The Meaning, Measurement and Modification of Hypoxia in the Laboratory and the Clinic

Hypoxia was identified as a microenvironmental component of solid tumours over 60 years ago and was immediately recognised as a potential barrier to therapy through the reliance of radiotherapy on oxygen to elicit maximal cytotoxicity. Over the last two decades both clinical and experimental studies have markedly enhanced our understanding of how hypoxia influences cellular behaviour and therapy response. Furthermore, they have confirmed early assumptions that low oxygenation status in tumours is an exploitable target in cancer therapy. Generally such approaches will be more beneficial to patients with hypoxic tumours, necessitating the use of biomarkers that reflect oxygenation status. Tissue biomarkers have shown utility in many studies. Further significant advances have been made in the non-invasive measurement of tumour hypoxia with positron emission tomography, magnetic resonance imaging and other imaging modalities. Here, we describe the complexities of defining and measuring tumour hypoxia and highlight the therapeutic approaches to combat it. 2014 The Royal College of Radiologists. Published by Elsevier Ltd. All rights reserved.


Introduction
The term hypoxia has been extensively used to describe a state of insufficient oxygen, which can be present in tumours as well as normal tissues and wounds (Table 1).
Central to this review is how oxygen levels are measured in tumours; the units of oxygen concentration and the relevant conversion factors are shown ( Table 2). The normal level of oxygen (tissue normoxia/physoxia) varies widely between different organs. This should not be confused with hypoxia, as even the lower levels are sufficient to support the tissue in question (Table 3). Hypoxia occurs to some degree in most solid tumours as a result of several factors, including the rapid growth rate of cancer cells and the highly disorganised/inefficient vasculature. Within any tumour, oxygen levels are extremely heterogeneous and can include mild hypoxia (2% O 2 ) and severe levels of hypoxia (<0.1% O 2 ). In addition, due to rapid changes in red cell flux, periods of cycling hypoxia occur, i.e. exposure to hypoxia levels followed by re-oxygenation. [11,12]. The conclusion of these studies is that less oxygenated/hypoxic tumour cells are more resistant to existing anticancer treatments, including chemotherapy, radiotherapy and surgery ( Figure 1) [17e19]. Hypoxic cells have been shown to be resistant to chemotherapy for several reasons. These include the distance from functional vessels, which makes the delivery and diffusion of drugs challenging, the low proliferation rates of cells in hypoxic areas and the selection pressure for the loss of p53-dependent apoptosis [20]. About 50% of all cancer patients receive radiotherapy either alone or in combination with other treatment modalities [21]. The efficacy of radiotherapy is determined by a number of factors, including the percentage of cells exposed to radiobiological levels of tumour hypoxia (Table 1) [18]. Many advances have been made in improving radiotherapy outcome and these can be broadly divided into those that increase the effect on the tumour and those that protect the normal tissue [13]. It is well known that the presence of oxygen at the time of ionising radiation will result in the production of free radicals that create stable organic peroxides, resulting in DNA damage, which is less easy to repair by the cell [1]. The consequence of this is an oxygen enhancement effect where hypoxic cells can be up to three times more resistant to radiation than those in normal oxygen concentrations. In addition, tumour hypoxia also predicts for poor outcome, even if the tumour is surgically removed, as the biological response to hypoxia includes increased invasion and metastasis [20].

Biological Responses to Hypoxia
The biological responses to hypoxia are extensive and well beyond the limits of this review. Some of the key pathways will be covered in brief.

Hypoxia Inducible Factor-mediated Signalling
The main players in the hypoxia response are a family of heterodimeric transcription factors named hypoxia inducible factors (HIFs), composed of an oxygen-labile alpha (a) subunit (HIF1a, HIF2a and HIF3a) and a constitutively expressed HIF1b subunit [22,23]. In the presence of oxygen, the HIFa subunits bind to the tumour suppressor protein von Hippel-Lindau, leading to ubiquitination and proteasomal degradation [24]. By contrast, when oxygen is limited, the enzymes that modify HIFa (HIF hydroxylases) become inactive and HIFa is stabilised and translocates to the nucleus. Once nuclear, HIFa heterodimerises with HIF-1b and binds to specific DNA sequences known as HIFresponse elements. Some of the most well-characterised target genes include the vascular endothelial growth factor A, carbonic anhydrase IX (CA-IX), the glucose transporter (GLUT1) and erythropoietin. Other HIF-regulated genes have roles in glucose metabolism, vasodilation, apoptosis, autophagy, oxygen sensing, invasiveness and metastasis [25,26].

