting pathways, such as MYC targets V1/2, mTORC1 signaling, and epithelial–mesenchymal transition (Supplementary Fig. 5), and negatively correlated with some immune-related pathways, including the interferon response, inflammatory response, and IL6-JAK-STAT3 signaling (Supplementary Fig. 5). This suggested a correlation between a high GR score and tumor progression and corroborated the reliability of the GR signature in prognosis prediction. Using the integrated TCGA and GSE datasets (GSE13212 and GSE31210) and the R package IOBR, we assessed immune-cell infiltration in all samples. The low-risk group had a significantly higher degree of critical immune-cell infiltration than the high-risk group as assessed by multiple methods, including T cells in MCP-counter; B and CD4 + T cells in EPIC; CD8 + T and dendritic cells in xCell; memory B and activated NK cells in CIBERSORT; B, NK and CD4 + T cells in quanTiseq; and B, CD4 + T, and dendritic cells in TIMER (Fig. 7A). Based on the expression levels of related meta-genes, we calculated the microenvironment, ESTIMATE, immune, stromal, and other immune microenvironment-related scores for each sample and compared them between the two risk groups (Fig. 7B). In both xCell and ESTIMATE, the microenvironment, ESTIMATE, and immune scores of the low-risk group were significantly higher than those of high-risk group, whereas in xCell, the epithelial score of the high-risk group was significantly higher than that of the low-risk group. The correlations between the GR score and various chemokines and their receptors are shown in a heatmap in Supplementary Fig. 6A. TGFBR2, TGFBR3, CCR2, CCR6, CX3CR1, and IL33 were the most strongly correlated with the resistance score (Supplementary Fig. 6B). Together, these results revealed a strong relationship between the GR signature-related genes and the immune microenvironment, which may affect tumor occurrence and development by affecting key immune-cell infiltration into the tumor microenvironment or interfering with chemokine/receptor expression and binding, ultimately affecting the prognosis of patients with LUAD. Exploration of the impact of the GR signature on immune and non-immune therapies To evaluate the effect of the GR signature on the treatment of patients with LUAD, we compared the treatment effects of immune and non-immune drugs in the high- and low-risk groups. We obtained TIDE data for the merged cohort (Fig. 8A). The immunotherapy response rate of the low-risk group was evidently higher than that of the high-risk group (Fig. 8B). The TIDE (Fig. 8C) and exclusion (Fig. 8D) scores were also significantly higher in the high-risk group than in the low-risk group, indicating that the immune escape potential is greater and the efficacy of immune checkpoint inhibition is worse in the high-risk group. By comparing samples from the immunotherapy-responsive and non-responsive groups, we found that the GR score was significantly higher in the non-responsive group than in the responsive group (Fig. 8E). GSE135222, an anti-PD-1/PD-L1 immunotherapy dataset, was used to evaluate the relationship between the GR signature and LUAD immunotherapy. Survival curves indicated that a higher GR score was associated with shorter survival after immunotherapy (Fig. 8F). However, because of the small sample size, the survival curve analysis was not statistically significant. Similarly, the immunotherapy response rate was substantially higher in the low GR-score group than in the high GR-score group (Fig. 8G). These results suggested that patients in the high-risk group are more likely to have a worse response to immunotherapy, which is consistent with the above finding that these patients are more likely to be immunosuppressed. In addition to immunotherapy, IC50 values of multiple chemotherapy agents for all samples were analyzed using the R package oncoPredict. The low-risk group was more sensitive to several agents, including doramapimod and axitinib (Fig. 9A–F). Notably, while patients in the high-risk group were more immunosuppressed and did not have an immunotherapeutic advantage, they were more sensitive to certain chemotherapeutic agents, including docetaxel and erlotinib (Fig. 9G–L). These results suggested that these chemotherapeutic agents may have a good effect in high-risk patients who do not respond well to gefitinib. Two predicted agents, AZD7762 (Fig. 9M) and BI.2536 (Fig. 9N), inhibited the viability of gefitinib-resistant PC-9 cells significantly more strongly than did gefitinib. This finding provides new therapeutic directions for gefitinib-resistant patients and confirms that the GR signature provides reliable guidance for the treatment of patients with LUAD. Discussion Lung cancer remains the leading cause of cancer mortality worldwide, and LUAD accounts for a considerable portion of cases (Zheng et al. 2024). As a first-generation EGFR-TKI, gefitinib is widely used in the treatment of patients with clinically advanced NSCLC (Cappuzzo et al. 2016). However, due to congenital and acquired drug resistance, its efficacy is greatly reduced (L'Hostis et al. 2023). Although new EGFR-TKIs have been successfully developed and applied in clinical practice in recent years, suitable targeted drugs are still lacking for many patients with NSCLC (Tan et al. 2024;Remon et al. 2024). Therefore, new treatment strategies based on knowledge of the drug resistance mechanisms are urgently needed. We screened genes that play an important role in LUAD and explored their roles in biological processes and their effects on immune and non-immune therapies so as to unravel the drug resistance mechanism and provide guidance for the prediction and treatment of GR in patients with LUAD. We identified 11 cell subtypes from the GSE117570 dataset (Song et al. 2019) and analyzed the expression of upregulated GR-related gene sets across these cell subtypes in LUAD. Subsequently, we calculated an expression-based resistance score for each cell type and divided the cell subpopulations into gefitinib-resistant and -sensitive groups. We found that the proportions of M2 macrophages and malignant cells were significantly higher in the high GR-score group than in the low GR-score group. M2 macrophages in the tumor microenvironment can secrete anti-inflammatory cytokines and chemokines that promote tumor growth, invasion, and metastasis (Weinhauser et al. 2023;Bied et al. 2023). Further, M2 macrophages play an important role in tumor chemoresistance (Jiang et al. 2023), and targeting M2 macrophages is a potential therapeutic approach to overcome antitumor-drug resistance (Wang et al. 2024). We further screened the GR-related gene set at the transcriptome level to narrow down the number of genes. Using univariate and multivariate Cox regression analyses, 22 genes with the highest influence on the survival of patients with LUAD were selected. Notably, seven of these 22 genes had a hazard ratio <1 (Fig. 3A), implying that their expression is positively correlated with the survival time of patients with LUAD. Notably, genes that promote GR do not necessarily coincide with oncogenic genes in LUAD. These genes are involved in multiple types and levels of complex biological processes to promote GR, and it cannot be excluded that some of these processes may indirectly inhibit tumor progression. One of the most prognostically significant genes was SLC47A1 , which belongs to the multidrug and toxin extrusion family and is an organic cation antiporter in the plasma membrane (Lin et al. 2022). SLC47A1 is downregulated in LUAD tissues compared to normal tissues (HUANG et al. 2023) and acts as a tumor-suppressor gene. While SLC47A1 is a GR-associated gene and appears to negatively impact the prognostic outcome of LUAD, targeting this gene enhanced the efficacy of temozolomide in glioma stem cells (Batara et al. 2023). In the validation analysis, most of the 22 genes showed higher expression in gefitinib-resistant PC-9 cells than in normal PC-9 cells (Fig. 3B, C); a few genes showed no statistically significant difference and three genes showed an opposite trend (lower expression in resistant cells). Such inconsistent results may be attributed to heterogeneity between cell lines and human LUAD tissues. We used 10 classical algorithms and more than 100 algorithm combinations to select the most suitable algorithm for prognostic prediction model construction. We thus obtained a random forest algorithm with the best fit. In previous studies, often, only one to three algorithms were assessed to build a prediction model (Huang et al. 2019), and the screening processes involved many complex procedures, which led to a high workload and the selected algorithm sometimes being suboptimal. In recent years, algorithms have been iteratively updated, and more than 10 mature algorithms are currently available (Zhang et al. 2024). Machine learning techniques, as an emerging technology in recent years, provide effective tools for the screening of algorithms. They not only have enhanced computing power and reduce the workload, but also allow selecting the most suitable one among various algorithms to build a model, which is a great progress. We applied only the 10 most commonly used algorithms, and algorithms more suitable than the random forest model for this signature may become available in future. Future studies should include new algorithms to update the signature constructed in this study. The reliability of our signature was successfully confirmed by survival and clinical-feature validation analyses in internal and external datasets. During the course of gefitinib treatment, dynamic alterations within the TIME have been detected, indicating a strong association with the GR pathway (Duan et al. 2020). Accordingly, we assessed immune-cell infiltration in the high- and low-risk groups. We found that the microenvironment, ESTIMATE, and immune scores were significantly higher in the low-risk group than in the high-risk group. A comprehensive pan-cancer multi-omics study identified diverse immune checkpoints and lymphocyte-depleted features specific to EGFR-positive NSCLC (Seo et al. 2018), suggesting a natural resistance to immunotherapy in such cancers. Although the suboptimal efficacy of immunotherapy in patients with EGFR-positive tumors has been confirmed in clinical studies (Yang et al. 2019), chemo-immunotherapy combinations have shown promising results in patients with acquired GR and without specific targets (Hayashi et al. 2022). Finally, we aimed to provide valuable treatment guidance for patients with GR. Large-scale data collection and analysis of immune and non-immune therapies confirmed that the signature we constructed predicted immunotherapy efficacy to a certain degree. Further, we identified some non-immunotherapeutic drugs that may have good efficacy in gefitinib-resistant patients, providing potential new treatment methods for these patients. Drug-prediction and cellular experiments indicated that AZD7762 and BI.2536 have good efficacy against drug-resistant cells, which may offer hope for patients with GR in the future. Unfortunately, most of these drugs are still in the research stage, and further studies are needed to confirm their efficacy in gefitinib-resistant patients and their application in clinical practice. This study had several limitations. We obtained single-cell data of LUAD from public databases and focused on the expression of gene sets highly expressed in GR in various cell types and the main biological processes involved. Single-cell research has become popular in recent years, and the data that can be explored are very extensive. Nevertheless, our exploration was preliminary, and further analysis may reveal more information on GR gene sets. In addition, as all data were derived from public databases, the sample conditions were determined, limiting the scope of the study. Future studies may address this issue by using single-cell data from purposefully collected samples. Finally, we only selected gefitinib-resistant and normal PC-9 LUAD cells for the experiments and did not verify these genes in other cell types. Data of a single cell line cannot completely represent the data of all LUAD cells in the database. Therefore, further experiments need to be carried out in other cell lines to confirm the results. Conclusions Using single-cell and transcriptome data, we established a 22 GR gene signature that performed well in survival prediction in multiple LUAD datasets and may serve as an independent clinical prognostic factor. Enrichment and TIME analyses demonstrated that the signature genes significantly