Background: It has been recognized that as a prostate cancer (PCa) metastasizes to bone, it begins to express bone-specific proteins such as osteopontin, bone sialoprotein, and osteocalcin. This process is known as osteomimicry. Liver metastases (LM) are associated with the poorest clinical outcomes and affect 25% of PCa patients at autopsy. We hypothesized that PCa that metastasize to the liver express liver-specific genes prior to liver metastasis – a process we have named “hepatomimicry”.

Methods: We have curated a list of 23 putative genes that are both primarily expressed in the liver and found in RNA-seq data from over 3000 PCa samples. Initial sources included the PCa transcription atlas (n = 2115), West Coast/SU2C Dream Team (n = 210), neuroendocrine PCa from cBioPortal (n = 49), and TEMPUS RNA-seq data generated specifically for Cedars-Sinai (n = 88). In each dataset, for the 23 hepatomimicry genes, we used recursive feature elimination and a t-test to discriminate between PCa LM and non-liver visceral metastases. In datasets with count-level data, we performed differential gene expression analysis with DeSeq2 and gene set enrichment analysis (GSEA) on identified differentially expressed genes.

Results: In three different datasets, MAT1A, ELF3, and WEE1 were found to be either optimal genes for discriminating between these two groups or significantly different by t-test and were more highly expressed in LM. 270 genes were downregulated and 437 genes were upregulated in liver metastases (FDR q-value < 0.05, log2 fold change > 1 for upregulated and < -1 for downregulated). Subsequent GSEA found that liver-specific gene sets are highly enriched in the genes upregulated in LM, and endothelial cell genes are strongly downregulated.

Conclusions: The initial results indicate that a three-gene subpanel of our initial set have predictive power for LM against other visceral metastases in PCa.While hepatocyte infiltration may impact the characterization of these samples by yielding a strong signal for liver-specific gene expression signature, confirmation of this finding remains important. Ongoing work is directed at confirming this signature in single-cell RNA-seq from PCa LM. Our work may be adaptable to developing diagnostic technologies such as liquid biopsies that may enhance the identification of patients who may have occult LM or that may develop LM as part of their natural history so that interventions may be enacted in a more timely manner.

VIEW THE PUBLICATION