Suppression of DNA Repair, Unfolded Protein Response and DNA Damage Response
There is increasing evidence that in an attempt to adapt to hypoxic conditions, tumour cells repress cellular processes that involve high-energy consumption. Under hypoxic conditions, many essential components of the DNA repair pathways have been shown to be repressed [27]. Homologous recombination, mismatch repair and nonhomologous end-joining have all been shown to be less effective under hypoxic conditions, suggesting that a general response to hypoxia is repression of DNA repair [28e30]. The mechanisms of DNA repair gene/protein repression are varied and include roles for HIF, micro-RNAs and epigenetic modifications [30e32]. For example, the  Also known as 'tissue normoxia'. The normal oxygen pressure in specific tissues/organs (see Table 3). Hypoxia Low oxygen levels, indicating that the tissue/organ has insufficient oxygen. Associated with the stabilisation of hypoxia inducible factor-1 (HIF-1).

Radiobiological hypoxia
The level of hypoxia where significant resistance to radiation is observed repression of the mismatch repair protein MLH1 by hypoxia correlates with increased levels of methylation of a particular histone (H3K9me2), which were attributed to increased levels and activity of the histone methyltransferase in hypoxia [33]. Key components of the homologous recombination pathway, such as RAD51 and BRCA1, have also been shown to be down-regulated under hypoxic conditions. The mechanism for RAD51 and BRCA1 down-regulation has been reported to be dependent on the formation of a repressive E2F4/p130 complex at the E2F site on the gene promoters, and is independent of both HIF-1a and cell cycle phase [34]. Importantly, the hypoxiamediated repression of DNA repair occurs at a variety of oxygen tensions and not just in regions of almost anoxia, indicating that larger proportions of tumours will have repressed DNA repair pathways. More severe levels of hypoxia (<0.1% O 2 ) have also been shown to induce specific responses not observed in milder hypoxia. One of the responses includes a reduction in protein synthesis due to ablation of the mRNA translation initiation through the unfolded protein response [35,36]. Under severe hypoxia, two responses (a rapid and a delayed one) affect members of the eukaryotic initiation factors (eIFs) family, which ablate the mRNA translation initiation process. While in the rapid response, the eIF2a is inhibited; in the delayed response, prolonged exposure to severe hypoxia leads to inactivation of eIF4 (reviewed in [35]). An additional specific response observed in severe levels of hypoxia is the DNA damage response ( Figure 2) [35,36,41e43]. In the absence of oxygen, the activity of ribonucleotide reductase is compromised, leading to depletion of dNTP levels and replication stress [44]. The accumulation of single-stranded DNA at the replication forks induces ataxia telangiectasia and rad3-related (ATR)-mediated signalling, which includes phosphorylation of H2AX, CHK1 and p53. In addition to the activation of ATR, ataxia telangiectasia mutated (ATM)-mediated signalling is also activated in hypoxia, despite the absence of detectable DNA damage [45,46].

DNA Damage Response and DNA Repair Inhibitors
Hypoxia is one of the major physiological differences between a tumour and normal tissue and therefore an attractive therapeutic target [13]. Given the key roles of ATR and ATM during hypoxia-induced replication stress and reoxygenation, these kinases are regarded as potential

Chemotherapy
Effective drug delivery and activity is compromised in hypoxic/acidic tumour regions. Adaption to hypoxia promotes chemoresistance.

