With the advancement of diagnosis and treatment methods, the death rate of lung cancer continues to decrease significantly, lung cancer remains the most fatal neoplastic disease all over the world, nowadays(18). Among which, LUAD is the most common histological type and treatment of LUAD has made important progress in chemotherapy, targeted therapy and other treatment methods. There are still a series of problems, such as drug resistance and identification of novel biomarkers(19, 20). Therefore, the development of biomarkers is particularly important in the diagnosis and treatment of LUAD.
EMT is a complex cellular process, during which the epithelial features decrease and epithelial cells obtain mesenchymal phenotype(21). At the cellular level, activation of EMT leads to the loss of cell polarity, the interruption of cell-cell junctions, the degradation of basement membrane and the reorganization of extracellular matrix (ECM)(22). At the molecular level, activation of EMT repress the expression of epithelial cadherin (E-cadherin) and certain cytokeratins and active the expression of mesenchymal markers, such as neural cadherin (N-cadherin), vimentin, fibronectin and β1 and β3 integrins(23). The activation of EMT and the above process are orchestrated by EMT-inducing transcriptions factors (EMT-TFs). EMT is closely related to tumor malignant progression, which includes tumor-initiating properties, motility and resistance to both chemotherapy and immunotherapy(24–26). These brought EMT to our attention.
However, the activation of EMT rarely as a spontaneous process in carcinoma cells. It is reported that signals from the tumor microenvironment (TME) act on cancer cells to induce the expression of EMT-TF, thereby coordinating the expression of various components in the EMT-related program(22). A large number of studies verified that the activation of EMT were closely associated to TME-associated cells, including immune cells, stromal cells, cancer-associated fibroblasts (CAFs)(11, 22). CAFs coordinately induce the activation of EMT in carcinoma cells by secreting a series of cytokines and growth factors, such as TGFβ, IL-6, EGF, VEGF and HGF(27–30). The activated effector T cells facilitate the EMT activation by releasing cytokines such as IL-6, TNF,TGFβ and other unknown mechanisms(31). Besides, TAMs secrete an array of factors to induce EMT(32–35). However, the effect of TME on EMT is not unilateral. Acting reciprocally, the tumor cells with EMT-related markers can regulate the activity of cellular components in the tumor microenvironment(36). Under the influence of EMT-TFs, the cellular components with antitumor functions, such as cytotoxic CD8 + T cells and etc. are overwhelmed by immunosuppressive cells (37–39). In addition to changes in cell composition, immunomodulatory markers on cell surface were also altered which induce immune evasion (40, 41). In summary, through interaction, EMT and TME influence each other and play an important role in the malignant progression of tumors.
A series of studies indicate the potential of EMT markers as tumoral high-grade malignancy markers in NSCLC and other malignant tumors(42, 43). However, due to the complexity, continuity and interaction with other phenotypes, a recent review indicated that one or a small number of markers maybe hard to assess EMT status. A combination of markers is more reliable(21). Therefore, the junction of EMT and TME markers is necessary for clinical oncology.
In our study, we selected 1184 EMT associated genes from dbEMT 2.0 website (http://dbemt.bioinfo-minzhao.org/), 102 immune associated genes and 48 stromal associated genes from published datasets to build our signature. We performed univariate Cox proportional hazard regression analysis on the 1334 genes and selected genes significantly associated to overall survival (OS) of TCGA-LUAD patients. After lasso regression analysis and multivariate Cox regression analysis, 23 genes were selected to build our EMT, immune and stromal associated genes, including KL, ECT2, FBLN5, TIMELESS, TERT, TYMS, FHL2, MNDA, TIMP1, IL11, CDH19, AOC4P, ETV1, KRT18, SATB2, EFNB2, TNFRSF11A, MEOX2, MEG3, ANGPTL4, WNT1, L1CAM, FURIN. The risk score was calculated of each patient to predict OS of LUAD patients. The K-M curves showed that the patients with high risk had a poorer outcome. More importantly prognostic power of the 23-gene signature then was validated in GEO cohorts. Gene signature is a powerful classifier to and has been widely used to predict prognosis of patients(44). In published research, different EMT-associated signatures were built in head and neck squamous cell carcinoma and pan-cancer, which were highly relevant to TME markers, respectively(45, 46). However, as far as we know, there is no signature based on EMT and TME markers.
Bioinformatic analysis demonstrated that the 23-gene signature mainly related to cell cycle and cell division, ubiquitin-mediated proteolysis and cell adhesion molecules. Interestingly, a large number of genes from 23-EIS signature are related to cell cycle, division and DNA replication. For example, Epithelial cell transforming 2 (ECT2), an activator of Rho family GTPases, was involved in a variety of cell process, including adhesion, migration, cell division, growth, and centromere maintenance(47). As a oncoprotein, ECT2 is reported to play an important role in the development of multiple cancers including LUAD(48). Moreover, a series of genes, such as TIMELESS(49), TERT(50) and FHL2(51) were also involved in cell cycle process, more than EMT, immune and stromal.
In addition, several key molecules of the EIS signature were also involved in classic signal pathways. EFNB2, cell surface transmembrane ligand for Eph receptors, is involved in Eph receptors and ephrins signaling, which affects tumor growth, invasion, metastasis and angiogenesis(52).Moreover, TIMP1(metalloproteinase inhibitor 1) binds to matrix metalloproteinases (MMPs) and inactivates them, which is closely associated to cell motility, invasion and angiogenesis. TIMP1 also acts as a growth factor to regulate cell differentiation, migration and cell death directly(53).
Importantly, we established a nomogram to predict the 1-, 3- and 5-year OS of LUAD patients. By integrating multiple risk factors, nomogram is considered as a reliable tool for risk prediction (54). As a predictive model, each factor associated with OS of patients was included into the nomogram, such as age, gender, and TNM stage. Besides, risk score based on gene signature can also incorporated as a predictive factor. According to the impact of its risk on survival, each factor was assigned points proportionally. In our study, after univariate analysis, the significant clinical factors (age, stage) and risk score was included into the nomogram. Due to the combination of the multi-gene signature, the predictive performance of the nomogram maybe more accurate than traditional nomogram with only clinical features. Since EMT is related to tumor metastasis and drug resistance, in addition to OS, this signature may have a potential predictive effect on metastasis possibility and medication nonadherence. Moreover, the detection method is simple and feasible, which is suitable to validate our gene-signature in biopsies.
However, certain limitations existed in our study. For example, the clinical features of the public cohort were not complete enough, such as neoadjuvant treatment information and smoking status of patients. Although the 23-gene signature can predict OS of patients accurately based on gene expression level, a deeper research of gene, such as gene mutation, was not included in our research. And also, the possibility of the genes being potential therapeutic targets was not clarified.
In conclusion, we developed and verified a 23-gene signature of EMT, immune and stromal based on TCGA and GEO datasets. These genes are all of great importance and expected to be new targets. Besides, a nomogram based on these genes and clinical features was developed, which could accurately predict a 1-, 3- and 5- year survival time of LUAD patients. More importantly, these findings provide new ideas for systematically evaluation of prognosis of LUAD patients and effective personalized treatment.