affected chromosomal processes and DNA replication, important immune-cell infiltration, and immune scores. Treatment prediction demonstrated that the signature predicted immunotherapy efficacy in patients with LUAD to a certain extent and identified several chemotherapy agents to which gefitinib-resistant patients may respond well. Clinical verification of the prognostic signature is required in future. Declarations Acknowledgements The authors thank the researchers who uploaded sequencing data to the public databases and everyone who helped in this study. Author Contributions Dong Zhou : Methodology, Writing–original draft. Zhi Zheng : Data curation, Writing–original draft. Yan-qi Li : Data curation, Methodology. Quan-Xing Liu : Data curation. Xu-Feng Deng : Data curation. Liang Chen : Data curation; Formal analysis. Man-Yuan Li : Investigation. Jiao Zhang : Data curation. Xiao Lu : Conceptualization. Hong Zheng: Conceptualization, Writing–review & editing. Ji-Gang Dai : Methodology, Supervision, Writing–review & editing. Funding This study was supported by grants from the National Natural Science Foundation of China for Dai (No. 81972190), Key Projects on Technological Innovation and Application Development of Chongqing (2022-195), and the Science and Technology Commission and the Chongqing Health Commission Joint Medical Research Program of Chongqing (2024MSXM090). Data availability The datasets generated and analyzed in the current study are available in the TCGA (https://portal.gdc.cancer.gov/) and GEO (https://www.ncbi.nlm.nih/) databases. Competing interests The authors declare no competing interests. Ethics approval Not applicable. References Batara DC, Park SW, Kim HJ, Choi SY, Ohn T, Choi MC, Park SI, Kim SH (2023) Targeting the multidrug and toxin extrusion 1 gene (SLC47A1) sensitizes glioma stem cells to temozolomide. Am J Cancer Res 13(9):4021-4038 Bied M, Ho WW, Ginhoux F, Bleriot C (2023) Roles of macrophages in tumor development: a spatiotemporal perspective. Cell Mol Immunol 20(9):983-992. 10.1038/s41423-023-01061-6 Cappuzzo F, Morabito A, Normanno N, Bidoli P, Del CA, Giannetta L, Montanino A, Mazzoni F, Buosi R, Burgio MA, Cerea G, Chiari R, Cortinovis D, Finocchiaro G, Foltran L, Migliorino MR, Tiseo M, Ferrari S, De Marinis F (2016) Efficacy and safety of rechallenge treatment with gefitinib in patients with advanced non-small cell lung cancer. Lung Cancer 99:31-37. 10.1016/j.lungcan.2016.06.008 Choi HY, Chang JE (2023) Targeted Therapy for Cancers: From Ongoing Clinical Trials to FDA-Approved Drugs. Int J Mol Sci 24(17). 10.3390/ijms241713618 Duan Q, Zhang H, Zheng J, Zhang L (2020) Turning Cold into Hot: Firing up the Tumor Microenvironment. Trends Cancer 6(7):605-618. 10.1016/j.trecan.2020.02.022 Fan J, Fang L, Cong S, Zhang Y, Jiang X, Wang N, Chen Y (2024) Potential pre-COPD indicators in association with COPD development and COPD prediction models in Chinese: a prospective cohort study. Lancet Reg Health West Pac 44:100984. 10.1016/j.lanwpc.2023.100984 Ge Z, Xu M, Ge Y, Huang G, Chen D, Ye X, Xiao Y, Zhu H, Yin R, Shen H, Ma G, Qi L, Wei G, Li D, Wei S, Zhu M, Ma H, Shi Z, Wang X, Ge X, Qian X (2023) Inhibiting G6PD by quercetin promotes degradation of EGFR T790M mutation. Cell Rep 42(11):113417. 10.1016/j.celrep.2023.113417 Haratake N, Ozawa H, Morimoto Y, Yamashita N, Daimon T, Bhattacharya A, Wang K, Nakashoji A, Isozaki H, Shimokawa M, Kikutake C, Suyama M, Hashinokuchi A, Takada K, Takenaka T, Yoshizumi T, Mitsudomi T, Hata AN, Kufe D (2024) MUC1-C Is a Common Driver of Acquired Osimertinib Resistance in NSCLC. J Thorac Oncol 19(3):434-450. 10.1016/j.jtho.2023.10.017 Hayashi H, Sugawara S, Fukuda Y, Fujimoto D, Miura S, Ota K, Ozawa Y, Hara S, Tanizaki J, Azuma K, Omori S, Tachihara M, Nishino K, Bessho A, Chiba Y, Haratani K, Sakai K, Nishio K, Yamamoto N, Nakagawa K (2022) A Randomized Phase II Study Comparing Nivolumab with Carboplatin-Pemetrexed for EGFR-Mutated NSCLC with Resistance to EGFR Tyrosine Kinase Inhibitors (WJOG8515L). Clin Cancer Res 28(5):893-902. 10.1158/1078-0432.CCR-21-3194 HUANG N, LIU Z, LI R (2023) Pan-cancer analysis reveals the relationship between SLC47A1 immune infiltration and clinical prognosis of human tumors. Journal of Holistic Integrative Pharmacy 4(1):103-118. 10.1016/S2707-3688(23)00098-5 Huang P, Lin CT, Li Y, Tammemagi MC, Brock MV, Atkar-Khattra S, Xu Y, Hu P, Mayo JR, Schmidt H, Gingras M, Pasian S, Stewart L, Tsai S, Seely JM, Manos D, Burrowes P, Bhatia R, Tsao MS, Lam S (2019) Prediction of lung cancer risk at follow-up screening with low-dose CT: a training and validation study of a deep learning method. Lancet Digit Health 1(7):e353-e362. 10.1016/S2589-7500(19)30159-1 Jiang S, Deng T, Cheng H, Liu W, Shi D, Yuan J, He Z, Wang W, Chen B, Ma L, Zhang X, Gong P (2023) Macrophage-organoid co-culture model for identifying treatment strategies against macrophage-related gemcitabine resistance. J Exp Clin Cancer Res 42(1):199. 10.1186/s13046-023-02756-4 Johnson M, Garassino MC, Mok T, Mitsudomi T (2022) Treatment strategies and outcomes for patients with EGFR-mutant non-small cell lung cancer resistant to EGFR tyrosine kinase inhibitors: Focus on novel therapies. Lung Cancer 170:41-51. 10.1016/j.lungcan.2022.05.011 L'Hostis A, Palgen JL, Perrillat-Mercerot A, Peyronnet E, Jacob E, Bosley J, Duruisseaux M, Toueg R, Lefevre L, Kahoul R, Ceres N, Monteiro C (2023) Knowledge-based mechanistic modeling accurately predicts disease progression with gefitinib in EGFR-mutant lung adenocarcinoma. NPJ Syst Biol Appl 9(1):37. 10.1038/s41540-023-00292-7 Lin Z, Liu J, Long F, Kang R, Kroemer G, Tang D, Yang M (2022) The lipid flippase SLC47A1 blocks metabolic vulnerability to ferroptosis. Nat Commun 13(1):7965. 10.1038/s41467-022-35707-2 Lu C, Wei XW, Wang Z, Zhou Z, Liu YT, Zheng D, He Y, Xie ZH, Li Y, Zhang Y, Zhang YC, Huang ZJ, Mei SQ, Liu JQ, Guan XH, Deng Y, Chen ZH, Tu HY, Xu CR, Chen HJ, Zhong WZ, Yang JJ, Zhang XC, Mok T, Wu YL, Zhou Q (2024) Allelic Context of EGFR C797X-Mutant Lung Cancer Defines Four Subtypes With Heterogeneous Genomic Landscape and Distinct Clinical Outcomes. J Thorac Oncol 19(4):601-612. 10.1016/j.jtho.2023.11.016 Noronha V, Menon NS, Patil VM, Chandrakanth MV, More S, Dhanawat A, Chowdhary OR, Singh AC, Goud S, Shah S, Karuvandan N, Jobanputra KN, Shah DK, Shah MJ, Sarma R, Patel D, Joarder R, Kumar P, John A, Kaur J, Bagra S, Purandare N, Janu A, Mahajan A, Prabhash K (2024) A Comparative Study Evaluating the Quality of Life and Survival Outcomes in Patients Receiving Chemotherapy Versus Oral Tyrosine Kinase Inhibitor in the Third Line and Beyond Setting for Advanced NSCLC. JTO Clin Res Rep 5(1):100622. 10.1016/j.jtocrr.2023.100622 Qu FJ, Zhou Y, Wu S (2024) Progress of immune checkpoint inhibitors therapy for non-small cell lung cancer with liver metastases. Br J Cancer 130(2):165-175. 10.1038/s41416-023-02482-w Reita D, Pabst L, Pencreach E, Guerin E, Dano L, Rimelen V, Voegeli AC, Vallat L, Mascaux C, Beau-Faller M (2021) Molecular Mechanism of EGFR-TKI Resistance in EGFR-Mutated Non-Small Cell Lung Cancer: Application to Biological Diagnostic and Monitoring. Cancers (Basel) 13(19). 10.3390/cancers13194926 Remon J, Besse B, Aix SP, Callejo A, Al-Rabi K, Bernabe R, Greillier L, Majem M, Reguart N, Monnet I, Cousin S, Garrido P, Robinet G, Campelo RG, Madroszyk A, Mazieres J, Curcio H, Wasag B, Pretzenbacher Y, Grillet F, Dingemans AC, Dziadziuszko R (2024) Overall Survival From the EORTC LCG-1613 APPLE Trial of Osimertinib Versus Gefitinib Followed by Osimertinib in Advanced EGFR-Mutant Non-Small-Cell Lung Cancer. J Clin Oncol 42(12):1350-1356. 10.1200/JCO.23.01521 Seo JS, Kim A, Shin JY, Kim YT (2018) Comprehensive analysis of the tumor immune micro-environment in non-small cell lung cancer for efficacy of checkpoint inhibitor. Sci Rep 8(1):14576. 10.1038/s41598-018-32855-8 Siegel RL, Giaquinto AN, Jemal A (2024) Cancer statistics, 2024. CA Cancer J Clin 74(1):12-49. 10.3322/caac.21820 Song Q, Hawkins GA, Wudel L, Chou PC, Forbes E, Pullikuth AK, Liu L, Jin G, Craddock L, Topaloglu U, Kucera G, O'Neill S, Levine EA, Sun P, Watabe K, Lu Y, Alexander-Miller MA, Pasche B, Miller LD, Zhang W (2019) Dissecting intratumoral myeloid cell plasticity by single cell RNA-seq. Cancer Med 8(6):3072-3085. 10.1002/cam4.2113 Tan L, Brown C, Mersiades A, Lee CK, John T, Kao S, Newnham G, O'Byrne K, Parakh S, Bray V, Jasas K, Yip S, Wong SQ, Ftouni S, Guinto J, Chandrashekar S, Clarke S, Pavlakis N, Stockler MR, Dawson SJ, Solomon BJ (2024) A Phase II trial of alternating osimertinib and gefitinib therapy in advanced EGFR-T790M positive non-small cell lung cancer: OSCILLATE. Nat Commun 15(1):1823. 10.1038/s41467-024-46008-1 Wang Q, Zhang J, Liu Z, Duan Y, Li C (2023) Integrative approaches based on genomic techniques in the functional studies on enhancers. Brief Bioinform 25(1). 10.1093/bib/bbad442 Wang S, Wang J, Chen Z, Luo J, Guo W, Sun L, Lin L (2024) Targeting M2-like tumor-associated macrophages is a potential therapeutic approach to overcome antitumor drug resistance. NPJ Precis Oncol 8(1):31. 10.1038/s41698-024-00522-z Weinhauser I, Pereira-Martins DA, Almeida LY, Hilberink JR, Silveira D, Quek L, Ortiz C, Araujo CL, Bianco TM, Lucena-Araujo A, Mota JM, Hogeling SM, Sternadt D, Visser N, Diepstra A, Ammatuna E, Huls G, Rego EM, Schuringa JJ (2023) M2 macrophages drive leukemic transformation by imposing resistance to phagocytosis and improving mitochondrial metabolism. Sci Adv 9(15):f8522. 10.1126/sciadv.adf8522 Yang JC, Gadgeel SM, Sequist LV, Wu CL, Papadimitrakopoulou VA, Su WC, Fiore J, Saraf S, Raftopoulos H, Patnaik A (2019) Pembrolizumab in Combination With Erlotinib or Gefitinib as First-Line Therapy for Advanced NSCLC With Sensitizing EGFR Mutation. J Thorac Oncol 14(3):553-559. 10.1016/j.jtho.2018.11.028 Yang JC, Srimuninnimit V, Ahn MJ, Lin CC, Kim SW, Tsai CM, Mok T, Orlando M, Puri T, Wang X, Park K (2016) First-Line Pemetrexed plus Cisplatin followed by Gefitinib Maintenance Therapy versus Gefitinib Monotherapy in East Asian Never-Smoker Patients with Locally Advanced or Metastatic Nonsquamous Non-Small Cell Lung Cancer: Final Overall Survival Results from a Randomized Phase 3 Study. J Thorac Oncol 11(3):370-379. 10.1016/j.jtho.2015.11.008 Yang Y, Hua Y, Zheng H, Jia R, Ye Z, Su G, Gu Y, Zhan K, Tang K, Qi S, Wu H, Qin S, Huang S (2024) Biomarkers prediction and immune landscape in ulcerative colitis: Findings based on bioinformatics and machine learning. Comput Biol Med 168:107778. 10.1016/j.compbiomed.2023.107778 Zhang L, Zhang X, Guan M, Zeng J, Yu F, Lai F (2024) Identification of a novel ADCC-related gene signature for predicting the prognosis and therapy response in lung adenocarcinoma. Inflamm Res. 10.1007/s00011-024-01871-y Zheng RS, Chen R, Han BF, Wang SM, Li L, Sun KX, Zeng HM, Wei WW, He J (2024) [Cancer incidence and mortality in China, 2022]. Zhonghua Zhong Liu Za Zhi 46(3):221-231. 10.3760/cma.j.cn112152-20240119-00035 Additional Declarations No competing interests reported. Supplementary Files Supplementarydata4.22.