Surgery
The biological response to hypoxia includes increased motility and invasion and therefore metastasis.
Tumour is more likely to have spread at time of treatment

Radiotherapy
In the presence of oxygen, radiation leads to the formation of reactive radicals which damage DNA.
Hypoxic tumour cells receive up to 3 times less DNA damage during radiotherapy

RESISTANCE TO THERAPY AND POOR PATIENT PROGNOSIS
Increased genomic instability and decreased DNA repair therapeutic targets in hypoxic tumour cells ( Figure 2). Disruption of these kinases through depletion or expression of kinase-dead proteins sensitises tumour cells to hypoxia/ re-oxygenation [42]. Recent studies using potent and specific inhibitors of cellular ATR activity, such as VE-821 and VE-822, showed promise in sensitising a variety of cancer cells to a range of oxygen concentrations [37,47,48]. In particular, VE-821 enhanced the effects of radiation, not only in normoxia, but also in hypoxic conditions [37]. Other ATR inhibitors include NU6027, which was shown to promote the genotoxic effects of several DNA-damaging agents, such as cisplatin, hydroxyurea, radiation and temozolomide [49]. It will be interesting to establish whether these combinations will also be effective under hypoxic conditions. Targeting CHK1, through gene silencing or chemical inhibition, has been shown to be a promising approach in enhancing tumour cell sensitivity to hypoxia/reoxygenation [50,51]. The CHK1 inhibitor AZD7762, although shown to have no effect under hypoxic conditions, significantly increases cell death after re-oxygenation and radiation treatment in several tumour cell lines [52e54]. Additionally, in xenograft models of lung cancer with brain metastasis, treatment with AZD7762 resulted in a prolonged survival in response to radiation [54]. More recently, ATM was also shown to be a promising target after hypoxia/ re-oxygenation. The ATM inhibitor KU-55933 induced DNA damage specifically under severe hypoxic conditions [46]. Moreover, in 2005, studies by Bryant et al. [55] and Farmer et al. [56] revealed that cancer cells with BRCA1 and BRCA2 deficiencies were extremely sensitive to poly ADP ribose polymerase (PARP) inhibitors. The explanation behind this sensitivity lies in the concept of synthetic lethality: when two repair pathways A (e.g. homologous recombination) and B (e.g. non-homologous end-joining) are abrogated alone, the cell remains viable; however, if both pathways are inhibited, this leads to cell death [57]. Recently, the novel concept of context synthetic lethality has emerged, and was shown to occur when one of the pathways is repressed due to the cellular context (e.g. hypoxia). In response to hypoxia, critical factors involved in homologous recombination are repressed, e.g. Rad51 and BRCA1, and this leads to a deficiency in repair and increased sensitivity to PARP inhibitors [58].

Hypoxia-activated Prodrugs
An elegant approach is to exploit the low levels of oxygen in hypoxic regions and design compounds that are only activated by enzymatic reduction in such conditions, the socalled hypoxia-activated prodrugs. Several of these agents have been described in the literature, including the wellestablished tirapazamine, AQ4N, PR-104A, SN30000, RH-1 and TH-302 [59e62]. However, most of these agents have been designed to induce DNA damage once activated under hypoxic conditions. The idea of designing hypoxia-activated prodrugs that can release an active inhibitor of the DNA damage response directly to the hypoxic regions is emerging. Recent proof-of-concept studies include the CHK1 inhibitor (CH-01) and a DNA-PK inhibitor (BCCA621C), each of which is masked by a bioreductive group. Under hypoxic conditions, reduction of the bioreductive group nitro moiety releases the active CHK1 inhibitor CH-01, leading to increased sensitivity of hypoxic cells [63]. In the case of BCCA621C, it was shown to be effectively released under hypoxic conditions and sensitise hypoxic radio-resistant cells [64].

Clinical Translation of Hypoxia-selective Therapies
Of course, a hypoxia-selective therapy is going to be of greater benefit in patients that have hypoxic tumours. The critical importance of integrating measures of tumour hypoxia into therapeutic trials is perhaps best exemplified by the phase III trial of tirapazamine. Substudies conducted within phase II trials in head-and-neck cancer had exemplified that patients with more hypoxic tumours (identified via positron emission tomography [PET] using [ 18 F]-fluoromisonidazole [[ 18 F]FMISO]) were more likely to show local regional failure after chemoradiotherapy than nonhypoxic tumours (8/13 versus 1/10). This was countered by the inclusion of tirapazamine into the treatment regimen (local regional failure hypoxic tumours 1/19 [65]). In the phase III trial, tirapazamine did not improve outcome in the overall patient population [66]. However, no measure of hypoxia was undertaken, so it was impossible to define cohorts who may have been showing an enhanced response based on the extent of tumour hypoxia. Additional complexities with the quality assurance of radiotherapy further marred the tirapazamine trial [67]. However, it is now generally recognised that hypoxia biomarkers are crucial for trials of hypoxia-selective therapies.

