The relation between hypoxia-inducible factor (HIF)-1α and HIF-2α expression with anemia and outcome in surgically treated head and neck cancer




Hypoxia promotes tumorigenesis through the hypoxia-inducible factor (HIF) pathway. There are 2 main homologues of the regulatory proteins, HIF-1α and HIF-2α, which have different effects in genetic knock-out experiments. Anemia may contribute to hypoxia by reducing oxygen delivery, but it is not known whether this influences HIF-α expression in tumors.


The expression of HIF-1α, HIF-2α, carbonic anhydrase-9 (CA-9), and peripheral hemoglobin (Hb) levels in 151 patients who underwent surgery for head and neck squamous cell carcinoma (HNSCC) were analyzed and related to outcome.


High HIF-1α was expressed in 45 of 140 tumors (30%), HIF-2α was expressed in 21 of 139 tumors (14%), and CA-9 was expressed in 56 of 149 tumors (62%). There was a positive correlation between HIF-1α expression and HIF-2α expression (P = .0001). HIF-1α alone was associated with a worse disease-specific survival (DSS) (P = .05) and disease-free survival (DFS) (P = .03) in multivariate analyses. Nine percent of tumors expressed both high HIF-1α and high HIF-2α. High HIF-1α/high HIF-2α expression was an independent prognostic factors in DSS (P = .04) and DFS (P = .005) in multivariate analyses. There was no correlation noted between Hb and HIF-1α, HIF-2α, or CA-9.


HIF-1α alone was correlated with DSS and DFS. The additive effect of HIF-2α on poor prognosis suggested that different pathways may be regulated by HIF-2α. Anemia that was not related to HIF-α expression suggests that tumor intrinsic factors regulate HIF-α; therefore, anemia may be a surrogate marker for other factors that affect outcome. Cancer 2006. © 2006 American Cancer Society.

Head and neck cancer represents the 5th most common cancer in men and 8th most common in women worldwide with approximately 600,000 new cases each year.1 Greater than 90% of the cancers are of squamous origin, arising from any part of the upper aerodigestive tract from the paranasal sinuses to the hypopharynx.2 Despite improvements in the treatment involving a combination of surgery, radiotherapy (RT), and chemotherapy. the 5-year survival rate for patients with advanced disease remains at approximately 50%.3, 4 New treatments that target molecular pathways are being developed in an attempt to improve the outcomes of these patients. However, a through understanding of the molecular pathways are required to direct these therapies successfully.

Head and neck squamous cell carcinomas (HNSCCs) are characterized by heterogeneous regions of hypoxia.5 Hypoxia results in cell death if it is severe or prolonged; however, cancer cells adapt to this hostile environment, enabling them to survive and proliferate. In part, it is this ability to adapt to a hostile environment that defines their malignant potential and characterizes a more aggressive phenotype.6

One way in which cells respond to reduced oxygen levels is through the hypoxia-inducible factor (HIF)-1 pathway. The HIF DNA-binding complex is a heterodimer of α and β subunits. The HIF-α subunits have a regulatory role in the oxygen response pathway; and, in the presence of oxygen, they undergo hydroxylation and subsequent degradation. However in hypoxic conditions, hydroxylation does not occur, and the stabilized HIF-α induces transcriptional activity. This regulates several important biologic pathways, including cell proliferation, angiogenesis, cell metabolism, apoptosis and migration (for review, see Harris, 20026). One of the genes regulated by the HIF-1 pathway is carbonic anhydrase-9 (CA-9), which catalyses the reversible hydration of carbon dioxide to carbonic acid (important in pH regulation). CA-9 also has been used as a marker of tumor hypoxia, and studies have indicated that it is important in tumor growth and survival.7

There are at least 2 members of the HIF-α family that share close sequence homology. Although studies have shown that the majority of solid tumors express nuclear HIF-1α, HIF-2α is expressed less frequently.8 There also are differences in cell distribution; and, whereas HIF-1α is the predominant form in epithelial cells, HIF-2α is predominant in endothelial cells. In vitro studies also have shown that the hypoxia response is dependent critically on the different isoforms in different tumor types.9, 10 Therefore, in the current study, we considered the importance of HIF-1α and HIF-2α expression in vivo.