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4573455","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":317622099,"identity":"3a1b0e7b-f992-401f-919d-4427d648603b","order_by":0,"name":"Dong Zhou","email":"","orcid":"","institution":"Xinqiao Hospital, Third Military Medical University (Army Medical University)","correspondingAuthor":false,"prefix":"","firstName":"Dong","middleName":"","lastName":"Zhou","suffix":""},{"id":317622100,"identity":"88948896-4f47-4dc4-91de-aa53ec38bf6e","order_by":1,"name":"Zhi Zheng","email":"","orcid":"","institution":"Xinqiao Hospital, Third Military Medical University (Army Medical University)","correspondingAuthor":false,"prefix":"","firstName":"Zhi","middleName":"","lastName":"Zheng","suffix":""},{"id":317622101,"identity":"9e436626-c839-4ce1-b0b7-5a1e0435fa1f","order_by":2,"name":"Yan-Qi Li","email":"","orcid":"","institution":"Xinqiao Hospital, Third Military Medical University (Army Medical University)","correspondingAuthor":false,"prefix":"","firstName":"Yan-Qi","middleName":"","lastName":"Li","suffix":""},{"id":317622102,"identity":"0681acce-8b54-4b07-902d-86eff31eb8de","order_by":3,"name":"Quan-Xing Liu","email":"","orcid":"","institution":"Xinqiao Hospital, Third Military Medical University (Army Medical University)","correspondingAuthor":false,"prefix":"","firstName":"Quan-Xing","middleName":"","lastName":"Liu","suffix":""},{"id":317622103,"identity":"a2c9259f-e762-4e03-82ae-90af519ef47c","order_by":4,"name":"Xu-Feng Deng","email":"","orcid":"","institution":"Xinqiao Hospital, Third Military Medical University (Army Medical University)","correspondingAuthor":false,"prefix":"","firstName":"Xu-Feng","middleName":"","lastName":"Deng","suffix":""},{"id":317622104,"identity":"23c348eb-183f-4a82-a9b4-fa74e75846dc","order_by":5,"name":"Liang Chen","email":"","orcid":"","institution":"Xinqiao Hospital, Third Military Medical University (Army Medical University)","correspondingAuthor":false,"prefix":"","firstName":"Liang","middleName":"","lastName":"Chen","suffix":""},{"id":317622106,"identity":"698679b8-82aa-4d4a-903e-6cdea70b80d4","order_by":6,"name":"Man-Yuan Li","email":"","orcid":"","institution":"Xinqiao Hospital, Third Military Medical University (Army Medical University)","correspondingAuthor":false,"prefix":"","firstName":"Man-Yuan","middleName":"","lastName":"Li","suffix":""},{"id":317622108,"identity":"4a6a08c3-edbc-4065-9cdb-dbf21e4a6c39","order_by":7,"name":"Jiao Zhang","email":"","orcid":"","institution":"Xinqiao Hospital, Third Military Medical University (Army Medical University)","correspondingAuthor":false,"prefix":"","firstName":"Jiao","middleName":"","lastName":"Zhang","suffix":""},{"id":317622110,"identity":"94771ff4-e40d-48f4-b8b9-a722a59a0484","order_by":8,"name":"Xiao Lu","email":"","orcid":"","institution":"Xinqiao Hospital, Third Military Medical University (Army Medical University)","correspondingAuthor":false,"prefix":"","firstName":"Xiao","middleName":"","lastName":"Lu","suffix":""},{"id":317622111,"identity":"f48ac2bb-b753-4424-b775-97973fcfcbba","order_by":9,"name":"Hong Zheng","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAv0lEQVRIiWNgGAWjYDACZjBpA+WxEa8lTYIELRBwmAQt8u3MBz98qDhfpzvtjAHDh7LDDPyzG/BrYWxmS5accea2hNntHAPGGecOM0jcOYBfCzMzjxkzbxtEC5BxmMFAIgG/FjZm/m/MvP/OQbT8JUYLDzMPGzNvwwGIFkZitEgwsxlLzjiWLLntdlrBwZ5z6TwSNwhoke8//PDDhxo7frPbyRsf/CizluOfQUALCjgAcikJ6kfBKBgFo2AU4AIAQGY7J9xDh7kAAAAASUVORK5CYII=","orcid":"","institution":"Xinqiao Hospital, Third Military Medical University (Army Medical University)","correspondingAuthor":true,"prefix":"","firstName":"Hong","middleName":"","lastName":"Zheng","suffix":""},{"id":317622116,"identity":"b1637200-853c-4373-9d9f-deba92bee70b","order_by":10,"name":"Ji-Gang Dai","email":"","orcid":"","institution":"Xinqiao Hospital, Third Military Medical University (Army Medical University)","correspondingAuthor":false,"prefix":"","firstName":"Ji-Gang","middleName":"","lastName":"Dai","suffix":""}],"badges":[],"createdAt":"2024-06-13 04:23:18","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4573455/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4573455/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":60447247,"identity":"cfc06e2c-3483-47e3-91f5-f3f77f1a64b3","added_by":"auto","created_at":"2024-07-16 21:59:47","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1210888,"visible":true,"origin":"","legend":"\u003cp\u003eSingle-cell data processing. (A) Uniform manifold approximation and projection map of cell identification and distribution. (B) Top 5 high- and low-expression genes in each cell group. (C) Heatmap of the top 10 marker genes in each cell group. KEGG (D) and Reactome (E) enrichment analyses of the top 100 genes expressed in each cell population.\u003c/p\u003e","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-4573455/v1/d1a7e81aebebd400cf28c6c3.png"},{"id":60447241,"identity":"67a3b21f-342d-41eb-8a1b-3c162b9c7e58","added_by":"auto","created_at":"2024-07-16 21:59:47","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":533253,"visible":true,"origin":"","legend":"\u003cp\u003eExpression and score of GR genes in single cells. (A) Expression levels of GR genes in single cells. (B) Gene set variation analysis of the GR gene set in single cells. (C) Cells were divided into high- and low-resistance groups based on the median score obtained from gene set variation analysis. Cell grouping (left), gene set variation analysis-based score for each cell (middle), grouping into high and low drug resistance of each cell (right). (D) Proportions of cell types in the high- and low-resistance groups\u003c/p\u003e","description":"","filename":"Figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-4573455/v1/e8c69fe5f24e3b5e9791ff6f.png"},{"id":60447244,"identity":"820b968e-908d-47d8-8b07-9bb3c670156b","added_by":"auto","created_at":"2024-07-16 21:59:47","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":565101,"visible":true,"origin":"","legend":"\u003cp\u003eScreening of GR genes with prognostic significance and signature construction using machine learning techniques. (A) Forest map of 22 GR genes with prognostic significance. (B) RT-qPCR verification of the expression levels of 22 genes in PC-9 normal and GR cell lines. (C) Machine learning techniques were used to identify the most appropriate algorithm among 10 algorithms and their combinations in the merged cohort. (D) A random forest algorithm was selected and a scoring model was constructed according to the importance of each gene.\u003c/p\u003e","description":"","filename":"Figure3.png","url":"https://assets-eu.researchsquare.com/files/rs-4573455/v1/2e63419d6bb77d4e84ee63ca.png"},{"id":60447243,"identity":"c577b152-1667-4470-9d1c-7ecb74e0451f","added_by":"auto","created_at":"2024-07-16 21:59:47","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":336676,"visible":true,"origin":"","legend":"\u003cp\u003eInternal and external validation of the constructed signature. The resistance score of each sample was calculated based on the algorithm corresponding to the constructed tags, and samples were divided into high- and low-risk groups based on the median. Survival curves and time-dependent ROC curves of TCGA (A), GSE13213 (B), GSE31210 (C), and merged (D) cohorts.\u003c/p\u003e","description":"","filename":"Figure4.png","url":"https://assets-eu.researchsquare.com/files/rs-4573455/v1/762c5726082f5eb6705c500a.png"},{"id":60448518,"identity":"dbe2c721-6c6e-47f3-9557-cb05bf1e24f7","added_by":"auto","created_at":"2024-07-16 22:15:47","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":220159,"visible":true,"origin":"","legend":"\u003cp\u003eAssociation analysis between GR signature and clinical characteristics. Comparison of T stage (A), N stage (B), M stage (C), tumor grade (D), recurrence or metastasis (E), survival status (F),and sex (G) between high- and low-risk groups.\u003c/p\u003e","description":"","filename":"Figure5.png","url":"https://assets-eu.researchsquare.com/files/rs-4573455/v1/c0ece96ec43a6d774a2606f0.png"},{"id":60448517,"identity":"14585606-9516-4c3a-8073-be15ab1de678","added_by":"auto","created_at":"2024-07-16 22:15:47","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":670965,"visible":true,"origin":"","legend":"\u003cp\u003eFunctional enrichment analysis of the GR signature. Heatmaps of the expression of the top 50 genes positively (A) or negatively (B) correlated with the 22 genes in the GR signature. (C) GO (left), KEGG (middle), Reactome (right) enrichment analyses of the GR signature.\u003c/p\u003e","description":"","filename":"Figure6.png","url":"https://assets-eu.researchsquare.com/files/rs-4573455/v1/5d3cfae4663f787f25135292.png"},{"id":60447251,"identity":"78885505-63b1-45cf-9662-faa96f1e1351","added_by":"auto","created_at":"2024-07-16 21:59:48","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":1258563,"visible":true,"origin":"","legend":"\u003cp\u003eAnalysis of the relationship between the GR signature and TIME. (A) Heatmap of the infiltration of various immune cells into the tumor microenvironment in the high- and low-risk groups based on eight methods (MCPcounter, EPIC, xCell, CIBERSORT, IPS, quanTIseq, ESTIMATE, TIMER). (B) Comparison of the microenvironment, ESTIMATE, immune, and stromal scores between two risk groups.\u003c/p\u003e","description":"","filename":"Figure7.png","url":"https://assets-eu.researchsquare.com/files/rs-4573455/v1/00b87f6b3f2da5c845272671.png"},{"id":60447249,"identity":"9ac62738-ba77-4a69-b9a6-09ec21b9e9d4","added_by":"auto","created_at":"2024-07-16 21:59:47","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":220983,"visible":true,"origin":"","legend":"\u003cp\u003eCorrelation analysis between the GR signature and the effect of immunotherapy in patients with LUAD. (A) TIDE values for each sample in the merged cohort. (B) Comparison of the immunotherapy response between high- and low-risk groups. Comparison of the TIDE (C) and exclusion (D) scores between high- and low-risk groups. (E) Differences in GR signature scores in patients with different responses to immunotherapy. (F) In the anti-PD-1/PD-L1 immunotherapy dataset GSE135222, the samples were divided into high- and low-risk groups according to the GR score (cut-off, 7.3). Survival curves were compared between the two groups. (G) Comparison of the proportion of immunotherapy response between high- and low GR-score groups in GSE135222. CR, complete response; PR, partial response; SD, stable disease; PD, progressive disease.\u003c/p\u003e","description":"","filename":"Figure8.png","url":"https://assets-eu.researchsquare.com/files/rs-4573455/v1/5b1f444d82c4cc99b22cfa7c.png"},{"id":60447250,"identity":"e5bef314-b51f-45b6-a63d-4755af46a0d0","added_by":"auto","created_at":"2024-07-16 21:59:47","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":240016,"visible":true,"origin":"","legend":"\u003cp\u003ePrediction of non-immune drug sensitivity. (A–L) Comparison of IC50 values of 12 chemotherapy agents between high- and low-risk groups. Agents (A–F) were more effective in the low-risk group, and agents (G–L) were more effective in the high-risk group. Effect of AZD7762 (M) and BI.2536 (N) on the viability of PC-9 FR cells.\u003c/p\u003e","description":"","filename":"Figure9.png","url":"https://assets-eu.researchsquare.com/files/rs-4573455/v1/3db4a3525e1abbd5dbc11a41.png"},{"id":61929153,"identity":"f42c77de-a38e-457c-897a-9bdecd328f53","added_by":"auto","created_at":"2024-08-07 07:48:53","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":5627027,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4573455/v1/5657f908-5f85-47f8-95c1-737545df14a0.pdf"},{"id":60447766,"identity":"d1bfb16c-8224-40ec-8f52-e68de0747c40","added_by":"auto","created_at":"2024-07-16 22:07:47","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":3137156,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementarydata4.22.docx","url":"https://assets-eu.researchsquare.com/files/rs-4573455/v1/2039c886cba22029525fee3a.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Identification of a gefitinib resistance-associated signature for predicting prognosis and therapeutic response in lung adenocarcinoma via integrated multi-omics analysis and machine learning","fulltext":[{"header":"Introduction","content":"\u003cp\u003eLung cancer remains one of the most common and deadliest malignant tumors worldwide, with a 5-year survival rate of\u0026nbsp;approximately 20%\u0026nbsp;(Siegel et al. 2024). Chemotherapy is the traditional treatment for advanced lung cancer, but it has limited efficacy, and its significant side effects greatly impact patients\u0026rsquo; quality of life\u0026nbsp;(Qu et al. 2024). With the advent of targeted therapies, the most common being epidermal growth factor receptor (EGFR) tyrosine kinase inhibitors (TKIs)\u0026nbsp;(Noronha et al. 2024), the survival of patients with\u0026nbsp;advanced driver gene-positive lung adenocarcinoma (LUAD) has greatly improved\u0026nbsp;(Choi, Chang 2023). Gefitinib, a first-generation EGFR-TKI, is widely used in the treatment of advanced EGFR mutation-positive patients with LUAD because of its improved efficacy and relatively low side effects\u0026nbsp;(Yang et al. 2016). However, inherent and acquired resistance to gefitinib substantially impair its clinical efficacy\u0026nbsp;(Johnson et al. 2022). Drug resistance and consequent\u0026nbsp;tumor metastasis are considered factors that directly affect patient prognosis. Therefore, there is an urgent\u0026nbsp;need for effective and tolerable regimens to address gefitinib resistance (GR).