Integration of osimertinib-targeted EGFR gene-associated differential gene expression in constructing a prognostic model for lung adenocarcinoma Authors: Haiwen Li, Li Yang, Quan Yang, Zhu Liang, Wenmei Su, Lili Yu Published: 2024 ABSTRACT Lung adenocarcinoma (LUAD) is one of the deadliest cancers. Epidermal growth factor receptor-tyrosine kinase inhibitor (EGFR-TKI)-targeted therapy is an important approach for treating LUAD. However, the development of acquired resistance poses a serious clinical challenge. Our objective was to explore the differentially expressed genes (DEGs) associated with EGFR and detect biomarkers for diagnosing and treating osimertinib resistance in LUAD patients. LUAD datasets were downloaded from public databases. Differential expression analysis was performed to screen DEGs, and prognostic modules were constructed by Cox regression. Enrichment analysis, gene regulatory network analysis and immune microenvironment analysis were employed to explore the underlying mechanisms in LUAD. Finally, the expression of prognosis module genes (PMGs) was validated in 8 LUAD tissue specimens and 5 cell lines by qRT-PCR. In total, 13 differential module genes (BIRC3, CCT6A, CPLX2, GLCCI1, GSTA1, HLA-DQB2, ID1, KCTD12, MUC15, NOTUM, NT5E, TCIM, and TM4SF4) were screened for the construction of a prognostic module. Notably, CCT6A and KCTD12 demonstrated excellent accuracy in the diagnosis of LUAD. Immune dysregulation and BIRC3, HLA-DQB2, KCTD12, and NT5E expression were significantly associated with invasive immune cells in LUAD patients. The expression level of CCT6A was highest in PC9-OR and H1975-OR cells, while the expression level of KCTD12 was highest in paracancerous tissue and HBE cells. The constructed prognostic model showed promise in predicting the survival of LUAD patients. Notably, KCTD12 and CCT6A might be candidate biomarkers for improving diagnostic performance and guiding individualized therapy for EGFR-TKI-resistant LUAD patients.Supplementary InformationThe online version contains supplementary material available at 10.1007/s10142-024-01499-5. Introduction Non-small cell lung cancer (NSCLC) constitutes 85% of all lung cancer cases and is a leading cause of cancer-related mortality worldwide. Lung adenocarcinoma (LUAD) is the most prevalent subtype and represents ∼ 48% of all NSCLCs, with a 5-year survival rate of only 15% (Bray et al. 2024). This low survival rate is most likely because more than 50% of NSCLC patients are diagnosed at an advanced stage (Siegel et al. 2023). Platinum-based doublet chemotherapy was previously the standard-of-care first-line treatment for patients with advanced NSCLC. The identification of driver mutations and the development of molecularly targeted therapies have led to a paradigm shift in the treatment of unresectable NSCLC (Riely et al. 2024). The upregulation and abnormal activation of epidermal growth factor receptor (EGFR) are key factors in LUAD development, with around 21% of patients having EGFR genetic alterations, making it a viable target for blockade therapy (Shigematsu et al. 2005). Anti-EGFR targeted therapies include antibodies like cetuximab (Saoudi Gonzalez et al. 2024) and panitumumab (Watanabe et al. 2023), as well as small-molecule tyrosine kinase inhibitors (TKIs) such as gefitinib (Zhou et al. 2024) and osimertinib (Le et al. 2024). Numerous randomized trials have confirmed that EGFR-TKIs are superior to cytotoxic agents in patients with EGFR-mutant NSCLC (John et al. 2018; Maemondo et al. 2010; Rosell et al. 2012). However, like patients receiving first- or second-generation EGFR-TKIs, those treated with osimertinib ultimately develop resistance. A considerable proportion of osimertinib-resistant patients exhibit an absence of unequivocally identified resistance pathways (Zeng et al., 2022b). It has been reported that the mechanism of EGFR resistance is gain-of-function with enhanced EGFR signaling in NSCLC (Laface et al. 2023). While the EGFR signaling pathway is generally well understood, there is insufficient research on how mutated EGFR interacts with specific signaling components. Consequently, there is a growing need to investigate targetable signaling components within the mutated EGFR signaling network and to identify useful biomarkers that could enhance the survival of patients with NSCLC. In this study, we performed systematic bioinformatics analyses to screen prognosis module genes (PMGs) related to the expression of EGFR in high- and low-risk groups of LUAD patients. Additionlly, we performed enrichment analysis, gene regulatory network analysis, immune microenvironment analysis, and mutation analysis to explore the underlying mechanisms in LUAD. Furthermore, the expression patterns of these PMGs were validated in 8 LUAD tissue specimens and PC9, PC9-OR, H1975, and H1975-OR cell lines by using qRT‒PCR. Overall, our study offers a novel perspective on the molecular mechanisms underlying LUAD and provides new ideas and potential targets for its diagnosis and treatment. Materials and methods Data download The LUAD datasets were downloaded from The Cancer Genome Project (TCGA, https://portal.gdc.cancer.gov/), and sequencing data from 539 LUAD cancer samples and 59 partially matched adjacent normal samples were obtained. The clinical data corresponding to the samples were obtained from the UCSC Xena database (http://genome.ucsc.edu), the microsatellite instability (MSI) and tumor mutation burden (TMB) data were obtained from the cBioPortal database (https://www.cbioportal.org/), and the tumor immune dysfunction and exclusion (TIDE) immune score results of the TCGA-LUAD dataset were calculated through the TIDE website (http://tide.dfci.harvard.edu). We also selected “Masked Somatic Mutation” data from the TCGA LUAD patient cohort (n = 618). Data were preprocessed using VarScan software, and SM status was visualized using the maftools R package. To analyze the copy number variations (CNVs) of LUAD, we used the TCGAbiolinks package in R to download the patients’ “Copy Number Segment” data (n = 1147). Then, we used GISTIC 2.0 to download and process the CNV fragments (Mermel et al. 2011). The parameter reference genome version was set to “hg38.UCSC.add_miR.160920.refgene.mat”, and all other parameters used the default parameter settings. In addition, we downloaded the expression profile datasets GSE27262 (Wei et al. 2012), GSE31210 (Okayama et al. 2012), and GSE32863 (Selamat et al. 2012) of LUAD patients from the GEO (https://www.ncbi.nlm.nih.gov/geo/). All expression profile data and grouping information corresponding to the samples in the GSE27262, GSE31210, and GSE32863 datasets were included in subsequent analysis. The GSE27262, GSE31210 and GSE32863 datasets were utilized as validation sets to verify the differences in the expression of the target gene EGFR among the different groups. See Table S1 for specific dataset information. Prognostic analysis and clinical correlation analysis of the target gene EGFR To explore the differential expression of the EGFR gene targeted by the drug osimertinib, we utilized the TCGA-LUAD dataset, which combines the expression of the EGFR target gene with the clinical prognostic data of LUAD patients, to analyze the relationship between the expression of the EGFR target gene and patient prognosis outcomes and constructed a Kaplan‒Meier (KM) survival curve by using the survminer R package to determine the optimal group (minimum P value). Then, based on the high and low expression of EGFR, we grouped the differences in the distributions of different clinical variables (T stage, N stage, M stage, pathological stage, sex, age, overall survival (OS), disease-specific survival (DSS), and progression-free intervals (PFIs) among the LUAD patients and constructed stacked column plots. Furthermore, to identify the potential mechanisms, as well as the associated biological characteristics and pathways of the differentially expressed genes (DEGs) in the high/low group of LUAD patients, we performed an analysis on the TCGA-LUAD dataset using the limma package (Ritchie et al. 2015). We employed the criteria of|logFC| > 1 and P.adj < 0.05 to screen DEGs for further study. Weighted gene association network analysis (WGCNA) We conducted WGCNA using the R package WGCNA package (Langfelder and Horvath 2008). Initially, we calculated the correlation coefficient between pairs of genes and utilized the weighted correlation coefficient to establish connections between genes in the network, following a scale-free network pattern. Next, a hierarchical clustering tree was constructed using the correlation coefficients among genes. In this article, WGCNA used the mRNA genes whose gene expression variance accounted for the top 25% of all genes in the TCGA-LUAD dataset as inputs. The module gene count was set at a minimum of 100, while the soft power was adjusted to 4. Both the module merging threshold and the minimum distance are set to 0.2. Once the module of interest was identified based on the correlation values, we considered all genes within the module to be highly relevant to LUAD. Then, by intersecting the module genes with the DEGs, we obtained a set of differential module genes (DMGs). Construction of a DMG prognosis model To construct the DMG prognosis model, we initially employed univariate Cox regression to evaluate the DMGs. DMGs that satisfied the criterion of P < 0.1 were selected. Subsequently, the glmnet package (Engebretsen and Bohlin 2019) was utilized to perform least absolute shrinkage and selection operator (LASSO) regression based on the aforementioned DMGs. The regression was executed with the parameter family = “cox” and tenfold cross-validation method. To prevent overfitting, LASSO regression was run for 1000 epochs.