Gene expression and hypoxia in breast cancer

Hypoxia is a feature of most solid tumors and is associated with poor prognosis in several cancer types, including breast cancer. The master regulator of the hypoxic response is the Hypoxia-inducible factor 1α (HIF-1α). It is becoming clear that HIF-1α expression alone is not a reliable marker of tumor response to hypoxia, and recent studies have focused on determining gene and microRNA (miRNA) signatures for this complex process. The results of these studies are likely to pave the way towards the development of a robust hypoxia signature for breast and other cancers that will be useful for diagnosis and therapy. In this review, we outline the existing markers of hypoxia and recently identified gene and miRNA expression signatures, and discuss their potential as prognostic and predictive biomarkers. We also highlight how the hypoxia response is being targeted in the development of cancer therapies.

Hypoxia is caused by several factors: inadequate vascularization (tumor angiogenesis is often charac ter ized by aberrant vessels that have altered perfusion); an increase in diffusion distances that is associated with tumor expansion (oxygen has to travel further to oxy genate tumor cells because of uncontrolled tumor growth); and tumor or therapyrelated anemia (caused by reduced oxygen transport capacity) [5]. Cancer cells can adapt to a hostile, lowoxygen environment and this contri butes to their malignancy and aggressive pheno type. This adaptation is governed by many factors, in clud ing transcriptional and posttranscriptional changes in gene expression. In this respect, up to 1.5% of the human genome is estimated to be transcriptionally responsive to hypoxia [6].
Several studies have attempted to characterize the tumor response to hypoxia and its prognostic impli cations. In particular, recent studies have identified gene and microRNA (miRNA) expression signatures (that is, lists of regulated genes or miRNAs) that are characteristic of this response. Here, we discuss these studies and focus on breast cancer as a type of cancer in which hypoxia has been shown to have clinical implications [5]. We then discuss the use of these signatures in attempts to identify predictive markers of disease. We also review the current approaches for targeting the master regulator of the hypoxic response, HIF1α, in cancer treatments and the potential use of miRNA and gene signatures in this context.

HIF, the hypoxia response and prognosis
The master transcriptional regulators of the hypoxic response are represented by the family of hypoxia inducible factors. HIFs are heterodimers formed by an oxygen and growthfactorsensitive subunit α and a constitutively expressed subunit β [7,8]. In normoxic cells, the α subunit is recognized by and forms a complex with the von HippelLindau protein (pVHL), which mediates its ubiquitination and degradation by the proteasome. In hypoxic cells, the α subunit is stabilized, it translocates to the nucleus where it dimerizes with the β subunit and activates the transcription of target genes by binding to the hypoxicresponse elements (HREs) present in their promoter region [7,8]. There are three isoforms of the α subunit, HIF1α, HIF2α and HIF3α, and one β subunit, HIF1β. HIF1α is the isoform most ubiquitously expressed in cells, whereas HIF2α and HIF3α are expressed in a tissuespecific manner. HIF2α is found mainly in endothelium, liver, lung and kidney, where it acts like HIF1α on target genes. HIF3α is highly expressed in thymus, cerebellum and cornea, where it acts in a dominantnegative fashion to inhibit HIF1α and HIF2α (for a review, see [9]).
HIF1 regulates key aspects of cancer biology, including cell proliferation and survival for example, through regulation of Cyclindependent kinase inhibitor 1A (CDKN1A) and Bcell lymphoma 2 (Bcl2)/adenovirus E1B 19 kDa proteininteracting protein 3 (BNIP3); metabolism for example, through Glucose transporter1 (GLUT1), GLUT3, Lactate dehydrogenase A (LDHA) and Pyruvate dehydrogenase kinase 1 (PDK1); pH regu lation, through Carbonic anhydrase 9 (CAIX); invasion and metastasis, through CXC chemokine receptor type 4 (CXCR4) and Mesenchymalepithelial transition factor (cMET); angiogenesis, through Vascular endothelial growth factor A (VEGFA); and stem cell maintenance, through Octamerbinding transcription factor 4 (OCT4) (Figure 1) [10]. In particular, GLUT1 and GLUT3 are trans porters that are involved in the uptake of glucose, the main source of ATP generation through glycolysis in tumor cells. HIF1 can induce many of the enzymes in this metabolic pathway, which culminates with the conversion of pyruvate into lactate by LDHA [11]. CAIX is a carbonic anhydrase located on the plasma membrane that hydrates CO 2 to form H + and HCO 3 extracellularly [12]. The secretion of VEGF by hypoxic cells stimulates endothelial cell proliferation and leads to the formation of new vessels from preexisting ones (that is, angio genesis), to provide additional perfusion [13].
