Gene expression profiling for the investigation of soft tissue sarcoma pathogenesis and the identification of diagnostic, prognostic, and predictive biomarkers

Abstract

Soft tissue sarcomas are malignant neoplasms derived from mesenchymal tissues. Their pathogenesis is poorly understood and there are few effective treatment options for advanced disease. In the past decade, gene expression profiling has been applied to sarcomas to facilitate understanding of sarcoma pathogenesis and to identify diagnostic, prognostic, and predictive markers. In this paper, we review this body of work and discuss how gene expression profiling has led to advancements in the understanding of sarcoma pathobiology, the identification of clinically useful biomarkers, and the refinement of sarcoma classification schemes. Lastly, we conclude with a discussion of strategies to further optimize the translation of gene expression data into a greater understanding of sarcoma pathogenesis and improved clinical outcomes for sarcoma patients.

Fig. 1

Schematic overview of the use of gene expression profiling for the investigation of sarcoma pathogenesis and the identification of diagnostic, prognostic, and predictive biomarkers. Following collection of biological samples for analysis, gene expression microarray experiments begin with the hybridization of cDNA or cRNA to a microarray chip, producing tens of thousands of gene expression measurements per sample. Following data collection, the most commonly used data analysis techniques are: unsupervised two-way clustering of samples and genes to discover novel biologic pathways observed in sarcoma subtypes and to discover new relationships within and between sarcoma subtypes. Supervised analyses complement the unsupervised approaches and are used to identify gene expression signatures and biomarkers to predict particular clinical phenotypes, such as: diagnostic subtype, prognosis, and response to treatment. Due to the complexity of microarray data and the large gene lists typically produced by both unsupervised and supervised analyses, it is useful to integrate the data with repositories of biomedical annotation information to optimize the translation of the findings into biomedically useful knowledge. A final step of independent validation (ideally on an independent platform with independent samples) is essential to ensure the validity and reproducibility of the findings

Fig. 2

Unsupervised hierarchical clustering of five sarcoma gene expression datasets. The cases are arranged along the columns and the genes along the rows. Within the heat map, red indicates relatively increased expression, black median expression, and green decreased expression. The color bar underlying the dendrogram indicates the diagnostic subtype of each case as indicated in the legend, which uses the following abbreviations: clear cell sarcoma (CCS), dermatofibrosarcoma protuberans (DFSP), desmoid-type fibromatosis (DTF), Ewing’s sarcoma (EWS), fibrosarcoma (FS), gastrointestinal stromal tumor (GIST), leiomyoma (LM), leiomyosarcoma (LMS), liposarcoma (LPS), malignant fibrous histiocytoma/pleomorphic sarcoma (MFH), malignant peripheral nerve sheath tumor (MPNST), neurofibroma (NF), pigmented villonodular synovitis (PVNS), schwannoma (SCH), rhabdomyosarcoma (RMS), and synovial sarcoma (SS). The clustering pattern suggests that one subset of sarcomas (including GIST, SS, EWS) shows relatively homogenous gene expression patterns and a second subset of sarcomas (including LMS, LPS, and MFH) shows more heterogeneous patterns of gene expression. The gene expression data presented in this figure were downloaded from the Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo/), Stanford Microarray Database (http://genome-www5.stanford.edu/), ArrayExpress (http://www.ebi.ac.uk/microarray-as/ae/), and supplemental materials accompanying the paper by Segal et al. [25]. Prior to clustering, the genes were mean-centered. Clustering was performed with average linkage and correlation (uncentered) as the distance metric using Cluster 3.0 software (http://bonsai.ims.u-tokyo.ac.jp/~mdehoon/software/cluster/software.htm#ctv)