\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&\text{R}\text{i}\text{s}\text{k}\text{s}\text{c}\text{o}\text{r}\text{e}\cr&\quad=\sum_{\text{i}}\text{C}\text{o}\text{e}\text{f}\text{f}\text{i}\text{c}\text{i}\text{e}\text{n}\text{t}\:\left({\text{h}\text{u}\text{b}\:\text{g}\text{e}\text{n}\text{e}}_{\text{i}}\right)\text{*}\text{m}\text{R}\text{N}\text{A}\:\text{E}\text{x}\text{p}\text{r}\text{e}\text{s}\text{s}\text{i}\text{o}\text{n}\:\left(\text{h}\text{u}\text{b}\:{\text{g}\text{e}\text{n}\text{e}}_{\text{i}}\right)\end{aligned}$$\end{document} Multivariate Cox regression analysis was conducted on the DMGs using LASSO regression to identify and construct a DMG prognosis model. The lambda coefficient of each prognostic DMG in the DMG prognosis model was obtained. Subsequently, the risk score of each LUAD patient was calculated, and patients were categorized into high and low groups based on the median. The genes in the DMG prognosis model are named prognosis module genes (PMGs). K‒M analysis was conducted using the survival package to examine the associations between the high- and low-risk score groups of the PMGs and OS. Additionally, we used time-dependent receiver operating characteristic (ROC) curves to assess survival predictions. The area under the ROC curve (AUC) values, which quantify the performance of the PMGs, were calculated using the timeROC R package (Beyene and El Ghouch 2022) in R (version 3.03). Group comparison of PMGs between the cancer and normal groups in four sets of datasets The Mann‒Whitney U test was used to compare the PMGs between the cancer and normal groups. The findings were visualized through group comparison diagrams. Additionally, the ROC curves of the PMGs in different groups (LUAD/Normal) were plotted using R’s proc package. The calculated AUC was utilized as an indicator to evaluate the diagnostic effectiveness of the PMGs in identifying cancer development in LUAD patients. Furthermore, the Mann‒Whitney U test was performed to analyze the differences in the expression of PMGs between the cancer group and the normal group in the GSE27262, GSE31210 and GSE32863 datasets. We determined the correlation coefficient between the expression levels of PMGs in each dataset using Spearman correlation analysis and subsequently presented the results. Functional enrichment and pathway enrichment analyses We used the R package clusterProfiler (Yu et al. 2012) to perform Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses. To identify statistically significant results, we applied the following screening criteria: a P value < 0.05 and an FDR value (q value) < 0.1. Gene set enrichment analysis (GSEA) and gene set variation analysis (GSVA) Based on the median risk score, the LUAD patients were divided into high- and low-risk score groups. To explore the enrichment of pathway genes, we conducted GSEA using the clusterProfiler package. The GSEA parameters employed were as follows: the seed number was set to 2022, a total of 1000 calculations were performed, and each gene set consisted of at least 10 genes and a maximum of 10 genes. The analysis focused on 500 genes, and the P value correction method utilized was Benjamini‒Hochberg correction. To evaluate whether different samples exhibited enriched pathways, we conducted GSEA utilizing the GSEA package. The gene set “c2.cp.all.v2022.1.Hs.symbols.gmt” from the MSigDB database was acquired for this purpose. GSVA was then performed on all genes within the high- and low-risk score groups of LUAD patients. We considered a P value < 0.05 as the threshold for significant enrichment, and the results presented the top 5 pathways (positive and negative) with a maximum of 10 pathways. Somatic mutation (SM) and copy number variation (CNV) analysis To analyze the SM status of PMGs in patients in the TCGA-LUAD dataset, “Masked Somatic Mutation” data from the TCGA website were selected. The SM data of PMGs in patients were preprocessed using VarScan software, and the SM data were visualized using the R package maftools. To examine the CNVs of the PMGs in the high- and low-risk score groups among LUAD patients, we used the TCGAbiolinks R package to download patients’ data, and obtained “masked copy number segment” data from 1147 patients and conducted GISTIC 2.0 analysis on the downloaded CNV segment data via GenePattern (https://cloud.genepattern.org). Immune assessment and clinical correlation analysis The Mann‒Whitney U test was used to assess the differences in tumor mutation burden (TMB) and microsatellite instability (MSI) between the high- and low-risk score groups in the PMGs of LUAD patients. In addition, we investigated the potential tumor treatment response to immune checkpoint blockade (ICB) by the TIDE score. Spearman correlation analysis was performed to investigate the relationships between the risk score and TMB, MSI, and TIDE in the high- and low-risk score groups of LUAD patients. The algorithm calculations and visualizations were carried out using the R package ggplot2. We used the R package ESTIMATE to evaluate differences in the immune activity of tumors between the high- and low-risk score groups in the TCGA-LUAD dataset. A quantitative analysis of the gene expression profiles of LUAD tumor samples was performed to derive an immune score for each sample. Results were visualized using group comparison charts with the ggplot2 R package. Then, we presented the differences in the distributions of different clinical variables (T stage, N stage, M stage, pathological stage, sex, age, OS, DSS, and PFI) among LUAD patients by comparing the high- and low-risk score groups. This analysis was conducted using a stacked histogram with the ggplot2 R package. Single-sample gene set enrichment analysis (ssGSEA) and immune infiltration analysis with the CIBERSORT algorithm We employed ssGSEA to quantify the relative abundance of each immune cell infiltrate. We constructed the TCGA-LUAD dataset DMG prognostic model through ssGSEA via the R package GSVA. The infiltration levels of different immune cell types in each sample were represented by the enrichment scores of the high- and low-risk score groups. We uploaded the gene expression matrix data of the TCGA-LUAD dataset’s high/low risk score groups, specifically the PGMs, to CIBERSORT. Subsequently, we obtained and showcased the matrix displaying the abundance of infiltrating immune cells. The difference in immune cell infiltration abundance between the high- and low-risk score groups according to the gene prognostic model is presented in the group comparison chart. The correlation between immune cells within the high/low risk score groups was calculated by the Spearman algorithm and presented visually using the R package ggplot2. Identification of LUAD disease subtypes in the TCGA dataset We utilized the ConsensusClusterPlus package (Wilkerson and Hayes 2010) in R to employ the consensus clustering method to identify distinct LUAD subtypes based on the PMGs. The differences in the expression of prognostic module genes between samples of different LUAD subtypes were calculated by the Mann‒Whitney U test. We also employed group comparison charts to present the results of the ESTIMATE score, immune score, stromal score, and tumor purity score by incorporating group information from various LUAD subtypes. Protein‒protein