\u003c/p\u003e\n\u003cp\u003eThe mechanism of GR is complex and involves multiple genes, layers, and dimensions, including genetic mutations in the EGFR pathway, abnormal\u0026nbsp;bypass pathway\u0026nbsp;activation, and changes in phenotypic characteristics\u0026nbsp;(Reita et al. 2021). For example, the EGFR T790M mutation, is a primary acquired GR mechanism in LUAD\u0026nbsp;(Ge et al. 2023). In tackling the resistance issue, the importance of exploring alternative treatments cannot be overstated. Osimertinib, a third-generation EGFR mutant-selective TKI, stands out as a promising option as it has\u0026nbsp;demonstrated high effectiveness in patients with non-small cell lung cancer (NSCLC)\u0026nbsp;carrying the T790M mutation\u0026nbsp;(Ge et al. 2023). However, the relatively small population with acquired resistance due to T790M and emergence of secondary resistance with the use of osimertinib strongly limit its clinical application\u0026nbsp;(Lu et al. 2024;Haratake et al. 2024). Therefore,\u0026nbsp;early identification of the\u0026nbsp;GR risk in patients with LUAD\u0026nbsp;and early intervention\u0026nbsp;are particularly important, enabling physicians to devise a treatment strategy based on disease progression prediction.\u003c/p\u003e\n\u003cp\u003eThe proliferation of public genome data has led to the rise of meta-analysis and computational modeling as pivotal methods for circumventing the constraints of insufficient statistical power in isolated studies\u0026nbsp;(Wang et al. 2023). In the past few years, regularized regression classifiers, such as the least absolute shrinkage and selection operator and elastic net,\u0026nbsp;have become increasingly recognized as\u0026nbsp;efficient tools for feature selection and prediction in high-dimensional datasets\u0026nbsp;(Yang et al. 2024;Fan et al. 2024).\u003c/p\u003e\n\u003cp\u003eIn the present study, we analyzed GR-related genes using integrated bulk and single-cell RNA-sequencing (scRNA-seq) data to a better understanding of the mechanism underlying GR, methods for predicting and preventing GR, and alternative therapies. First, we analyzed the expression characteristics of GR genes in various cell subgroups based on scRNA-seq data. Next, multiple supervised machine learning methods were used to identify GR signature genes, and a clinical prediction model was built based on the results. Finally, following a thorough evaluation of the biological mechanisms of the signature genes identified, their involvement in LUAD was validated based on database information and cell experiments.\u003c/p\u003e"},{"header":"Materials And Methods","content":"\u003cp\u003e\u003cstrong\u003eData collection and processing\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWhole-gene mRNA expression matrices, clinical data (phenotypes and overall survival), and single-nucleotide polymorphism mutation data of Cancer Genome Atlas (TCGA)-LUAD samples were downloaded from the UCSC website (https://xenabrowser.net/). In total, 513 tumor samples and 59 normal samples were obtained, and samples with missing information were excluded. The GSE117570 dataset, which comprises 10\u0026times; scRNA-seq data of 11,485 cells, was obtained from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/), and data on P1, P3, and P4 were selected for analysis. Expression data of signature genes in particular cells were downloaded from the Tumor Immune Single-cell Hub database (http://tisch.comp-genomics.org/home/). Two GEO datasets (GSE31210 and GSE13213) with complete mRNA expression and clinical information were obtained from the GEO database to construct the validation cohort. Genes upregulated in GR were obtained from the Gene Set Enrichment Analysis (GSEA) database (https://www.gsea-msigdb.org/gsea/index.jsp, \u0026ldquo;COLDREN_GEFITINIB_RESISTANCE_UP\u0026rdquo; dataset). \u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003escRNA-seq data collection and analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe R package Seurat was used to convert raw scRNA-seq data into Seurat objects, followed by quality control and the exclusion of unqualified cells. The gene expression in core cells was standardized using a linear regression algorithm, and the top 2,000 highly variable genes were identified using analysis of variance. Uniform manifold approximation and projection was used to identify cell subpopulations. The FindAllMarkers function in the R package Seurat and the COSG R package were employed to identify marker genes for all cell clusters and differentially expressed genes among the screened marker genes. The R package clusterProfiler was used for Kyoto Encyclopedia of Genes and Genomes (KEGG) and Reactome enrichment analyses of top 100 marker genes in each cell cluster. A single-sample GSEA score was calculated for each cell cluster based on the expression levels of GR-related genes, and all cell samples were divided into high- and low-score groups based on the median score. Data of 50 hallmark gene sets were downloaded from the GSEA database and the R package GSVA was employed to calculate hallmark scores for each cell cluster.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eIdentification and validation of a prognostic signature\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eUnivariate Cox proportional risk regression analysis was conducted on GR-related genes in the TCGA-LUAD training cohort, using \u003cem\u003ep \u003c/em\u003e\u0026lt; 0.05 as the screening criterion. Ten classical algorithms (single or combined), including random forest, least absolute shrinkage and selection operator, gradient boosting machine, survival support vector machine, supervised principal components, ridge regression, partial least squares regression for Cox, CoxBoost, Stepwise Cox, and elastic network, were used to construct a prognostic signature based on the screened genes, and the C-index was calculated for all signatures to select the most suitable one. The R packages survival, pROC, and ggplot2 were used for survival analysis and time-dependent receiver operating characteristic (ROC) analysis using the TCGA-LUAD training cohort, GSE13213 and GSE31210 validation cohorts, and their combination.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePrognostic analysis of risk groups and clinical phenotypes\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe proportions of clinical phenotype (sex, T stage, N stage, M stage, stage, recurrence, metastasis, and overall survival) subcategories in the high- and low-risk groups were calculated, and resistance scores were systematically compared among the subcategories. Univariate and multivariate Cox regression analyses were used to assess the influence of the prognostic signature and other common clinical phenotypic features on the prognosis of patients with LUAD.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunctional enrichment analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCorrelation coefficients between the resistance score and all genes were calculated, and the 100 genes with the highest positive and negative correlations were screened out. The R package clusterProfiler was used for Gene Ontology (GO), KEGG, and Reactome functional enrichment analyses of the screened genes. After calculating the score for each sample based on the constructed signature, all samples were divided into high- and low-risk groups, and GSEA was applied to the differentially expressed genes between the two groups. The cancer-related hallmark gene set file (h.all.v2023.2.Hs.symbols.gmt) was downloaded from the Molecular Signatures Database (https://www.gsea-msigdb.org/gsea/msigdb). The R package GSEA (v1.38.2) was used to calculate the normalized enrichment score and false discovery rate, and the R package ggplot2 (v3.3.6) was used to display the GSEA results.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAnalysis on tumor immune microenvironment (TIME) and tumor stemness\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eLUAD samples were ordered according to stemness indices (mRNAsi and mDNAsi) and their overall survival and distribution of high- and low-risk groups were shown. Pearson correlation coefficients between stemness indices and resistance scores were calculated. Immune-cell infiltration was analyzed using the MCPCOUNTER, EPIC, XCELL, CIBERSORT, IPS, QUANTISEQ, ESTIMATE, and TIMER algorithms in the TIMER database (https://cistrome.shinyapps.io/timer/), and the infiltration scores of all types of immune cells were compared between the high- and low-risk groups using the R package IOBR. Multiple immune microenvironment-related indices, including the microenvironment, ESTIMATE, immune, and stroma scores, were calculated and compared between the high- and low-risk groups. The expression levels of various chemokines and their receptors were compared between the high- and low-risk groups, and Pearson correlation coefficients between the expression levels of significantly differentially expressed chemokines and the resistance scores were calculated.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAnalysis of drug efficacy prediction in gefitinib-resistant populations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData of patients who received immunotherapy was downloaded from the Tumor Immune Dysfunction and Exclusion (TIDE) database (http://tide.dfci.harvard.edu), and true or false response to immunotherapy and TIDE and exclusion scores were compared between the high- and low-risk groups. GSE135222, an anti-programmed cell death protein 1 (PD-1)/PD-1 ligand (PD-L1) immunotherapy dataset from GEO, was used to evaluate the relationship between the GR signature and LUAD immunotherapy response. Treatment data of several non-immune chemotherapy drugs were downloaded from the Genomics of Drug Sensitivity in Cancer database (https://www.cancerrxgene.org/), and the R package oncoPredict was employed to evaluate the drug half-maximal inhibitory concentration (IC50) values for each sample.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCell culture and reagents\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePC-9 normal NSCLC cells were obtained from the Cell Bank of the Chinese Academy of Science (Shanghai, China). PC-9 gefitinib-resistant cells (strain CTCC-ZHYC-0141) were purchased from Meisen Cell Technology (Zhejiang, China). All cells were maintained at 37 \u0026deg;C in a 5% CO\u003csub\u003e2\u003c/sub\u003e humidified atmosphere. PC-9 normal and gefitinib-resistant cells were cultured in RPMI-1640 medium supplemented with 10% fetal bovine serum and penicillin-streptomycin (100 U/ml and 100 mg/ml, respectively) (all from Thermo Fisher Scientific China, Shanghai, China). Before the start of this study, the resistance of the PC-9 resistant strain was confirmed using cell viability assays. The PC-9 gefitinib-resistant strain was maintained in medium supplemented with gefitinib at concentrations of 2.5 mM, 5 mM, and 10 mM for the first to third generations, respectively, and 10 mM from the fourth generation onward. Gefitinib (184475-35-2), AZD 7762 (1246094-78-9), and BI.2536 (755038-02-9) were purchased from Sigma-Aldrich China (Shanghai, China) and dissolved in dimethylsulfoxide at a stock concentration of 50 mM. All drugs were stored at \u0026ndash;20 \u0026deg;C in 5-mL aliquots.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRNA extraction and reverse transcription quantitative (RT-q)PCR\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eResistance gene expression in PC9 normal and gefitinib-resistant cells was assessed using RT-qPCR. The gene-specific qPCR primers are listed in Table S1. Total RNA was extracted from the cells using TRIzol reagent (Invitrogen Life Technologies) and reverse-transcribed into cDNA using a Takara kit (Takara Biotechnology). qPCR analysis was performed using SYBR Green master mix (Thermo Scientific).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCell proliferation assay\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCCK8 kits purchased from Sigma-Aldrich China (Shanghai, China) were used to detect the inhibitory effects of gefitinib, AZD7762, and BI.2536 on the viability of PC-9 normal and gefitinib-resistant cells. The cells were seeded in 96-well culture plates at 6 \u0026times; 10\u003csup\u003e3\u003c/sup\u003e cells/well, cultured for 24 h, and exposed to increasing concentrations of gefitinib, AZD7762, and BI.2536 for an additional 24 h. Ten microliters of CCK-8 solution was added to each well, and the plates were incubated for 2 h. The absorbance at 450 nm was measured using a microplate reader (Thermo Scientific).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStatistical analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData are presented as mean \u0026plusmn; SD or SEM and were analyzed using SPSS 26.0. Differences were considered significant at *\u003cem\u003ep\u003c/em\u003e \u0026lt; 0.05, **\u003cem\u003ep\u003c/em\u003e \u0026lt; 0.01, and ***\u003cem\u003ep\u003c/em\u003e \u0026lt; 0.001. The R packages used were mentioned above. Plots were generated using R Studio 4.3.3 or GraphPad Prism 8.0.