Tumor type has an important bearing on hypoxia response; in breast cancer, evidence suggests that the expression of HIF-1α and its targets are key determinants of prognosis. High HIF-1α expression has been associated with poorer prognosis in several studies (Table 1) and a recent metaanalysis confirmed this [3]. CAIX upregu la tion has also been associated with aggressive features and poor overall and relapsefree survival [1416]. High expression of the HIF1α target gene VEGF has also been associated with poor prognosis [1719]. GLUT1 upregu lation has been associated with increased risk of recur rence, highergrade tumors and proliferation [20], and the expression of this gene is associated with perinecrotic (in close proximity to the necrotic core) HIF-1α expres sion [21]. Increased expression of Lactate dehydrogenase-5 (LDH-5) has been associated with poor prognosis in endometrial, colorectal, head and neck and nonsmall cell lung cancer [2226], and the expression of this gene in breast cancer has been linked to HIF-1α expression [27]. Interestingly, Rademakers et al. [28] described a strictly cytoplasmic expression pattern for LDH-5 in head and neck carcinomas, which showed a strong correlation with hypoxia. On the other hand, Koukourakis and colleagues [2227] have repeatedly described a mixed cytoplasmic and nuclear expression pattern for LDH-5 in different types of tumor, including head and neck cancer. Nuclear LDH-5 reactivity was linked with high HIF-1α expression, poorer survival and more aggressive tumors [23,24], but its biological significance is still unknown.
Other hypoxia signaling pathways have also been iden ti fied; examples are pathways activated by the mamma lian target of rapamycin (mTOR) kinase and independent signals regulated by the unfolded protein response (UPR) in the adaptive response to low O 2 conditions. In particular, mTOR is a sensor of metabolic signals that can influence cell survival and growth through changes in several signaling pathways that are involved in protein synthesis, autophagy, apoptosis and metabolism [29]. Intriguingly, mTOR and HIF1 are reciprocally regulated, meaning that the deriving signaling pathways cannot be considered totally independent. Specifically, HIF1α can inhibit mTOR through its targets Regulated in develop ment and DNA damage responses 1 (REDD1) and BNIP3 [30,31], whereas mTOR inhibition can result in increased HIF1-α translation, resulting in a regulatory loop [32]. Hypoxia, as a negative regulator of mTOR signaling, could potentially act as a suppressor of tumor growth, but recent evidence suggests that this response to hypoxia is less pronounced in tumor cells than in normal cells, especially when the hypoxia is moderate (1% O 2 ). Conversely, in the presence of more severe (≤0.1% O 2 ) or prolonged hypoxia, protein synthesis and proliferation are inhibited in most cells as a possible way to preserve energy [29].

Hypoxia and treatment resistance
Although there is still a paucity of goodsized clinical studies and there have been discrepancies between findings, a tendency of hypoxic tumor cells to be drug and radioresistant has been identified [33]. Mechanisms of resistance include lack of oxidation of DNA free radicals by O 2 (giving rise to resistance to ionizing radia tion and antibiotics that induce DNA breaks), cell cycle arrest (giving rise to drug resistance), compromised drug exposure because distance from vasculature is increased (causing drug resistance) and extracellular acidification (also leading to drug resistance) (reviewed in [34]). HIF1α activation has also been associated with resis tance to endocrine therapy and chemotherapy [35].
In a study involving 187 breast cancer patients treated with either neoadjuvant epirubicin chemotherapy or combined epirubicin and tamoxifen, both HIF1α and its target CAIX were associated with treatment resistance [36]. A further study of 114 breast cancers, which were treated preoperatively with aromatase inhibitor, showed that HIF1α expression was an independent factor that was associated with treatment resistance [37]. This concurs with earlier evidence that tumors with low CAIX expression benefit from adjuvant endocrine or chemo therapy treatment [38]. In a study of 45 malignant astrocytomas, elevated CAIX was associated with poor response to combined treatment with bevacizumab and irinotecan [39]. Elevated serum CAIX has been asso ciated with reduced progressionfree survival in meta static breast cancer patients treated with trastuzumab [40].
