Expression of Hypoxia-inducible Factor-1α

INTRODUCTION

Hypoxia is a common feature of many malignant neoplasms and exacerbates the disease in two main ways: it promotes both local and systemic tumor progression (1) and may compromise the beneficial effects of chemotherapeutic drugs (2) and ionizing radiation. The deleterious effects of hypoxia on radiotherapy outcome have long been established, based upon a formidable body of experimental evidence. That cells maintained under anaerobic conditions are relatively insensitive to radiation was first demonstrated in the 1930s (3) . This pioneering work laid the foundation for a large number of subsequent experimental studies in vivo, which have shown unequivocally that hypoxia interferes with the response of tumors to radiation (4) . Various methodological approaches have been developed to measure tumor hypoxia, including exogenous markers for magnetic resonance spectroscopy and immunohistochemical staining as well as polarographic oxygen electrodes (5) . Indeed, using this latter technique, pretreatment oxygenation levels have been found to be predictive of the radiation response and survival of patients with cancer of the uterine cervix (6 , 7) and of the head and neck (8 , 9) . Widespread application of the technique is precluded by the invasiveness of the method, its use being restricted to superficially located tumors.

HIF 3 -1 plays a pivotal role in essential adaptive responses to hypoxia, its expression and transcriptional activity increasing exponentially with decreases in levels of cellular oxygen (10, 11, 12) . Several dozen target genes are known to be transactivated by HIF-1, including those encoding erythropoietin, glucose transporters, glycolytic enzymes, and vascular endothelial growth factor (13) . HIF-1 is a heterodimeric basic-helix-loop-helix-PER-ARNT-SIM transcription factor that is composed of two subunits, HIF-1α and HIF-1β (10 , 14) . Whereas HIF-1β can dimerize with several basic-helix-loop-helix-PER-ARNT-SIM transcription factors, HIF-1α is unique to HIF-1; the levels of its expression are the primary determinant of HIF-1 DNA binding and transcriptional activity (11 , 15) . Induction of HIF-1α thus appears to be a critical step in the hypoxic response and occurs via increased mRNA expression, protein stabilization, nuclear localization, and augmented activity of its transcriptional activation domains (13) . Nuclear accumulation of the protein can be detected immunohistochemically and has been shown to occur not only in human malignancies and their metastases (16, 17, 18, 19) but also in ischemic myocardial tissue (20) . The present study reveals HIF-1α to be expressed in the vast majority of patients with squamous cell cancer of the oropharynx. In the same population of individuals, the local cure rate upon radiation therapy has already been shown to depend inversely on the IMD, a marker for angiogenic activity (21) . We now show that elevated levels of HIF-1α are closely and independently correlated with reduced rates of local failure-free, disease-free, and overall survival, thereby indicating radioresistant and aggressive disease.

PATIENTS AND METHODS

Patients.

For inclusion in this study, patients were required to have presented at the Department of Radiation Oncology, Inselspital Bern, between October 1991 and December 1997 with histologically proven squamous cell cancer of the oropharynx and to have been treated by radical external beam radiotherapy. A review of the radiation records identified 139 patients satisfying entry criteria. Forty-one of these were subsequently excluded because of the small size of the biopsy (18 patients), previous or synchronous malignancies (12 patients), irradiation after neck dissection (7 patients), intercurrent death (2 patients), lack of follow-up (1 patient), or the existence of distant metastasis at the onset of treatment (1 patient). The principal clinical characteristics of the patient cohort are summarized in Table 1 ⇓ . With the approval of the regional board of the Medical Ethics Commission, paraffin-embedded tissue samples were obtained from the archives of the Pathology Department. The histopathological diagnosis and grading of biopsies were reviewed by an experienced pathologist (J. L.). Clinicopathological covariables were derived from patient charts.