Anemia long has been considered an important factor in treatment outcome in a number of solid tumors, including HNSCC.11 Initially, resistance to RT was attributed to the reduced oxygen diffusion within tumors associated with reduced Hb. However, anemia is associated with a worse outcome in patients with HNSCC who undergo surgery.12 Therefore, it has been suggested that the generalized reduced oxygen carrying capacity of the blood and reduced oxygen delivery to tumors may produce intratumor hypoxia and activation of the HIF-1 pathway.13, 14

The role of hypoxia in tumor development and its significance to outcome have directed researchers into developing treatments that target the HIF pathway or modify the hypoxic microenvironment by correcting anemia.15 Thus, it has become very important to understand the functional roles of both HIF-1α and HIF-2α in vivo and to understand whether anemia can influence hypoxic gene expression. Therefore, we investigated the relation between anemia; the hypoxic markers HIF-1α, HIF-2α, and CA-9; and long-term clinical outcomes in patients who underwent surgery for HNSCC.



This study was carried out after approval was obtained from the Oxford Ethics Committee. One hundred fifty-one consecutive patients with HNSCC from the Otorhinolaryngology Department at Radcliffe Infirmary, Oxford, were analyzed. All patients underwent primary surgery with curative intent. The decision regarding postoperative RT was made in a multidisciplinary setting after surgery in conjunction with the histology results and the patient's wishes. One hundred twenty-eight patients (85%) received postoperative RT. Of 23 patients who did not receive RT, 16 patients (69%) with Stage I and II disease made an elective decision not to receive RT, 5 patients (22%) died before the completion of RT, and 2 patients (9%) declined additional treatment.

Patients who had previously been treated for HNSCC or who were to receive RT prior to surgery were excluded from this study. Clinical data, including date of birth, gender, tumor subsite, tumor stage, preoperative Hb level, postoperative RT, disease recurrence, and survival, were recorded. Tumors were staged according to the American Joint Committee on Cancer/International Union Against Cancer Tumor, Lymph Node, Metastasis (TNM) classification system,16 with advanced-stage disease defined as Stage III/IVA-IVB (Stage III: T3N0M0 or T1-T3N1M0; Stage IVA: T4a,N0-N1,M0 or T1-T3N2M0; Stage IVB: T4bN1-N3M0 or T1-T4N3M0). Outcome data included all events that occurred immediately after surgery. Anemia was defined as a Hb level ≤11 g/dL based on previously published studies in patients with HNSCC17 and because, in several clinical studies, patients with Hb levels lower than that had poorer outcomes after RT.18

Tissue Microarray

Formalin fixed, paraffin embedded specimens derived by surgical resection were obtained from the Pathology Department, John Radcliffe Hospital, Oxford. At least 2 representative areas of viable tumor and, when present, tumor adjacent to necrotic regions were identified for each patient. Cores (1.0 mm) from the paraffin embedded blocks were taken, and 4-μm sections of cores were adhered to Micro Slides in an 8 × 15 grid arrangement and were provided with a unique tissue microarray (TMA) identifier.


Our group and others previously described the techniques and immunohistochemical expression of HIF-1α, HIF-2α, and CA-9 by using these antibodies.19, 20 Briefly, after dewaxing and rehydration, an endogenous peroxidase block (DAKO Envision) was applied for 5 minutes and then washed in Tris-buffered saline (TBS). In brief, for HIF-1α and HIF-2α immunostaining, samples were incubated for 16 hours in Tris-ethylenediamine tetraacetic acid, pH 9.0, at 70°C. A mouse monoclonal antihuman antibody to HIF-1α (ESEE122) was applied at 1 in 30 dilution and, for HIF-2α (EP190), a mouse monoclonal antihuman antibody at 1 in 1 dilution was applied, both for 4 hours at room temperature. For CA-9 immunostaining, the samples were incubated with 10% human serum in TBS for 10 minutes. A mouse monoclonal antihuman antibody (M75)21 was applied at 1 in 50 dilution for 30 minutes at room temperature. Secondary polymer from the Envision kit (DAKO) was applied for 30 minutes at room temperature. Visualization of the immunostaining was by diaminobenzidine substrate. After rinsing in water, slides were lightly counterstained with hematoxylin, dehydrated, and mounted. HNSCC tumors with known expression of HIF-1α, HIF-2α, and CA-9 and renal cancer sections with known expression of HIF-1α were used as positive controls. Substitution of primary antibody with phosphate-buffered saline was used as a negative control.