The HIF target GLUT1 exerts a cytoprotective effect by allowing increased glucose transport into hypoxic cancer cells, and its overexpression is common in breast cancer [41]. In vitro studies with antibodies that block GLUT1 function, in conjunction with cytotoxic agents commonly used in breast cancer treatment, abolish proliferation in cancer cell lines, indicating a role for GLUT1 in treatment resistance [42].The HIF target gene VEGF has been associated with resistance to both hormonal and chemo therapies for breast cancer [43]. There is a lack of general agreement on the effect of antiangiogenic therapy on tumor perfusion and hypoxia (reviewed in [44]), but some evidence suggests that antiangiogenic agents might reduce tumor oxygenation, inducing the activation of HIF1 and its downstream targets and subsequently lead ing to tumor escape [45,46].
These studies highlight the importance of assessing hypoxia. Although several studies have been performed on single genes, we could identify only one study that  looked at the role of a hypoxia geneexpression signature in treatment response [47]. This highlights the need for more comprehensive studies to investigate the expression of multiple hypoxia markers and of gene and miRNA signatures before and after treatment. Careful pharmaco kinetic and pharmacodynamic analyses are also needed to derive markers of treatment efficacy or resistance. The finding of such research could not only allow the selec tion of patients who would benefit most from treat ments, but could also avoid the use of specific treatments in cases where they might be detrimental [45].

Targeting hypoxia in cancer treatment
Given the role of HIF1 in resistance to cancer treat ments, the inhibition of this protein is an attractive therapeutic approach (Table 2). In vitro data suggest that small molecule inhibitors of HIF1α in combination with adenovirusdelivered gene therapy might reverse the hypoxic chemoresistance of cancer cells [48]. Concerted attempts have thus been made to identify HIF1 inhibi tors using highthroughput screens. A better under stand ing of the HIF activation pathway could inform the choice of therapy, the individualization of treatments and the development of novel agents. Several of the cancer treat ments already licensed for use, including the Topoiso merase 1 inhibitor topotecan, have been shown to inhibit HIF1α protein accumulation in cell lines and xenograft studies [49,50]. It may be that, in the clinical setting, such agents will have synergy with drugs such as bevacizumab, which is thought to cause treatmentinduced hypoxia and subsequent HIF1α activation that lead to drug resistance [46].
Several novel compounds are under investigation. Bortezomib is a proteasome inhibitor already approved for the treatment of hematological malignancies. A pharma codynamic study in a metastatic colorectal cancer phase II trial observed downregulation of CAIX in response to bortezomib, suggesting a disrupted hypoxia response to this compound [51]. Another novel com pound, PX478, inhibits HIF-1α transcription and HIF1α protein levels in a p53 and pVHLindependent manner [52]. YC1, a synthetic compound, has been widely used in the laboratory setting to investigate the physiological and pathological role of HIF. In cancer cell lines, YC1 inhibits HIF through factor inhibiting HIF (FIH)depen dent inactivation of the carboxyterminal transactivation domain (CAD) of HIF1α [53].
A highthroughput cellbased screen has shown that another compound, DJ12, inhibits HIFinducible trans cription [54]. Another approach demonstrated that ascor bate increases the activity of prolyl hydroxylase enzymes, leading to HIF downregulation, in cells treated with antisurface transferrin receptor (TFR) antibody [55]. The antiHIF activity of two other novel anticancer drugs, AJM290 and AW464, has also been examined; both compounds inhibit HIF1α transcription at the CAD and DNAbinding domains, although they also inhibit HIF degradation [56]. Gene therapy that utilizes HIF1α expression and the promoter regions of its downstream target genes (that is, HREs) would be an attractive approach. This might allow the targeted delivery of anticancer agents to tumor tissue. For example, it has been shown that hypoxic cells can be targeted by combining a HIFresponsive promoter with an oncovirus that is armed with the interleukin4 gene. Treatment of xenografts using this technique led to main tained tumor regression [57]. One group demon strated that HIF1αbased gene therapy can eradicate small EL4 xenografts and also that this therapy augments the efficacy of the antiangiogenic agent angiostatin [58]. Nevertheless, the great variability in the level of hypoxia, and hence HIF1α expression, within a single tumor presents a challenge to such approaches.

