| Summary: | <p>Low levels of cellular oxygen, also known as hypoxia, is a major characteristic of solid tumours. Cancer hypoxia is associated with poor prognosis, resulting in prometastatic and angiogenic effects as well as inducing resistance to both chemo- and radiotherapeutic treatments. Despite the critical role that low oxygen tensions play in cancer progression, the clinical use of hypoxia-targeting treatments is still extremely low due to the absence of both accurate and cost-effective methods available to measure hypoxia.</p>
<p>One promising technique to measure hypoxia are gene expression signatures. As opposed to other methods, usually more expensive or invasive, gene expression signatures can be used to measure hypoxia from cancer biopsies, and even in retrospective cohorts. During the last twenty years, more than fifty hypoxia gene expression signatures have been developed. These expression signatures were derived using a variety of computational methods and different experimental conditions. For example, some signatures have been derived <i>in vitro</i>, others <i>in vivo</i> only and others using both approaches including one or more cancer types. As a result, the scientific community has to decide which of the plethora of signatures to use in their experiments and clinical trials, but which signature and which way of summarising a signature into a “score” is most appropriate for which scenario is currently unknown.</p>
<p>This work is the largest and most comprehensive analysis and validation of the 53 published hypoxia gene expression signatures to date. Hypoxia gene expression scores were calculated on publicly available gene expression data from ~1,000 cell line samples (the Gene Expression Omnibus) and ~6,000 clinical samples (The Cancer Genome Atlas) and compared to the same score on normoxic samples. These hypoxia signature scores were compared to the scores derived from more than 7.5 million Random Gene Signatures (RGS) to determine whether they were truly able to measure hypoxia. The overall most effective signature and score combination on cell line samples was the Sorensen 2010 signature using the median score. This achieved an impressive 92.84% accuracy in identifying hypoxic samples in 98 cell lines. In clinical samples, the Buffa 2010 signature using the mean score appears most appropriate as it fulfils the three key criteria of a) being on average higher in tumour samples than in normal tissue samples, b) differing in performance compared to random gene signatures c) serving as a strong prognostic marker (using a number of different thresholds) across 10 important tumour types. However, large prospective clinical studies with multiple measures of hypoxia are urgently needed to affirm our recommendation. This work lays down a new benchmark in how to measure hypoxia and the novel method used may transform how gene expression signatures in multiple fields might be evaluated in years to come.</p>
|
|---|