Hypoxia-Inducible Factor 1α Expression as an Intrinsic Marker of Hypoxia: Correlation with Tumor Oxygen, Pimonidazole Measurements, and Outcome in Locally Advanced Carcinoma of the Cervix

It is well established that the microenvironment plays an important role in the development and growth of a tumor. In particular, the tissue hypoxia that results from an inadequate supply of oxygen is now recognized as a key factor driving tumorigenesis (1, 2). Hypoxia leads to the up-regulation of a number of genes associated with increasing the intrinsic biological aggressiveness of a tumor (1), both locally and by increasing the likelihood of distant spread. Hypoxia may also reduce the effectiveness of cancer treatment because hypoxic tumors are inherently resistant to radiation (3) and some chemotherapeutic agents (1).

All cells, whether normal or abnormal, sense a decrease in oxygen and respond by activating a number of physiologic responses such as erythropoiesis (4), glycolysis, and angiogenesis (5, 6). This response is mediated by activation of the hypoxia-inducible factor (HIF) family of genes, which code for heterodimeric basic helix-loop-helix proteins composed of α and β subunits. HIF-1α is manufactured in the cytoplasm of cells but is rapidly degraded under normoxia. However, the intracellular content of HIF-1α increases immediately after a decrease in oxygen tension. This increase in HIF-1α occurs as part of a redox-sensitive stabilization (7). The stabilized HIF-1α dimerizes with HIF-1β, and the heterodimers bind to DNA on the hypoxia response elements contained within the promotor regions of target genes. These target genes such as carbonic anhydrase 9 (CA9), Glut-1, vascular endothelial growth factor, and erythropoietin are involved in the adaptation of a tumor to its environment and thus in the development of a more malignant phenotype (4, 5).

One of the current interests in HIF is to evaluate its potential as an intrinsic marker of tumor hypoxia. Several groups have shown the value of measuring hypoxia as a prognostic factor in solid human tumors (8, 9, 10). Although oxygen electrodes have been widely used (8), the method is only suitable for accessible tumors. In view of the increasing number of strategies being tested for therapeutically targeting hypoxic cells (11, 12), there is an obvious need for a reliable method for measuring tumor hypoxia on a routine clinical basis. One approach being studied is to use biopsies taken at the time of initial diagnosis and examine the expression of hypoxia-inducible proteins as potential intrinsic markers of hypoxia. For example, previous work by our group showed that in carcinoma of the cervix expression of CAIX (13) and Glut-1 (14) showed some correlation with pimonidazole binding (15) and with the level of tumor oxygenation measured using Eppendorf oxygen electrodes and treatment outcome (13, 14). In this study, we have now examined the expression of HIF-1α in relation to data obtained with other methods for measuring/assessing tumor hypoxia (oxygen electrode measurements, pimonidazole binding, and CAIX expression) and to prognosis. The hypothesis behind the work was that, in view of its key role in oxygen sensing, HIF-1α might be a better intrinsic marker of hypoxia than other proteins studied and so would be a strong prognostic factor for treatment outcome.

The work was carried out after approval from the South Manchester Ethics Committee, and all patients gave informed consent. The study comprised patients registered at the Department of Oncology at the Christie Hospital between 1997 and 2000 (used for comparison of biological markers) and between 1987 and 1993 (used for correlations with outcome) with histologically proven carcinoma of cervix. The patients had International Federation of Gynecologists and Obstetricians stage Ib–IVa disease and were treated with curative radiotherapy according to the standard techniques of the Manchester School (16). Tumor biopsies were taken at the time of the staging examination under anesthetic. Information on tumor differentiation was obtained from the hospital pathology department. Tumor diameter was measured either by magnetic resonance imaging to obtain an average diameter from three planes or by clinical examination. For the 1987 to 1993 series of patients treated with radiation alone, the sites of any disease relapse were identified clinically and radiologically and, where appropriate, confirmed on biopsy. Recurrences were classified as local (i.e., within the radiation field) or metastatic (i.e., outside the radiation field). When recurrences were both local and metastatic, both were taken into account. The median follow-up in surviving patients was 60 months (range, 27–115 months).