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData collection and processing\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWhole-gene mRNA expression matrices, clinical data (phenotypes and overall survival), and single-nucleotide polymorphism mutation data of Cancer Genome Atlas (TCGA)-LUAD samples were downloaded from the UCSC website (https://xenabrowser.net/). In total, 513 tumor samples and 59 normal samples were obtained, and samples with missing information were excluded. The GSE117570 dataset, which comprises 10\u0026times; scRNA-seq data of 11,485 cells, was obtained from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/), and data on P1, P3, and P4 were selected for analysis. Expression data of signature genes in particular cells were downloaded from the Tumor Immune Single-cell Hub database (http://tisch.comp-genomics.org/home/). Two GEO datasets (GSE31210 and GSE13213) with complete mRNA expression and clinical information were obtained from the GEO database to construct the validation cohort. Genes upregulated in GR were obtained from the Gene Set Enrichment Analysis (GSEA) database (https://www.gsea-msigdb.org/gsea/index.jsp, \u0026ldquo;COLDREN_GEFITINIB_RESISTANCE_UP\u0026rdquo; dataset). \u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003escRNA-seq data collection and analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe R package Seurat was used to convert raw scRNA-seq data into Seurat objects, followed by quality control and the exclusion of unqualified cells. The gene expression in core cells was standardized using a linear regression algorithm, and the top 2,000 highly variable genes were identified using analysis of variance. Uniform manifold approximation and projection was used to identify cell subpopulations. The FindAllMarkers function in the R package Seurat and the COSG R package were employed to identify marker genes for all cell clusters and differentially expressed genes among the screened marker genes. The R package clusterProfiler was used for Kyoto Encyclopedia of Genes and Genomes (KEGG) and Reactome enrichment analyses of top 100 marker genes in each cell cluster. A single-sample GSEA score was calculated for each cell cluster based on the expression levels of GR-related genes, and all cell samples were divided into high- and low-score groups based on the median score. Data of 50 hallmark gene sets were downloaded from the GSEA database and the R package GSVA was employed to calculate hallmark scores for each cell cluster.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eIdentification and validation of a prognostic signature\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eUnivariate Cox proportional risk regression analysis was conducted on GR-related genes in the TCGA-LUAD training cohort, using \u003cem\u003ep \u003c/em\u003e\u0026lt; 0.05 as the screening criterion. Ten classical algorithms (single or combined), including random forest, least absolute shrinkage and selection operator, gradient boosting machine, survival support vector machine, supervised principal components, ridge regression, partial least squares regression for Cox, CoxBoost, Stepwise Cox, and elastic network, were used to construct a prognostic signature based on the screened genes, and the C-index was calculated for all signatures to select the most suitable one. The R packages survival, pROC, and ggplot2 were used for survival analysis and time-dependent receiver operating characteristic (ROC) analysis using the TCGA-LUAD training cohort, GSE13213 and GSE31210 validation cohorts, and their combination.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePrognostic analysis of risk groups and clinical phenotypes\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe proportions of clinical phenotype (sex, T stage, N stage, M stage, stage, recurrence, metastasis, and overall survival) subcategories in the high- and low-risk groups were calculated, and resistance scores were systematically compared among the subcategories. Univariate and multivariate Cox regression analyses were used to assess the influence of the prognostic signature and other common clinical phenotypic features on the prognosis of patients with LUAD.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunctional enrichment analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCorrelation coefficients between the resistance score and all genes were calculated, and the 100 genes with the highest positive and negative correlations were screened out. The R package clusterProfiler was used for Gene Ontology (GO), KEGG, and Reactome functional enrichment analyses of the screened genes. After calculating the score for each sample based on the constructed signature, all samples were divided into high- and low-risk groups, and GSEA was applied to the differentially expressed genes between the two groups. The cancer-related hallmark gene set file (h.all.v2023.2.Hs.symbols.gmt) was downloaded from the Molecular Signatures Database (https://www.gsea-msigdb.org/gsea/msigdb). The R package GSEA (v1.38.2) was used to calculate the normalized enrichment score and false discovery rate, and the R package ggplot2 (v3.3.6) was used to display the GSEA results.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAnalysis on tumor immune microenvironment (TIME) and tumor stemness\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eLUAD samples were ordered according to stemness indices (mRNAsi and mDNAsi) and their overall survival and distribution of high- and low-risk groups were shown. Pearson correlation coefficients between stemness indices and resistance scores were calculated. Immune-cell infiltration was analyzed using the MCPCOUNTER, EPIC, XCELL, CIBERSORT, IPS, QUANTISEQ, ESTIMATE, and TIMER algorithms in the TIMER database (https://cistrome.shinyapps.io/timer/), and the infiltration scores of all types of immune cells were compared between the high- and low-risk groups using the R package IOBR. Multiple immune microenvironment-related indices, including the microenvironment, ESTIMATE, immune, and stroma scores, were calculated and compared between the high- and low-risk groups. The expression levels of various chemokines and their receptors were compared between the high- and low-risk groups, and Pearson correlation coefficients between the expression levels of significantly differentially expressed chemokines and the resistance scores were calculated.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAnalysis of drug efficacy prediction in gefitinib-resistant populations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData of patients who received immunotherapy was downloaded from the Tumor Immune Dysfunction and Exclusion (TIDE) database (http://tide.dfci.harvard.edu), and true or false response to immunotherapy and TIDE and exclusion scores were compared between the high- and low-risk groups. GSE135222, an anti-programmed cell death protein 1 (PD-1)/PD-1 ligand (PD-L1) immunotherapy dataset from GEO, was used to evaluate the relationship between the GR signature and LUAD immunotherapy response. Treatment data of several non-immune chemotherapy drugs were downloaded from the Genomics of Drug Sensitivity in Cancer database (https://www.cancerrxgene.org/), and the R package oncoPredict was employed to evaluate the drug half-maximal inhibitory concentration (IC50) values for each sample.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCell culture and reagents\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePC-9 normal NSCLC cells were obtained from the Cell Bank of the Chinese Academy of Science (Shanghai, China). PC-9 gefitinib-resistant cells (strain CTCC-ZHYC-0141) were purchased from Meisen Cell Technology (Zhejiang, China). All cells were maintained at 37 \u0026deg;C in a 5% CO\u003csub\u003e2\u003c/sub\u003e humidified atmosphere. PC-9 normal and gefitinib-resistant cells were cultured in RPMI-1640 medium supplemented with 10% fetal bovine serum and penicillin-streptomycin (100 U/ml and 100 mg/ml, respectively) (all from Thermo Fisher Scientific China, Shanghai, China). Before the start of this study, the resistance of the PC-9 resistant strain was confirmed using cell viability assays. The PC-9 gefitinib-resistant strain was maintained in medium supplemented with gefitinib at concentrations of 2.5 mM, 5 mM, and 10 mM for the first to third generations, respectively, and 10 mM from the fourth generation onward. Gefitinib (184475-35-2), AZD 7762 (1246094-78-9), and BI.2536 (755038-02-9) were purchased from Sigma-Aldrich China (Shanghai, China) and dissolved in dimethylsulfoxide at a stock concentration of 50 mM. All drugs were stored at \u0026ndash;20 \u0026deg;C in 5-mL aliquots.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRNA extraction and reverse transcription quantitative (RT-q)PCR\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eResistance gene expression in PC9 normal and gefitinib-resistant cells was assessed using RT-qPCR. The gene-specific qPCR primers are listed in Table S1. Total RNA was extracted from the cells using TRIzol reagent (Invitrogen Life Technologies) and reverse-transcribed into cDNA using a Takara kit (Takara Biotechnology). qPCR analysis was performed using SYBR Green master mix (Thermo Scientific).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCell proliferation assay\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCCK8 kits purchased from Sigma-Aldrich China (Shanghai, China) were used to detect the inhibitory effects of gefitinib, AZD7762, and BI.2536 on the viability of PC-9 normal and gefitinib-resistant cells. The cells were seeded in 96-well culture plates at 6 \u0026times; 10\u003csup\u003e3\u003c/sup\u003e cells/well, cultured for 24 h, and exposed to increasing concentrations of gefitinib, AZD7762, and BI.2536 for an additional 24 h. Ten microliters of CCK-8 solution was added to each well, and the plates were incubated for 2 h. The absorbance at 450 nm was measured using a microplate reader (Thermo Scientific).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStatistical analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData are presented as mean \u0026plusmn; SD or SEM and were analyzed using SPSS 26.0. Differences were considered significant at *\u003cem\u003ep\u003c/em\u003e \u0026lt; 0.05, **\u003cem\u003ep\u003c/em\u003e \u0026lt; 0.01, and ***\u003cem\u003ep\u003c/em\u003e \u0026lt; 0.001. The R packages used were mentioned above. Plots were generated using R Studio 4.3.3 or GraphPad Prism 8.0.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003eAnalysis of LUAD scRNA-seq data\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSingle-cell data of LUAD samples comprising 11,485 cells were obtained from GSE117570, and samples P1, P3,\u0026nbsp;and P4 were selected for analysis. The R package\u0026nbsp;SingleR was used for cell-cluster dimension reduction, visualization, and annotation, identifying 11 major cell subgroups, including B cells, T helper 2 (Th2) cells, CD8\u003csup\u003e+\u0026nbsp;\u003c/sup\u003eT effector cells, plasmacytoid dendritic cells, endothelial cells, malignant cells, M1 macrophages, M2 macrophages, monocytes, natural killer (NK) cells, and plasma cells (Fig. 1A). The top five highest and lowest differentially expressed genes in each cell subgroup are shown in Fig. 1B. Fig. 1C shows the top 10 expressed marker genes in each cell subgroup identified using the R package COSG. To intuitively visualize the biological processes the core cell types were involved in, we conducted KEGG and Reactome pathway functional enrichment analyses of the top 100 marker genes (Fig. 1D, E). Notably,\u0026nbsp;the marker genes in B, CD8\u003csup\u003e+\u0026nbsp;\u003c/sup\u003eT effector, and NK cells were significantly enriched in\u0026nbsp;chemokine signaling (Fig. 1D), whereas Th2 cells were strongly associated with PD-L1 expression and the PD-1 checkpoint pathway in cancer (Fig. 1D). Marker genes in Th2 cells, plasmacytoid dendritic cells, M1 macrophages, and monocytes were significantly enriched in interleukin signaling (Fig. 1E).