Use of gene expression to identify heterogeneity of metastatic behavior among high-grade pleomorphic soft
tissue sarcomas
Keith Skubitz1, Princy Francis2, Amy Skubitz1, Xianghua Luo1,
and Mef Nilbert2,3
1University of Minnesota, 2Lund University,
3Hvidovre Hospital
Sarcomas are heterogeneous
• Heterogeneity of biological behavior exists even within histologic subtypes of sarcomas, complicating clinical care, clinical trials, and drug development.
Example
• Assume treatment A has no adverse effect
• Assume benefit of treatment A is all or none in a certain percentage of patients
• Some biological behaviors that do not correlate well with morphology may be determined by gene expression patterns
• A common approach to identify prognostic factors is to search for differences in gene expression between 2 groups defined by an outcome (eg survival)– Requires defining 2 groups– Irrelevant genes may obscure important
patterns– Different genes could be important in different
subsets
• Alternatively, identification of subsets independent of clinical information could be useful
• We used PCA with a variety of gene sets in an attempt to identify heterogeneity– Clear cell renal carcinoma (RCC)– Serrous ovarian carcinoma (OVCA)– Aggressive fibromatosis (AF)
PCA with 604 probes up or down >/=5-fold in ccRCC vs normal kidney
B
PCA with probes from ubiquitylation in control of cell cycle pathway
A
• Gene expression patterns that distinguished 2 subsets of RCC (RCC gene set), OVCA (OVCA gene set), and AF (AF gene set) were identified
Question
• Do the RCC-, OVCA-, and AF-gene sets identify subsets of high-grade pleomorphic STS?
Samples
• 73 Samples obtained from Lund University
• 40 MFH
• 20 LMS
• 9 other high-grade pleomorphic STS
Data
• cDNA microarray slides with ~16,000 unique UniGene clusters
• About 50% of the genes in the RCC-, OVCA-, and AF- gene sets were present in this data set
Methods
• Data were pooled to form a set of 234 genes present in at least one of the RCC-, OVCA-, or AF-gene sets
• Hierarchical clustering using this gene set was performed
Hierarchical Clustering
Hierarchichal Clustering
1 2 3 4
Important Caveats
• Clustering pattern depends on composition of sample set
• Many types of clustering and ways to modify data
Conclusions
• Analysis of a set of STS using a gene set derived from other tumor systems without regard to clinical data, identified differences in time to metastasis
• Thus, an approach to subcategorizing samples before searching for variables that correlate with clinical behavior may be useful
Conclusions
• Although no confirmation of clinical relevance is available, stratifying patients entering trials by a similar approach could be useful, and would not result in loss of information
Conclusions
• Although no confirmation of clinical relevance is available, stratifying patients entering trials by a similar approach could be useful, and would not result in loss of information
• Banked samples should be obtained for all STS patients entering clinical trials for later analysis