Prediction of metastatic behavior in high-grade pleomorphic soft tissue sarcomas by gene expression.

Abstract

10561 Background: The biologic heterogeneity of soft tissue sarcomas (STS) complicates treatment. Metastatic propensity may be determined by gene expression patterns that do not correlate well with morphology. In earlier studies, gene expression patterns were identified that distinguish 2 subsets of clear cell renal carcinoma (RCC), serous ovarian carcinoma (OVCA), and aggressive fibromatosis (AF). We reported the use of a gene set derived from these three studies to separate 73 high grade STS into groups with different probabilities of developing metastatic disease (PrMet). We wished to confirm our findings using an independent data set. Methods: We utilized these gene sets, hierarchical clustering (HC), Kaplan-Meier, and log-rank analyses to examine the Affymetrix HU_133 expression profiles of 309 STS. Results: HC using a pooled gene set derived from the AF-, RCC-, and OVCA-gene sets identified subsets of the STS samples. Kaplan-Meier analysis revealed differences in PrMet between the clusters defined by the first branch point of the clustering dendrogram (p=0.048), and also among the 4 different clusters defined by the second branch points (p<0.0001). Analysis also revealed differences in PrMet between the leiomyosarcomas (LMS), dedifferentiated liposarcomas (LipoD), and undifferentiated pleomorphic sarcomas (UDS) (p=0.0004). HC of the LipoD and UDS samples with the pooled probe set divided the samples into 2 groups with different PrMet (p=0.013, and 0.0002, respectively). HC of the UDS samples also showed 4 groups with different PrMet (p=0.0007). In contrast, HC found no subgroups of the LMS samples. Each individual gene set (AF-, RCC-, and OVCA-) separated the UDS samples into subsets of different metastatic outcome, but only the AF- gene set separated the LipoD samples, and no gene set identified LMS subsets. Conclusions: These data confirm our earlier studies and suggest that this approach may allow the identification of more than 2 subsets of high grade STS, each with distinct clinical behavior, and may be useful to stratify STS in clinical trials and in patient management.