Identification of heterogeneity among soft tissue sarcomas by gene expression profiles from different tumors

Abstract

The heterogeneity that soft tissue sarcomas (STS) exhibit in their clinical behavior, even within histological subtypes, complicates patient care. Histological appearance is determined by gene expression. Morphologic features are generally good predictors of biologic behavior, however, metastatic propensity, tumor growth, and response to chemotherapy may be determined by gene expression patterns that do not correlate well with morphology. One approach to identify heterogeneity is to search for genetic markers that correlate with differences in tumor behavior. Alternatively, subsets may be identified based on gene expression patterns alone, independent of knowledge of clinical outcome. We have reported gene expression patterns that distinguish two subgroups of clear cell renal carcinoma (ccRCC), and other gene expression patterns that distinguish heterogeneity of serous ovarian carcinoma (OVCA) and aggressive fibromatosis (AF). In this study, gene expression in 53 samples of STS and AF [including 16 malignant fibrous histiocytoma (MFH), 9 leiomyosarcoma, 12 liposarcoma, 4 synovial sarcoma, and 12 samples of AF] was determined at Gene Logic Inc. (Gaithersburg, MD) using Affymetrix GeneChip U_133 arrays containing approximately 40,000 genes/ESTs. Gene expression analysis was performed with the Gene Logic Genesis Enterprise System Software and Expressionist software. Hierarchical clustering of the STS using our three previously reported gene sets, each generated subgroups within the STS that for some subtypes correlated with histology, and also suggested the existence of subsets of MFH. All three gene sets also recognized the same two subsets of the fibromatosis samples that we had found in our earlier study of AF. These results suggest that these subgroups may have biological significance, and that these gene sets may be useful for sub-classification of STS. In addition, several genes that are targets of some anti-tumor drugs were found to be differentially expressed in particular subsets of STS.

Figure 1

Clustering of gene expression in the aggressive fibromatosis samples using the RCC gene set (top) and OVCA gene set (bottom) and the Eisen clustering software Cluster. The 12 AF samples were clustered using the Eisen clustering software Cluster and the set of 167 gene fragments from the U_133 microarray set most differentially expressed between two groups of ccRCC previously described [11] (top) and the set of 200 gene fragments most differentially expressed between borderline and invasive OVCA [13] (bottom) as described in the text. Samples AF-1 to AF-5 formed a cluster, while samples AF-6 to AF-12 formed another cluster. The tissue samples in the tree are joined by very short branches if they have gene expression patterns that are very similar to each other, and by increasingly longer branches as their similarity decreases.

Figure 2

Clustering of gene expression of the STS and AF samples with the RCC gene set (A), OVCA gene set (B), and AF gene set (C). The 12 AF samples and the 25 other STS samples were clustered using the Eisen clustering software Cluster as described in the text. The 16 samples that cluster with AF-1 to AF-5 (open squares) using all 3 gene sets are indicated by open circles. The 6 samples that cluster with AF-6 to AF-12 (solid squares) using all 3 gene sets are indicated by closed circles. The clustering of 3 samples (solid triangles) varied with the gene set. The tissue samples in the tree are joined by very short branches if they have gene expression patterns that are very similar to each other, and by increasingly longer branches as their similarity decreases.

Figure 3

Clustering of gene expression of the STS samples with the RCC gene set (A), OVCA gene set (B), and AF gene set (C). The 25 STS samples were clustered using the Eisen clustering software Cluster as described in the text and are labeled as in Figure 2. The tissue samples in the tree are joined by very short branches if they have gene expression patterns that are very similar to each other, and by increasingly longer branches as their similarity decreases.

Figure 4

Clustering of gene expression of the MFH samples with the RCC gene set (A), OVCA gene set (B), AF gene set (C), and the protein kinase gene set (D) as described in the text. The 16 MFH samples were clustered using the Eisen clustering software Cluster as described in the text. MFH-1 to MFH-9 grouped together in panel A and are indicated by an asterisk. The tissue samples in the tree are joined by very short branches if they have gene expression patterns that are very similar to each other, and by increasingly longer branches as their similarity decreases.