TNBC is still the serious subtype of breast cancer due to its complex molecular and cellular heterogeneity with a increasing incidence worldwider [24]. Previous studies have shown that TME is highly correlated with the occurrence, progression and prognosis of breast cancer [25]. TME is where the immune system interacts with the tumor, indicating that the plastic cells of the tumor and immune system are an important part of TME. Undoubtably, the TME play a vital role in the development of various kinds of cancer, expecially in breast cancer [13, 26]. Therefore, in our present study, we used the TNBC data obtained from the open-access cBioPortal website to identify specific TME-related genes. Those genes represent the bioactivities of immune and stromal components in TME and might pose a great impact on the prognosis of the TNBC.
In our present study, we calculated the immune and stromal scores in TME by the ESTIMATE algorithm to investigate the infiltrated level of immune and stromal cells in TME of TNBC. Our results illustrated that in the claudin subtype, the immune score of Claudin-low subtype was significantly higher than that of other subtypes (p < 0.05). Kaplan-Meier analysis also showed a higher immune score predicted OS in TNBC patients Good (p < 0.05). This outcomes was consistent with the results published by Kay Dias that Claudin-low cancer has different clinical pathology and prognostic characteristics from other types of breast cancer. In terms of disease free survival (DFS), Claudin-low cancer has the best 10-year prognosis (72.5%, p = 0.002) [27].
In order to explore the potential mechanism of TME changes, we performed GO function enrichment analysis on the screened 289 differentially expressed genes, and found that majorities were related to the TME. Then, a PPI network was constructed assess the interactions between the corresponding TME-related proteins. Multiple critical genes was determined as they had higher number of node connections in PPI. Interesting, most of the genes are found to be immune-related, indicating that TNBC is highly immunogenic with strong immune characteristics. In the identified genes, COL1A1 promote the metastasis of breast cancer especially in TNBC cell lines as previous reported. In metastasis, extracellular matrix (ECM) secreted more COL1A1 than usual to regulate the bioprocess of cells and ECM, subsequently resulting in an invasive and metastatic phenotype [28]. Pre-immune markers CXCL9, CXCL13 were positively correlated with enhanced number of tumor infiltrating lymphocyte (TIL), and were significantly associated with the longer DFS and (pathologic complete response, pCR) [29, 30].
In the study, a survival analysis was performed to explore the potential prognostic value of 289 DEGs and establish a risk model for predicting the prognosis of TNBC. We identified eight TME-related genes (ACAP1, DUSP1, LYZ, GZMA, SASH3, CCL5, CD74, DPT). The expression levels of these genes in TNBC patients had significant correlations with poor prognosis, indicating that our huge data analysis via ESTIMATE algorithm on the cBioPortal were greatly successful [31–34]. Among these eight genes, we are particularly interested in ACAP1, DUSP1, and GZMA. ACAP1 (ArfGAP With Coiled-Coil, Ankyrin Repeat And PH Domains 1) is a gene encoding protein. Hoffman et al. reported that the expression of the six genes identified by the genes was related to the risk of breast cancer, including the expression of three genes in breast tissue (RCCD1, DHODH and ANKLE1), and three in whole blood genes (RCCD1, ACAP1 and LRRC25). ACAP1 played a role in cell proliferation and activation of Arf6 protein [20]. In addition, ACAP1 might be correlated with EEC / PI, and regulated the recycling of integrin β1 during cell migration [7]. Dual-specificity phosphatase-1 (DUSP1 / MKP1) was a member of the triple-tyrosine dual-specificity phosphatase family. As a protein phosphatase, DUSP1 downregulates p38 MAPK and JNKs signals by directly dephosphorylating threonine and tyrosine. DUSP1 participated in multiple bioprocess such as cell proliferation, differentiation and apoptosis. It was showed that[33] TNBC had the highest frequency of DUSP1 methylation if comparing with other breast cancer subtypes. Therefore, DUSP1 methylation was considered as a unique subtype-specific marker in TNBC patients. Granzyme A, one of the five granzymes encoded in the human genome, was a human protein encoded by GZMA, which was closely related to immunity. In a meta-analysis, it was found that at least half of malignant tumors had low or missing GZMA protein expression. High levels of GZMA and PRF1 synergistically affected the survival rate of tumors [16].
By using the established prognostic model to predict the 5-year survival rate of TNBC patients, we concluded that the AUC of the ROC curve was 0.737. It meant that the prognostic model had a good survival prediction performance. In the coming future practices, TNBC patients will be divided into high-risk groups and low-risk groups as the mRNA-based risk score prognostic models suggested. Clinicians can determine the therapies based on the predicted outcomes of the model, so as to achieve personalized treatments of TNBC patients. Particularly in high-risk populations, positive strategies should be adopted to prevent TNBC recurrence. Meanwhile, high-risk populations should also be followed up more frequently, and breast MRI scans should be performed routinely to detect the TNBC recurrence earlier. We also demonstrated that the prognostic model was independent of other clinical factors in TNBC. In the GEO dataset (No. GSE103091) to predict the patients’ 5-year survival rate, in order to verify its predictive ability, the AUC value of the ROC curve reaches 0.636. This shows that our model has significance in wide application.
As far as we concerned, these eight biomarkers have not yet been studied in TNBC before. Hence, our findings can provide a solid basic foundation for the development of these new prognostic factors for TNBC in clinical practices, particularly in diagnostic kits exploration. The advantages of the predictive genes we identified is that no further requirement of needed somatic mutation assessment were needed in patients. In addition, our method greatly reduces the cost of sequencing, which makes targeted sequencing applications more cost-effective and routine. Accurate prognosis was essential for appropriate treatment selection. In routine clinical practice, pathological stage classification was common evaluations of prognosis in TNBC patients [35]. However, the clinical outcomes of patients at same stages were usually various with each other due to the known tumor heterogeneity, which indicated that the current staging classification was far from sufficient to a comprehensive prognosis of TNBC [36]. Obviously, the current-used pathological stages in TNBC was entirely based on the anatomical scope of the diseases. This limited property indicated that they were unable to fully represent the biological heterogeneity of TNBC [36]. The tumor heterogeneity, as demonstrated in the previous report, is not only represent the numerous genetic mutations happening in tumor cells, but also dynamic changes of the TME. Those changes of TME mediated by large number of the recruited immune cells somehow determined the occurrence, development and prognosis of the TNBC [37]. As the conventional classifications failed to estimate the tumor heterogeneity, the prognostic model we proposed was expected to improve the prognosis accuracy in TNBC patients.
Honestly, our research also suffered from some limitations. First, the population of the TNBC samples obtained cBioPortal website was mainly limited to white and black people, so it is necessary to expend our study to other nationalities. Secondly, the AUC of the prognostic model we evaluated by GEO dataset (No. GSE103091) was not high enough. Therefore, more verifications in multicenter clinical trials and prospective studies were needed. In the future, we will also explore the possibility of more predictors to improve the predictive performance of our model. Other regression modeling methods will be used to determine whether the prediction accuracy can be improved or not.
In summary, it was clearly demonstrated the eight-gene prognostic model was a considerably reliable tool for predicting the OS of TNBC patients, and can help clinicians selecting personalized treatments for their TNBC patients.