Sept 11, 2003 | Two exciting studies published last month in The Lancet and Cell, respectively, provide excellent examples that DNA microarrays are not merely passive platforms to survey gene expression data but, in the right hands, can shed important light on aspects of disease, in these cases cancer, that have previously proven intractable — the prediction of drug response and the underlying cellular mechanisms.
Square dance: Highly expressed genes (red squares) correlate with sensitivity to the chemotherapeutic drug docetaxel.
Writing in the Aug. 2 issue of The Lancet
, Jenny Chang and colleagues at the Baylor College of Medicine describe a 92-gene signature in breast cancer biopsies that predicts, with impressive accuracy, the sensitivity or resistance to neoadjuvant chemotherapy with the drug docetaxel. This is perhaps the first in what will become a flood of clinical studies using DNA chips to monitor chemosensitivity in patients.
Although docetaxel (produced by Aventis, which also funded the Baylor study) is a widely used chemotherapeutic agent, many patients are either partially or completely resistant. Frustratingly, there is no a priori means of predicting whether a given patient will respond until after the treatment begins.
Chang's team set out to see if the expression patterns of genes in the breast tumor itself might correlate with resistance to the drug. They used needle core biopsies in 24 women to extract RNA from tumors averaging 8 cm in size. Then, using the Affymetrix GeneChip HG-U95Av2 array, which consists of probe sets for more than 12,500 genes, Chang's team measured the gene expression patterns from the 24 samples. Eleven women were sensitive to docetaxel, as evidenced by a reduction in the tumor mass, whereas 13 were resistant.
After filtering out those genes with either low expression levels or minor differences between the two patient groups, the Baylor team was left with a subset of 1,628 genes. From these, they identified a group of 92 genes that were differentially expressed under stringent statistical criteria when comparing the resistant to the sensitive tumors. Based on the gene signatures, 10 of the 11 sensitive tumors (91 percent) and 11 of the 13 resistant tumors (85 percent) were correctly classified (88-percent accuracy overall). The results were confirmed in an independent group of patients.
The drug-resistant tumors tended to exhibit increased activity of genes involved in protein and RNA synthesis and the cell cycle; the sensitive tumors featured genes with functions in cell death, cell structure, and signaling. But the results do little to clarify the mode of action of docetaxel, in part because some genes that might be important, but were expressed at low levels, were deliberately set aside in the analysis.
"If validated," Chang and colleagues conclude, "these molecular profiles could allow development of a clinical test for docetaxel sensitivity, thus reducing unnecessary treatment for women with breast cancer." The raw data have been submitted to the NCBI gene expression omnibus (www.ncbi.nlm.nih.gov/geo), allowing their reuse in follow-up studies. As University of Cambridge's James Brenton and Carlos Caldas say in an accompanying editorial, "Data from clinical studies and molecular profiling need to be tightly coupled and widely accessible if the promise of predictive cancer genomics is to be achieved."
KSS and Tell
Meanwhile, investigators at the Dana-Farber Cancer Institute writing in Cell have applied data mining to DNA microarray data, revealing surprising clues to the role of a key oncogene. Levels of cyclin D1, a well-known cog in the cell cycle, are commonly increased in a wide variety of cancers, but its precise mode of action remains cloudy. Mark Ewen, Todd Golub, and colleagues set about scrutinizing the activity of thousands of genes in hundreds of cancer samples to divine a mechanism of action for this oncogene.
Using Affymetrix chips (Hu6800), Ewen's team first investigated the effects of artificially raising the level of normal and mutant cyclin D1 in cultured cells. This identified a group of 21 target genes, the activity of which increases significantly in response to cyclin D1. Next, they compared the expression of this 21-gene set with more than 16,000 genes, in the archived data sets of more than 190 tumors, in a resource called the Global Cancer Map.
Applying a statistical method called Kolmogorov-Smirnov Scanning (KSS), Ewen's team ranked these 16,000 genes to find which best correlated with the 21-gene signature — genes, Ewen surmises, that may work with cyclin D1 to switch on the signature. Among the top 50 genes revealed in this way were a group of three transcription factors, but of these, only one — a protein called CCAAT enhancer binding protein (C/EBP) ß, not previously associated with cyclin D1 — held up when the expression analysis was expanded further. A series of conventional wet-lab experiments further supports the notion of a direct physical interaction between cyclin D1 and C/EBP ß.
The new data-mining approach to the expanding digital compendium of gene expression signatures reveals "previously unappreciated relationships between genes involved in the development of cancer," Ewen says. Look for many more examples to follow.
DIAGRAM SOURCE: CHANG, J.C. ET AL. THE LANCET 362, 362-369; 2003. REPRODUCED WITH PERMISSION FROM ELSEVIER.