Assessment of Protein Expression

All TMA-immunostained slides were reviewed under light microscopy by 2 observers (K.A.S. and S.C.W.), scored independently, then reviewed at a conference microscope, and a was consensus determined. Fields were assessed at low magnification (× 100) and at intermediate magnification (× 200). A score was obtained only if >10% of the surface area of the sample contained tumor cells. The intensity and percentage of cancer cells with nuclear HIF-α expression were assessed in all optical fields. Staining was scored according to intensity from 0 to 3 (0, absent staining; 1, weak staining; 2, moderate staining; and 3, strong staining), and the percentage of cells involved was scored from 0 to 4 (0, 0% of cells involved; 1, 1–10% of cells involved; 2, 11–50% of cells involved; 3, 51–80% of cells involved; and 4, >80% of cells involved). The expression score was derived from the product of the intensity and percentage scores to give value between 0 and 12. The assessment of CA-9 has been described previously17 and involved recording the percentage of membrane staining on tumor cells by using the categories described above. For HIF-1α, HIF-2α, and CA-9, patients were divided into categories of low and high reactivity by calculating the median expression score.

TMA cores from perinecrotic regions were available for immunohistochemical expression scoring from 55 patients (36%). Because of the possibility of artifactual expression from these regions, the values were excluded if similar levels of expression were not observed in the other tumor samples.

To compare TMA immunohistochemical expression with whole-section analysis, we compared the expression of CA-9 and HIF-1α on conventional size sections from samples that were used in the construction of the TMA. CA-9 tumor membrane and nuclear HIF-1α expression were assessed in 10 separate fields under intermediate magnification (× 200). Intensity was graded as described above, and the percentage of tumor cells that expressed CA-9 or HIF-1α in each field was recorded as a continuous variable. The overall score for each tumor was the mean percentage score across the 10 fields. This was converted into the percentage score described above, and the expression score was calculated as the product of the intensity and percentage scores (0–12). Whole-section analysis ‘predicted’ the TMA grouping in 11 of 14 patients (79%) for CA-9 and in 9 of 10 patients (90%) for HIF-1α.

Statistical Analysis

Statistical analysis and graphs were generated using by GraphPad Prism (version 3.0) and SPSS software. Locoregional control and survival were analyzed by using the Kaplan–Meier method, and prognostic factors were assessed with log-rank analysis. Correlations between continuous variables were obtained by linear regression analysis and Spearman rank correlation. Univariate and multivariate analyses were performed, and a Cox proportional hazards model was used to analyze the effects of patient, tumor, and immunohistochemical results on overall survival (OS), disease-specific survival (DSS), and disease-free survival (DFS). Only factors that were significant in the multivariate model are reported. Two-tailed P values are given and were considered statistically significant when P < .05.


The clinicopathologic characteristics of the patients who were included in the current study are detailed in Table 1. The mean follow-up of this cohort was 33 months (range, 0–108 months) at the time of this analysis. To date, 71 patients remain alive and disease free, 4 patients remain alive with disease, and 76 patients have died.

Table 1. Patient and Tumor Characteristics for the Whole Group
CharacteristicNo. of patients (%)
  1. HIF indicates hypoxia inducible factor; Hb, hemoglobin; RT, radiotherapy.

