Background: Presently, a 50-gene expression model (PAM50) serves as a breast cancer (BC) subtype classifier that is insufficient to distinguish, within each single PAM50-classified subtype, patient subpopulations having different prognosis. There is a pressing need for inexpensive and minimally invasive biomarker tests to easily and accurately predict individuals' clinical outcomes and response to treatments. Although quantitative proteomic approaches have been developed to identify/profile proteins secreted (secretome) from various cancer cell lines in vitro, missing are the clinicopathological relevance and the associated prognostic value of these secretomic identifications.
Methods: To discover biomarkers to predict individualized prognosis we introduce a new multi-omics (secreto-transcriptomics) method that identifies, in their oncogenically secreted states, candidate markers of BC subtypes whose genes bear patient-specific mRNA expression alterations of prognostic significance. First, we used label-free quantitative (LFQ) proteomics to identify the proteins showing BC-subtypic secretion from a series of BC cell lines representing major BC-subtypes. To determine and externally validate the prognostic value of these secreted proteins, we developed a secreto-transcriptomic approach that discovered a PAM50-subtypic Secretion-Correlated mRNA Expression Pattern (SeCEP) wherein the PAM50-subtypic secretion of select proteins statistically correlated with cis-mRNA expression of their encoding genes in patients of the corresponding PAM50-subtypes. Kaplan-Meier analysis of SeCEP genes was used to identify new liquid biopsy biomarkers for predicting individualized prognosis.
Results: The mRNA expression-to-secretion correlation (SeCEP) pinpointed multiple genes that are fully translated into the oncogenically active secretome in a PAM50-subtypic manner. Further, multiple SeCEP genes in distinct combinations or panels of multiple SeCEP genes were identified as 'systems prognostic markers' that showed mRNA co-overexpression patterns in the distinct subpopulations of PAM50-subtypic patients with poor prognosis or high-risk of relapse. Thus, our secreto-transcriptomic approach statistically linked BC subtypic secretome genes with patient-specific information about their mRNA expression alterations and significantly improved the sensitivity and specificity in patient stratification in the context of clinical outcomes or prognosis.
Conclusions: By combining LFQ secretome screening with proteo-transcriptomic retrospective analysis of patient data our integrated multi-omics approach bypasses costly, tedious, genome-wide fishing and predictive modeling that are commonly required to distinguish a few prognostically altered genes from thousands of other non-BC related genes in a genome.
Keywords: Label-free quantitative proteomics; Multi-omics correlations; Patient survival analysis; Protein secretion; Secretion-correlated expression pattern; TCGA.