|
|
Tumor classifier
for human gliomas based on gene expression profiles Sophie Godard, Gad Getz, Hiroyuki Kobayashi,
Pierre Farmer, Mauro Delorenzi, Annie-Claire Diserens, Marie-France
Hamou, Roger Stupp, Robert Janzer, Philipp Bucher, Eytan Domany,
Nicolas de Tribolet, Monika E. Hegi
University Hospital (CHUV),
Lausanne, Switzerland; Weizmann Institute, Rehovot, Israel; Swiss Institute of
Bioinformatics & ISREC, Epalinges, Switzerland
The
design of optimal treatment strategies tailored to individual patients and
identification of novel molecular targets for therapy requires further insight
into molecular aspects of glioblastoma development.
Here we sought to classify 51 gliomas according to their gene expression
profiles, comprising 24 low grade astrocytomas (LGA), 9 respective recurrent
high grade tumors, termed secondary glioblastoma (ScGBM), and 18 newly diagnosed
primary glioblastomas (PrGBM).
Glioblastoma multiforme may progress over years from LGA (WHO grade II) before
culminating in glioblastoma multiforme (WHO grade IV), but more frequently
arises rapidly without clinical or histological evidence of a less malignant
precursor lesion.
Gene expression profiles obtained from cDNA arrays (1200 genes) were analyzed by
Coupled Two-Way Clustering (CTWC), a method based on the identification of
subsets of genes or samples, such that when one is used to cluster the other,
stable and significant partitions emerge.
Stable clusters are identified by means of the underlying clustering method,
SuperParamagnetic Clustering (SPC).
Stable gene clusters separating the tumors according to their subtype emerged
that revealed interesting biological features implicating differences in
biological behavior of the tumors: One such gene cluster best discriminating
LGAs and ScGBM from PrGBMs comprises genes involved in angiogenesis such as
VEGF, VEGFR, but also IGFBP2 that has not been directly linked to
angiogenesis.
Relative upregulation of such genes in PrGBMs may reflect more severe hypoxic
conditions triggering angiogenic activity.
This may have important implications for therapy.
Despite the fact that PrGBM and ScGBMs are indistinguishable by classical
histology, response may differ due to their inherent distinct biology.
A glioma classifier based on gene expression was constructed by combining CTWC
with supervised statistical analysis (ranksum and t-test; FDR, [false discovery
rate] q<=0.05).
This allowed identification of four clusters rich in genes discriminating tumor
subtypes.
The discriminant power of this selected set of gene clusters as a glioma
classifier was successfully evaluated using a new set of gliomas and the k
Nearest-Neighbor method.
In conclusion, identification of subgroups of patients by means of molecular
diagnosis who are most likely to benefit from a targeted therapy will have great
clinical impact.
Copyright © 2003 American Association for Cancer Research. All rights
reserved.
Source: http://aacr03.agora.com/planner/displayabstract.asp?presentationid=8173
|