\u003c/p\u003e\n\u003cp\u003eThe expression levels of\u0026nbsp;GR-related genes in each cell cluster are shown in bubble plots in Fig. 2A. The\u0026nbsp;R package GSVA was used to score each cell cluster based on GR-related gene expression, and the clusters were divided into two groups with high and low resistance scores using the median as a threshold (Fig. 2B). Fig. 2C displays the distribution of the cell clusters,\u0026nbsp;resistance score values, and the distribution of cell clusters in the high- and low-score groups. The proportions of all cell clusters in the\u0026nbsp;high- and low-score groups are shown in a bar chart in Fig. 2D. The proportions\u0026nbsp;of all cell clusters in samples P1, P3, and P4\u0026nbsp;are shown in Supplementary Fig. 1A. Supplementary Fig. 1B shows the distributions of cell populations in the\u0026nbsp;high- and low-drug resistance score groups separately, and the line chart in\u0026nbsp;Supplementary Fig. 1C\u0026nbsp;more intuitively displays the differences in the proportions of the clusters between the two groups. The proportions of M2 macrophages and malignant cells were significantly higher in the high-resistance score group than in the low-resistance score group, whereas the number of Th2 cells was the highest in the low-score group. These results indicated that a\u0026nbsp;high resistance score may predict a tumor microenvironment more favorable for tumor growth, which is associated with\u0026nbsp;poor prognosis, suggesting the feasibility of using GR-related genes for predicting the prognosis of LUAD.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eIdentification of candidate genes and GR signature establishment\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe obtained 71 GR-related genes with expression profiles from merged TCGA and GEO data. Univariate Cox analysis identified 22 prognostically significant\u0026nbsp;genes (\u003cem\u003ep\u003c/em\u003e \u0026lt; 0.05): \u003cem\u003eSLC47A1\u003c/em\u003e, \u003cem\u003eUBE2M\u003c/em\u003e, \u003cem\u003eOLFM1\u003c/em\u003e, \u003cem\u003eSLC7A1\u003c/em\u003e, \u003cem\u003eTRAF5\u003c/em\u003e, \u003cem\u003eTIMM50\u003c/em\u003e, \u003cem\u003ePPIF\u003c/em\u003e, \u003cem\u003eTMEM158\u003c/em\u003e, \u003cem\u003ePRELID1\u003c/em\u003e, \u003cem\u003eCDH2\u003c/em\u003e, \u003cem\u003eIKBIP\u003c/em\u003e, \u003cem\u003ePPP2R1B\u003c/em\u003e, \u003cem\u003eTUB\u003c/em\u003e, \u003cem\u003eCEBPG\u003c/em\u003e, \u003cem\u003eUCHL1\u003c/em\u003e, \u003cem\u003eNFKBIZ\u003c/em\u003e, \u003cem\u003eMDM4\u003c/em\u003e, \u003cem\u003eUBA2\u003c/em\u003e, \u003cem\u003eMAP1B\u003c/em\u003e, \u003cem\u003eLIX1L\u003c/em\u003e, \u003cem\u003eALDH1B1\u003c/em\u003e, and \u003cem\u003eNAA10\u003c/em\u003e. Of these, 15 genes were negatively correlated with prognosis, and seven were positively correlated. The hazard ratio and \u003cem\u003ep\u003c/em\u003e-value of each gene are shown in a forest plot in Fig. 3A. The expression levels of these 22 genes were generally higher in PC-9 gefitinib-resistant LUAD cells than in normal PC-9 cells (Fig. 3B).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo construct the prognostic prediction model, we employed 10 traditional algorithms for precision calculation, using TCGA data as the training set and GSE31210, GSE13213, and merged data as the validation sets. By comparing the algorithms and more than 100 algorithm combinations, we found that the random forest algorithm (mean C-index, 0.758) had the best fit (Fig. 3C). Subsequently, we constructed a GR signature scoring model based on the weights and coefficients for the 22 GR genes provided by the random forest algorithm (Fig. 3D). Among the genes screened by the random forest algorithm, the expression of the top 10 most important genes in multiple cells in multiple datasets is shown in a heatmap in Supplementary Fig. 2. The genes were highly expressed in different cell types, suggesting that the GR signature may affect the prognosis of patients with LUAD via multiple pathways.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eInternal and external validation of the GR signature using TCGA, GEO, and merged cohorts\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe first\u0026nbsp;validated GR signature performance\u0026nbsp;in the TCGA cohort. Based on the gene expression weighting algorithm mentioned above, we calculated GR-related scores for all samples, which were then divided into high-\u0026nbsp;and low-resistance score groups\u0026nbsp;based on the median score. As shown in Fig. 4A, the survival time of patients in the high-score group was significantly shorter than that in the low-score group (\u003cem\u003ep\u003c/em\u003e \u0026lt;\u0026nbsp;0.0001), and the time-dependent ROC curve showed that the GR signature maintained high precision in the TCGA cohort for 7 years (Fig. 4A). Subsequently, two suitable LUAD-GSE datasets (GSE13213 and GSE31210) were selected from the GEO database for external validation of signature reliability. The GR scores of all\u0026nbsp;samples were calculated\u0026nbsp;using the same algorithm, and the samples were divided into high- and low-score groups. In both external cohorts, survival in the high-score group was significantly worse than that in the low-score group (Fig. 4B, C, \u003cem\u003ep\u0026nbsp;\u003c/em\u003e= 0.0036 and\u003cem\u003e\u0026nbsp;p\u0026nbsp;\u003c/em\u003e= 0.0019), and the accuracy of the GR signature\u0026nbsp;remained high for 5 years, as shown by the time-dependent ROC curve (Fig. 4B, C). In a merged\u0026nbsp;cohort\u0026nbsp;of the TCGA and two GEO cohorts, the GR signature was reliable for prognosis prediction (Fig. 4D). Thus, the GR signature was demonstrated to be reliable in its ability to predict survival in both internal and external cohorts.\u003c/p\u003e\n\u003cp\u003eNext, we collected clinical phenotypic data for all samples in the TCGA cohort and explored their correlation with the GR score. The data showed that the higher the T stage (Fig. 5A), N stage (Fig. 5B), M stage (Fig. 5C), and grade (Fig. 5D), the higher\u0026nbsp;the GR score (all\u0026nbsp;\u003cem\u003ep\u003c/em\u003e \u0026lt; 0.05).\u0026nbsp;GR scores were also significantly higher in patients with recurrence or metastasis than in those without (Fig. 5E). Among the patients with complete followup data, the scores were significantly higher among deceased patients than among the survivors (Fig. 5F, \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.0001). Collectively, these results demonstrated the accuracy and reliability of the GR signature in predicting clinical phenotypes and patient outcomes. Interestingly, the mean GR score was higher in men than that in women (Fig. 5G, \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.001), suggesting\u0026nbsp;that the GR signature may be related to sex hormones or other genes; however, further studies are needed to verify this. Univariate (Supplementary Fig. 3A) and multivariate (Supplementary Fig. 3B) Cox regression analyses of the resistance score\u0026nbsp;and clinical characteristics, including age, sex, TNM stage, recurrence, and metastasis, revealed that the resistance score\u0026nbsp;performed well in both analyses and served as an independent prognostic factor, which further confirmed the reliability of the\u0026nbsp;GR signature.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunctional enrichment and TIME analysis of GR signature\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo confirm the prognostic value of the\u0026nbsp;GR signature in patients with LUAD\u0026nbsp;from another perspective, we subjected the 100 genes with the highest positive (Fig. 6A) and negative (Fig. 6B) correlations\u0026nbsp;with the\u0026nbsp;GR signature\u0026nbsp;resistance score\u0026nbsp;to GO, KEGG,\u0026nbsp;and\u0026nbsp;Rectome functional enrichment analyses. The top 20 enriched pathways for each analysis (Fig. 6C\u0026ndash;E) revealed that the\u0026nbsp;GR signature was closely associated with cell cycle (particularly, chromosomal processes) and DNA replication, suggesting that\u0026nbsp;the signature genes may affect the proliferation and survival of LUAD cells by interfering with mitotic processes, thus affecting patient prognosis.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eBased on epigenetic and transcriptome data, we calculated the mDNA stemness index (si) and mRNAsi for each sample and assessed their correlation with overall survival and risk (Supplementary Fig. 4A, B). Pearson correlation analysis revealed a positive correlation between the GR score and the mDNAsi (Supplementary Fig. 4A, r = 0.11, \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.05) and mRNAsi (Supplementary Fig. 4B, r = 0.35, \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.001) in the TCGA cohort. By applying GSEA to the high- and low-risk groups, we found that in many cancers, particularly LUAD, high GR scores were positively correlated with some cancer-promoting pathways, such as MYC targets V1/2, mTORC1 signaling, and epithelial\u0026ndash;mesenchymal transition (Supplementary Fig. 5), and negatively correlated with some immune-related pathways, including the interferon response, inflammatory response, and IL6-JAK-STAT3 signaling (Supplementary Fig. 5). This suggested a correlation between a high GR score and tumor progression and corroborated the reliability of the GR signature in prognosis prediction.\u003c/p\u003e\n\u003cp\u003eUsing the integrated TCGA and GSE datasets (GSE13212 and GSE31210) and the R package IOBR, we assessed immune-cell infiltration in all samples. The low-risk group had a significantly higher degree of critical immune-cell infiltration than the high-risk group\u0026nbsp;as assessed by multiple methods, including T cells in\u0026nbsp;MCP-counter; B and CD4\u003csup\u003e+\u0026nbsp;\u003c/sup\u003eT cells in EPIC; CD8\u003csup\u003e+\u0026nbsp;\u003c/sup\u003eT and dendritic cells in xCell; memory B and activated NK cells in CIBERSORT; B, NK and CD4\u003csup\u003e+\u0026nbsp;\u003c/sup\u003eT cells in quanTiseq; and\u0026nbsp;B, CD4\u003csup\u003e+\u0026nbsp;\u003c/sup\u003eT, and dendritic cells in TIMER (Fig. 7A). Based on the expression levels of related meta-genes, we calculated the microenvironment, ESTIMATE, immune, stromal, and other immune microenvironment-related scores for each sample and compared them between the two risk groups (Fig. 7B). In both xCell and ESTIMATE, the microenvironment, ESTIMATE, and immune scores of the low-risk group were significantly higher than those of high-risk group, whereas in xCell, the epithelial score of the high-risk group was significantly higher than that of the low-risk group. The correlations between the GR score and various chemokines and their receptors are shown in a heatmap in Supplementary Fig. 6A. TGFBR2, TGFBR3, CCR2, CCR6, CX3CR1, and IL33 were the most strongly correlated with the resistance score (Supplementary Fig. 6B). Together, these results revealed a strong relationship between the GR signature-related genes and the immune microenvironment, which may affect tumor occurrence and development by affecting key immune-cell infiltration into the tumor microenvironment or interfering with chemokine/receptor expression and binding, ultimately affecting the prognosis of patients with LUAD.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eExploration of the impact of the GR signature on immune and non-immune therapies\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo evaluate the effect of the\u0026nbsp;GR signature on the treatment of patients with LUAD, we compared the treatment effects of\u0026nbsp;immune and non-immune drugs\u0026nbsp;in the high-\u0026nbsp;and low-risk groups. We obtained TIDE data for the merged cohort (Fig. 8A). The immunotherapy response rate of the low-risk group was evidently higher than that of the high-risk group (Fig. 8B). The TIDE (Fig. 8C) and exclusion\u0026nbsp;(Fig. 8D)\u0026nbsp;scores\u0026nbsp;were also significantly higher in the high-risk group than in the low-risk group, indicating that the immune escape potential is greater and the efficacy of immune checkpoint inhibition is worse in the high-risk group.\u0026nbsp;By comparing samples from the immunotherapy-responsive and non-responsive groups, we found that the GR score was significantly higher in the non-responsive group than in the responsive group (Fig. 8E). GSE135222, an anti-PD-1/PD-L1 immunotherapy dataset, was used to evaluate the relationship between the GR signature and LUAD immunotherapy. Survival curves indicated that a higher GR score\u0026nbsp;was associated with shorter survival after immunotherapy (Fig. 8F). However, because of the small sample size, the survival curve analysis was not statistically\u0026nbsp;significant. Similarly, the immunotherapy response rate was substantially higher in the low GR-score group than in the high GR-score group (Fig. 8G). These results suggested that patients in the high-risk group are more likely to have a worse response to immunotherapy, which is consistent with the above finding that these patients are more likely to be immunosuppressed.