With the development of systemic chemotherapy and radiotherapy in clinic, the survival of BC patients has been greatly improved. However, survival rates of patients with metastasis and recurrence are still unsatisfied. Therefore, identifying new biomarkers to predict the prognosis of BC patients is helpful for the customization of personalized treatment. High-throughput RNA-seq contributes to exploring various biomarkers for diagnosis and prognosis prediction of many cancers, including BC, along with the development of technology20–23.
In this study, we firstly found that the high infiltration degree of immune cells in TME was beneficial to the prognosis of BC patients. Further, machine learning analysis was used to systematically analyze the potential role of tumor immunity-related DEGs as prognostic predictors, construct a prognostic gene signature-based model and evaluate the value of the feature genes as prognostic markers of BC. In order to solve the “curse of dimensionality” (the combination of small sample and a large number of genes), which is common in high-throughput biological data, we analyzed the expression data using LASSO regression model and cross-validation method. LASSO regression is adept at dealing with high-dimensional regression variables by reducing all regression coefficients to zero to force many regression variables to be completely zero, thus, LASSO regression model has high stability and prediction accuracy24,25. The BC survival-related DEGs with a better predictive performance were identified by LASSO regression model and cross validation in this study, and a 6-mRNA signature model was constructed to predict the prognosis of BC.
There are several important findings during the analysis. Firstly, we found that most DEGs influenced the prognosis of BC patients by affecting the proliferation and activation of immune cells in TME. Secondly, 154 survival-related DEGs were screened using univariate Cox analysis and most of them were risk factors that may cause cancer (97/154). Finally, we constructed a prognostic gene signature model by using multivariate Cox analysis, and screened 6 BC immunity-related feature genes (PXDNL, FABP7, STX11, PIGR, SHISAL2A, WDR17) which could be used as independent prognostic factors. PXDNL (Peroxidasin like), a member of the peroxidase gene family, has a variety of biological functions, including hormone biosynthesis, host defense and cell movement26. A recent study by Li et al. has reported that PXDNL is closely related to the development of BC, and the high expression of PXDNL is a potential independent prognostic indicator of BC27. Pongor L et al. have found that PXDNL is a super-tumor suppressor gene in BC after analysis on gene chip data28. FABP7 (Fatty acid-binding protein 7) is a member of multi-gene fatty acid binding protein family and has been extensively studied in BC. Alshareeda AT et al. have discovered that FABP7 is poorly expressed in BC, and it can inhibit BC cell proliferation and promote differentiation through the JAK/STAT pathway29. In the study of Liu RZ et al., FABP7 is significantly up-regulated in triple-negative BC, and increased FABP7 level is associated with poor prognosis, absence of estrogen and progesterone receptors (ER, PR) and HER 2, increased cell proliferation and high tumor grade30. The expression of PIGR (Polymeric Immunoglobulin Receptor) is down-regulated in BC tissues14. Bin Xiao et al. have screened PIGR as a feature gene to construct a risk scoring system related to the prognosis of BC31. Through these prior studies, we found that PXDNL, FABP7 and PIGR of the 6 feature genes were closely related to the occurrence and development of BC and they could serve as prognostic markers of BC. While STX11, SHISAL2A and WDR17 had not been reported in BC, indicating that these three genes were newly discovered, and the survival analysis results of our study indicated that these three genes may be favorable factors for the prognosis of BC patients.