 Male107 (71)
 Female44 (29)
 High45 (30%)
 Low95 (63%)
 Unavailable11 (7%)
 High21 (14)
 Low118 (78)
 Unavailable12 (8)
Carbonic anhydrase-9
 High92 (62)
 Low57 (38)
 Unavailable2 (<1)
Tumor classification
 T126 (17)
 T236 (24)
 T328 (19)
 T4a,T4b61 (40)
Lymph node status
 N055 (36)
 N131 (21)
 N257 (38)
 N38 (5)
 Hypopharynx20 (13)
 Larynx33 (22)
 Oral cavity51 (34)
 Oropharynx47 (31)
 Yes135 (89)
 No16 (11)
Pretreatment Hb
 ≤11.0 g/dL20 (13)
 >11.0 g/dL126 (83)
 Unknown5 (3)
Postoperative RT
 Yes128 (85)
 No23 (15)
 Alive disease free71 (47)
 Alive with disease4 (3)
 Cancer death55 (36)
 Death from other cause21 (14)

Expression of HIF-1α, HIF-2α, and CA-9

Within tumors, the pattern of HIF-1α and HIF-2α expression was predominantly nuclear, and there was variable cytoplasmic expression. The nuclear expression of both HIF-1α and HIF-2α ranged from weak to strong staining in up to 100% of cells. In addition, strong HIF-2α expression was seen in tumor-associated macrophages. Tissue adjacent to cancer tissue, when present, expressed varying degrees of both HIF-1α and HIF-2α cytoplasmic reactivity, whereas nuclear expression rarely was seen. CA-9 expression was observed on the cell membrane in up to 100% of tumor cells with minimal cytoplasmic staining (Fig. 1). Tissue adjacent to cancer tissue did not express CA-9.

Figure 1.

These are representative tissue microarray sections that were used in of immunohistochemical analyses. (A) Nuclear hypoxia-inducible factor (HIF)-1α expression. Inset: Renal cell-positive control. (B) Carbonic anhydrase-9 (CA-9) membranous expression. (C) Nuclear HIF-2α expression. (D) Nuclear HIF-2α expression and tumor-associated macrophages that express HIF-2α (arrow). Original magnification × 200 (A, C, D); × 400 (B).

Failure to obtain a result from some patients was caused by an inadequate tumor sample (<10% of the surface area contained tumor cells) or loss of tumor sample from the array slide. This is shown as ‘unavailable’ in Table 1. Results were obtainable in 140 patients (93%) for HIF-1α, 139 patients (92%) for HIF-2α, and 149 patients (99%) for CA-9. By using the expression score described above, we scored 45 patients (30%) with high HIF-1α expression, 21 patients (14%) with high HIF-2α expression, and 56 patients (62%) with high CA-9 expression.

Using the expression score as a continuous variable, there was a high degree of correlation between HIF-1α expression and HIF-2α expression (correlation coefficient [r] = .34; P < .0001) and between CA-9 expression and HIF-1α (r = .17; P = .04) and HIF-2α (r = .25; P = .003) expression. Furthermore, HIF-1α expression was correlated with International Union Against Cancer stage (r = .22; P = .01), although HIF-2α expression was not (r = .09; P = .27).

HIF-α Expression and Clinical Outcome

The initial analysis of the results considered HIF-1α and HIF-2α expression separately. Univariate analysis of HIF-1α expression and Kaplan–Meier survival analysis revealed a significantly worse DSS (P = .02) and DFS (P = .02) but only a trend toward a worse OS (P = .08) for patients who had high HIF-1α expression compared with patients who had low HIF-1α expression (Fig. 2). In the univariate analysis of HIF-2α expression, there was no association with outcome for OS (P = .43), DSS (P = .16), or DFS (P = .10) (Fig. 3).

Figure 2.

Kaplan–Meier survival curves were calculated for hypoxia-inducible factor (HIF)-1α according to (A) overall survival, (B) disease-specific survival, and (C) disease-free survival (DFS). Pts indicates patients.

Figure 3.

Kaplan–Meier survival curves were calculated for hypoxia-inducible factor (HIF)-2α according to (A) overall survival, (B) disease-specific survival, and (C) disease-free survival (DFS). Pts indicates patients.