Needle Electrodes and Tissue-based Biomarkers
Oxygenation in tumours can be directly measured using needle electrodes, a technique that was invaluable in first proving the associations with hypoxia and treatment response [67e73]. However, this technique is limited to accessible tumours and, as the equipment went out of production, is available in very few centres. Alternative approaches were therefore needed and there are a number of potential ways of identifying hypoxia in tumour samples. Some exploit the reducing nature of the hypoxic microenvironment, such as the use of pimonidazole, a nitroimidazole that binds to macromolecules in cells under hypoxic conditions. Pimonidazole can be safely given to patients and binding ascertained in biopsies using specific antibodies [74]. It has been used in trials of hypoxiamodifying therapies (e.g. accelerated radiotherapy with carbogen and nicotinamide) and successfully tallies with patients showing the most marked improvements in response [75]. Pimonidazole has also proved invaluable in validating the hypoxic regulation of other potential endogenous tissue markers of hypoxia. From studies using material from cervical carcinoma patients, expression of the proteins CA-IX and GLUT1 were confirmed as potential surrogates for hypoxia through their coincidental expression with pimonidazole binding [76]. With the advent of array-based technologies, hypoxic signatures have been developed by identifying genes whose expression clusters with known hypoxia-regulated genes, including CA-IX. Such signatures have proven highly prognostic [77,78] and predictive of response to agents that specifically radiosensitise hypoxic cells (nimorazole [79,80]). Clearly, tissuebased biomarkers have extensive utility in clinical studies. However, an obvious issue with these approaches is the need for a biopsy. Biopsies are challenging or impossible in some tumour types and repeat measurements would be restricted. Serological markers, such as osteopontin, have shown some utility [81] and would enable repeat sampling. However, in both cases, these methods do not reveal heterogeneity in hypoxia across a tumour volume, which may be of relevance in terms of overall response and may also be a prerequisite of a therapeutic approach, e.g. if you wanted to increase the radiotherapy dose specifically to hypoxic cells.

Non-invasive Imaging of Tumour Hypoxia
Imaging affords a means of evaluating hypoxia across an entire tumour volume and lends itself well to repeat measurements. The most widely established methods for imaging hypoxia are based on PET. Magnetic resonance imaging (MRI) methods are, in theory, attractive options for serial imaging of tumour hypoxia, as it is widely available and avoids imparting ionising radiation to patients. However, at present, its role is less well established than PETbased methods. Other methods, including bioluminescence and photo-acoustic imaging [82], may play a role in the future, but remain research techniques at present.

Measuring Tumour Hypoxia Using Positron Emission Tomography Imaging
Ideally, a hypoxia PET tracer should target cellular rather than vascular pO 2 at clinically relevant oxygen concentrations only in viable cells and possess uniform and rapid cell entry (lipophilic molecule), rapid clearance from normoxic cells (hydrophilic molecule), yielding a high target-tobackground ratio independent of perfusion and be resistant to non-hypoxia-dependent metabolism [83]. None of the radiotracers that have been used in clinical studies ( Figure 3) fulfil all properties and development of better radiotracers is still on-going. The radiotracers proposed to date for PET imaging fall broadly into two categories, nitroimidazoles and copper-complexed dithiosemicarbazone (Cu-ATSM) derivatives [86], both of which are subject to futile cycling in the presence of oxygen, but enzymatically reduced in hypoxic cells leading to cellular retention. For nitroimidazoles, the mechanism of retention is via the formation of radical anions that bind to intracellular macromolecules, whereas Cu(II)-ATSM undergoes reduction to Cu(I)-ATSM, which is retained in hypoxic cells. However, through in situ metabolism, Cu(I) can become dissociated from ATSM and be washed out from cells, preventing its retention, despite the presence of hypoxia [87]. Nitroimidazoles have been generally labelled with 18 F (halflife of 1.83 h) for PET imaging, whereas Cu-ATSM has been radiolabelled with four different positron emitters of copper, two very short-lived isotopes ( 62 Cu and 60 Cu with halflives of 0.16 and 0.40 h, respectively), allowing repeated administrations but limiting the imaging time to less than 1 h, and two longer lived isotopes ( 61 Cu and 64 Cu having half-lives of 3.3 and 12.7 h, respectively) for delayed measurement, more likely to reflect hypoxia rather than perfusion, but at the expense of a higher radiation dose to the patient (Table 4) [88].

