There existed rare studies on the biological function of PTGDS in breast cancer. based on tumour-infiltrating immune cell subsets have been established, aiming to provide new clues regarding prognostic prediction and precision therapy for breast cancer. The key differentially expressed gene between different breast cancer immunotypes has also been identified. We performed unsupervised clustering analysis and construct a novel immunotyping which could classify breast cancer cases into immunotype A (B_cellhigh NKhigh CD8+_Thigh CD4+_memory_T_activatedhigh Tlow Mast_cell_activatedlow Neutrophillow) and immunotype B (B_celllow NKlow CD8+_Tlow CD4+_memory_T_activatedlow Thigh Mast_cell_activatedhigh Neutrophilhigh) in luminal B, HER2-enriched and basal-like subtypes. The NT157 5-year (85.7% 73.4%) and 10-year OS (75.60% 61.73%) of immunotype A population were significantly higher than those of immunotype B. A novel tumour-infiltrating immune cell-based prognostic model had also been established and the result immunorisk score (IRS) could serve as a new prognostic factor for luminal B, NT157 HER2-enriched and basal-like breast cancer. The higher IRS was, the worse prognosis was. We further screened the differentially expressed genes NT157 NT157 between immunotype A and B and identified a novel breast cancer immune-related gene, prostaglandin D2 synthase (PTGDS) and higher PTGDS mRNA expression level was positively correlated with earlier TNM stage. Immune-related signaling pathways analysis and immune cell subsets correlation analysis revealed that PTGDS expression was related with abundance of B cells, CD4+ T cells and CD8+ T cells, which was finally validated by immunohistochemical and immunofluorescence staining. We established a novel immunotyping and a tumour-infiltrating immune cell-based prognostic prediction model in luminal B, HER2-enriched and basal-like breast cancer by analyzing the prognostic significance of multiple immune cell subsets. A novel breast cancer immune signature gene PTDGS was discovered, which might serve as a protective prognostic factor and play an important role in breast cancer development and lymphocyte-related immune response. value for the deconvolution of each sample using Monte Carlo sampling, providing measurement confidence for each estimation. Samples with < 0?05 were considered accurate and could be included for further analysis. Histological validation and clinical data collection We collected formalin-fixed paraffin-embedded sections from 98 breast cancer patients who underwent surgical treatment at the Second Affiliated Hospital of Zhejiang University School of Medicine from August 2014 to August 2017. The related basic clinicopathological and survival information was also collected after receipt of informed consent and approval from the ethics committee. Gene expression and co-localization were validated by monoclonal antibody-based immunohistochemistry and immunofluorescence. Immunohistochemical staining by Envision method was performed on formalin-fixed paraffin-embedded slides, which had been dewaxed and rehydrated before antigen retrieval step. The intensity and frequency were used as evaluation indexes based on the brown staining of PTGDS. The intensity was divided into: negative (0), weak positive (1), positive (2), strong positive (3). The frequency was divided into: 0% ~ 10% (1), 11% ~ 30% (2), 31% ~ 50% (3), 51% ~ 75% (4), 76% ~ 100% (5). Comprehensive score = intensity*frequency. For immunofluorescence staining, formalin-fixed paraffin-embedded slides were heat-repaired by citrate buffer for 2 minutes, incubated with primary antibody at 4 overnight, incubated with fluorescein-labelled secondary antibody at room temperature, stained with DAPI and photographed by laser confocal microscopy. Bioinformatical and statistical analysis All statistical analyses were conducted using R studio software (Version 1.1.414; http://www.rstudio.com/products/rstudio). This study was conducted and reported in accordance with the TRIPOD guidelines. The molecular subtyping of breast cancer in patients were all determined with a PAM50 identifier function provided by the genefu package. Unsupervised hierarchical clustering analysis was conducted within breast cancer samples and cell subsets with the hclust function. Unsupervised hierarchical clustering analysis could discriminate breast cancer samples based on different immunotypes. Survival analysis was performed by the survival and survminer packages. Survival curves were constructed by the Kaplan-Meier method and compared by the log-rank test. Hazard ratios (HRs) were calculated using both univariable and multivariable Cox proportional hazards regression models. The LASSO-Cox regression model with LASSO penalty was used to select the most specific prognostic cell subpopulations among the 22 immune cell subsets, and the optimal values of the penalty parameter were determined by tenfold cross-validations. A new prognostic variable, immunorisk score, was then established based on the abundance of the selected immune cells using Cox regression coefficients in the integrated GEO dataset, which was further validated in the TCGA-BRCA and METABRIC cohorts. A multivariable Cox regression model was used to determine independent prognostic factors. Group comparisons were performed for continuous and categorical variables using one-way ANOVA and the test, respectively. Correlations among cell subsets were analysed by Pearson's correlation test. All statistical checks were two-sided, and < 0?05 was considered statistically significant. Results Overview of included breast tumor cohorts After data incorporation and filtration, 801 breast Igfbp5 cancer samples and 964 normal tissue samples from 12.