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp; \u0026nbsp; In addition to immunotherapy, IC50 values of multiple chemotherapy agents for all samples were analyzed using the R package oncoPredict. The low-risk group was more sensitive to several agents, including doramapimod and axitinib (Fig. 9A\u0026ndash;F). Notably, while patients in the high-risk group were more immunosuppressed and did not have an immunotherapeutic advantage, they were more sensitive to certain chemotherapeutic agents, including docetaxel and erlotinib (Fig. 9G\u0026ndash;L). These results suggested that these chemotherapeutic agents may have a good effect in high-risk patients who do not respond well to gefitinib. Two predicted agents, AZD7762 (Fig. 9M) and BI.2536 (Fig. 9N), inhibited the viability of gefitinib-resistant PC-9 cells significantly more strongly than did gefitinib. This finding provides new therapeutic directions for gefitinib-resistant patients and confirms that the GR signature provides reliable guidance for the treatment of patients with LUAD.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eLung cancer remains the leading cause of cancer mortality worldwide, and LUAD accounts for a considerable\u0026nbsp;portion of cases\u0026nbsp;(Zheng et al. 2024). As a first-generation EGFR-TKI, gefitinib is widely used in the treatment of patients with clinically advanced NSCLC\u0026nbsp;(Cappuzzo et al. 2016). However, due to congenital and acquired drug resistance, its efficacy is greatly reduced\u0026nbsp;(L\u0026apos;Hostis et al. 2023). Although new EGFR-TKIs have been successfully developed and applied in clinical practice in recent years,\u0026nbsp;suitable targeted drugs are still lacking for\u0026nbsp;many patients with NSCLC\u0026nbsp;(Tan et al. 2024;Remon et al. 2024). Therefore, new treatment strategies based on knowledge of the drug resistance mechanisms are urgently needed. We screened genes that play an important role in LUAD and explored their roles in biological processes\u0026nbsp;and their effects on immune and non-immune therapies\u0026nbsp;so as to unravel the drug resistance mechanism and provide guidance for the prediction and treatment of GR in patients with LUAD.\u003c/p\u003e\n\u003cp\u003eWe identified 11 cell subtypes from the GSE117570 dataset\u0026nbsp;(Song et al. 2019)\u0026nbsp;and analyzed the expression of upregulated GR-related gene sets across these cell subtypes in LUAD. Subsequently, we calculated an expression-based resistance score for each cell type and divided the cell subpopulations into gefitinib-resistant and -sensitive groups. We found that the proportions\u0026nbsp;of M2 macrophages and malignant cells were significantly higher\u0026nbsp;in the high\u0026nbsp;GR-score group than in the low GR-score group. M2 macrophages in the tumor microenvironment can secrete anti-inflammatory cytokines and chemokines\u0026nbsp;that\u0026nbsp;promote tumor growth, invasion, and metastasis\u0026nbsp;(Weinhauser et al. 2023;Bied et al. 2023). Further, M2 macrophages play an important role in tumor chemoresistance\u0026nbsp;(Jiang et al. 2023), and targeting M2 macrophages is a potential therapeutic approach to overcome antitumor-drug resistance\u0026nbsp;(Wang et al. 2024).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe further screened the GR-related gene set at the transcriptome level to narrow down the number of genes. Using univariate and multivariate Cox regression analyses,\u0026nbsp;22 genes with the highest influence on the survival of patients with LUAD were selected.\u0026nbsp;Notably,\u0026nbsp;seven of these 22 genes had a\u0026nbsp;hazard ratio\u0026nbsp;\u0026lt;1 (Fig. 3A), implying that their expression is positively correlated with the survival time of patients\u0026nbsp;with LUAD. Notably, genes that promote GR do not necessarily coincide with oncogenic genes in LUAD. These genes are involved in multiple types and levels of complex biological processes to promote GR, and it cannot be excluded that some of these processes may indirectly inhibit tumor progression. One of the most prognostically significant genes was \u003cem\u003eSLC47A1\u003c/em\u003e, which belongs to the multidrug and toxin extrusion family and is an organic cation antiporter in the plasma membrane\u0026nbsp;(Lin et al. 2022). \u003cem\u003eSLC47A1\u003c/em\u003e is downregulated in LUAD tissues compared to\u0026nbsp;normal tissues\u0026nbsp;(HUANG et al. 2023)\u0026nbsp;and acts as a tumor-suppressor gene. While \u003cem\u003eSLC47A1\u003c/em\u003e is a GR-associated gene and appears to negatively impact the prognostic outcome of LUAD, targeting this gene enhanced the efficacy of temozolomide in glioma stem cells\u0026nbsp;(Batara et al. 2023).\u003c/p\u003e\n\u003cp\u003eIn the validation analysis, most of the 22 genes showed higher expression in gefitinib-resistant PC-9 cells than in normal PC-9 cells (Fig. 3B, C); a few genes showed no statistically significant difference and\u0026nbsp;three genes showed an opposite trend (lower expression in resistant cells). Such inconsistent results may be attributed to heterogeneity between cell lines and human LUAD tissues.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe used 10 classical algorithms and more than 100 algorithm combinations to select the most suitable algorithm for prognostic prediction model construction. We thus\u0026nbsp;obtained a random forest algorithm with the best fit. In previous studies, often, only one to three algorithms were assessed to build a prediction model\u0026nbsp;(Huang et al. 2019), and the screening processes involved many complex procedures, which led to\u0026nbsp;a high workload and the selected algorithm sometimes being suboptimal. In recent years, algorithms have been iteratively updated, and\u0026nbsp;more than 10 mature algorithms are currently\u0026nbsp;available\u0026nbsp;(Zhang et al. 2024). Machine learning techniques, as an emerging technology in recent years, provide\u0026nbsp;effective tools for the screening of algorithms. They not only have enhanced computing power and reduce the workload, but also allow selecting the most suitable one among various algorithms to build a model, which is a great progress. We applied only the 10 most commonly used algorithms, and algorithms more suitable than the random forest model for this signature may become available in future. Future studies should include new algorithms to update the signature constructed in this study.\u0026nbsp;The reliability of our signature was successfully confirmed by survival and clinical-feature validation analyses in internal and external datasets.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eDuring the course of gefitinib treatment, dynamic alterations within the TIME have been detected, indicating a strong association with the\u0026nbsp;GR pathway\u0026nbsp;(Duan et al. 2020). Accordingly, we assessed immune-cell infiltration in the high- and low-risk groups. We found that the microenvironment, ESTIMATE, and immune scores\u0026nbsp;were significantly higher in the low-risk group than in the high-risk group. A\u0026nbsp;comprehensive pan-cancer multi-omics study identified diverse immune checkpoints and lymphocyte-depleted features specific to EGFR-positive NSCLC\u0026nbsp;(Seo et al. 2018), suggesting a natural resistance to immunotherapy in such cancers. Although the suboptimal efficacy of immunotherapy in patients with EGFR-positive tumors has been confirmed in clinical studies\u0026nbsp;(Yang et al. 2019), chemo-immunotherapy combinations have shown promising results in patients with acquired GR and without specific targets\u0026nbsp;(Hayashi et al. 2022).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFinally, we aimed to provide valuable treatment guidance for patients with GR. Large-scale data collection and analysis of immune\u0026nbsp;and non-immune therapies\u0026nbsp;confirmed that the signature we constructed\u0026nbsp;predicted immunotherapy efficacy to a certain degree. Further, we identified some non-immunotherapeutic drugs that may\u0026nbsp;have good efficacy in gefitinib-resistant patients, providing potential new treatment methods for these patients. Drug-prediction and cellular experiments indicated that AZD7762 and BI.2536 have good efficacy against drug-resistant cells, which may offer hope for patients with GR in the future. Unfortunately, most of these drugs are still in the research stage, and further studies are needed to confirm their efficacy in gefitinib-resistant patients and their application\u0026nbsp;in clinical practice.\u003c/p\u003e\n\u003cp\u003eThis study had several limitations. We obtained single-cell data of LUAD from public databases and focused on the expression of gene sets highly expressed in GR in various cell types and the main biological processes involved. Single-cell research has become popular in recent years, and the data that can be explored are very extensive. Nevertheless, our exploration was preliminary, and further analysis may reveal more information on GR gene sets. In addition, as all data were derived from public databases, the sample conditions were determined, limiting the scope of the study. Future studies may address this issue by using single-cell data from purposefully collected samples. Finally, we only selected gefitinib-resistant and normal PC-9 LUAD cells for the experiments and did not verify these genes in other cell types. Data of a single cell line cannot completely represent the data of all LUAD cells in the database. Therefore, further experiments need to be carried out in other cell lines to confirm the results.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eUsing single-cell and transcriptome data, we established a 22 GR gene signature that performed well in survival prediction in multiple LUAD datasets and may serve as an independent clinical prognostic factor. Enrichment and TIME analyses demonstrated that the signature genes significantly affected chromosomal processes and DNA replication, important immune-cell infiltration, and immune scores. Treatment prediction demonstrated that the signature predicted immunotherapy efficacy in patients with LUAD\u0026nbsp;to a certain extent\u0026nbsp;and identified several chemotherapy agents to which gefitinib-resistant patients may respond well. Clinical verification of the prognostic signature is required in future.\u003c/p\u003e\n"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe authors thank the researchers who uploaded sequencing data to the public databases and everyone who helped in this study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDong Zhou\u003c/strong\u003e: Methodology, Writing\u0026ndash;original draft. \u003cstrong\u003eZhi Zheng\u003c/strong\u003e: Data curation, Writing\u0026ndash;original draft. \u003cstrong\u003eYan-qi Li\u003c/strong\u003e: Data curation, Methodology. \u003cstrong\u003eQuan-Xing Liu\u003c/strong\u003e: Data curation. \u003cstrong\u003eXu-Feng Deng\u003c/strong\u003e: Data curation.\u003cstrong\u003e\u0026nbsp;Liang Chen\u003c/strong\u003e: Data curation; Formal analysis. \u003cstrong\u003eMan-Yuan Li\u003c/strong\u003e: Investigation. \u003cstrong\u003eJiao Zhang\u003c/strong\u003e: Data curation. \u003cstrong\u003eXiao Lu\u003c/strong\u003e: Conceptualization.\u003cstrong\u003e\u0026nbsp;Hong Zheng:\u003c/strong\u003e Conceptualization, Writing\u0026ndash;review \u0026amp; editing. \u003cstrong\u003eJi-Gang Dai\u003c/strong\u003e: Methodology, Supervision, Writing\u0026ndash;review\u0026nbsp;\u0026amp; editing.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was supported by grants from the National Natural Science Foundation of China for Dai (No. 81972190),\u0026nbsp;Key Projects on Technological Innovation and Application Development of Chongqing (2022-195), and the Science and Technology Commission and the Chongqing Health Commission Joint Medical Research Program\u0026nbsp;of\u0026nbsp;Chongqing\u0026nbsp;(2024MSXM090).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets generated and analyzed in the current study are available in the TCGA (https://portal.gdc.cancer.gov/) and GEO (https://www.ncbi.nlm.nih/) databases.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eBatara DC, Park SW, Kim HJ, Choi SY, Ohn T, Choi MC, Park SI, Kim SH (2023) Targeting the multidrug and toxin extrusion 1 gene (SLC47A1) sensitizes glioma stem cells to temozolomide. Am J Cancer Res 13(9):4021-4038\u003c/li\u003e\n\u003cli\u003eBied M, Ho WW, Ginhoux F, Bleriot C (2023) Roles of macrophages in tumor development: a spatiotemporal perspective. Cell Mol Immunol 20(9):983-992. 