With the deepening of the study of COAD, it has been confirmed that immunotherapy can significantly improve the prognosis of patients with COAD. The results of relevant clinical trials also showed that multiple immune regulatory pathways in the tumor immune microenvironment might play an important role in the survival of tumor cells, and could greatly influence the response of relevant immunotherapy (20–22). In this study, Our main purpose was to establish an effective and stable prediction model based on the TME related genes of COAD. As we already know that the heterogeneity of TME is very important in predicting the prognosis of tumor patients and the effectiveness of targeted therapy (23–25). At the same time, due to the heterogeneity and immune correlation of COAD, the genetic information of TME has become an important link in optimizing the treatment related to COAD. We got DEGs according to the TME gene profile of COAD and constructed prognostic indicators through the association of these DEGs in our study. After various ways of validation, we found that this indicator was superior to the commonly used clinical indicators, which could provide lower sampling cost and better repeatability of prognosis prediction methods for the individualized treatment of patients with COAD.
NMF method is a promising new clustering method. In this study, we used this method to distinguish two molecular clusters and we found that C1 and C2 had significant differences in immune scores of TME infiltrating cell types. What’s more, the correlation analysis about OS and PFS did indicate that C1 and C2 were significantly different as well. All of these proved the heterogeneity of TME in COAD. And we further constructed the prognostic characteristics of patients with COAD based on 15 TME related genes, which were verified using univariate and lasso analysis.
Patients were turned into low-risk group and high-risk group on the basis of risk score for further research and analysis. It was not difficult to find that the higher risk score COAD patients had, the worse prognosis they would experience. Hierarchical Analysis further showed that the risk value of patients increased with the progression of COAD. The AUC value of the prognostic features constructed by us was basically greater than 0.65, which confirmed its effectiveness in predicting the prognosis of patients with COAD. Previous studies have proved that overexpression of FOXD1, TNM staging and tumor differentiation were independent prognostic risk factors for OS and DFS (26). Therefore, patients with high FOXD1 expression were more likely to have poorer overall survival and disease-free survival. Quantitative RT-PCR analysis of EPHB4 gene expression showed that compared with matched normal tissues from the same patient, the expression of EPHB4 gene increased in 82% of COAD tumor samples, which proved that the amplification of EPHB4 gene might also be closely related to the poorer prognosis of COAD (27).
As we all know, nomogram is a simple and accurate evaluation tool in clinical work. Our nomogram could more accurately predict COAD patients’ prognosis, and the ROC curve, DCA curve and calibration curve had fully proved its effectiveness and accuracy. Besides, univariate regression and multivariate Cox regression analysis together with relevant clinicopathological parameters did prove that our risk score was an independent prognostic factor for COAD patients. The risk scores of clinical subgroups of patients with COAD were also different, and the patients with high TNM stage had poorer prognosis. Our results proved that risk score could provide flexible risk stratification for patients with COAD. We further compared our risk prognosis characteristics with the prognosis signature established by Qiu-Yun Luo et al (28), Changjing Wang et al (29) and Xu Wang et al (30), and found that its overall performance was better than the latter three.
In recent years, more and more attention has been paid to immune checkpoints in the research related to tumor immunotherapy (31–32). At present, we have found that tumor cells might generate immunosuppressive environment through lots of mechanisms, including increasing the expression of PDCD1, CD274 and CTLA4 (immunosuppressive molecules) in TME (33–35) and the occurrence of immune escape (36). The levels of expression for PDCD1 and CTLA4 genes were up-regulated with the increase of risk value, which might lead to poorer prognosis in COAD patients with high risk. In addition, the related risk value was positively correlated with the EMT related gene (FAP). Generally speaking, a greater degree of immune infiltration might enhance the immune defense of low-risk patients to a certain extent, thus indicating a good prognosis. However, the expression of EMT gene in this study increased with the increase of risk score, which might lead to higher risk of metastasis and worse prognosis in patients with high risk of COAD.
The parameters of clinical staging often need to be obtained in combination with imaging and pathological results, and then evaluated and analyzed by professionals. Our gene signature only needed to score specific genes after sequencing. The scoring process did not require the participation of professionals, but just required simple calculation, which could greatly improve the repeatability and accessibility of risk scoring with COAD patients. The information of gene could be obtained from a litttle tissues or even exocrines, which might greatly reduce the cost to obtain COAD prognostic indicators in clinical work. In this research, the risk score we established was proved to be significantly related to the prognosis of patients with COAD, and the area under the AUC curve of our signature was significantly larger than that of the other three. In low-risk group COAD patients, the net profit was also better than the cyclical indicators. Therefore, our model had more advantages in clinical work and was more worthy of the choice of first-line doctors.
Our research still had some deficiencies. Firstly, all our research data were obtained from TCGA and GEO databases, and the size of study samples was not large enough, which might lead to a certain deviation in the relevant results of this study. Secondly, our research conclusions also lacked the validation of relevant internal and external experiments. So it was important to conduct multi-center, large sample and prospective double-blind trials for subsequent relevant studies.