After the placement of a thermoplast mask, all patients underwent planning computer tomography, as well as two- or three-dimensional based planning and control using simulator and portal vision imaging devices, prior to radiotherapy. Megavoltage radiation was delivered by means of a linear accelerator in daily fractions, five times/week for 5–8 weeks. A median total dose of 74 Gy (range, 54–80.5) was administered; 97% of the patients received at least 66 Gy. Twenty-seven patients were subjected to concomitant cisplatin-based chemotherapy, 12 patients underwent cisplatin monotherapy, and 13 were treated with both cisplatin and 5-fluorouracil; 1 individual received methotrexate, and 1 patient received carboplatin. In patients with CR of the primary tumor and with clinical or radiological evidence of persistent lymph node metastasis, a neck dissection was performed.

Baseline studies included physical examination, chest X-rays, panendoscopy of the upper aerodigestive tract, and magnetic resonance imaging or computed tomography of the neck. During treatment, patients were examined on a weekly basis. The response to treatment was assessed 2–4 weeks after the end of therapy; the primary tumor and lymph node metastases were evaluated separately. After treatment, all patients underwent clinical examination and imaging on a regular basis. Fifty-five patients (56%) were observed until their death; the median follow-up period for survivors was 2.6 years.

Immunohistochemistry.

From the 98 patients, 148 paraffin-embedded biopsies were processed for immunohistochemistry. Two or more biopsies were available in 20 patients. Sections (3-μm thick) were transferred to gelatinized microslides and air-dried overnight at 37°C. They were dewaxed in xylene (three changes), rehydrated in a graded series of decreasing ethanol concentration, and rinsed then in Tris-buffered saline [50 mm Tris/HCl (pH 7.4) containing 100 mm sodium chloride]. Immunostaining was performed according to the Catalyzed Signal Amplification System (Dako, Carpinteria, CA), which uses a streptavidin-biotin-horseradish peroxidase complex. The slides were initially immersed in target retrieval solution (Dako) at 97°C for 15 min and thereafter treated in accordance with the manufacturer’s instructions. They were exposed to a monoclonal antibody against HIF-1α (H1α67, used at a dilution of 1:10,000; Ref. 17 ) for 30 min. The biotinyl tyramide amplification reagent was diluted 1:10 in protein blocking solution (Dako). The reaction product was visualized by exposing sections to 3,3-diaminobenzidine for 1 min. Nuclei were lightly counterstained with hematoxylin. Sections were then mounted in Aquatex (Merck, Darmstadt, Germany). Tissue samples incubated with nonimmune serum or with the antibody diluent (Dako) served as negative controls.

Tumor cell immunoreactivity was scored according to the nuclear staining. Both the extent of staining (relative number of HIF-1α-positive cells) and the intensity of the reaction were taken into account: −, not detected; +, <1% positive cells; ++, 1–10% positive cells with slight to moderate staining or 10–50% positive cells with slight staining; +++, 10–50% positive cells with moderate to marked staining; ++++, >50% positive cells with moderate to marked staining. For all statistical tests, the five grades of staining were reduced to three: 0, not detected; I, weak (+/++); II, strong (+++/++++). When more than one biopsy/patient was available, the highest score was selected for further evaluation. Assessment was performed in a blinded fashion and independently by two investigators (D. M. A., P. B.). Conflicting scores were resolved at a discussion microscope.

Statistics.

Bivariate association between ordinal variables was assessed using Spearman’s correlation (exact version), which yielded the correlation coefficient, rho. For categorical data, Pearson’s χ² test was used. All tests of statistical significance were two-sided.

Correlations between variables and the response to treatment as well as distant metastasis (time-independent outcome analysis) were determined using the logistic regression method, implemented in both the univariate and multivariate fashion and including a backward elimination procedure to remove variables with P ≥ 0.1. The qualifying criteria for inclusion in the multivariate analysis were P < 0.1, or OR <0.5 or >2, in the univariate analysis.