interaction (PPI) network To construct a PPI network related to the DMGs, we utilized the STRING database. Subsequently, we employed Cytoscape 3.6 (version 3.9.1) for visualization purposes. Within the PPI network, densely connected local regions likely signify molecular complexes. Notably, all nodes present in the PPI network were considered significant genes (particularly hub genes and mRNAs). Construction of mRNA–miRNA and mRNA-transcription factor (TF) interaction networks The miRDB database was used for miRNA prediction of interactions with hub genes, data instances with a target score ≥ 95 in the miRDB database were retained, and an mRNA‒miRNA interaction network was visualized. Moreover, we conducted a search in the CHIPBase database (version 3.0) and the hTFtarget database (https://guolab.wchscu.cn/hTFtarget/) to identify TFs that bind to key genes. The overlapping results from both databases were retained, enabling the creation of an mRNA-TF interaction network. Finally, the mRNA‒miRNA and mRNA-TF interaction networks were visualized using Cytoscape software. Construction of a multifactor prognostic risk score model To develop a prognostic model for the multifactor risk score, we initially conducted univariate Cox regression analysis involving the risk score and clinical variables. The results of this analysis were visualized through forest plots. All variables with P < 0.05 were included in the multivariate Cox regression analysis, which enabled the construction of a comprehensive prognostic risk score model. We utilized the R package rms to construct a nomogram based on the results of multivariate Cox regression analysis. To assess the accuracy and precision of the nomogram, a calibration curve was generated. The accuracy and resolution of the multifactor risk score prognostic model were evaluated using decision curve analysis (DCA). To construct the DCA plots, we employed the R package ggDCA. Cell culture Different human lung NSCLC cell lines were used in this study, including PC9 and H1975 sensitive to osimertinib were obtained from ATCC, PC9-OR and H1975-OR resistant to osimertinib were maintained in our laboratory, and the human bronchial epithelial (HBE) was obtained from ATCC. The cells were incubated in a humidified atmosphere at 37 °C with 5% CO2 in RPMI-1640 medium (Gibco) supplemented with 10% fetal bovine serum (Invitrogen). Clinical samples In total, 8 LUAD patients were recruited for the sampling of LUAD cancer samples and matched with adjacent normal samples from the Affiliated Guangdong Medical University Hospital from May 2023 to October 2023. The detailed clinical information is shown in Supplementary Table S2. The study was approved by the Ethics Committee of the Affiliated Guangdong Medical University (reference number: PJKT2024-023) on February 1, 2024. RNA extraction and qRTPCR Total RNA was extracted from cells or tissues using TRIzol reagent (TaKaRa, Dalian, China) following the manufacturer’s instructions. The detailed steps were described in our previous study (Wu et al. 2024). The sequences of primers used are listed in Supplementary Table S3. Target gene expression relative to that of the GAPDH gene was determined as 2ΔΔCt. Statistical analysis The specific technical roadmap of our study was depicted in the figure (Figure. S1). All data processing and analysis in this article were conducted using R software (version 4.1.2). To assess the statistical significance between two groups of continuous variables, we employed the independent Student’s t test for normally distributed variables and the Mann‒Whitney U test for nonnormally distributed variables. Survival analysis was conducted using the R survival package. K‒M survival curves were generated to assess differences in survival, while the log-rank test was used to assess the significance of survival time disparities between the two patient groups. Unless otherwise noted, Spearman correlation analysis was employed to calculate the correlation coefficient between various molecules. All the statistical P values were two-sided, with P < 0.05 considered to indicate statistical significance. Results LUAD dataset processing We first normalized the four LUAD datasets, TCGA-LUAD, GSE27262, GSE31210 and GSE32863, using the R package limma package (Fig. 2S). The TCGA-LUAD dataset contains 539 LUAD cancer samples and 59 partially matched adjacent normal samples. The GSE27262 dataset contains a total of 25 LUAD samples and 25 corresponding matched normal lung tissue samples; the GSE32863 dataset contains 58 LUAD tumor tissue samples from patients and 58 corresponding matched normal lung tissue samples from patients; and the GSE31210 dataset contains 226 LUAD tissue samples and 20 normal lung tissue samples from normal controls. After normalization of the expression profile data of the four LUAD datasets, TCGA-LUAD, GSE27262, GSE32863 and GSE31210, the batch effect between samples was removed, and the expression of the sample data became more consistent (Fig. 2S). Prognostic analysis and clinical correlation analysis of EGFR We conducted a prognostic KM curve analysis utilizing the survival data (OS.event) of patients with LUAD. Subsequently, we determined the best grouping (minimum P value) of the target genes in LUAD patients within the TCGA-LUAD dataset (P = 0.039, Fig. 1A). Fig. 1Construction and clinical correlation analysis of EGFR. (A) Prognostic KM curve results of EGFR in the TCGA-LUAD dataset. (B–J) Clinical T stage (B), N stage (C), M stage (D), sex (E), age (F), pathological stage (G), OS event (H), DSS event (I), and PFI event (J) stacked histogram results. (K-L) Volcano plot of DEG analysis (K) and heatmap of the positive and negative top 20 DEGs (L) between different groups in the TCGA-LUAD dataset. LUAD, lung adenocarcinoma; TCGA, The Cancer Genome Atlas; OS, overall survival; DSS, disease-specific survival; PFI, progression-free interval; DEGs, differentially expressed genes Construction and clinical correlation analysis of EGFR. (A) Prognostic KM curve results of EGFR in the TCGA-LUAD dataset. (B–J) Clinical T stage (B), N stage (C), M stage (D), sex (E), age (F), pathological stage (G), OS event (H), DSS event (I), and PFI event (J) stacked histogram results. (K-L) Volcano plot of DEG analysis (K) and heatmap of the positive and negative top 20 DEGs (L) between different groups in the TCGA-LUAD dataset. LUAD, lung adenocarcinoma; TCGA, The Cancer Genome Atlas; OS, overall survival; DSS, disease-specific survival; PFI, progression-free interval; DEGs, differentially expressed genes Then, we analyzed the T stage, N stage, M stage, sex, age, pathological stage, and OS in LUAD patients in the high and low EGFR expression groups. The proportions of DSS events and PFI events are displayed through stacked histograms (Fig. 1B–J). The proportions of sex (Fig. 1E) and OS events (Fig. 1H) were significantly different, while the proportions of T stage (Fig. 1B), N stage (Fig. 1C), M stage (Fig. 1D), age (Fig. 1F), pathological stage (Fig. 1G), DSS event (Fig. 1I), and PFI event (Fig. 1J) were not significantly different between the high and low groups of LUAD patients in the TCGA-LUAD dataset. To analyze the DEGs between the high and low LUAD groups, we used the limma package to perform differential expression analysis. A total of 19,538 DEGs were obtained from the TCGA-LUAD dataset, and 356 DEGs (159 upregulated and 197 downregulated) met the thresholds of|logFC| > 0.5 and P.adj < 0.05 in the high and low EGFR expression groups, respectively, in LUAD (Fig. 1K). A heatmap was generated to visualize the 40 DEGs of interest (top 20 DEGs in the high group and top 20 DEGs in the low group) between the high and low groups in the TCGA-LUAD dataset (Fig. 1L). The specific DEG information is shown in Table S4. WGCNA to determine coexpression modules in the TCGA-LUAD dataset In the process of WGCNA, we first selected genes whose variance was in the top 25% of all genes as the total input genes. Then, we clustered the samples in the TCGA-LUAD dataset through a clustering tree, and we set the screening criterion to 0.9 to determine the optimal power threshold (Fig. 2A). The optimal power threshold for WGCNA was 4, and the genes were clustered into 10 modules: MEturquoise, MEyellow, MEmagenta, MEblue, MEgreen, MEpink, MEred, MEblack, MEbrown, and MEgray (Fig. 2B-C). In addition, based on the expression patterns of the module genes and the grouping information of the TCGA-LUAD dataset, we obtained 10 modules that correlated with the normal/LUAD groups in the TCGA-LUAD