Causal Bayesian gene networks associated with bone, brain and lung metastasis of breast cancer.

Using a machine learning method, this study aimed to identify unique causal networks of genes associated with bone, brain, and lung metastasis of breast cancer. Bayesian network analysis identified differentially expressed genes in primary breast cancer tissues, in bone, brain, and lung breast cancer metastatic tissues, and the clinicopathological features of patients obtained from the Gene Expression Omnibus microarray datasets. We evaluated the causal Bayesian networks of breast metastasis to distant sites (bone, brain, or lung) by (i) measuring how well the structures of each specific type of breast cancer metastasis fit the data, (ii) comparing the structures with known experimental evidence, and (iii) reporting predictive capabilities of the structures. We report for the first time that the molecular gene signatures are specific to the different types of breast cancer metastasis. Several genes, including CHPF, ARC, ANGPTL4, NR2E1, SH2D1A, CTSW, POLR2J4, SPTLC1, ILK, ALDH3B1, PDE6A, SCTR, ADM, HEY1, KCNF1, and UVRAG, were found to be predictors of the risk for site-specific metastasis of breast cancer. Expression of POLR2JA, SPTLC1, ILK, ALDH3B1, and the estrogen receptor was significantly associated with breast cancer bone metastasis. Expression of PDE6A and NR2E1 was causally linked to breast cancer brain metastasis. Expression of HEY1, KCNF1, UVRAG, and the estrogen and progesterone receptors was strongly associated with breast cancer lung metastasis. The causal Bayesian network structures of these genes identify potential interactions among the genes in distant metastases of breast cancer, including to the bone, brain, and lung, and may serve as target candidates for treatment of breast cancer metastasis.