HIF-1α and HIF-2α expression levels were included in a multivariate analysis of survival as separate variables. Other variables that were included in the models were advanced disease (Stage III/IVA-IVB), anemia (Hb ≤ 11 g/dL), gender, age, smoking history, lymph node status, tumor subsite, and tumor grade. From this analysis, HIF-1α expression continued to be an independent and significant factor for DSS (P = .05) and DFS (P = .03). HIF-2α expression was not significant for OS (P = .37), DSS (P = .39), or DFS (P = .17). Factors that remained in the model and were significant with regard to outcome are shown in Table 2.

Table 2. Multivariate Analyses Examining Overall Survival, Disease-Specific Survival, and Disease-Free Survival
HR (95% CI)PHR (95% CI)PHR (95% CI)P
  • OS indicates overall survival; DSS, disease-specific survival; DFS, disease-free survival; HR, hazards ratio; 95% CI, 95% confidence interval; HIF, hypoxia-inducible factor; Hb, hemoglobin.

  • *

    Each analysis was repeated with HIF-1α and HIF-2α as individual variables and with HIF-α expression combined. For the combined HIF-1α/HIF-2α analysis, the comparator was low expression of both markers.

  • P values <.05 were considered significant.

HIF-1α and HIF-2α as individual variables
 HIF-1α1.11 (0.63–1.98).721.84 (0.99–3.40).051.79 (0.99–3.25).03
 HIF-2α1.34 (0.70–2.57).371.39 (0.66–2.89).391.57 (0.82–3.01).17
 Stage III/IVA-IVB3.20 (1.59–6.43).0011.63 (0.80–3.30).181.69 (0.89–3.23).11
 Hypopharyngeal tumors2.42 (1.24–5.96).013.57 (1.22–10.47).022.35 (0.96–5.73).06
 Anemia (Hb ≤11 g/dL)1.52 (0.75–3.10).251.91 (0.75–4.91).181.19 (0.54–2.62).66
HIF-1α and HIF-2α expression combined
 High HIF-1α/high HIF-2α1.57 (0.73–3.38).252.39 (1.00–5.70).043.11 (1.39–6.96).006
 High HIF-1α/low HIF-2α1.30 (0.70–2.41).401.11 (0.51–2.38).801.53 (0.79–2.94).20
 Low HIF-1α/high HIF-2α1.37 (0.48–3.92).550.63 (0.14–2.75).540.89 (0.27–2.98).85
 Stage III/IVA-IVB3.02 (1.53–5.96).0011.53 (0.75–3.15).241.28 (0.72–2.29).40
 Hypopharyngeal tumors2.42 (1.11–5.38).023.74 (1.27–11.03).022.39 (0.97–5.90).06
 Anemia (Hb ≤11 g/dL)1.52 (0.74–3.09).251.96 (0.75–5.08).171.22 (0.56–2.67).62

We subsequently considered the combined expression of HIF-1α and HIF-2α for each patient. One hundred thirty-six patients (90%) had results available for HIF-1α or HIF-2α. A small subset of patients (12 patients; 9%) had high HIF-1α expression and high HIF-2α expression, whereas 81 patients (60%) had both low HIF-1α expression and low HIF-2α expression. Forty-three patients (32%) had either high HIF-1α expression or high HIF-2α expression. Among the latter subset, 34 patients (79%) had high HIF-1α/low HIF-2α expression, and 9 patients (21%) had low HIF-1α/high HIF-2α expression.

Considering the combined expression pattern of HIF-1α and HIF-2α, univariate analysis of HIF-α expression and Kaplan–Meier survival analysis revealed that the group that had high HIF-1α/high HIF-2α expression had significantly worse DSS (P = .008) and DFS (P = .001), but not OS (P = .1), compared with the group that had low HIF-1α/low HIF-2α expression (Fig. 4). The groups with high HIF-1α/low HIF-2α expression and low HIF-1α/high HIF-2α expression were not associated with a survival difference.

Figure 4.