dataset (Fig. 2D). We conducted a comparative analysis of five modules that exhibited statistically significant differences (P < 0.05, absolute value of correlation,|COR| ≥0.3) among the 9 modules (excluding irrelevant gray modules: MEgray) and various groups from the TCGA-LUAD dataset. The modules MEturquoise (|COR|=0.51, P = 3e-40), MEblue (|COR|=0.85, P = 1e-168), MEgreen (|COR|=0.31, P = 7e-15), MEpink (|COR|=0.38, P = 2e-22) and MEbrown (|COR|=0.43, P = 7e-28) were included in the subsequent analysis. Moreover, we compared the DEGs between the TCGA-LUAD dataset and the MEturquoise (Fig. 2E), MEblue (Fig. 2F), MEgreen (Fig. 2G), MEpink (Fig. 2H), and MEbrown (Fig. 2I) modules. The module genes provided were intersected, and a Venn diagram was generated to obtain the DMGs. Finally, a total of 257 DMGs were obtained (Fig. 2E–I). Table S5 lists the specific DMG names. Fig. 2WGCNA determines coexpression modules in the TCGA-LUAD dataset. (A) Scale-free network display of sample module filtering thresholds in the TCGA-LUAD dataset. (B) Display of the aggregation results of the gene modules in the TCGA-LUAD dataset. (C) Correspondence between genes and modules in the TCGA-LUAD dataset. (D) Correlation analysis results between the gene clustering module and different groups (normal/LUAD) of the TCGA-LUAD dataset. (E-F) Venn diagram of the DMGs and TCGA-LUAD dataset MEturquoise module (Fig. 4E), MEblue (Fig. 4F), MEgreen (Fig. 4G), MEpink (Fig. 4H), and MEbrown (Fig. 4I). WGCNA, weighted gene coexpression network analysis; TCGA, The Cancer Genome Atlas; LUAD, lung adenocarcinoma; DEGs, differentially expressed genes; DMGs, differentially expressed genes WGCNA determines coexpression modules in the TCGA-LUAD dataset. (A) Scale-free network display of sample module filtering thresholds in the TCGA-LUAD dataset. (B) Display of the aggregation results of the gene modules in the TCGA-LUAD dataset. (C) Correspondence between genes and modules in the TCGA-LUAD dataset. (D) Correlation analysis results between the gene clustering module and different groups (normal/LUAD) of the TCGA-LUAD dataset. (E-F) Venn diagram of the DMGs and TCGA-LUAD dataset MEturquoise module (Fig. 4E), MEblue (Fig. 4F), MEgreen (Fig. 4G), MEpink (Fig. 4H), and MEbrown (Fig. 4I). WGCNA, weighted gene coexpression network analysis; TCGA, The Cancer Genome Atlas; LUAD, lung adenocarcinoma; DEGs, differentially expressed genes; DMGs, differentially expressed genes Construction of the DMG prognostic model and difference analysis between the high- and low-risk score groups To assess the prognostic value of the 257 DMGs in the TCGA-LUAD dataset, we conducted univariate Cox regression analysis with a significance threshold of P < 0.05. We identified 72 DMGs that satisfied our screening criteria (Table S6). Next, employing LASSO regression analysis, we constructed a LASSO regression model based on these 72 DMGs (Fig. 3A-B). Considering the minimum partial likelihood bias, we retained 26 DMGs out of the initial 72 DMGs as diagnostic markers for LUAD. Moreover, we visually represented the LASSO regression outcomes and generated a LASSO variable trajectory diagram (Fig. 3A). The 26 DMGs included in the LASSO regression model were AMN, BIRC3, CCT6A, CLEC3B, CPLX2, CPS1, FNIP2, FURIN, GLB1L3, GLCCI1, GSTA1, HLA-DQB2, ID1, INHA, KCNK3, KCTD12, KLK11, LGR4, MUC15, NOTUM, NT5E, PCSK9, SEMA3A, SLC34A2, TCIM, and TM4SF4. To further validate the prognostic significance of the 26 DMGs in the TCGA-LUAD dataset, we employed multifactor Cox regression to analyze the correlation between the expression levels of the 26 DMGs and the clinical prognosis of LUAD patients and ultimately developed a DMG prognostic model consisting of 13 promising PMGs (BIRC3, CCT6A, CPLX2, GLCCI1, GSTA1, HLA-DQB2, ID1, KCTD12, MUC15, NOTUM, NT5E, TCIM, and TM4SF4) (Table S7). Fig. 3Construction of the differential module gene prognosis model. (A-B) Variable trajectory diagram of the LASSO regression model (A) and prognostic model diagram (B). (C) Forest plot of multivariate Cox regression analysis of PMGs in the TCGA-LUAD dataset. (D) Risk score factor plot of the differential module gene prognostic model. (E-F) The prognostic KM curve (E) and time-dependent ROC curve (F) of the risk score of the DMG prognostic model in the TCGA-LUAD dataset. (G–I) Volcano plot of DEG analysis (G), heatmap of PMGs (H) and correlation analysis (I) in the high- and low-risk score groups of the DMG prognostic model of the TCGA-LUAD dataset. LASSO, least absolute shrinkage and selection operator; TCGA, The Cancer Genome Atlas; LUAD, lung adenocarcinoma \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}Riskscore&=BIRC3*0.1928+CCT6A*0.182\\&\quad+CPLX2*0.1259\\&\quad+GLCCI1*-0.1659+GSTA1*-0.0836\\&\quad+HLA-DQB2*-0.08225\\&\quad+ID1*0.1283+KCTD12*-0.3015\\&\quad+MUC15*-0.0865+NOTUM*-0.1285\\&\quad+NT5E*0.1502+TCIM*-0.101\\&\quad+TM4SF4*0.0659\end{aligned}$$\end{document} Construction of the differential module gene prognosis model. (A-B) Variable trajectory diagram of the LASSO regression model (A) and prognostic model diagram (B). (C) Forest plot of multivariate Cox regression analysis of PMGs in the TCGA-LUAD dataset. (D) Risk score factor plot of the differential module gene prognostic model. (E-F) The prognostic KM curve (E) and time-dependent ROC curve (F) of the risk score of the DMG prognostic model in the TCGA-LUAD dataset. (G–I) Volcano plot of DEG analysis (G), heatmap of PMGs (H) and correlation analysis (I) in the high- and low-risk score groups of the DMG prognostic model of the TCGA-LUAD dataset. LASSO, least absolute shrinkage and selection operator; TCGA, The Cancer Genome Atlas; LUAD, lung adenocarcinoma The fundamental depiction of the DMG prognostic model using a forest plot is shown in Fig. 3C, while Fig. 3D provides a visual representation of the sample grouping within the constructed DMG prognostic model using a risk score factor map. The analysis of the prognostic KM curve demonstrated a highly significant difference (P < 0.001) in the survival outcomes of LUAD patients between the high- and low-risk score groups in the TCGA-LUAD dataset (Fig. 3E). The time-dependent ROC curve analysis demonstrated that the risk score of the DMG prognostic model exhibited considerable accuracy (AUC1 = 0.761, AUC3 = 0.704, AUC5 = 0.718; Fig. 3F) in predicting survival outcomes in patients with LUAD. There were 50 DEGs (24 up- and 26 downregulated genes) between the high- and low-risk score groups (Fig. 3G). The differential expression of the 13 PMGs in the high- and low-risk score groups was significantly different. A heatmap was generated to visually present the specific outcomes of the differential analysis (Fig. 3H). There was a positive correlation between the risk score and the expression levels of BIRC3, CCT6A, CPLX2, ID1, NT5E, and TM4SF4. Conversely, a negative correlation was observed between the risk score and the expression levels of GLCCI1, GSTA1, HLA-DQB2, KCTD12, MUC15, NOTUM, and TCIM (Fig. 3I). Analysis of the differences in the expression of PMGs The differential expression of 13 PMGs across different groups (LUAD/normal) in the TCGA-LUAD dataset was analyzed. The expression levels of 10 PMGs (CCT6A, NT5E, CPLX2, GLCCI1, GSTA1, HLA-DQB2, ID1, KCTD12, MUC15, and NOTUM) in different groups (LUAD/Normal) were significantly different (P < 0.05) between the LUAD and normal groups. Specifically, CCT6A and NT5E exhibited higher expression levels in the LUAD group than in the normal group. Conversely, CPLX2, GLCCI1, GSTA1, HLA-DQB2, ID1, KCTD12, MUC15, and NOTUM were expressed at lower levels in the LUAD group than in the normal group (Fig. 4A). We further investigated the diagnostic potential of these 10 PMGs in both the LUAD and normal groups. Additionally, we generated ROC curves to demonstrate the outcomes, as depicted in Fig. 4B-K. Notably, CCT6A (AUC = 0.949, Fig. 4B) and KCTD12 (AUC = 0.927, Fig. 4H) exhibited high accuracy in accurately identifying the LUAD and normal groups. Fig. 4Expression difference analysis of PMGs. Group comparison diagram of PMGs between different groups in the TCGA-LUAD dataset (LUAD/Normal). (B–K) ROC curves of the prognostic module genes CCT6A (B), CPLX2 (C), GLCCI1 (D), GSTA1 (E), HLA-DQB2 (F), ID1 (G), KCTD12 (H), MUC15 (I), NOTUM (J), and NT5E (K) in the TCGA-LUAD dataset. ns, not significant; TCGA, The Cancer Genome Atlas; LUAD, thyroid carcinoma; ROC, receiver operating characteristic curve; AUC, area under the curve Expression difference analysis of PMGs. Group comparison diagram of PMGs