Head and neck cancer
The average overall survival for head and neck cancer patients is around 50% [1], but this varies greatly among different groups of patients. Applying clinical (TNM) staging to create different prognostic groups can only explain survival variation for 25% [2, 3, 4, 5]. We have previously shown that the addition of a prognostic gene expression profile can improve outcome prediction, suggesting that a substantial part of the survival variation is explained by tumor biology [6].

Hypoxia affects treatment outcome/prognosis
One of the most studied biological factors affecting prognosis of head and neck cancers is tumor hypoxia [7]. Tumor cells can become hypoxic by chronic (diffusion limited) and acute (perfusion limited) mechanisms, which can have different effects on tumor cells and their microenvironment. Which of the two has the most prognostic implications is still unclear [8].

Because oxygen is essential to cause DNA-damage upon irradiation, hypoxic cells respond poorly to radiotherapy [9, 10, 11, 12]. Since approximately two third of all head and neck cancer patients is (partly) treated with radiotherapy, hypoxia can be a great obstacle in the treatment of these tumors [13]. A meta-analysis of clinical trials showed that in vivo modification of the acute and/or chronic oxygen status during radiotherapy can improve survival of head and neck cancer patients, demonstrating that hypoxia is an important factor in radioresistance [14]. Unfortunately, the hypoxia modification therapy comes with added toxicity and the benefit from hypoxia modification was modest in the whole series [14]. This led to the hypothesis that only patients with hypoxic tumors profit from such a therapeutic intervention, which was shown to be correct in two recent studies [15,16].

Selection of hypoxic patients
Since selection of patients appears to be of importance, a robust approach to quantify hypoxia is essential. Different techniques have been applied to evaluate the level of hypoxia in a tumor and its impact on radiotherapy response [7], including an oxygen-sensitive needle probe inserted into the tumor [17, 18, 19, 20], exogenous immunohistochemical markers (e.g. pimonidazole [21]), endogenous markers (e.g. HIF1-alpha [22, 23, 24] or carbonic anhydrase IX [16, 22, 25, 26]) and imaging techniques like MRI [27] and PET [28]. None of these techniques is currently used in clinical practice.

Hoping to better reflect the intricate cellular response to hypoxia, there have been reports of panels of markers or gene expression sets that correlate hypoxia status with prognosis [15, 22, 29, 30,  31, 32, 33]. Several published signatures have been validated to be prognostic or even predictive in head and neck cancer [15, 29, 30, 31]. These signatures appear to have only a few genes in common, raising the question which signature performs best for the assessment of the level of hypoxia within a tumor. In none of these series a distinction was made between acute and chronic hypoxia.

With the intent to better select patients for hypoxia modification the NIMRAD study was recently initiated, aiming to ‘prospectively validate a gene signature that can be used in clinical practice to personalize treatment and select appropriate patients for hypoxia modifying treatment’ [34].

Study goals
We aimed to study the differences between the published hypoxia signatures that have been validated in head and neck cancer. First, we identified hypoxia signatures that have been validated to be prognostic or even predictive in head and neck cancer. We compared the genes included in these signatures and next the uniformity of these signatures in the classification of head and neck cancer patients into a ‘hypoxic’ and ‘less hypoxic’ group. In addition, we sought to compare these signatures with expression data of cell lines subjected to chronic/acute hypoxia. Lastly, the ability of the different signatures to predict radiotherapy response was tested in a series of 91 head and neck cancer patients who underwent chemoradiotherapy.