Selection of Radiotracers
PET tracers are at varying stages of validation and evaluation, with [ 18 F]FMISO being the most established and [ 18 F]FAZA increasingly being used. A comparison of radiotracers (Table 4) is hampered by the use of tumour models and tumour types, with varying degrees of hypoxia and radiotracer uptake being quantified using different methods. For the radiotracers used in clinical studies, validation in rodents showed tumour uptake above background level that correlates with pO 2 electrode measurements and/ or pimonidazole immunohistochemistry and that is sensitive to changes in oxygen tissue concentration [84,89]. Few [93], but the imaging time of both radiotracers was suboptimal in this study [94,95]. There are also limited reproducibility data that have been collected only for [ 18 F]FMISO in head and neck cancer [96,97]. Intertumoural differences in hypoxia selectivity of Cu-ATSM [98,99] have prevented its widespread acceptance as a hypoxia PET tracer.

Quantification of Tumour Hypoxia
Radiotracer uptake in tumours is heterogeneous in time and in space ( Figure 4). The time window for static imaging should be selected to decouple radiotracer retention from perfusion, which differ between tumour types and may vary in different parts of a tumour. High uptake is assumed to indicate retention in hypoxic cells, but could also result from slow clearance from normoxic cells. Conversely, low uptake is assumed to result from rapid washout from normoxic cells, but may also arise from slow entry into hypoxic cells, especially those poorly perfused. Tissue uptake will be further confounded by the presence of radiolabelled metabolites, which can differ between species (Table 4), between individuals and with treatment. Alternatively, dynamic imaging with different modelling approaches can potentially characterise both perfusion and retention and allow hypoxia perfusion patterns to be identified [100,101].
Hypoxia in PET images is most often quantified using the tumour to background ratio at a given time and requires estimation of first the tumour uptake, which can be done using the maximum value or an average over a given volume, and second the background uptake, which can be taken from a neighbouring muscle, the contralateral side (brain) or blood (sampled or extracted from a blood pool in PET image). The hypoxic volume is the volume of the tumour that is hypoxic and its measurement from PET images requires selection of an uptake threshold, which depends on the radiotracer, the tumour type and the imaging time, e.g. tumour to background ratio >1.2e1.4 for [ 18 F] FMISO [102,103]. The hypoxic fraction is the fraction of the tumour that is hypoxic and its derivation requires that the whole tumour volume be delineated in addition to the estimation of the hypoxic volume and this is typically extracted from other images, either computed tomography or [ 18 F] 2-fluoro-2-deoxy-glucose PET [104].

Application to Hypoxia-modifying Treatments
PET imaging of hypoxia can be used to identify patients with hypoxic tumours who could benefit most from additional treatment, either by prescribing hypoxia-modifying drugs or by increasing the radiation dose to the tumour [105]. PET imaging with fluorinated nitroimidazoles is being integrated into clinical trials of hypoxia-modifying treatments. Significantly fewer head and neck cancer patients with hypoxic tumours as measured with [ 18 F]FMISO had locoregional failure after radiochemotherapy with tirapazamine compared with those who were not given the hypoxic cell cytotoxin [65]. Less convincing results were reported with [ 18 F]FAZA in a similar clinical imaging study [106], despite promising results in mice bearing a murine breast cancer xenograft (EMT6) [107]. [ 18 F]EF5 uptake in a human non-small cell lung cancer xenograft model (H460) was reduced in rats treated with the hypoxia-activated prodrug SN30000 alone or with radiotherapy and the changes correlated with baseline hypoxia levels [108]. The distribution of hypoxia in the images could also be used to boost the radiation dose to hypoxic subvolumes of the tumour (called dose painting), exploiting the conformal and accurate radiation delivery made possible with the development of intensity-modulated radiotherapy and imageguided radiotherapy. Clinical implementation and evaluation await clarification in the treatment dose delivery [105].