Because several authors have reported deterioration in immunohistochemical staining in stored sections (17, 18), sections from formalin-fixed, paraffin-embedded tumor biopsies were used within 3 months of cutting. Immunohistochemistry was carried out using the Tyramide Signal Amplification System (NEN Life Sciences, Boston, MA), which uses a streptavidin-biotin-horseradish peroxidase complex. Sections (4 μm thick) were dewaxed in xylene, passed through graded alcohols to water, microwaved in EDTA for 25 minutes, and then processed according to the manufacturer’s protocol. The primary antibody (mouse monoclonal ab463; Novus Biologicals, Littleton, CO) was applied at a dilution of 1:1,000 with overnight incubation at 4°C. The secondary antibody (rabbit antimouse immunoglobulin/biotin; Dako EO413; Dako, Carpinteria, CA) was applied at a 1:400 dilution for 30 minutes at room temperature. The sections were then incubated with streptavidin-horseradish peroxidase according to the manufacturer’s instructions and incubated for 10 minutes with the biotinyl tyramide amplification reagent. The antibody was visualized using 3,3′-diaminobenzidine, and the sections were counterstained with Gills No. 1, dehydrated, and coverslipped. A total of 177 tumors were stained for HIF-1α: 99 from the series of patients treated between 1987 and 1993 and for whom radiotherapy outcome data were obtained, and another 78 from patients treated between 1997 and 2000 for whom tumor oxygenation data were available.

For the series of patients treated between 1997 and 2000, the sections were immunostained with a mixture of antibodies to CD31 (Dako M0283; Dako) and CD34 (Dako M7165; Dako) after HIF-1α staining but before counterstaining and coverslipping. The method has been described in detail elsewhere (19), with the exception of the goat antimouse immunoglobulin being applied at a 1:25 dilution followed by a 1:50 dilution. Color was developed using the alkaline phosphatase anti-alkaline phosphatase method (19).

Only tumor nuclear HIF-1α staining was scored using a method described elsewhere (20). Both the percentage area stained and the intensity of the staining were scored as described by others (20): 0, no nuclear staining; 1, ≤1% staining; 2, 1% to 10% slight to moderate staining or 10% to 50% slight staining; 3, 10% to 50% moderate staining; and 4, ≥50% moderate to marked staining. Scoring was performed in a double-blind manner by two independent investigators (G. J. H. and H. R. V.). Any disagreement in score was resolved by discussion, and a final score was agreed on.

Sterile polarographic needle electrodes were used to perform oxygen measurements. The methods are described in detail elsewhere (21). Measurements were made with the patients under general anesthetic (propofol infusion and nitric oxide). The individual patient oxygen measurements were expressed as the median pO₂ (mm Hg) and HP5 (the percentage of pO₂ values <5 mm Hg).

The methods used for the administration of pimonidazole (Hypoxyprobe-1; NPI Inc., Belmont, MA) to patients, taking tumor biopsies, and staining sections for the level of pimonidazole staining have been described elsewhere (22). Patients were biopsied 12 to 20 hours after receiving pimonidazole. A semiquantitative scoring system was used to quantify pimonidazole binding (23). Whole sections were scored on a field-by-field basis to estimate the area of immunostained tumor cells using a grading scale of 0–4 (0, 0%; 1, >0% to 5%; 2, >5% to 15%; 3, >15% to 30%; 4, >30%). Areas of necrosis, stroma, normal epithelium, or obvious edge artifact were excluded. The level of pimonidazole binding in the tumor was calculated as the average score for all of the fields.

The method has been described elsewhere (13). The primary antibody was a mouse antihuman antibody raised to the external domain of CA9 (24). CAIX staining was scored semiquantitatively, as follows: 0, <1%; 1, 1% to 10%; 2, 10% to 30%; and 3, >30%.

The nonparametric Spearman’s rank correlation was used to examine the relationships between the end points studied. Kaplan-Meier survival curves and Cox regression analysis were used to assess outcome. For the survival analyses, small groups were pooled (stages III and IV; HIF-1α scores of 0, 1, and 2). Variables were treated either as a factor (taking no account of the ordering in the levels) or as a numeric variable (to test for trend). The hazard ratio (HR) is an estimate of the relative risk in the group compared with the reference group, which is indicated with a HR of 1. A HR of >1 implies the variable increases the risk. Bivariate analyses were carried out testing for the effect of HIF-1α after allowing for potential confounding factors, with HIF-1α being entered as a numerical variable (trend test). Tests were performed for interactions between HIF-1α and the confounding factors to investigate the possibility that the effect of HIF-1α differed between subgroups. All statistical tests were two-sided at the 0.05 significance level, with no allowance made for multiple testing.