between different groups in the TCGA-LUAD dataset (LUAD/Normal). (B–K) ROC curves of the prognostic module genes CCT6A (B), CPLX2 (C), GLCCI1 (D), GSTA1 (E), HLA-DQB2 (F), ID1 (G), KCTD12 (H), MUC15 (I), NOTUM (J), and NT5E (K) in the TCGA-LUAD dataset. ns, not significant; TCGA, The Cancer Genome Atlas; LUAD, thyroid carcinoma; ROC, receiver operating characteristic curve; AUC, area under the curve Difference analysis and correlation analysis of PMGs The expression patterns of 13 PMGs in various groups (LUAD/normal) were analyzed in three LUAD datasets: GSE27262, GSE32863 and GSE31210 (Fig. 5A-C). Figure 5A shows significant differences (P < 0.05) in the expression levels of the 8 PMGs (CCT6A, HLA-DQB2, NT5E, TM4SF4, ID1, KCTD12, MUC15, and NOTUM) between the different groups in the GSE27262 dataset. Significant differences were observed in the expression levels of the 8 PMGs (BIRC3, CCT6A, NT5E, TM4SF4 GSTA1, ID1, KCTD12, and NOTUM) among the different groups in the GSE32863 dataset (P < 0.05, Fig. 5B). Similarly, in the GSE31210 dataset, significant differences were found in the expression levels of the 7 PMGs (CCT6A, HLA-DQB2, NT5E, TM4SF4, ID1, KCTD12, and MUC15) across the different groups (P < 0.05, Fig. 5C). Fig. 5Difference analysis and correlation analysis of the PMGs. (AC) Group comparison diagram of PMGs in different groups (normal/LUAD) of the GSE27262 dataset (A), GSE32863 dataset (B) and GSE31210 dataset (C). (D) Correlation dot plot results of PMGs in the TCGA-LUAD dataset. (E-F). Correlation scatter plot results of the PMGs CPLX2 and GSTA1 (E) and ID1 and CCT6A (F) in the TCGA-LUAD dataset. TCGA, The Cancer Genome Atlas; LUAD, Lung Adenocarcinoma Difference analysis and correlation analysis of the PMGs. (AC) Group comparison diagram of PMGs in different groups (normal/LUAD) of the GSE27262 dataset (A), GSE32863 dataset (B) and GSE31210 dataset (C). (D) Correlation dot plot results of PMGs in the TCGA-LUAD dataset. (E-F). Correlation scatter plot results of the PMGs CPLX2 and GSTA1 (E) and ID1 and CCT6A (F) in the TCGA-LUAD dataset. TCGA, The Cancer Genome Atlas; LUAD, Lung Adenocarcinoma We then performed correlation analysis using the expression data of the 13 PMGs in the TCGA-LUAD dataset, and the results were visually represented as a correlation heatmap (Fig. 5D). GLCCI1, GSTA1, ID1, TCIM, and CPLX2 were negatively correlated with the other PMGs. Specifically, the expression levels of CPLX2 and GSTA1 were positively correlated (R = 0.4, P < 2.2e-16; Fig. 5E), whereas the expression levels of ID1 and CCT6A (R = -0.29, P = 5.2e-13; Fig. 5F) were negatively correlated within the TCGA-LUAD dataset. Functional and pathway enrichment analysis of PMGs To examine the correlation between molecular functions, cellular components, biological pathways and LUAD, we initially conducted GO analysis and KEGG analysis on the 13 PMGs (Table S8). The results of GO analysis (Fig. 6A-C) revealed that the 13 PMGs were primarily associated with various biological processes, such as NIK/NF-kappaB signaling, cell activation involved in the immune response, positive regulation of protein kinase A signaling, MHC class II protein complex assembly, negative regulation of necroptotic processes, phosphoric ester hydrolase activity, MHC class II receptor activity, 5’-nucleotidase activity, proteasome binding, syntaxin-1 binding, intramolecular oxidoreductase activity and other molecular functions. KEGG pathways such as the platinum resistance pathway and the Hippo signaling pathway were also enriched (Fig. 6D). Next, we performed a combined logFC GO/KEGG enrichment analysis on the 13 PMGs (Fig. 6E). Thirteen PMGs were mainly concentrated in the MF pathway. Fig. 6GO and KEGG analyses of the PMGs. (A) GO and KEGG enrichment analysis of PMGs. (B-C) Circular network diagram of the BP pathway (B) and MF pathway (C) according to the GO functional enrichment analysis. (D) Network diagram of the KEGG enrichment analysis results. (E) Bubble chart displaying the results of GO and KEGG enrichment analysis of PMGs combined with logFC in the TCGA-LUAD dataset. GO, Gene Ontology; BP, biological process; CC, cellular component; MF, molecular function; KEGG, Kyoto Encyclopedia of Genes and Genomes GO and KEGG analyses of the PMGs. (A) GO and KEGG enrichment analysis of PMGs. (B-C) Circular network diagram of the BP pathway (B) and MF pathway (C) according to the GO functional enrichment analysis. (D) Network diagram of the KEGG enrichment analysis results. (E) Bubble chart displaying the results of GO and KEGG enrichment analysis of PMGs combined with logFC in the TCGA-LUAD dataset. GO, Gene Ontology; BP, biological process; CC, cellular component; MF, molecular function; KEGG, Kyoto Encyclopedia of Genes and Genomes GSEA and GSVA enrichment analysis The GSEA results of genes between the high- and low-risk score groups in the TCGA-LUAD dataset are presented in a mountain plot (Fig. 7A). The genes in the high- and low-risk score groups were significantly enriched in cell cycle checkpoints (Fig. 7B), glycolysis (Fig. 7C), diseases of programmed cell death (Fig. 7D), apoptosis (Fig. 7E) and other pathways (Table S9). The differences in the c2.cp.all.v2022.1.Hs.symbols.gmt gene set between the high- and low-risk score groups were analyzed by GSVA (Fig. 7F-G, Table S10). The top 5 logFC positive and negative pathways that met the requirements are displayed. Fig. 7GSEA and GSVA enrichment analysis. (A) GSEA enrichment analysis of genes between different groups (high/low risk score group) in the TCGA-LUAD dataset revealed four main biological characteristics. (B–G). Genes in the TCGA-LUAD dataset were significantly enriched in the cell cycle checkpoint (B), glycolysis (C), programmed cell death (D), and apoptosis (E) pathways. (F–G) Heatmap (F) and group comparison chart (G) showing the GSVA enrichment analysis of genes between different groups. TCGA, The Cancer Genome Atlas; LUAD, lung adenocarcinoma; GSEA, gene set enrichment analysis; GSVA, gene set variation analysis GSEA and GSVA enrichment analysis. (A) GSEA enrichment analysis of genes between different groups (high/low risk score group) in the TCGA-LUAD dataset revealed four main biological characteristics. (B–G). Genes in the TCGA-LUAD dataset were significantly enriched in the cell cycle checkpoint (B), glycolysis (C), programmed cell death (D), and apoptosis (E) pathways. (F–G) Heatmap (F) and group comparison chart (G) showing the GSVA enrichment analysis of genes between different groups. TCGA, The Cancer Genome Atlas; LUAD, lung adenocarcinoma; GSEA, gene set enrichment analysis; GSVA, gene set variation analysis Multifactor Cox regression model for mutation analysis and immune assessment We analyzed the mutations of 13 PMGs in the low-risk score group and the high-risk score group of LUAD patients from the TCGA-LUAD dataset and visualized the results (Fig. 3S.A-B). The mutation types of 13 PMGs in LUAD patients were mainly single nucleotide polymorphisms (SNPs) in both groups. Additionaly, we analyzed the CNVs of 13 PMGs in LUAD patients in the TCGA-LUAD dataset. GISTIC 2.0 was used to analyze and visualize the results (Fig. 3S.E-H). The results showed that 13 PMGs exhibited gene amplification and gene deletion in the low- and high-risk score groups of LUAD patients. We examined the EGFR mutation sites in LUAD patient samples from the TCGA-LUAD dataset, which specifically targets the EGFR gene, using the cBioPortal database. The obtained results are shown in Fig. 4S.A. The analysis revealed that EGFR gene mutations occurred in 16% of the total samples within the TCGA-LUAD dataset, with mutations primarily falling into 7 distinct types. Additionally, we conducted a detailed investigation of the specific mutation sites of EGFR in the TCGA-LUAD dataset and determined that the predominant mutation site was L858R (Fig. 4S.B). Furthermore, the StromalScore, ImmuneScore, ESTIMATEScore, and Tumor Purity score in both the high- and low-risk score groups of the PMGs within the TCGA-LUAD dataset were analyzed and are displayed in Fig. 4S.C-F; the results revealed statistically significant differences between the various groups in terms of the StromalScore, ImmuneScore, ESTIMATEScore, and Tumor Purity score. Notably, a negative correlation was observed between the risk score and the stromal score (R=-0.17, P = 6.8e-05), immune score (R=-0.081, P = 0.063), and ESTIMATEScore (R = -0.14, P = 0.0018), while a positive correlation was observed with tumor purity (R = 0.14, P = 0.0018, Fig. 4S.H-J). Clinical correlation