Magnetic Resonance Imaging and Measurement of Hypoxia
Most MRI studies to evaluate tumour hypoxia to date have used either dynamic contrast-enhanced MRI (DCE-MRI) or blood oxygen level-dependent (BOLD) imaging. In DCE-MRI, a gadolinium-based contrast agent bolus is injected into a vein and the passage of the contrast agent is tracked through the tumour microvasculature. The presence of gadolinium species increases the longitudinal  n/a n/a n/a n/a Pancreas n/a 2.0 (120e150 min) n/a n/a n/a n/a Kidneys n/a >1.2y (180 min) n/a n/a n/a n/a Oesophagus n/a n/a >1.3{ (120 min) n/a n/a n/a Metabolismk (measurement time) Mouse 0.37 (120 min) 0.50 (120 min) n/a n/a n/a n/a Rat 0.80 ( relaxation rate (termed R 1 ) of protons, causing signal change, from which estimates of perfusion can be derived [109]. Several DCE-MRI studies have related such measurements to genetic readouts of hypoxia [110], oxygen tension [111] and immunohistochemical measurement of hypoxic fraction [112,113]. However, the strength of these relationships is inconsistent. This is explained partially by the fact that different studies measure slightly different MRI parameters, all of which have an incomplete and varying relationship to perfusion. Furthermore, chronic hypoxia and acute transient variations in oxygen tension are affected not only by oxygen delivery (perfusion), but also by diverse factors, including haemoglobin saturation, vascular geometry and consumption (oxidative phosphorylation). Therefore, DCE-MRI estimates of perfusion are at best indirect estimates of hypoxia and in some circumstances bare little relationship to hypoxia at all. BOLD techniques have been developed to exploit the differences in magnetic resonance transverse relaxation (termed R 2 *) induced by regional differences in deoxygenated haemoglobin levels [114]. BOLD measurements are typically carried out in combination with gas challenge, designed to alter the oxy-and deoxygenated haemoglobin ratio dynamically and therefore help map hypoxia. However, as BOLD measurements are sensitive to changes in vessel calibre they are subject to ambiguities in interpretation [115,116]. Although the technique has been carried out in cancer patients for nearly two decades [117], substantial artefact is found in some body regions, the extent of which depends on organ site, the control of visceral motion and the precise protocol adopted. This method has some success in mapping chronic hypoxic regions (delineated by pimonidazole adduct formation) in pelvic tumours such as prostatic carcinoma [118], but signal changes do not correlate well with absolute pO 2 levels [116]. Despite some initial encouraging data, there is little current evidence that either DCE-MRI or BOLD will provide validated, translational biomarkers of hypoxia in the near future. There is, therefore, a need to develop MRI methods of quantifying hypoxia that are more closely coupled to tissue chronic low oxygen tension and acute fluctuations in pO 2 and are free from the technical difficulties found with BOLD.
One promising method is oxygen-enhanced MRI (OE-MRI) using a T 1 -weighted contrast mechanism. It has long been recognised that oxygen dissolved in blood and tissue plasma increases R 1 in a similar mechanism to gadolinium [115,116], but the signal change is very small and so this contrast mechanism has only been exploited more recently, enabled by improvements in MRI scanner hardware and analysis methods. Human studies have shown that signal changes of up to 20% can be measured reliably in a range of well-vascularised normal tissues and that the technique is well tolerated [119]. Preclinical experiments have shown significant increases in tumour R 1 in a variety of tumour models [120,121] that relate to hypoxic fraction and vessel density [122]. A handful of clinical studies across a range of solid tumours have shown that this method is translational [122,123], but further studies are required to determine whether this or other MRI methods provide useful biomarkers for monitoring therapy and/or determining clinical outcome for patients undergoing radiotherapy and hypoxia-modifying therapy.

Conclusion
We have appreciated for many years that tumour hypoxia poses a significant problem for effective therapy. This knowledge has led to the development of strategies to target tumour hypoxia and equally importantly the preclinical testing of novel therapies in conditions that mimic the tumour microenvironment. It is clear that in order to enhance our ability to target hypoxic tumour cells we need to know where they are and how they respond to therapy. The advances in application of tissue biomarkers and hypoxia imaging discussed here highlight this real and exciting possibility.