Kaplan–Meier survival curves were calculated by using double stratification for (A) overall survival, (B) disease-specific survival, and (C) disease-free survival (DFS). HIF indicates hypoxia-inducible factor; Pts, patients; NS, nonsignificant.

Repeating the multivariate analysis with combined HIF-α expression revealed that high HIF-1α/high HIF-2α expression was associated with significantly decreased DSS (P = .04) and DFS (P = .006). Again, the groups with high HIF-1α/low HIF-2α expression and low HIF-1α/high HIF-2α expression were not associated with a survival difference. Other factors that remained in the models and were significant with regard to outcome are shown in Table 2.

Because the combined expression of both HIF-1α and HIF-2α was associated with decreased DSS and DFS in multivariate testing, we also examined the group with low HIF-1α expression (n = 90 patients) to determine the impact of HIF-2α expression on survival outcomes. Within this group, 81 patients had low HIF-2α expression, and 9 patients had high HIF-2α expression. There was no survival difference with high HIF-2α expression for OS (P = .69), DSS (P = .44), or DFS (P = .65). Furthermore, CA-9 expression was not associated with a survival difference for OS (P = .3), DSS (P = .1), or DFS (P = .2) in the univariate or multivariate analyses.

Hb, Hypoxia, and Survival

Pretreatment Hb levels ranged from 8.6 g/dL to 16.8 g/dL. One hundred twenty-six patients presented with a normal Hb level (>11 g/dL), 20 patients were anemic (Hb ≤11 g/dL), and 5 patients did not have pretreatment Hb levels recorded.

Using linear regression analysis of Hb and the expression scores for HIF-1α, HIF-2α, and CA-9 as continuous variables, there was no statistically significant association between Hb and HIF-1α (P = .45; r = − .07), HIF-2α (P = .82; r = − .02), or CA-9 (P = .35, r = .08). In univariate analysis, the presence of pretreatment anemia had a significant impact on OS (P = .01) and DSS (P = .04). In anemic patients, the 5-year OS was 24% compared with 50% in nonanemic patients, and the DSS was 34% compared with 59% in nonanemic patients. There was no significant difference in DFS (P = .09) between anemic and nonanemic patients. However, in multivariate testing, the association with survival was lost, and there was no significant association between anemia and outcome.


In this study, HIF-1α and HIF-2α tumor expression was predominantly nuclear with some cytoplasmic expression, and CA-9 was expressed on the cell membrane as reported previously in patients with HNSCC.22, 23 The cytoplasmic expression of HIF-α has been reported in other studies and was incorporated into the expression analysis in some studies but not in others,20, 24 perhaps reflecting a lack of understanding regarding the significance of the finding. In the current study, we did not assess cytoplasmic expression, because it generally was weak and difficult to differentiate from background staining in contrast to the nuclear expression, which was visible clearly when it was present. We also observed HIF-2α expression in tumor-associated macrophages, which was reported previously.25 TMAs provided sufficient material to allow the assessment of 93% of tumors for HIF-1α, 92% of tumors for HIF-2α, and 99% of tumors for CA-9, rates that compare favorably with other reports.26, 27 Furthermore, by using the expression scoring system, 45% of tumors had high HIF-1α expression, 14% of tumors had high HIF-2α expression, and 62% of tumors had high CA-9 expression, rates that compare with reported rates in patients HNSCC of from 37% to 87% for HIF-1α expression,20, 22, 24 from 7% to 89% for HIF-2α expression,19, 20, 22 and from 26% to 89% for CA-9 expression.23, 28, 29

It is well recognized that the HIF-α proteins and CA-9 are regulated by hypoxia and by the HIF-1 pathway.6 Both HIF-1α and HIF-2α, which are part of the nuclear transcriptional response to hypoxia, demonstrated a high degree of correlation, as reported previously in patients with HNSCC.20 HIF-1α expression also was correlated with disease stage, a finding that also has been reported in patients with esophageal squamous cell carcinomas,30 and reflects an aggressive phenotype. However, HIF-2α expression did not correlate with disease stage, suggesting that there is not a simple relation between HIF-2α expression and disease behavior. Furthermore, although CA-9 demonstrated a positive correlation with HIF-1α and HIF-2α, the correlation coefficient was low. This may reflect the finding that CA-9 is a downstream gene target in the HIF-1 pathway and also may be influenced by the markedly different half-lives of CA-9 (days) and HIF-α (minutes).31 Therefore, it is possible that HIF-α expression may reflect more accurately the hypoxic profile at the time of sample harvesting.