Staining was predominantly nuclear, but cytoplasmic staining was seen occasionally. Dual staining with the endothelial cell markers CD31/CD34 enabled the visualization of HIF-1α staining in relation to blood vessels. Staining varied between and within sections, and both perivascular and perinecrotic staining patterns were observed (Fig. 1). Scoring was repeatable with good intraobserver (r = 0.76; P < 0.0001; n = 78) and interobserver (r = 0.86; P < 0.0001; n = 78) reproducibility. Any difference in the score between observers was usually only by a single point. In the 13 cases in which there was disagreement, the slides were reanalyzed in conference, and a final consensus score was agreed on.

In 72 patients, a median of 4 oxygen electrode tracks (range, 1–7 oxygen electrode tracks) was made per tumor, giving a median of 128 oxygen measurements (range, 32–300 oxygen measurements). The median pO₂ was 4 mm Hg (range, 0–45 mm Hg), and the median HP5 was 54% (range, 0–97%). There was a weak negative correlation between HIF-1α expression in the tumors and oxygenation measured as median pO₂ (r = −0.26; P = 0.03; n = 72) but no relationship with HP5 (r = 0.13; P = 0.26; n = 72). Data for tumor pimonidazole staining were available for 36 patients. There was a weak positive correlation between the level of HIF-1α expression and pimonidazole binding in the tumors (r = 0.34; P = 0.040; n = 36; Fig. 2). In 160 patients, tumor biopsies were also stained for CAIX (13). There was a weak but statistically significant positive correlation between HIF-1α and CA9 expression (r = 0.27; P = 0.001).

In the 1987 to 1993 series of patients, an analysis was made of the relationship between the expression of HIF-1α in tumors and the outcome of patients after radiotherapy. There was no significant correlation between HIF-1α expression and disease stage (r = 0.18; P = 0.077; n = 99), tumor size (r = 0.14; P = 0.30; n = 55), or patient age (r = 0.03; P = 0.76; n = 99). In 88 squamous cell carcinomas, there was also no significant relationship between HIF-1α expression and tumor grade (r = −0.18; P = 0.10). Data were also available from a study measuring the intrinsic radiosensitivity of tumors as surviving fraction at 2 Gy [SF2 (25)]. There was no significant relationship between HIF-1α expression and SF2 (r = 0.022; P = 0.87; n = 57).

For the examination of HIF-1α expression in relation to treatment outcome, small groups were pooled (HIF-1α scores of 0, 1, and 2; clinical stage 3 and 4). Table 1 summarizes the distribution of patients according to HIF-1α expression and clinical parameters. Table 1 also shows the distribution of patients according to HIF-1α expression and SF2. Table 2 summarizes the univariate analyses of treatment outcome. In the series of patients studied as a whole, the level of HIF-1α expression in tumors did not emerge as a prognostic factor for disease-free survival (P = 0.56), metastasis-free survival (P = 0.54), or local recurrence-free survival (P = 0.27; Fig. 3). Tumor stage was a significant prognostic factor, and tumor size was prognostic for metastasis-free survival. Cox regression analyses were carried out to examine for the potential influence of confounding factors on HIF-1α expression. After allowing for tumor stage, grade, and patient age, HIF-1α was not a significant prognostic factor for treatment outcome. However, there was a difference in the association between HIF-1α expression and treatment outcome depending on tumor size, with high HIF-1α being associated with increased risk in small tumors and decreased risk in large tumors. A formal test for interaction in the 54 patients reached statistical significance for local control [P = 0.009; HR = 2.89 (95% CI, 0.87–9.65) for small tumors compared with HR = 0.39 (0.15–1.03) for large tumors] and survival [P = 0.046; HR = 1.46 (0.54–3.99) for small versus 0.40 (0.19–0.85) for large tumors]. There were no significant interactions of HIF-1α with stage, grade, or age. Although the difference between small and large tumors was based on a formal test of interaction, the small numbers of cases and the lack of a prespecified hypothesis suggest that this result should be treated with some caution until confirmed by an independent study. Because the numbers of patients were small in the bivariate analyses, for graphical presentation HIF-1α expression was also grouped as two variables (Fig. 4), and for the two groups, the same significant interaction was seen for both local control (P = 0.001) and survival (P = 0.042).