Variables were correlated with local failure-free survival, disease-free survival, and overall survival. The analysis of local failure-free survival considered both local tumor progression after incomplete remission and local relapse after CR as adverse events, whereas that of disease-free survival took both local failure and distant metastasis into account. The analysis of overall survival included death from any cause. Survival was measured from the time when therapy was initiated to that when the first adverse event was detected or to the date of the last follow-up. Deaths attributable to nontumor-related causes were censored, except in the analysis of overall survival. Survival curves were plotted according to the Kaplan-Meier method; the log rank test was used to determine significant differences between these. A Cox regression was performed to calculate the RRs. Qualifying criteria for inclusion in the multivariate Cox regression analysis were P < 0.1, or RR <0.5 or >2, in the univariate analysis. A backward elimination procedure was then performed to eliminate nonsignificant variables (P ≥ 0.1).

Because IMD has been shown for the same patient cohort to be closely correlated with the therapeutic outcome (21) , this parameter was included in all multivariate analyses. To obtain ORs and RRs comparable with the parameters reported in the present study, data pertaining to this variable were grouped into four categories, according to the number of microvessels/high power field: I, 0–80; II, 81–110; III, 111–130; and IV, >130.

Statistical analyses were performed using the SPSS package (version 9.0.0; SPSS, Inc., Chicago, IL).

RESULTS

Descriptive Statistics.

A total of 98 patients were eligible for this study. Patients and disease characteristics are presented in Table 1 ⇓ . Most of the individuals had advanced T stage (88% T_3–4) or positive N stage (66% N_1–3); tumors of the tonsillar fossa (39%) and base of tongue (30%) predominated. In 79 cases (81%), CR of the primary tumor was achieved, and in 33 of the 65 patients (51%) with nodal spread, CR of lymph node metastases was accomplished. In 42 patients (43%), local failure occurred; this was because of either relapse at the same site after CR (23 patients) or progression after incomplete remission (19 patients). Fourteen patients developed distant metastasis.

In all HIF-1α-positive tumor samples, immunostaining was nuclear; in rare instances, weak cytoplasmic reactivity was also observed. Two predominant patterns of nuclear staining were encountered: focal expression, with the most intense reaction occurring distal to the closest blood vessel and surrounding necrotic areas (Fig. 1, A and B) ⇓ ; and diffuse expression, independent of vessel proximity (Fig. 1, C and D) ⇓ . Predominant focal expression was found in 60 of 92 HIF-1α-positive tumors (65%), whereas diffuse staining was encountered in 32 tumors (35%). Negative controls manifested no immunoreactivity. Six patients registered negative for HIF-1α, 56 (57%) exhibited weak staining (category I), and 36 (37%) manifested strong immunoreactivity [category II (Table 1) ⇓] . In 17 of the 20 patients (85%) with more than one biopsy/tumor, the immunoreactivity scores of the biopsies were equal.

The only significant association between HIF-1α and patient or disease characteristics was a reverse correlation between the immunoreactivity and histological grade (r = −0.30; P = 0.01). The distribution of HIF-1α staining between smokers and nonsmokers and between patients receiving or not receiving chemotherapy was equally balanced. No significant correlation existed between immunostaining for HIF-1α and IMD (r = 0.07; P = 0.48).

Time-independent Outcome Analysis.

In the univariate analysis, HIF-1α was inversely correlated with both CR of the primary tumor (OR=0.33; P = 0.03) and of lymph node metastases [OR, 0.34; P = 0.02 (Table 2) ⇓] . IMD was predictive, as reported previously (21) . Of the four variables (age, T stage, HIF-1α, and IMD) qualifying for inclusion in the multivariate analysis of CR of the primary tumor, HIF-1α (OR, 0.30; P = 0.01) and IMD (OR, 0.44; P = 0.03) retained their significance. Age and smoking habits were eliminated from the model, with P ≥ 0.1. Of the four variables (T stage, N stage, HIF-1α, and IMD) qualifying for inclusion in the multivariate analysis of CR of lymph node metastasis, N stage (OR, 0.14; P = 0.002) and HIF-1α (OR, 0.30; P = 0.03) retained their significance. T stage and IMD were removed from the model, with P ≥ 0.1. In the univariate analysis of distant metastasis, no variable attained statistical significance; but T stage and HIF-1α, having OR values of 2.21 and 2.71, respectively, qualified for inclusion in the multivariate analysis. T stage was then eliminated (P ≥ 0.1), leaving HIF-1α with the same OR and borderline significance (P = 0.07) as in the univariate analysis.

