Globally, breast cancer is a primary cause of death due to cancer in women and is the most frequently diagnosed form of cancer in a large number of countries(1). Despite striking progress have been made in the early diagnosis, therapeutic process monitoring, and prognostic evaluation, clinical outcome of BRCA still remains poor. The high heterogeneity of breast cancer exists not only in the genotypes and phenotypes of tumor cells but also in the tumor immune microenvironment, resulting in patients with the same clinical stage presented distinct responses to treatments and clinical outcomes. Traditionally clinical and pathological classification methods, such as tumor size, regional lymph node metastasis status, and distant metastasis, are too general to accurately determinate the prognosis of an individual patient(21). Thus, the novel prognostic biomarkers based on risk stratification may be the most effective strategy for precise prognostic prediction, contributing into optimal tailored treatment. In the past decades, accumulating studies suggested that lncRNAs participate in tumorigenesis, progression, metastasis and the prognosis of breast cancer via variety of ways(22, 23). Also, lncRNAs were reported as key regulators in regulating cancer immunity(11). Emerging evidence has supported that immune-related lncRNAs may serve as new therapeutic targets and disease molecular biomarkers for cancer clinic management and possess predictive value for survival prognosis(24–26).
However, most of previous researches aimed at prognostic signatures based on noncoding RNAs, focusing on prognostic prediction of cancer patients, are constructed by adopting the exact expression levels(27–29). This research was inspired by gene pairing and designed to develop a novel and reliable risk signature based on combinations of two-lncRNAs, without requirement of specific expression levels.
In our study, raw data of lncRNAs from TCGA-BRCA project was introduced into differential co-expression analysis to determine differential expressed immune-related lncRNAs (DEirlncRNAs), and lncRNA-pairs were validated by improved algorithm of cyclically single pairing along with a 0-or-1 matrix. Then, univariate and multivariate regression analysis followed by Lasso penalized regression to identify DEirlncRNAs pairs. Next, each AUC value of ROC curve was counted to obtain the ideal risk signature and the AIC value of each point on the AUC curve was calculated to determine the best cut-off point to stratify BRCA patients into the low- or high-risk-group. Additionally, prognostic value of risk model was validated using survival analysis, ROC curve, and univariable and multivariable regression analysis. Additionally, prognostic nomogram was constructed and confirmed to facilitate clinical extension. Furthermore, the potential role of risk score in TIME characterization and immunotherapy was investigated. Finally, the synergistic effect of risk score with TMB in term of prognostic prediction was demonstrated.
It was well established that lncRNAs with high abundance functioned as pivotal players in tumorigenesis, progression, and prognosis of cancer(30). This algorithm was designed to determination of DEirlncRNAs followed by establishment the most significant pairing of irlncRNA. As such, irlncRNA pairs with lower or higher expression value rather than each lncRNA expression level had to be examined. It was worthwhile to mentioned that as-constructed risk model harbored significant superiority of analysis cost relative to prognostic signature dependent on specific expression value of genes.
To enhance the efficacy and accuracy of prognostic prediction, the AIC values were employed to obtain the best cut-off point for risk stratification instead of simple median value. Additionally, AUC value of 1-/2-/3-year OS ROC curve was calculated then compared with various clinical variables to exhibit prognostic advantage of risk model. The excellent prognostic performance of risk model was validated by K-M analysis. Furthermore, risk signature was demonstrated to perform well as an independent prognostic predictor in both univariable and multivariable regression analysis. Besides, risk signature remained powerful prognostic ability in clinical variables stratified survival curves. Finally, risk-clinical nomogram that integrated risk score, age and clinical stage was established for clinical transformation.
Since risk model was established on irlncRNAs, this risk model was potentially mediated in modeling of TIME or suppression of immune-relevant cells. The results of TIME context indicated that risk score was negatively related with activated immune cell (i.e., M1 Macrophages, activated NK cell, CD8 + T cells, etc.,), whereas positively correlated with immunosuppressive cells (i.e., M2 Macrophages, etc.,), implying subjects with high risk was well characterized as immune suppressive phenotype, which was coincident with lower risk score suggested longer overall survival time.
Immunotherapy was treatment employing immune system fight against cancer cells, thus, infiltrating immune cells could affect clinical outcome of immune checkpoint blockades administration. The results showed that risk score was significantly and negatively correlated with the immunotherapy hub targets (i.e., PDCD1, etc.,), suggesting samples with low-risk score might be more affected by immune checkpoint blockade pathways, then inhibited anti-tumor immune activation and deteriorate prognosis accordingly. Since no immunotherapy data in BRCA cohort, it was unable to further explore the correlation of risk score with response of immunotherapy.
Currently, several clinical data pointed out a correlation between genetic alternations with responsiveness to immunological treatment (31, 32). We calculated and determined the TMB, which is a predictive indicator of sensitivity to immunological treatment, increased significantly with risk score elevated. Subsequent stratified survival curve demonstrated that risks score held prognostic predictive capability which was independent of TMB, suggesting that TMB and risk score represent different aspects of immunobiology. Besides, risk score together with mutation data revealed the significant distinction of genes variant frequency between high and low risk score group from the level of transcriptome.