Elsevier

The Lancet

Volume 359, Issue 9314, 13 April 2002, Pages 1301-1307
The Lancet

Mechanisms of Disease
Molecular characterisation of soft tissue tumours: a gene expression study

https://doi.org/10.1016/S0140-6736(02)08270-3Get rights and content

Summary

Background

Soft-tissue tumours are derived from mesenchymal cells such as fibroblasts, muscle cells, or adipocytes, but for many such tumours the histogenesis is controversial. We aimed to start molecular characterisation of these rare neoplasms and to do a genome-wide search for new diagnostic markers.

Methods

We analysed gene-expression patterns of 41 soft-tissue tumours with spotted cDNA microarrays. After removal of errors introduced by use of different microarray batches, the expression patterns of 5520 genes that were well defined were used to separate tumours into discrete groups by hierarchical clustering and singular value decomposition.

Findings

Synovial sarcomas, gastrointestinal stromal tumours, neural tumours, and a subset of the lelomyosarcomas, showed strikingly distinct gene-expression patterns. Other tumour categories—malignant fibrous histiocytoma, liposarcoma, and the remaining leiomyosarcomas—shared molecular profiles that were not predicted by histological features or immunohistochemistry. Strong expression of known genes, such as KIT in gastrointestinal stromal tumours, was noted within gene sets that distinguished the different sarcomas. However, many uncharacterised genes also contributed to the distinction between tumour types.

Interpretation

These results suggest a new method for classification of soft-tissue tumours, which could improve on the method based on histological findings. Large numbers of uncharacterised genes contributed to distinctions between the tumours, and some of these could be useful markers for diagnosis, have prognostic significance, or prove possible targets for treatment.

Introduction

Soft-tissue tumours are neoplasms that show morphological and immunophenotypical characteristics of mesenchymal cells such as fibroblasts, adipocytes, muscle cells, or peripheral nerve-sheath cells. These rare tumours have a wide range of histological appearances, within which over 100 different entities have been defined.1 Although many soft-tissue tumours can be distinguished from each other, the boundaries between several diagnostic groups are vague and can overlap. The classification is further complicated by the fact that there are few reliable immunohistochemical markers to aid in tumour subclassification or to help predict a patient's outcome. The specialty of soft-tissue tumour pathology is therefore hampered—perhaps more than in any other group of tumours—by uncertainty surrounding diagnosis.

Over 30 000 expressed genes in man have (at least partly) been sequenced, and we expect that the sequence of all expressed genes will be available soon. Level of expression of such large numbers of genes, impossible by old methods, can now be studied with cDNA expression microarrays.2 Furthermore, hierarchical clustering analysis recognises groups of genes that are co-expressed, providing a new level of insight into their possible functions. Microarray hybridisation technologies have begun to identify new molecular subclassifications in breast carcinomas,3, 4 lymphomas,5 leukaemias,6 melanomas,7 and prostate cancer.8

We did cDNA gene microarray analysis on a set of 41 soft-tissue tumours to identify gene clusters that define tumour families on a molecular level, to relate these families to histological diagnoses and known molecular markers, and to highlight new markers of potential diagnostic value.

Section snippets

Specimens and RNA isolation

Frozen tissue samples were obtained from soft-tissue tumour specimens resected at the Vancouver Hospital and Health Sciences Centre, the Stanford University Medical Centre, and the Hospital of the University of Pennsylvania between 1993 and 2000. 41 specimens were used for this study: these included eight gastrointestinal stromal tumours; eight monophasic synovial sarcomas; four liposarcomas (one dedifferentiated, one myxoid, two pleomorphic); 11 leiomyosarcomas (including one primary and

Results

46 specimens from 41 soft-tissue tumours were analysed for geneexpression profiles by a combination of 22K and 42K cDNA microarrays. Histological sections of representative tumours are shown in figure 1; sections of all specimens used can be viewed on the accompanying website.9 The relation between tumour type and gene-expression profile was analysed for 5520 well defined genes that showed variation in expression across the 46 arrayed specimens (figure 2) by hierarchical cluster analysis13 and

Discussion

We have reported gene-expression profiles of 41 soft-tissue tumours with cDNA microarrays; the complete dataset is available in a searchable format on the website accompanying this report.9 We have shown that singular value decomposition analysis can be used to overcome bias introduced by use of different batches of arrays. The two methods used for removal of array bias showed strikingly similar results: 5520 genes survived reselection after removal of array bias by singular value

GLOSSARY

eigengene
A trend in gene expression. If an eigengene correlates with a suspected source of artifact, than it can be deduced from the dataset.
eigenarray
Represents a similar trend in array types to an eigengene.
hierarchical clustering
Clustering in data mining is a statistical discovery process that groups a set of data in such a way that the intracluster similarity is kept to a maximum and the intercluster similarity is kept to a minimum. In the clustering process, two clusters are merged only if

References (36)

  • M Bittner et al.

    Molecular classification of cutaneous malignant melanoma by gene expression profiling

    Nature

    (2000)
  • SM Dhanasekaran et al.

    Delineation of prognostic biomarkers in prostate cancer

    Nature

    (2001)
  • Molecular portraits of soft-tissue tumours

  • The Brown lab

  • Molecular portraits of human breast tumours

  • Stanford microarray database

  • MB Eisen et al.

    Cluster analysis and display of genome-wide expression patterns

    Proc Natl Acad Sci USA

    (1998)
  • O Alter et al.

    Singular value decomposition for genome-wide expression data processing and modeling

    Proc Natl Acad Sci USA

    (2000)
  • Cited by (506)

    • Learning vector quantized representation for cancer subtypes identification

      2023, Computer Methods and Programs in Biomedicine
    View all citing articles on Scopus
    View full text