analysis of the PMGs The proportions of T-, N-, M-, pathological-stage, sex, age, OS, DSS, and PFI differences between the high- and low-risk score groups of LUAD patients are displayed through stacked histograms (Fig. 8). The results showed that the T-, N-, pathological-stage, sex, OS, DSS, and PFI populations in the high- and low-risk score groups of LUAD patients within the TCGA-LUAD dataset were significantly different. Fig. 8Clinical correlation analysis of the differential module gene prognostic model. The stacked histogram results of clinical T stage (A), N stage (B), M stage (C), age (D), sex (E), pathological stage (F), OS event (G), DSS event (H), and PFI event (I) of the high/low risk score group of the TCGA-LUAD dataset of the differential module gene prognostic model. TCGA, The Cancer Genome Atlas; LUAD, Lung Adenocarcinoma; OS, Overall Survival; DSS, Disease-Specific Survival; PFI, Progression-Free Interval Clinical correlation analysis of the differential module gene prognostic model. The stacked histogram results of clinical T stage (A), N stage (B), M stage (C), age (D), sex (E), pathological stage (F), OS event (G), DSS event (H), and PFI event (I) of the high/low risk score group of the TCGA-LUAD dataset of the differential module gene prognostic model. TCGA, The Cancer Genome Atlas; LUAD, Lung Adenocarcinoma; OS, Overall Survival; DSS, Disease-Specific Survival; PFI, Progression-Free Interval Immune infiltration analysis of PMGs We conducted an analysis using ssGSEA to investigate the abundance of 28 immune cell types and the variations in immune infiltration among the high- and low-risk score groups of LUAD patients within the TCGA-LUAD dataset (Fig. 9A) and found that 11 types of immune cells exhibited significant differences in infiltration abundance between these two groups (P < 0.05). In addition, a notable positive correlation was observed between the infiltration levels of 11 immune cell types and others in both the low- and high-risk score groups (Fig. 9B-C). Moreover, 11 immune cell types were found to be positively correlated with BIRC3, HLA-DQB2, KCTD12, and NT5E in both the low- and high-risk score groups (Fig. 9D-E). Conversely, these 11 immune cell types exhibited a negative correlation with CPLX2, GSTA1, and ID1 in the high-risk score group. Fig. 9ssGSEA immune infiltration analysis of the high- and low-risk score groups in the DMG prognostic model of the TCGA-LUAD dataset. (A) ssGSEA immune infiltration analysis results of the high- and low-risk score groups of the differential module gene prognostic model in the TCGA-LUAD dataset. (B-C) Correlation analysis of immune cell infiltration in the TCGA-LUAD cohort in the low-risk score group (B) and high-risk score group (C). (D-E) Correlation point plots between immune cells and prognostic module genes in the low-risk score group (D) and high-risk score group (E) of the TCGA-LUAD dataset. TCGA, The Cancer Genome Atlas; LUAD, lung adenocarcinoma; ssGSEA: single-sample gene-set enrichment analysis ssGSEA immune infiltration analysis of the high- and low-risk score groups in the DMG prognostic model of the TCGA-LUAD dataset. (A) ssGSEA immune infiltration analysis results of the high- and low-risk score groups of the differential module gene prognostic model in the TCGA-LUAD dataset. (B-C) Correlation analysis of immune cell infiltration in the TCGA-LUAD cohort in the low-risk score group (B) and high-risk score group (C). (D-E) Correlation point plots between immune cells and prognostic module genes in the low-risk score group (D) and high-risk score group (E) of the TCGA-LUAD dataset. TCGA, The Cancer Genome Atlas; LUAD, lung adenocarcinoma; ssGSEA: single-sample gene-set enrichment analysis CIBERSORT analysis revealed significant differences in the abundances of nine types of immune cells between the high- and low-risk score groups (P < 0.05, Fig. 10A). The correlations between the infiltration abundance of these immune cells and the risk score groups are illustrated in Fig. 10B-C. Moreover, a notable correlation was observed between the infiltration abundance of the nine immune cell types and the 13 PMGs in the TCGA-LUAD dataset for both the low- and high-risk score groups (P < 0.05, Fig. 10D-E). Fig. 10CIBERSORT immune infiltration analysis of the high- and low-risk score groups in the differential module gene prognostic model of the TCGA-LUAD dataset. (A) CIBERSORT immune infiltration analysis results of the high- and low-risk score groups of the differential module gene prognostic model of the TCGA-LUAD dataset. (B-C) Correlation analysis results of immune cell infiltration abundance in the low-risk score group (B) and high-risk score group (C) of the differential module gene prognostic model in the TCGA-LUAD dataset. (D-E) Correlation point plots between immune cells and prognostic module genes in the low-risk score group (D) and high-risk score group (E) of the differential module gene prognostic model in the TCGA-LUAD dataset. TCGA, The Cancer Genome Atlas; LUAD, Lung adenocarcinoma CIBERSORT immune infiltration analysis of the high- and low-risk score groups in the differential module gene prognostic model of the TCGA-LUAD dataset. (A) CIBERSORT immune infiltration analysis results of the high- and low-risk score groups of the differential module gene prognostic model of the TCGA-LUAD dataset. (B-C) Correlation analysis results of immune cell infiltration abundance in the low-risk score group (B) and high-risk score group (C) of the differential module gene prognostic model in the TCGA-LUAD dataset. (D-E) Correlation point plots between immune cells and prognostic module genes in the low-risk score group (D) and high-risk score group (E) of the differential module gene prognostic model in the TCGA-LUAD dataset. TCGA, The Cancer Genome Atlas; LUAD, Lung adenocarcinoma Construction of LUAD-related subtypes Two LUAD subtypes were ultimately identified (Fig. 11A): subtype 1 contained 351 patients, while subtype 2 contained 184 patients. The different cluster numbers observed in the consistent clustering results of the cumulative distribution function (CDF) plot and the area under the CDF curve delta plot are displayed in Fig. 11B-C. For unsupervised clustering with k = 2 clusters, the TCGA-LUAD dataset yielded the most robust consistent clustering results. Moreover, principal component analysis (PCA) highlighted significant differences between the two LUAD subtypes (Fig. 11D). The expression levels of the 13 PMGs exhibited noteworthy differences between the two LUAD subtypes (P < 0.05, Fig. 11E-F). We also found that the StromalScore, ImmuneScore, ESTIMATEScore, and Tumor Purity score were significantly different between the two LUAD subtypes (P < 0.05, Fig. 11G-J). Fig. 11Construction of LUAD-related subtypes. (A) Consistent clustering (K = 2) result diagram of the TCGA-LUAD dataset. (B-C) Consistent clustering cumulative distribution function (CDF) plot (B) for different cluster numbers in consistent clustering, and area under the CDF curve delta plot (C). (D) PCA results of two LUAD subtypes in the TCGA-LUAD dataset. (E) Heatmap of prognostic module genes between two LUAD subtypes in the TCGA-LUAD dataset. (F) Group comparison diagram of prognostic module genes between two LUAD subtypes in the TCGA-LUAD dataset. The group comparison chart shows the stromal score (G), immune score (H), ESTIMATEScore (I), and tumor purity score (J) between the two LUAD subtypes in the TCGA-LUAD dataset. TCGA, The Cancer Genome Atlas; LUAD, Lung adenocarcinoma; CDF, Cumulative distribution function Construction of LUAD-related subtypes. (A) Consistent clustering (K = 2) result diagram of the TCGA-LUAD dataset. (B-C) Consistent clustering cumulative distribution function (CDF) plot (B) for different cluster numbers in consistent clustering, and area under the CDF curve delta plot (C). (D) PCA results of two LUAD subtypes in the TCGA-LUAD dataset. (E) Heatmap of prognostic module genes between two LUAD subtypes in the TCGA-LUAD dataset. (F) Group comparison diagram of prognostic module genes between two LUAD subtypes in the TCGA-LUAD dataset. The group comparison chart shows the stromal score (G), immune score (H), ESTIMATEScore (I), and tumor purity score (J) between the two LUAD subtypes in the TCGA-LUAD dataset. TCGA, The Cancer Genome Atlas; LUAD, Lung adenocarcinoma; CDF, Cumulative distribution function Constructed PPI network, mRNA‒miRNA network, and mRNA-TF interaction network The analysis of 13 PMGs was conducted using the STRING database, resulting in the construction