Although HIF-1α and HIF-2α share close sequence homology, and the majority of solid tumors express both HIF-1α and HIF-2α, albeit at different levels,8 there is evidence that they have differential roles both between cell types and within cells that express both factors. Using RNA interference, it has been shown that the hypoxic-mediated gene response depends critically on HIF-1α but not HIF-2α in endothelial and breast cancer cells but is dependent on HIF-2α alone in renal carcinoma cells.9 Furthermore, in vitro studies have shown that HIF-2α has unique gene targets,32 and animal studies indicate that it has a role in tumor development.33 In the current study, we considered the expression of HIF-α in a large group of surgically treated tumors. We also examined the combined HIF-α expression and the separate expression of HIF-1α/HIF-2α to understand the significance of HIF-2α expression in vivo.

Analysis of HIF-α expression revealed that HIF-1α was associated with significantly worse DSS and DFS in univariate and multivariate analyses but that HIF-2α was not associated with a worse outcome. Several groups examined the correlations between HIF-1α expression and outcomes in patients with HNSCC and reported that high HIF-1α expression was associated with poor OS in patients with HNSCC who received RT.20, 24 In addition, in patients with esophageal, nasopharyngeal, and cervical tumors who received treatment with a variety of modalities, HIF-1α expression has been correlated with a poor outcome.19, 30, 34 Although HIF-2α expression has been used as a marker of tumor hypoxia, there is equivocal evidence linking its expression with clinical outcome. Whereas it has been shown to be an independent indicator of survival in patients with HNSCC20 and patients with early-stage nonsmall cell lung cancers35, other studies in patients with HNSCC and nasopharyngeal cancers have found that HIF-2α does not correlate with outcome.19, 22

Considering the combined HIF-α expression profile of the tumors, we observed that 32% of tumors expressed either HIF-1α or HIF-2α, which is consistent with the findings from Koukourakis et al.,20 who reported that HIF-1α and HIF-2α were expressed independently in 30% to 45% of HNSCCs. When we considered tumors that expressed both high HIF-1α and high HIF-2α, we observed that there was a statistically stronger association with a worse outcome for DSS and DFS in both univariate and multivariate analyses. We also examined the subgroup of patients that had low HIF-1α expression and observed that the addition of HIF-2α did not alter the survival outcome. This suggests that, as an independent factor, HIF-2α has a minor role but that it may have an additive effect when considering outcome.

In a previous study from our center that involved patients with HNSCC who underwent surgery, it was concluded that the overexpression of HIF-1α was associated with an improved DFS and OS.22 From that original study, 73 of 79 patients were included in the current study. Six patients were excluded, including 4 patients because of inadequate pathologic tumor material, 1 patient because they had been treated for a cutaneous squamous cell carcinoma, and 1 patient because they had received chemotherapy for an unrelated tumor type. Several differences between the studies exist that may explain the contradictory results. Technical differences in the 2 studies included the use of an improved antigen-retrieval method that superseded the pressure-cooking method that was used originally. The study designs also differed: In the current study, TMAs were used to assess immunohistochemical expression in contrast to the whole sections used in the original study. TMAs have been developed to facilitate and accelerate the analysis of immunohistochemical expression36; and, despite the small diameter of the tissue samples, they typically offer a high degree of concordance with whole-section analysis.37, 38 Also, in the original study, a scoring system was used that involved a simple positive or negative result based on the presence of any staining. This scoring system almost certainly over-scored tumors, and this is reflected in the high numbers that were classified as positive (87%) for HIF-1α. More recent studies have stratified the degree of staining based on the degree of reactivity.19, 24, 30 In addition, in the current study, a higher proportion of patients received postoperative RT (85%) compared with the original study (35%). The contradictory findings between the 2 studies, thus, may be explained in part by design differences, larger patient numbers, and the longer follow-up available for patients from the original study. Despite an increase in the literature in this area, the conclusions of Beasley et al.22 have not been replicated; therefore, we conclude that the overexpression of HIF-1α should be considered to be associated with decreased survival in patients with HNSCC.

