Investigating a Novel Molecular Classification of Breast Cancer Based on the Tumor Proteome

Background Advancements in sequencing technologies have led to useful molecular classifications of breast cancer—specifically HER2, Basal-Like, Luminal-A, and Luminal-B classifications of the Prosigna PAM50 qPCR assay. These classifications allow a better prediction of disease outcome and have led to the development of more class-specific therapies. Towards a better understanding of the “proteo-genomic” relationship between somatic mutation and signaling in breast cancer, recent quantitative proteomic analyses suggest that breast-cancer may be subdivided into three, rather than four, distinct subgroups—Stromal, Luminal, and Basal. This opens the door for developing therapies based on the aberrant protein signaling of the tumor, especially for patients with significant stromal infiltrates, a phenotype that is strongly associated with poor clinical outcomes. Objective A critical constraint to applying this proteomic classification to every breast cancer patient is the time and cost of whole proteome and analyses. We sought to identify an RNA signature similar to the PAM50 that predicts the newly classified proteomic subgroups based on differential RNA expression. Furthermore, we want to see if protein expression levels of the 50 genes in the PAM50 can be used to classify stromal subtype that emerged in the whole-proteome level. Methods We began with RNA differential expression analysis of approximately 60,000 genes from 77 patients using The Cancer Genome Atlas (TCGA) database. We are now using k-means consensus clustering to explore the predictive value of the PAM50 protein expression in identifying the three groups identified in previous proteomics studies. Preliminary Findings We found that RNA differential expression could not stratify the patients into the proteomic subgroups, with the stromal group being the most heterogeneous patient population at the RNA level. This finding points to the vast amount of modification occurring between transcription and translation as well as the potential inaccuracy of classifications based on transcriptional clustering. We are in the process of conducting the consensus clustering