HIF-1α expression is a common feature of solid human tumors and has been reported in many different tumor types (26, 27, 28, 29, 30, 31). In the study described here, 96% of the tumors stained positively for HIF-1α, a figure that is similar to the 81% (31) and 100% (32) reported by others in carcinoma of the cervix. The scoring system used (20) was reproducible with excellent inter- and intra-scorer agreement. An association was found between the level of HIF-1α staining and median pO₂ but not HP5. The lack of correlation between HP5 and HIF-1α differs from the recently published data from Haugland et al. (32), where an association was found with HP5 (r = 0.40; P < 0.01; n = 42). Although the same staining method was used, the scoring system was different. Haugland et al. (32) used computerized image analysis software to score the slides, whereas we used a simple subjective scoring system as used by others to obtain statistically significant prognostic data (20). Together, the work of Haugland et al. (32) and that reported here suggest there is only a weak relationship between HIF-1α expression in a tumor and the level of oxygenation measured using oxygen electrodes.

The weak relationship between HIF-1α expression and tumor oxygenation probably reflects a number of factors. Tumor heterogeneity may play some part with sampling difference between the methods. A punch biopsy taken from the tumor surface might not reflect the level of oxygenation measured by an Eppendorf electrode as it passes through a tumor. On the other hand, oxygen electrodes will pass through normal and necrotic tissue, areas that can be excluded from immunohistochemistry measurements. In addition to the influence of tumor heterogeneity and sampling differences, differences in the type of hypoxia measured might be important. That is, whether the different approaches for measuring tumor hypoxia reflect acute or chronic hypoxia and low or intermediate hypoxia. Although oxygen electrodes will measure all types of hypoxia, HIF-1α stabilization (and therefore expression) occurs when pO₂ levels decrease below 10 mm Hg, and HIF-1α is rapidly degraded (within seconds) when normoxia returns (7, 33).

The third factor that might lead to a weak correlation between oxygen electrode and HIF-1α data is the up-regulation of HIF-1α by mechanisms other than hypoxia (34). HIF-1α protein can be stabilized by the phosphatidylinositol 3′-kinase pathway (35, 36) and also by heavy metals such as iron and cobalt ions (37, 38). In support of the latter suggestion, HIF-1α was expressed diffusely throughout tumor sections (Fig. 1), i.e., it was up-regulated close to blood vessels (possibly reflecting acute hypoxia), adjacent to necrosis and overlapping with the level of pimonidazole staining (chronic hypoxia), and in other areas of the tumors.

In addition to the weak relationship between HIF-1α expression and pO₂, a correlation was found with the level of pimonidazole staining and CAIX expression. The latter finding extends the results reported previously in nasopharyngeal (39) and non–small-cell lung carcinomas (40). The finding is consistent with the observation of a weak relationship between HIF-1α expression and oxygen electrode data, pimonidazole being a hypoxia specific marker, and with the known role of HIF1-α in the induction of CAIX under hypoxia (41). Together, these data suggest that HIF-1α expression is an intrinsic but weak marker of hypoxia in locally advanced carcinoma of the cervix.

In the study reported here, HIF-1α expression was not a prognostic factor for survival in carcinoma of the cervix. This finding differs from other studies, in which HIF-1α expression was a significant prognostic factor in cervical (42, 43) and other (20, 29, 39, 44, 45, 46) cancers. It could be that hypoxia is not predictive of outcome in the series of patients studied. Analysis of the prospectively collected pO₂ and pimonidazole data in relation to treatment outcome is awaiting maturation of the clinical data. However, an interim analysis of the pO₂ data was described by Loncaster et al. (47) and showed that the pO₂ data yielded significant prognostic information for local control. However, our finding agrees with that of Swinson et al. (48), who found no correlation with prognosis in non–small-cell lung cancer. It may be that the role of HIF-1α as a prognostic factor is tumor type specific. Our exploratory, hypothesis-generating analyses suggest that HIF-1α may act in different ways at different stages or sizes of carcinoma of the cervix. In small tumors, high HIF-1α expression was associated (as expected) with a poor outcome, consistent with a tumor cell adaptation to hypoxia, e.g., increased angiogenesis. The possible switch in large tumors to an association between high HIF-1α expression and good outcome might relate to changes in the balance of genes up-regulated by HIF-1α, e.g., proapoptotic molecules might be preferentially induced (49).

In summary, the transactivation of HIF-1α is a complex process and involves several different pathways and molecules. The potentially complex relationship between HIF-1α expression and treatment outcome implied by the work reported here suggests it has no role as a prognostic or predictive factor in locally advanced carcinoma of the cervix.