The association of hypoxia with cancer is two-way: tumours typically experience low oxygen levels that lead to a hypoxia response, and mutations that cause an inappropriate hypoxia response can lead to cancer. Now, Chi et al. have shown that expression of a set of hypoxia-induced genes is strongly indicative of poor prognosis.

Cells respond to hypoxia through the hypoxia-inducible transcription factors (HIFs). Under normal conditions the HIF α-subunit is degraded by polyubiquitylation, but in normal hypoxic conditions and in tumours the rate of degradation decreases.

The authors set out to define a hypoxia gene-expression signature in primary cultured normal cells to compare it with expression in tumour cells. Because the hypoxia response varies widely between cell types, they used several different epithelial and mesenchymal samples. They obtained expression profiles from each cell type at several time points over a 24-hour response to hypoxia. Clustering analysis showed classes of genes that were up- or down-regulated in response to hypoxia, some of which were common and some of which were cell-type specific.

As carcinomas are derived from epithelial cells, the authors used the data from the epithelial cell cultures to define a set of 168 genes that were consistently induced by hypoxia. Using several previously published data sets from breast and ovarian cancer they found that tumours with high expression of hypoxia-response genes were of higher grade, were more likely to have p53 and oestrogen-receptor deficiencies, and, most importantly, correlated with significantly shorter survival times.

For the simple analysis of new tumour samples, the authors defined a 'hypoxia-response score' from the average expression of the hypoxia-induced genes. This score was compared with other prognostic indicators using the data from one of the breast cancer studies. It was found to be a better indicator of survival than traditional factors such as age, tumour size, tumour grade or response to chemotherapy. The addition of the hypoxia-response score increased the prognostic power of a multivariate analysis by 10%. Also included in the analysis was a similar signature for the wound-healing response. The two signatures showed only a weak correlation with each other, and between them accounted for 40% of the predictive power of the analysis.

The nature of the observed hypoxia response in tumours and the precise mechanistic role that it has in affecting clinical outcome are still to be elucidated. However, this study is of considerable clinical importance and further demonstrates the power of hypothesis-led gene-expression studies for prognosis. Furthermore, such studies will help in identifying the causes and consequences of hypoxia responses in human cancers and in discovering whether the degrees of these responses correspond to the actual tumour oxygen tensions.