of a PPI network (Fig. 12A). The functional similarity analysis results indicated that within the group of 11 hub genes, TCIM exhibited the highest functional similarity with the other key genes (Fig. 12B). According to the GeneMANIA database analysis results, the 9 key genes primarily demonstrated coexpression and physical interactions with other genes (Fig. 12C). A set of 11 hub genes (BIRC3, CCT6A, CPLX2, GLCCI1, GSTA1, ID1, KCTD12, MUC15, NOTUM, NT5E, and TCIM) and 58 miRNAs were identified for the establishment of the mRNA‒miRNA interaction network (Fig. 12D). The detailed mRNA‒miRNA interaction relationships are provided in Table S11. Additionally, 7 hub genes (BIRC3, CCT6A, GLCCI1, ID1, KCTD12, NOTUM, and TM4SF4) and 33 TFs were screened to construct the mRNA-TF interaction network, which is visualized in Fig. 12E. The detailed mRNA-TF interaction relationships are illustrated in Table S12. Fig. 12Construction of the protein‒protein interaction network (PPI network). (A) Prognosis module gene PPI network. (B) The results of functional similarity analysis of key genes. (C) PPI network of key genes identified via GeneMANIA database analysis. (D-E) mRNA‒miRNA (D) and mRNA-TF (E) interaction networks of the hub genes. The sky-blue oval block in the mRNA‒miRNA (D) interaction network is mRNA; the light yellow diamond-shaped block is miRNA. The sky-blue oval block in the mRNA-TF (E) interaction network is mRNA; the light green triangle block is transcription factor (TF). TCGA, The Cancer Genome Atlas; LUAD, lung adenocarcinoma; PPI network, protein‒protein interaction network; TF, transcription factor Construction of the protein‒protein interaction network (PPI network). (A) Prognosis module gene PPI network. (B) The results of functional similarity analysis of key genes. (C) PPI network of key genes identified via GeneMANIA database analysis. (D-E) mRNA‒miRNA (D) and mRNA-TF (E) interaction networks of the hub genes. The sky-blue oval block in the mRNA‒miRNA (D) interaction network is mRNA; the light yellow diamond-shaped block is miRNA. The sky-blue oval block in the mRNA-TF (E) interaction network is mRNA; the light green triangle block is transcription factor (TF). TCGA, The Cancer Genome Atlas; LUAD, lung adenocarcinoma; PPI network, protein‒protein interaction network; TF, transcription factor Construction of the multivariate risk score prognostic model To assess the prognostic significance of the risk score within the TCGA-LUAD dataset, we conducted univariate and multivariate Cox regression analyses involving the DMG prognostic model and pertinent clinical variables, including T stage, N stage, M stage, pathological stage, sex, and age, to construct a multifactor prognostic risk score model (Table S13). The findings are presented in a forest plot (Fig. 13A). Comprehensive clinical information for all patients with LUAD in the TCGA-LUAD dataset can be found in Table S14. Fig. 13Preconstruction of the multifactor risk score prognostic model. (A) Forest plot of the differential module gene prognostic model in the TCGA-LUAD dataset: risk score combined with univariate Cox regression analysis of clinical variables. (B) Nomogram of the multifactor risk score prognostic model. (C-E) 1-year (C), 3-year (D), and 5-year (E) calibration curves of the multifactor risk score prognostic model nomogram analysis. (F-H) 1-year (F), 3-year (G), and 5-year (H) DCA plots of the multifactor risk score prognostic model. The cancer genome atlas; LUAD, lung adenocarcinoma; DCA, decision curve analysis \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}\text{R}\text{i}\text{s}\text{k}\text{s}\text{c}\text{o}\text{r}\text{e}&=\text{C}\text{O}\text{X}-\text{R}\text{i}\text{s}\text{k}\text{s}\text{c}\text{o}\text{r}\text{e}*0.358+\text{S}\text{t}\text{a}\text{g}\text{e}\_\text{T}\text{T}2*0.106\\&\quad+\text{S}\text{t}\text{a}\text{g}\text{e}\_\text{T}\text{T}3*0.799+\text{S}\text{t}\text{a}\text{g}\text{e}\_\text{T}\text{T}4*0.362\\&\quad+\text{S}\text{t}\text{a}\text{g}\text{e}\_\text{N}\text{N}1*0.764+\text{S}\text{t}\text{a}\text{g}\text{e}\_\text{N}\text{N}2*0.82\\&\quad+\text{S}\text{t}\text{a}\text{g}\text{e}\_\text{N}\text{N}3*-14.5+\text{S}\text{t}\text{a}\text{g}\text{e}\_\text{M}\text{M}1*0.635\\&\quad+\text{S}\text{t}\text{a}\text{g}\text{e}\_\text{M}\text{M}\text{X}*0.0139-0.4821\end{aligned}$$\end{document} Preconstruction of the multifactor risk score prognostic model. (A) Forest plot of the differential module gene prognostic model in the TCGA-LUAD dataset: risk score combined with univariate Cox regression analysis of clinical variables. (B) Nomogram of the multifactor risk score prognostic model. (C-E) 1-year (C), 3-year (D), and 5-year (E) calibration curves of the multifactor risk score prognostic model nomogram analysis. (F-H) 1-year (F), 3-year (G), and 5-year (H) DCA plots of the multifactor risk score prognostic model. The cancer genome atlas; LUAD, lung adenocarcinoma; DCA, decision curve analysis Following the nomogram analysis, we assessed the prognostic capability of the model and constructed a nomogram (Fig. 13B). Our findings indicated that the inclusion of N stage variables in the multivariate risk score prognostic model demonstrated significantly superior utility compared to the inclusion of other variables. Furthermore, we conducted 1-year, 3-year, and 5-year prognostic calibration analyses on the nomogram of the multivariate risk score prognostic model, followed by the generation of a calibration curve (Fig. 13C-E). Subsequently, decision curve analysis (DCA) was employed to evaluate the clinical utility of the developed multivariate risk score prognostic model at 1 year (Fig. 13F), 3 years (Fig. 13G), and 5 years (Fig. 13H). Our results indicated that the clinical predictive efficacy of the constructed multivariate Cox regression model ranked highest at 5 years, followed by 3 years and then 1 year (Fig. 13C-H). Validation of the expression levels of selected PMGs To further validate the expression patterns of the 13 selected PMGs, we detected their mRNA expression by qRT‒PCR in 8 samples of LUAN cancer tissue specimens and matched paracancerous tissue. The results showed that the mRNA expression level of CCT6A was significantly greater in LUAD tissue than in paracancerous tissue (P < 0.05). Conversely, the mRNA expression levels of BIRC3, CPLX2, ID1, KCTD12, MUC15, and NOTUM were lower in LUAD tissue than in paracancerous tissue (P < 0.05, Fig. 14A-H). The mRNA expression profiles of CCT6A and KCTD12 in LUAN cancer tissue specimens and matched paracancerous tissue specimens were consistent with those in the four LUAD datasets analyzed. Additionaly, we detected the mRNA expression levels of CCT6A and KCTD12 in HBE, H1975, PC9, H1975-OR, and PC9-OR cells. Among them, the mRNA expression level of CCT6A was highest in the H1975-OR and PC9-OR cell lines, followed by the H1975 and PC9 cell lines, and lowest in HBE cells. Conversely, the mRNA expression level of KCTD12 was found to be highest in HBE cells, followed by H1975 and PC9 cells, and lowest in H1975-OR and PC9-OR cells (Fig. 14I-J). Fig. 14Validation of the expression levels of 13 selected PMGs. (A–H) The expression levels of 13 PMGs and EGFR in LUAN cancer tissue specimens (T) and matched paracancerous tissue specimens (P) from Patient 825 (A), Patient 828 (B), Patient 829 (C), Patient 831 (D), Patient 832 (E), Patient 834 (F), Patient 837 (G) and Case 838 (H). The expression levels of KCTD12 (I) and CCT6A (J) in the HBE, PC9, PC9-OR, H1975 and H1975-OR cell lines. (*P < 0.05, **P < 0.01, ***P < 0.001, and ns: not significant) Validation of the expression levels of 13 selected PMGs. (A–H) The expression levels of 13 PMGs and EGFR in LUAN cancer tissue specimens (T) and matched paracancerous tissue specimens (P) from Patient 825 (A), Patient 828 (B), Patient 829 (C), Patient 831 (D), Patient 832 (E), Patient 834 (F), Patient 837 (G) and Case 838 (H). The expression levels of KCTD12 (I) and CCT6A (J) in the HBE, PC9, PC9-OR, H1975 and H1975-OR cell lines. (*P < 0.05, **P < 0.01, ***P < 0.001, and ns: not significant) Discussion LUADs are an important part of NSCLC and seriously threaten human health. Despite recent advancements in the management of NSCLC, including molecular targeted therapy and immunotherapy, the prognosis for NSCLC patients is still unsatisfactory. Molecular targeted therapy, particularly EGFR-TKI monotherapy, has emerged as a pivotal treatment modality and a foundational element in the combined therapeutic approach for NSCLC patients with EGFR