Methods for detecting hypoxia
Methods that can reliably detect hypoxic tumors are crucial because of the roles of hypoxia in tumor prognosis and in resistance to specific treatments. Various methods are used to detect hypoxia in solid cancer tumors, but contrasting results have been reported [5]. O 2 measure ment with a polarographic O 2 needle electrode is the most direct method, but it has limitations, including its invasiveness, its inability to represent the whole tumor, and the possibility that it can generate false positive determinations as a result of oxygen consumption by the electrodes. In the clinic, the assessment of hypoxia is moving towards the evaluation of endogenous and exo genous markers. Immunohistochemistry is widely used in patient biopsies, and this method can detect both endogenous and exogenous markers of hypoxia. Among the endogenous markers, particular interest has been paid to HIF-1α and some of its target genes, including GLUT1, CAIX and VEGF. One limitation that is asso ciated with these markers is their potential regulation by nonhypoxiarelated factors (for example, pH or the concen trations of metabolites such as glucose and gluta mine). Exogenous markers of hypoxia include nitroimi dazole compounds derived from imidazole (for example, pimonidazole, 2(2nitro1Himidazol1yl)N(2,2,3,3,3 pentafluoropropyl)acetamide (EF5)). These compounds need to be systemically administered to patients and generate stable adducts with proteins in hypoxic conditions; these can be detected by the use of specific antibodies on tumor biopsies. The main limitations of these methods are their invasiveness (they are performed on tumor biopsies), nonrepresentative sampling (the tumor can be very heterogeneous and biopsies can be nonrepresentative of the whole tumor), and the inability to perform multiple evaluations so as to follow changes in tumor oxygenation after treatment [59].
A more recently developed technique for imaging hypoxic tumors that is now being implemented in the clinic is the use of nitroimidazole derivatives in combi nation with positron emission tomography (PET). Several derivatives of nitroimidazole are now being studied in order to identify the best tracer with high uptake and low toxicity [60,61]. Among these, 18 Ffluoromisonidazole ( 18 FMISO) is the most extensively studied, and it has an investigational new drug (IND) authorization from the Food and Drug Administration (FDA) as an investiga tional product for use in humans. Although the 18 F MISOPET technique is noninvasive and allows the serial imaging of hypoxia, the accumulation of 18 FMISO in hypoxic tumors is relatively low. This results in a low signaltonoise ratio and hence a poor contrast between hypoxic tumors and surrounding normal tissues (for a detailed review, see [62]).
The imaging of tumor hypoxia by blood oxygen level dependent magnetic resonance imaging (BOLD MRI) is also being investigated. This modality relies on the detection of paramagnetic deoxyhemoglobin within red blood cells, and does not require administration of exoge nous tracers. The main limitations of this technique are the fact that it does not measure tissue pO 2 directly and could be influenced by blood flow, tumor perfusion and other vascular parameters.
In addition to these difficulties, it is becoming clear that assessing one single factor, such as HIF1, does not reflect the complexity of a tumor response to hypoxia, and hence is unlikely to be a reliable marker [3,5]. More comprehensive approaches for the detection and selec tion of hypoxic tumors for therapy have therefore been investigated.

Gene signatures of hypoxia
The identification by global expression analysis of multi ple genes (that is, gene signatures) and pathways that are responsive to hypoxia might overcome most of the limitations of current markers and other detection methods. Such gene expression signatures also have the potential to reflect the complexity of the tumor hypoxia response. They could, therefore, be used to reveal the nature of the hypoxic response to a specific therapy in terms of gene networks and hence improve our under standing of mechanisms of resistance. This would enable not only the identification of prognostic and predictive markers but also the selection of novel targets for thera peutics.
