Principal component analysis of RNA-seq data unveils a novel prostate cancer-associated gene expression signature

Abstract

Prostate cancer is a highly heterogeneous disease and the second more common tumor in males. Molecular and genetic profiles are currently used to identify subtypes and to guide therapeutic intervention. However, roughly 26% of primary prostate cancers of both good and poor prognosis is driven by unknown molecular lesions. Thus, ongoing research aims at providing better prognostic biomarkers and causal molecular alterations to intervene. Here, we use Principal Component Analysis (PCA) and custom RNAseq-data normalization to identify a gene expression signature which segregates primary PRAD from normal tissues. This Core-Expression Signature (PRAD-CES) includes 33 genes and accounts for 39% of data complexity along what we call the PC1-cancer axis. The PRAD-CES is populated by protein-coding (AMACR, TP63, HPN) and RNA-genes (PCA3, ARLN1) sparsely found in previous studies, others with validated/predicted roles as biomarkers (HOXC6, TDRD1, DLX1), and/or cancer drivers (PCA3, ARLN1, PCAT-14). Of note, the PRAD-CES also comprises six over-expressed LncRNAs with no previous association with Pca biology. Overall, the PCA allows us to capture roughly 3/4 of data complexity in the first eight PCs, whereas PC1, PC2 and PC3 altogether comprise 57% of such variability. Finally, GO enrichment analysis and PCs correlation with major PRAD clinical features (i.e. Gleason Score, AR Score, TMPRSS2-ERG fusions and Tumor Cellularity) suggest that PC2 and PC3 gene signatures might describe more aggressive and inflammation-prone transitional forms of PRAD with infiltrating immunosystem. The list of surfaced genes may entail putative novel prognostic biomarkers and/or molecular alterations to intervene.

On the gene expression landscape of cancer

“…In Ref. [ 32 ] we performed an analysis of the top 33 genes in the ranking for PRAD. Some of these genes have been already validated in the literature, but there is also a number of promising, yet not validated, indications for biomarkers or target genes.…”

“…That is, tumors in initial states show coordinates in the intermediate region whereas advanced stages of tumors correspond to coordinate values close to the center of the cancer attractor. Additionally, in a separate analysis of prostate adenocarcinoma (PRAD) [ 32 ], we show that the coordinate along PC1 correlates with clinical data on tumor cellularity of samples. In other words, the fraction of tumor cells in samples increases as we move along PC1 towards the tumor attractor.…”

“…Besides the aforementioned 5 major conclusions, additional results are presented as Supporting Information. Other specific lines of work have been developed in separated publications, such as the analysis of PRAD mentioned above [ 32 ], the concept of smooth and abrupt transitions and their relation to GE rearrangements [ 33 ], the computation of volumes of the atractors basins and their relation to the transition rates [ 34 ], the elaboration of a 1D model for carcinogenesis based on the coordinate along PC1 [ 35 ], etc.…”