Abstract P5-08-12: Co-expression modules identified from published immune signatures reveals five distinct immune subtypes in breast cancer

Abstract

Abstract Introduction Immune modulating therapies offer an attractive novel approach in the treatment of breast cancer. There is a growing body of literature demonstrating that immune-related expression signatures predict breast cancer prognosis and chemo- and/or targeted therapy responsiveness. However, it is unclear how these signatures relate to one another. Here we evaluated 58 immune signatures in breast cancer and generated co-expression modules to classify patients into immune subtypes. Methods We evaluated 58 published expression signatures related to immune function in 5 breast cancer gene expression datasets (TCGA (n=817), METABRIC (n=1992), EMC344 (n=344), pooled triple negative: GSE31519 (n=579), pooled neoadjuvant chemotherapy treated: GSE25066 (n=508)). For each dataset, consensus clustering was used to subset the signatures based on their co-expression pattern. Signatures in the same consensus cluster across all 5 datasets were used to define immune modules. Module scores were computed as the average across their constituent signatures. Patients were classified into immune subtypes based on their module scores using consensus clustering. Overall survival (OS) differences between immune subtypes were assessed using Cox proportional hazard modeling in basal breast cancers from the METABRIC dataset (n=329). Results Consensus clustering of the 58 expression signatures consistently yields four distinct co-expression modules across the five datasets. These modules appear to represent distinct immune components and signals, with constituent signatures relating to: 1) T-cells and/or B-cells (T/B-cell), 2) interferon (IFN), 3) transforming growth factor beta (TGFB), 4) core serum response, dendritic cells and/or macrophages (CSR). Of note, the T/B-cell module contains 20 of the 58 signatures evaluated; and the CSR module is highly correlated to proliferation (r=0.81). Subtyping of patients based on these co-expression modules consistently yields subsets that fall into five major immune subtypes. The expression pattern of the four modules within each immune subtype is summarized below: T/B-cellIFNTGFBCSRT/B-cell/IFN HighHighHighIntermediateIntermediateIFN/CSR HighLowHighLowHighImmune LowLowLowLowLowCSR HighLowLowLowHighTGFB HighLowLowHighLowImmune Co-expression Modules (columns); Immune Subtypes (rows) These immune subtypes are associated with differences in overall survival in the METABRIC basal breast cancer cases, where the CSR High subtype has the worst outcome (10-year OS: 23%). In comparison, the subsets corresponding to the T/B-cell/IFN High subtype have better outcomes (Hazard ratio: 0.43, p = 0.018). In contrast, no significant outcome differences were observed between the poor outcome CSR-High subtype and the remaining three immune subtypes (p>0.05). Conclusion Our exploratory study identified four distinct immune co-expression modules (T/B-cell, IFN, TGFB, or CSR) from a collection of published immune signatures. Using these modules, we identified 5 immune subtypes with prognostic significance in basal breast cancers. We propose to test representative signatures from the 4 modules and the combined immune subtypes as predictive biomarkers of response to immunotherapies. Citation Format: Amara D, Wolf D, van 't Veer L, Esserman LJ, Campbell MJ, Yau C. Co-expression modules identified from published immune signatures reveals five distinct immune subtypes in breast cancer. [abstract]. In: Proceedings of the Thirty-Eighth Annual CTRC-AACR San Antonio Breast Cancer Symposium: 2015 Dec 8-12; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2016;76(4 Suppl):Abstract nr P5-08-12.