Several groups have derived hypoxia gene expression profiles that have prognostic significance in breast cancer [47,6367] (Table 3). For example, Winter et al. [47] defined an in vivo hypoxia 'metagene' (signature) in head and neck squamous cell carcinomas (HNSCCs) by clustering (that is, by finding) genes whose expression pattern was similar to that of a set of wellknown hypoxiaregulated genes, including CAIX, GLUT1 and VEGF. The metagene contained 99 genes, several of which were previously described as hypoxiaresponsive in vitro. These genes included Aldolase A (ALDOA), Glyceraldehyde 3phosphate dehydrogenase (GAPDH), Placental growth factor (PGF) and BNIP3 as well as some new genes that could play an important role in the hypoxic response in vivo, such as Metaxin 1 (MTX1), Breast cancer antiestrogen resistance 1 (BCAR1), Protea some subunit α type7 (PSMA7) and Solute carrier organic anion transporter family member 1B3 (SLCO1B3). This signature proved to be prognostic in independent HNSCC and breast cancer series [47]. Some of these genes are being studied in ongoing followup studies. An example is Iron sulfur cluster scaffold homolog (ISCU), a gene that was downregulated in the hypoxia signature; this gene was subsequently found to be a target of the hypoxiaregulated hsa-miR-210 and a good prognostic factor [68].Chi et al. [65] analyzed the gene expression profiles of mammary and renal tubular epithelial cells that were exposed to low O 2 levels. They derived a signa ture called 'epithelial hypoxia signature' that presented coordinated variation in several human cancers. Of particular note, they found that a set of renal tumors could be stratified into two groups, one with high and one with low expression of the hypoxiaresponse genes. The highhypoxiaresponse group included clearcell renal cell carcinomas, which frequently present high levels of HIF1α and/or HIF2α because of the loss of functional pVHL. The signature could also differentiate between low and highsignatureexpression groups in a set of ovarian cancer samples and two different sets of breast cancer samples. In one of the breast cancer sets, Chi et al. [65] found a significant association between high expression of the hypoxia signature and mutation in p53, negative estrogen receptor status and high grade tumors. In all of these sample sets, those patients assigned to the highexpression group had the worse prognosis. Finally, Chi et al. [65] also showed that the generated signature was an independent predictor of poor prognosis, proving its potential in clinical decision making. Seigneuric et al. [67] used the data from Chi et al.'s study [65] to distinguish gene signatures in human mammary epithelial cells that are associated with early (1, 3 and 6 hours) hypoxic exposure rather than late (after 12 and 24 hours) hypoxic exposure. They showed that only the earlyexposure gene signature had significant prognostic power, allowing the stratification of a cohort of patients with breast cancer into two groups: those with low expression of the early hypoxic response signature (better prognosis) and those with high expression of this signature (worse prognosis).
More recently, Buffa et al. [63] derived a hypoxia signature that is common to HNSCC and breast cancers. They used a metaanalysis approach to generate a more general and robust signature that might better reflect tumor response to hypoxia in vivo and be better suited for clinical use. They showed that a reduced metagene including as few as three genes (VEGFA, Solute carrier family 2 member 1 (SLC2A1; also known as GLUT1) and Phosphoglycerate mutase 1 (PGAM1)) had prognostic power similar to that of a large signature in independent breast cancer, HNSCC and lung cancer series. But they also validated a networkbased approach that considers multiple hypoxia prototype genes, builds a coexpression network of hypoxiarelated genes across clinical series, and then uses the network to generate biologically and clinically relevant hypotheses. For example, Buffa et al. [63] showed that genes involved in angiogenesis (VEGFA), glucose metabolism (SLC2A1, PGAM1, Enolase I (ENOI), LDHA, Triosephosphate isomerase II (TPII) and ALDOA) and cell cycling (CDKN3) were among those most likely to be overexpressed both in hypoxic HNSCC and hypoxic breast cancers. These genes could all contribute to global survival pathways triggered by hypoxia in vivo.
Despite cellline diversity, the derivation of gene signa tures using in vitro model systems can be powerful because some of the fundamental processes are con served and clean experimental design can be easily applied. Conversely, the in vivo tumor system requires consideration of multiple cell types, microenvironmental changes and threedimensional complexity. Approaches that integrate knowledge of gene function garnered from in vitro experiments with the analysis of expression in vivo might deliver signatures that better represent the hypoxia response that occurs in cancer.