Time-dependent Survival Analysis.

The univariate survival analysis revealed HIF-1α to be inversely correlated with local failure-free survival (P = 0.006, log rank), disease-free survival (P = 0.008, log rank), and overall survival [P = 0.001, log rank (Fig. 2) ⇓] ; the RRs were 2.15, 2.01, and 2.17, respectively, per increase of one staining grade (Table 3) ⇓ . The staining pattern (focal versus diffuse) had no significant influence on local tumor control or survival (Tables 2 ⇓ and 3 ⇓ ). The correlation of local tumor control with IMD has been reported previously (21) . According to the inclusion criteria (P < 0.1 or 0.5 > RR > 2.0), T stage, smoking habits, HIF-1α, and IMD qualified for entry into the multivariate Cox regression model for all three survival analyses. T stage, HIF-1α, and IMD remained in the models for local failure-free survival and disease-free survival, but smoking habits were eliminated (P ≥ 0.1). In the multivariate models, the RR of HIF-1α increased to 2.43 for local failure-free survival (P = 0.002), to 2.20 for disease-free survival (P = 0.004), and to 2.21 for overall survival (P = 0.0009), thereby establishing the independence of HIF-1α from T stage and IMD.

DISCUSSION

Prediction of the probability of successful treatment is of paramount importance for the individualization of therapy and the avoidance of either unnecessary toxicity or treatment failure. Because tumor hypoxia has been shown repeatedly to thwart the therapeutic effects of ionizing radiation, considerable efforts have been invested in the development of an easily applied means of measuring pretherapeutic tumor oxygenation levels (5) . The most promising results have been achieved using polarographic oxygen electrodes. Readings thus obtained confirmed the close relationship between tumor oxygenation and radiation response in cancer of the uterine cervix (6 , 7) and that of the head and neck (8 , 9 , 22) . Studies have focused on these two tumor entities because the method is invasive. The need for specific hypoxia markers that would be quantifiable either by noninvasive imaging (positron emission tomography and single photon emission computed tomography) or on routine tumor biopsies is therefore obvious. Several immunologically detectable 2-nitroimidazoles, which are metabolized predominantly in hypoxic areas, have been proposed as potential candidates and tested in various animal and human tumors (23) , including squamous cell cancer of the head and neck (24) . However, the use of such agents involves their i.v. injection prior to the removal of the biopsy, thus rendering necessary a second intervention after the primary diagnosis.

In the present study, we investigated the predictive and prognostic potential of HIF-1α, which is a subunit of the HIF-1 heterodimer, a protein that plays an essential role in oxygen homeostasis (13) . Our data indicate that HIF-1α is overexpressed in the neoplastic tissue of the vast majority of squamous cell cancers of the oropharynx, a tumor that is often treated by radical irradiation. Hypoxia is a principal determinant of HIF-1α accumulation, and immunostaining for this protein was encountered at some distance from tumor vessels (Fig. 1, A and B) ⇓ , reflecting the decrease in oxygen concentration with increasing distance from the capillaries. In other cases, however, immunoreactivity was observed diffusely throughout the entire tumor tissue (Fig. 1, C and D) ⇓ . These latter findings may reflect the existence of alternative regulatory modes of HIF-1α expression. The expression of HIF-1α has been shown to be enhanced by v-src (25) and in response to several growth factors, including insulin-like growth factors 1 and 2, basic fibroblast growth factor, and epidermal growth factor (26) . Activation of the phosphatidylinositol 3-kinase/AKT/FRAP pathway, which mediates signals from a broad range of growth factors, has likewise been demonstrated to increase HIF-1α expression (27) . In addition, p53 and the von Hippel-Lindau tumor suppressor protein have been implicated in the degradation of HIF-1α; the loss of their function results in augmented HIF-1α levels (28 , 29) .