mutations. However, the development of acquired drug resistance limits the long-term efficacy of EGFR-TKIs, leading to treatment failure and poor prognosis. Therefore, understanding the underlying mechanisms of drug resistance, the tumor microenvironment, tumor heterogeneity, and the intricate interplay between EGFR and various molecular players is crucial for exploring novel treatment strategies to improve patient outcomes and guide personalized therapeutic approaches. The objective of this study was to investigate the potential mechanisms of action of EGFR in LUAD and provide new ideas and potential targets for its diagnosis and treatment. Previous research has demonstrated that CCT6A is located in close proximity to EGFR on chromosome 7p11. Additionally, in both LUAD and glioblastoma patients, CCT6A and EGFR were found to be coamplified and coexpressed. Notably, CCT6A amplification was associated with unfavorable outcomes (Hallal et al. 2019; Wang et al. 2022). In this study, we identified 13 PMGs—BIRC3, CCT6A, CPLX2, GLCCI1, GSTA1, HLA-DQB2, ID1, KCTD12, MUC15, NOTUM, NT5E, TCIM, and TM4SF4—and among them, the expression levels of CCT6A and NT5E were upregulated, whereas the expression level of KCTD12 was downregulated in the LUAD group compared to the normal group. Furthermore, CCT6A and KCTD12 displayed high accuracy in differentiating between the two groups, indicating their potential diagnostic significance. These findings imply that these PMGs may play a crucial role in the occurrence and development of LUAD and might become potential biomarkers for diagnosis and prognostic evaluation. Ying Z et al. reported that CCT6A suppressed SMAD2 function and promoted metastasis in NSCLC cells (Ying et al. 2017). Similarly, Zhu M et al. reported that high expression of CCT6A significantly decreased the survival of patients with NSCLC (Zhu et al. 2017). In this study, we also found that CCT6A was significantly overexpressed in LUAD tissue compared to paracancerous tissue, correlating with poor OS, while KCTD12 was downregulated in LUAD and associated with better OS, as validated by qRT-PCR and Kaplan-Meier survival analysis. Previous studies have also indicated a significant decrease in the mRNA and protein expression levels of KCTD12 in LUAD (Shi et al. 2022). These findings indicated that CCT6A and KCTD12 are likely involved in sensitivity to targeted therapy and are promising candidates for the diagnosis of LUAD. GO analysis revealed that 13 PMGs were enriched in a series of functions related to immune response, signal transduction, and biological pathway regulation in terms of BP and MF. KEGG analysis revealed enrichment of signaling pathways associated with platinum drug resistance and the Hippo signaling pathway, and GSEA revealed significant enrichment of genes related to cell cycle checkpoints, glycolysis, programmed cell death, and apoptosis pathways. These results offer crucial insights to enhance our understanding of the functional roles of these genes in the initiation and progression of LUAD, which could aid in the discovery of potential biomarkers and therapeutic targets. Luo L et al. demonstrated that KCTD12 effectively suppressed the growth of human uveal melanoma OCM-1 cells and delayed the progression from G2/M to G1 phase in OCM-1 cells overexpressing KCTD12 (Luo et al. 2017). Adefovir dipivoxil enhanced the sensitivity of colon cancer cells to vemurafenib by disrupting the KCTD12-CDK1 interaction, leading to G2 phase arrest in the cell cycle (Yang et al. 2019). KCTD12 binds to CDC25B and activates CDK1 and Aurora A to promote the G2/M transition and enhance tumorigenesis (Zhong et al. 2017). Zeng W et al. (Zeng et al. 2022a) reported that CCT6A knockdown suppressed cell growth, induced apoptosis, and halted cell cycle progression at the G0/G1 phase in OS cells. These findings indicated that KCTD12 and CCT6A likely play crucial roles in cycle checkpoint and apoptosis pathways for targeted therapy in LUAD. We also constructed a protein–protein interaction (PPI), mRNA–miRNA, and mRNA–TF interaction network to predict the miRNAs and TFs that interact with the hub genes. Seven hub genes (BIRC3, CCT6A, GLCCI1, ID1, KCTD12, NOTUM, and TM4SF4), 33 TFs, and a total of 59 pairs of mRNA‒miRNA interactions were identified. Notably, interactions such as those between CCT6A and miR-152-3p, as well as between CCT6A and the miR-148a/152 regulatory network, were established. These insights enhance our understanding of the roles and interaction networks of these genes in lung adenocarcinoma (LUAD) development, offering valuable leads for further exploration into the molecular mechanisms underlying LUAD. Previous studies have reported (Li et al. 2020) that the regulatory network involving miR-152-3p and CCT6A could play a significant role in glioma diagnosis and treatment. Specifically, NEAT1 was found to upregulate CCT6A expression in glioma tissues and cell lines, and this effect was reversed by miR-152-3p. Additionally, in colon cancer, research by Peng et al (Peng et al. 2022) revealed that CCT6A was upregulated while miR-148a/152 was downregulated in tumor tissues. Further investigations demonstrated that increasing miR-148a/152 expression suppressed migration, invasion, and proliferation, whereas reducing CCT6A expression induced cell apoptosis in colon cancer. These collective findings suggested that CCT6A may have implications for the diagnosis and treatment of LUAD. Although this study successfully identified prognosis module genes (PMGs) associated with LUAD and validated their expression patterns using qRT-PCR, several limitations exist. Firstly, the study did not conduct in vivo (e.g., mouse models) or in vitro experiments (e.g., cell culture) to explore the functions of these potential targets, limiting our understanding of PMG’ biological roles and mechanisms in LUAD. Secondly, the small sample size of only eight LUAD samples used for expression validation may affect the generalizability and statistical significance of the results, not fully reflecting PMGs’ performance in larger cohorts. Additionally, reliance on various public databases for data analysis, such as TCGA and GEO, introduces heterogeneity that may impact result interpretation and limit conclusions. Lastly, due to time and resource constraints, we were unable to comprehensively investigate the functions of all identified PMGs across different LUAD subtypes, potentially leaving important biological characteristics and mechanisms undiscovered. To address these limitations, future research should include larger sample validations, functional experiments, and the integration of multi-omics data to further elucidate the relationship between EGFR expression and the role of PMGs in LUAD. Conclusions In summary, we identified DEGs and DMGs related to the expression of EGFR in the high/low and LUAD/normal groups by comprehensive bioinformatics analysis of LUAD tissues. In total, 13 PMGs (BIRC3, CCT6A, CPLX2, GLCCI1, GSTA1, HLA-DQB2, ID1, KCTD12, MUC15, NOTUM, NT5E, TCIM, and TM4SF4) were screened and verified to be able to predict survival prognosis. Notably, CCT6A and KCTD12 exhibited high diagnostic accuracy in LUAD. Furthermore, in vivo and in vitro, the expression levels of CCT6A and KCTD12 were consistent among the four LUAD datasets. These findings suggest that KCTD12 and CCT6A might be candidate biomarkers for diagnostic performance and could guide individualized therapy for EGFR-TKI-resistant LUAD patients. This study revealed that CCT6A and KCTD12 are dependable predictors of LUAD efficacy, and the targeting of CCT6A and KCTD12 represents a novel avenue for further research and therapeutic approaches in LUAD. Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2 Supplementary Material 3 Supplementary Material 3 Supplementary Material 4 Supplementary Material 4 Supplementary Material 5 Supplementary Material 5 Supplementary Material 6 Supplementary Material 6 Supplementary Material 7 Supplementary Material 7 Supplementary Material 8 Supplementary Material 8 Supplementary Material 9 Supplementary Material 9 Supplementary Material 10 Supplementary Material 10 Supplementary Material 11 Supplementary Material 11 Supplementary Material 12 Supplementary Material 12 Supplementary Material 13 Supplementary Material 13 Supplementary Material 14 Supplementary Material 14 Supplementary Material 15 Supplementary Material 15 Supplementary Material 16 Supplementary Material 16 Supplementary Material 17 Supplementary Material 17 Supplementary Material 18 Supplementary Material 18