The molecular determinants of metastasis are one of the great unsolved mysteries of cancer research. But expression analysis using microarrays is exposing some of the daemons that lurk within this black box. Tobey MacDonald and colleagues, reporting in the October issue of Nature Genetics, have used microarrays to identify consistent changes in gene expression in metastatic medulloblastoma — a brain cancer that most commonly affects children. The genes identified leave a remarkably consistent trail that might lead down some new therapeutic avenues.

The authors used some statistical methods that are fast becoming the gold standard for the molecular characterization of cancer. Having measured the expression levels of over 1900 transcripts for 10 metastatic and 13 non-metastatic medulloblastoma samples using commercially available oligonucleotide microarrays, they identified 85 genes — the predictor set — that differed significantly between metastatic and non-metastatic tumours. To determine the ability of these genes to categorize a tumour as metastatic or non-metastatic, the expression profile for each tumour was compared against those of the other 22 — a technique known as the 'leave one out' approach. The predictor set got it right for 90% of the non-metastatic tumours, but only for 50% of the metastatic ones, leading the authors to speculate that there might be subsets with different expression profiles within the metastatic group. Nevertheless, it correctly predicted the outcomes for four of another five blinded tumour samples, but categorized only one of three metastatic cell lines — the Daoy cell line — as metastatic.

But more exciting than the predictor's mixed successes was the finding that many of the genes that are overexpressed in the metastatic group are in the same growth-factor signalling pathway: the platelet-derived growth factor receptor PDGFRA was overexpressed in 85% of metastatic, but no non-metastatic tumours. Migration assays revealed that PDGFA (PDGFRA's ligand) stimulated movement of Daoy cells on fibronectin matrices, and this was blocked by neutralizing antibodies to PDGFRA. Activation of PDGFRA increased the phosphorylation of the mitogen-activated protein kinases MAPK1 and MAPK3, and their upstream kinases, MAPKK1 and MAPKK2. The antibodies blocked this increase in MAPK signalling, and pretreatment with a specific inhibitor of these MAPKKs, U0126, blocked PDGFA-mediated migration and MAPK activation in these cells.

Children with medulloblastoma are given radiotherapy to prevent or treat metastasis, but this has serious neurological consequences. Identifying individuals at low risk of metastasis would allow clinicians to tailor this aggressive therapy to children for whom the benefits would outweigh the risks. The prospect of treating metastatic medulloblastoma with tyrosine kinase inhibitors such as STI-571 (Gleevec), which inhibits PDGFRA's kinase activity as well as ABL's, is another potential outcome of this work. Further analysis of gene-expression patterns in metastatic medulloblastoma will doubtless improve the predictor set, perhaps yielding further promising therapeutic targets.