10.1038/s41423-023-01061-6\u003c/li\u003e\n\u003cli\u003eCappuzzo F, Morabito A, Normanno N, Bidoli P, Del CA, Giannetta L, Montanino A, Mazzoni F, Buosi R, Burgio MA, Cerea G, Chiari R, Cortinovis D, Finocchiaro G, Foltran L, Migliorino MR, Tiseo M, Ferrari S, De Marinis F (2016) Efficacy and safety of rechallenge treatment with gefitinib in patients with advanced non-small cell lung cancer. Lung Cancer 99:31-37. 10.1016/j.lungcan.2016.06.008\u003c/li\u003e\n\u003cli\u003eChoi HY, Chang JE (2023) Targeted Therapy for Cancers: From Ongoing Clinical Trials to FDA-Approved Drugs. Int J Mol Sci 24(17). 10.3390/ijms241713618\u003c/li\u003e\n\u003cli\u003eDuan Q, Zhang H, Zheng J, Zhang L (2020) Turning Cold into Hot: Firing up the Tumor Microenvironment. Trends Cancer 6(7):605-618. 10.1016/j.trecan.2020.02.022\u003c/li\u003e\n\u003cli\u003eFan J, Fang L, Cong S, Zhang Y, Jiang X, Wang N, Chen Y (2024) Potential pre-COPD indicators in association with COPD development and COPD prediction models in Chinese: a prospective cohort study. Lancet Reg Health West Pac 44:100984. 10.1016/j.lanwpc.2023.100984\u003c/li\u003e\n\u003cli\u003eGe Z, Xu M, Ge Y, Huang G, Chen D, Ye X, Xiao Y, Zhu H, Yin R, Shen H, Ma G, Qi L, Wei G, Li D, Wei S, Zhu M, Ma H, Shi Z, Wang X, Ge X, Qian X (2023) Inhibiting G6PD by quercetin promotes degradation of EGFR T790M mutation. Cell Rep 42(11):113417. 10.1016/j.celrep.2023.113417\u003c/li\u003e\n\u003cli\u003eHaratake N, Ozawa H, Morimoto Y, Yamashita N, Daimon T, Bhattacharya A, Wang K, Nakashoji A, Isozaki H, Shimokawa M, Kikutake C, Suyama M, Hashinokuchi A, Takada K, Takenaka T, Yoshizumi T, Mitsudomi T, Hata AN, Kufe D (2024) MUC1-C Is a Common Driver of Acquired Osimertinib Resistance in NSCLC. J Thorac Oncol 19(3):434-450. 10.1016/j.jtho.2023.10.017\u003c/li\u003e\n\u003cli\u003eHayashi H, Sugawara S, Fukuda Y, Fujimoto D, Miura S, Ota K, Ozawa Y, Hara S, Tanizaki J, Azuma K, Omori S, Tachihara M, Nishino K, Bessho A, Chiba Y, Haratani K, Sakai K, Nishio K, Yamamoto N, Nakagawa K (2022) A Randomized Phase II Study Comparing Nivolumab with Carboplatin-Pemetrexed for EGFR-Mutated NSCLC with Resistance to EGFR Tyrosine Kinase Inhibitors (WJOG8515L). Clin Cancer Res 28(5):893-902. 10.1158/1078-0432.CCR-21-3194\u003c/li\u003e\n\u003cli\u003eHUANG N, LIU Z, LI R (2023) Pan-cancer analysis reveals the relationship between SLC47A1 immune infiltration and clinical prognosis of human tumors. Journal of Holistic Integrative Pharmacy 4(1):103-118. 10.1016/S2707-3688(23)00098-5\u003c/li\u003e\n\u003cli\u003eHuang P, Lin CT, Li Y, Tammemagi MC, Brock MV, Atkar-Khattra S, Xu Y, Hu P, Mayo JR, Schmidt H, Gingras M, Pasian S, Stewart L, Tsai S, Seely JM, Manos D, Burrowes P, Bhatia R, Tsao MS, Lam S (2019) Prediction of lung cancer risk at follow-up screening with low-dose CT: a training and validation study of a deep learning method. Lancet Digit Health 1(7):e353-e362. 10.1016/S2589-7500(19)30159-1\u003c/li\u003e\n\u003cli\u003eJiang S, Deng T, Cheng H, Liu W, Shi D, Yuan J, He Z, Wang W, Chen B, Ma L, Zhang X, Gong P (2023) Macrophage-organoid co-culture model for identifying treatment strategies against macrophage-related gemcitabine resistance. J Exp Clin Cancer Res 42(1):199. 10.1186/s13046-023-02756-4\u003c/li\u003e\n\u003cli\u003eJohnson M, Garassino MC, Mok T, Mitsudomi T (2022) Treatment strategies and outcomes for patients with EGFR-mutant non-small cell lung cancer resistant to EGFR tyrosine kinase inhibitors: Focus on novel therapies. Lung Cancer 170:41-51. 10.1016/j.lungcan.2022.05.011\u003c/li\u003e\n\u003cli\u003eL\u0026apos;Hostis A, Palgen JL, Perrillat-Mercerot A, Peyronnet E, Jacob E, Bosley J, Duruisseaux M, Toueg R, Lefevre L, Kahoul R, Ceres N, Monteiro C (2023) Knowledge-based mechanistic modeling accurately predicts disease progression with gefitinib in EGFR-mutant lung adenocarcinoma. NPJ Syst Biol Appl 9(1):37. 10.1038/s41540-023-00292-7\u003c/li\u003e\n\u003cli\u003eLin Z, Liu J, Long F, Kang R, Kroemer G, Tang D, Yang M (2022) The lipid flippase SLC47A1 blocks metabolic vulnerability to ferroptosis. Nat Commun 13(1):7965. 10.1038/s41467-022-35707-2\u003c/li\u003e\n\u003cli\u003eLu C, Wei XW, Wang Z, Zhou Z, Liu YT, Zheng D, He Y, Xie ZH, Li Y, Zhang Y, Zhang YC, Huang ZJ, Mei SQ, Liu JQ, Guan XH, Deng Y, Chen ZH, Tu HY, Xu CR, Chen HJ, Zhong WZ, Yang JJ, Zhang XC, Mok T, Wu YL, Zhou Q (2024) Allelic Context of EGFR C797X-Mutant Lung Cancer Defines Four Subtypes With Heterogeneous Genomic Landscape and Distinct Clinical Outcomes. J Thorac Oncol 19(4):601-612. 10.1016/j.jtho.2023.11.016\u003c/li\u003e\n\u003cli\u003eNoronha V, Menon NS, Patil VM, Chandrakanth MV, More S, Dhanawat A, Chowdhary OR, Singh AC, Goud S, Shah S, Karuvandan N, Jobanputra KN, Shah DK, Shah MJ, Sarma R, Patel D, Joarder R, Kumar P, John A, Kaur J, Bagra S, Purandare N, Janu A, Mahajan A, Prabhash K (2024) A Comparative Study Evaluating the Quality of Life and Survival Outcomes in Patients Receiving Chemotherapy Versus Oral Tyrosine Kinase Inhibitor in the Third Line and Beyond Setting for Advanced NSCLC. JTO Clin Res Rep 5(1):100622. 10.1016/j.jtocrr.2023.100622\u003c/li\u003e\n\u003cli\u003eQu FJ, Zhou Y, Wu S (2024) Progress of immune checkpoint inhibitors therapy for non-small cell lung cancer with liver metastases. Br J Cancer 130(2):165-175. 10.1038/s41416-023-02482-w\u003c/li\u003e\n\u003cli\u003eReita D, Pabst L, Pencreach E, Guerin E, Dano L, Rimelen V, Voegeli AC, Vallat L, Mascaux C, Beau-Faller M (2021) Molecular Mechanism of EGFR-TKI Resistance in EGFR-Mutated Non-Small Cell Lung Cancer: Application to Biological Diagnostic and Monitoring. Cancers (Basel) 13(19). 10.3390/cancers13194926\u003c/li\u003e\n\u003cli\u003eRemon J, Besse B, Aix SP, Callejo A, Al-Rabi K, Bernabe R, Greillier L, Majem M, Reguart N, Monnet I, Cousin S, Garrido P, Robinet G, Campelo RG, Madroszyk A, Mazieres J, Curcio H, Wasag B, Pretzenbacher Y, Grillet F, Dingemans AC, Dziadziuszko R (2024) Overall Survival From the EORTC LCG-1613 APPLE Trial of Osimertinib Versus Gefitinib Followed by Osimertinib in Advanced EGFR-Mutant Non-Small-Cell Lung Cancer. J Clin Oncol 42(12):1350-1356. 10.1200/JCO.23.01521\u003c/li\u003e\n\u003cli\u003eSeo JS, Kim A, Shin JY, Kim YT (2018) Comprehensive analysis of the tumor immune micro-environment in non-small cell lung cancer for efficacy of checkpoint inhibitor. Sci Rep 8(1):14576. 10.1038/s41598-018-32855-8\u003c/li\u003e\n\u003cli\u003eSiegel RL, Giaquinto AN, Jemal A (2024) Cancer statistics, 2024. CA Cancer J Clin 74(1):12-49. 10.3322/caac.21820\u003c/li\u003e\n\u003cli\u003eSong Q, Hawkins GA, Wudel L, Chou PC, Forbes E, Pullikuth AK, Liu L, Jin G, Craddock L, Topaloglu U, Kucera G, O\u0026apos;Neill S, Levine EA, Sun P, Watabe K, Lu Y, Alexander-Miller MA, Pasche B, Miller LD, Zhang W (2019) Dissecting intratumoral myeloid cell plasticity by single cell RNA-seq. Cancer Med 8(6):3072-3085. 10.1002/cam4.2113\u003c/li\u003e\n\u003cli\u003eTan L, Brown C, Mersiades A, Lee CK, John T, Kao S, Newnham G, O\u0026apos;Byrne K, Parakh S, Bray V, Jasas K, Yip S, Wong SQ, Ftouni S, Guinto J, Chandrashekar S, Clarke S, Pavlakis N, Stockler MR, Dawson SJ, Solomon BJ (2024) A Phase II trial of alternating osimertinib and gefitinib therapy in advanced EGFR-T790M positive non-small cell lung cancer: OSCILLATE. Nat Commun 15(1):1823. 10.1038/s41467-024-46008-1\u003c/li\u003e\n\u003cli\u003eWang Q, Zhang J, Liu Z, Duan Y, Li C (2023) Integrative approaches based on genomic techniques in the functional studies on enhancers. Brief Bioinform 25(1). 10.1093/bib/bbad442\u003c/li\u003e\n\u003cli\u003eWang S, Wang J, Chen Z, Luo J, Guo W, Sun L, Lin L (2024) Targeting M2-like tumor-associated macrophages is a potential therapeutic approach to overcome antitumor drug resistance. NPJ Precis Oncol 8(1):31. 10.1038/s41698-024-00522-z\u003c/li\u003e\n\u003cli\u003eWeinhauser I, Pereira-Martins DA, Almeida LY, Hilberink JR, Silveira D, Quek L, Ortiz C, Araujo CL, Bianco TM, Lucena-Araujo A, Mota JM, Hogeling SM, Sternadt D, Visser N, Diepstra A, Ammatuna E, Huls G, Rego EM, Schuringa JJ (2023) M2 macrophages drive leukemic transformation by imposing resistance to phagocytosis and improving mitochondrial metabolism. Sci Adv 9(15):f8522. 10.1126/sciadv.adf8522\u003c/li\u003e\n\u003cli\u003eYang JC, Gadgeel SM, Sequist LV, Wu CL, Papadimitrakopoulou VA, Su WC, Fiore J, Saraf S, Raftopoulos H, Patnaik A (2019) Pembrolizumab in Combination With Erlotinib or Gefitinib as First-Line Therapy for Advanced NSCLC With Sensitizing EGFR Mutation. J Thorac Oncol 14(3):553-559. 10.1016/j.jtho.2018.11.028\u003c/li\u003e\n\u003cli\u003eYang JC, Srimuninnimit V, Ahn MJ, Lin CC, Kim SW, Tsai CM, Mok T, Orlando M, Puri T, Wang X, Park K (2016) First-Line Pemetrexed plus Cisplatin followed by Gefitinib Maintenance Therapy versus Gefitinib Monotherapy in East Asian Never-Smoker Patients with Locally Advanced or Metastatic Nonsquamous Non-Small Cell Lung Cancer: Final Overall Survival Results from a Randomized Phase 3 Study. J Thorac Oncol 11(3):370-379. 10.1016/j.jtho.2015.11.008\u003c/li\u003e\n\u003cli\u003eYang Y, Hua Y, Zheng H, Jia R, Ye Z, Su G, Gu Y, Zhan K, Tang K, Qi S, Wu H, Qin S, Huang S (2024) Biomarkers prediction and immune landscape in ulcerative colitis: Findings based on bioinformatics and machine learning. Comput Biol Med 168:107778. 10.1016/j.compbiomed.2023.107778\u003c/li\u003e\n\u003cli\u003eZhang L, Zhang X, Guan M, Zeng J, Yu F, Lai F (2024) Identification of a novel ADCC-related gene signature for predicting the prognosis and therapy response in lung adenocarcinoma. Inflamm Res. 10.1007/s00011-024-01871-y\u003c/li\u003e\n\u003cli\u003eZheng RS, Chen R, Han BF, Wang SM, Li L, Sun KX, Zeng HM, Wei WW, He J (2024) [Cancer incidence and mortality in China, 2022]. Zhonghua Zhong Liu Za Zhi 46(3):221-231. 10.3760/cma.j.cn112152-20240119-00035\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"LUAD, gefitinib resistance, single-cell RNA-sequencing, signature, immune microenvironment, treatment","lastPublishedDoi":"10.21203/rs.3.rs-4573455/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4573455/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"Gefitinib resistance (GR) is widespread; therefore, alternative treatments for lung adenocarcinoma (LUAD) are needed. The study of gefitinib-resistance gene sets may lead to a better understanding of the mechanism underlying GR, methods for predicting and preventing GR, and alternative therapies. GR gene sets, single-cell data, and transcriptome data were obtained from public databases. Univariate and multivariate regression analyses and machine learning techniques were used to screen genes and construct a signature, respectively. Survival analysis and time-dependent receiver operating characteristic (ROC) curves were used to assess signature performance in internal and external data sets. Enrichment and tumor immune-microenvironment analyses were used to explore the mechanism of the signature genes in GR. Novel immunological and non-immunological therapies were explored. A signature consisting of 22 genes was successfully constructed in LUAD cohort, which performed well in both internal and external validation. The signature was closely related to chromosomal processes, DNA replication, important immune-cell infiltration, and multiple immune scores in enrichment and tumor microenvironment analyses. Further, the signature predicted immunotherapy efficacy in patients with LUAD to a certain extent, and we identified various agents other than gefitinib that may have better treatment effects in high-risk and low-risk groups, providing treatment guidance for gefitinib-resistant patients. The 22-gene signature can predict the prognosis of gefitinib-resistant patients with LUAD and immunotherapy efficacy, and provides new guidance for non-immunotherapy.","manuscriptTitle":"Identification of a gefitinib resistance-associated signature for predicting prognosis and therapeutic response in lung adenocarcinoma via integrated multi-omics analysis and machine learning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-07-16 21:59:43","doi":"10.21203/rs.3.rs-4573455/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"a671a524-e35b-4ba2-be1c-960e9c73aaf8","owner":[],"postedDate":"July 16th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-08-07T07:40:37+00:00","versionOfRecord":[],"versionCreatedAt":"2024-07-16 21:59:43","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4573455","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4573455","identity":"rs-4573455","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}