Gene signatures reflect the hypoxic response at the transcriptional level, which is only part of the story of the overall effect of hypoxia. miRNA signatures are therefore under investigation as posttranscriptional regulators of the hypoxic response. miRNA signatures of hypoxia miRNAs are small noncoding RNAs that control gene expression posttranscriptionally by regulating mRNA translation and stability [69,70]. The expression of miRNAs in tumors and normal tissues has been com pared, and the differences have been found to affect cellular processes, including proliferation, apoptosis and metabolism, with the miRNAs acting as either oncogenes or tumor suppressors [71,72]. Furthermore, changes in miRNA expression have been associated with clinico pathological features and disease outcome in different tumor types, including breast cancer [7376].
Several hypoxiainducible miRNAs have been identi fied and two studies have focused their attention on breast cancer [77,78]. Kulshreshtha et al. [78] compiled a list of miRNAs that were consistently upregulated across a panel of breast and colon cancer cell lines exposed to hypoxia. Moreover, several of the miRNAs that were included in this signature were also overexpressed in breast cancer and other solid tumors, suggesting that hypoxia could be a key factor in miRNA modulation in  [77,78] and its expression levels significantly corre lated with a hypoxia gene expression signature in breast cancer [47], suggesting that it is also regulated by hypoxia in vivo. Furthermore, hsa-miR-210 expression was prog nostic in a study of 210 breast cancers [77]. Great effort is now being directed towards unveiling targets that contribute to tumor aggressiveness. Com parative analysis of hypoxiaregulated miRNAs using gene expression profiles might add valuable information to the interrogation of targetprediction algorithms. Several targets have been investigated to date ( Figure 2 and Table 4) showing roles for hsa-miR-210 in cellcycle regulation, apoptosis, iron accumulation, the production of reactive oxygen species, cell metabolism, DNA repair, tumor initiation, and the survival, migration and differen tiation of endothelial cells (Figure 2) [68,7987]. Of parti cular note, our group recently showed the major biological effects of miR-210 in targeting ISCU, all of which are likely to contribute to important phenotypes in cancer. By downregulating ISCU, miR-210 decreases the activity of Kreb's cycle enzymes and mitochondrial function, contributes to an increase in free radical generation in hypoxia, increases cell survival under hypoxia, induces a switch to glycolysis in both normoxia and hypoxia, and upregulates the iron uptake required for cell growth. Importantly, analysis of more than 900 patients with different tumor types, including breast cancer, showed that the suppression of ISCU was corre lated with a worse prognosis [68].
Although most studies on miRNAs have focused their attention on miR-210, other miRNAs could contribute to the hypoxic response. For example, experimental evidence suggests that miR-26 and miR-107 might have roles in cell survival in a lowoxygen environment [78]. A recent study has shown that miR-495 is robustly up regulated in a subset of a breast cancer stem cell population, both in stabilized cancer cell lines and in primary cells [88], where it promotes colony formation and tumorigenesis. Moreover, miR-495 is involved in main tenance of the cancer stem cell phenotype, in invasion by suppression of Ecadherin, and in hypoxia resistance through modulation of the REDD1 mTOR pathway.
Finally, the ability to detect miRNAs (for example, hsa-miR-210) in plasma and urine, as well as in tumor tissues, further increases the clinical potential of these small molecules [89].
Although this young field is undergoing rapid development, there are as yet no signatures that can be used in the clinical setting, but the results show that this area of research has great potential.

Conclusions
Hypoxia occurs in most solid tumors, and has been associated not only with malignant progression and poor  Table 4 for a full list of targets, full names and related references.  prognosis but also with specific resistance to anticancer therapies. Many biomarkers have been suggested for hypoxia, but they all have limitations. Furthermore, it is unlikely that a singlegene biomarker will be sufficient to characterize the complexity of a tumor's response to hypoxia. Several gene and miRNA expression signatures have also been suggested, and these have revealed common alities and specificities of the hypoxia response in different experimental cancer systems both in vitro and in vivo. These signatures promise greater prognostic and therapeutic potential than singlegene markers, but the specific interactions between these signatures, the HIF response and responses to treatments remain unclear. A full understanding of these interactions is of paramount importance both when assigning the most beneficial treat ment to patients and when designing new thera peutic strategies, such as combined modality treatments and multitarget or multiplehit strategies. In this respect, the validation, optimization and assessment of these potential biomarkers in prospective clinical studies and randomized trials are increasingly needed to trans form them into useful clinical tools.