In contrast to published series demonstrating an association between high CA-9 expression and decreased survival and DFS in patients with HNSCC,19, 23 in the current study, we did not observe a correlation between CA-9 and outcome. However, because hypoxia is focal, there is a possibility of missing sections on TMAs; therefore, the number of cores required to demonstrate concordance with ‘large’ sections depends on the heterogeneity of the tissue and the expression pattern of the protein being studied. Therefore, we compared the staining of conventional size sections from tumors that were used in construction of the TMA for CA-9 with the array results. We observed that 4 cores were capable of ‘predicting’ the whole section result in 11 of 14 tumors (79%). This result is lower than the result for HIF-1α expression and is lower than reported for other antibodies (99% p53 expression in colon cancer using 3 cores26 and 95% estrogen receptor expression and 75% progesterone receptor expression in breast cancer using 1 core39) but nevertheless represents substantial concordance and suggests that this is not a strong prognostic factor. This finding also may explain why there was only a weak correlation between CA-9 expression and HIF-α expression.

Oxygen delivery to tumors is regulated in regulated by the O2 carrying capacity of the blood; therefore, it has been suggested that anemia,40 which is common among cancer patients, may contribute to tumor hypoxia. Alternatively, anemia may represent an epiphenomenon that is associated with chronic or advanced disease. It has been shown that pretreatment Hb levels equate with a poor OS and local control in patients with HNSCC, although the definition of anemia varies widely.41, 42 Although the majority of studies in HNSCC have examined the outcome after RT,11 1 study in patients with laryngeal cancer retrospectively reviewed 258 patients who underwent surgery. The authors reported that 10% of their patients presented with pretreatment anemia and that this was associated with decreased locoregional control in a multivariate analysis.12 Although, in the current study, anemia was associated with worse outcome and disease control in univariate analysis, those results were not confirmed in multivariate analysis. This finding may reflect the heterogeneity of the anatomic subsite examined and the inherent variability in surgical techniques, and further studies will be needed to confirm it.

Anemia in animal models reduces intratumor oxygen levels.13 To our knowledge, this is the largest study to date that examines the relation between anemia and hypoxic markers in patients with HNSCC. Our findings support a recent study by Koukourakis et al. in which they failed to find an association between anemia and levels of HIF-1α, HIF-2α, CA-9, and vascular endothelial growth factor (VEGF) in HNSCC in a cohort of 45 patients who received RT.17 Furthermore, in patients with endometrial cancer, no association was found between Hb and levels of HIF-1α, HIF-2α, or VEGF,43 although HIF-1α was correlated with Hb in cervical cancers.34 This finding has implications for the therapeutic approach of correcting anemia to alter tumor oxygenation and suggests that altering Hb levels may not modify the molecular response to hypoxia. However, it is possible that hypoxic protein expression reflects the response of a molecular pathway that can be modified by other signaling pathways or that anemia may be interacting with other pathways that obscure its independent assessment.

Tumor hypoxia is recognized increasingly as an important therapeutic target with bioreductive drugs and anti-HIF drugs entering clinical trials.15 The current results support the importance of the HIF-1 pathway in predicting outcome in surgically treated HNSCC and suggest an additional role for HIF-2α in the HIF signaling pathway. However, no association between the hypoxic markers and Hb were found, suggesting that anemia, although it may predict a poor outcome, does not influence the HIF-1 signaling pathway directly. Thus, modifications of both factors represent independent strategies for therapeutic intervention.


We thank Dr. S. Pastorekova for the gift of antibody M75