We have demonstrated immunostaining for HIF-1α to be a powerful tool in predicting the success probability of radiotherapy in patients with squamous cell cancer of the oropharynx. To distinguish between its predictive value in short- and long-term disease control, CR of the primary disease and survival were analyzed separately. Both in the univariate and multivariate analysis, an inverse correlation was found to exist between HIF-1α and CR of the primary tumor (Table 2) ⇓ . This association between HIF-1α immunoreactivity and locoregional tumor control by radiotherapy was even more pronounced when local failure-free survival was considered, with P = 0.006 (Table 3) ⇓ . Again, the multivariate analysis confirmed HIF-1α expression to be predictive, irrespective of the T stage or IMD. In concert with this, HIF-1α and IMD were not correlated with one another. The simultaneous application of hypoxic cell-specific compounds, such as tirapazamine, represents a promising approach to overcome hypoxia-induced radiation resistance (reviewed in Ref. 30 ). Whether the assessment of HIF-1α expression can identify those patients who would benefit most from such drugs remains to be determined.

In addition to the association with a decreased local tumor control rate, HIF-1α expression was significantly correlated with impaired disease-free and overall survival (Fig. 2 ⇓ and Table 3 ⇓ ). Because locoregional control is crucial for the overall survival of head and neck cancer patients (31 , 32) , the prognostic power of HIF-1α staining lies mainly in its association with local tumor control. However, the OR for predicting distant metastasis according to HIF-1α expression was 2.71, with a borderline significance of P = 0.07 (Table 2) ⇓ . That P failed to fall below 0.05 may reflect the weak statistical power of the small patient group with distant metastasis (n = 14). Remarkably, all patients with systemic disease manifested HIF-1α immunoreactivity (Table 1) ⇓ , which supports the existence of a relevant association between the expression of this factor and distant metastasis. Indeed, a number of clinical studies have shown that low oxygen levels in tumors are associated with increased metastasis (33 , 34) . Experimental studies suggest at least three possible explanations for these findings:

• Hypoxia-induced cell death depends upon the presence of wild-type p53 (35) . Because cells with mutated p53 are far less susceptible to hypoxia-induced apoptosis than those with the wild-type form, intratumoral hypoxia will exert a strong selection pressure for p53-mutated cells, thereby increasing the likelihood of tumor progression.

• Hypoxia increases the mutation frequency (36) .

• Hypoxia is a major stimulus for angiogenesis, which also promotes distant metastasis (37) .

The only significant correlation that existed between HIF-1α expression and patient or disease characteristics was a reverse association with histological grade. Clinical studies using polarographic oxygen electrodes have failed to demonstrate the existence of a relationship between hypoxia and histological grade, either in cancers of the uterine cervix (7) or in those of the head and neck (38) . The significance of the association between HIF-1α and histological differentiation thus remains to be clarified.

In the present study, we have shown that HIF-1α is overexpressed in the vast majority of patients with squamous cell cancer of the oropharynx and, even more important, that the degree of HIF-1α immunoreactivity has strong predictive and prognostic significance in individuals treated by curative radiation therapy. The extent to which hypoxia accounts for this association with treatment failure remains to be clarified by simultaneous assessment of HIF-1α expression and tumor oxygenation. Because HIF-1α expression is also induced by tumor-specific genetic alterations (25, 26, 27, 28, 29) , it may provide greater prognostic information than measurement of intratumoral hypoxia.