Overexpression of Hypoxia-inducible Factor 1α in Common Human Cancers and Their Metastases¹

Altered glucose metabolism and cellular adaptation to hypoxia are fundamental to the basic biology and treatment of cancer. Four lines of evidence support this thesis: (a) clonal expansion of cancer cells depends on enhanced glucose transport and glycolysis (the Warburg effect; Refs. 1, 2); (b) tumors cannot grow beyond several mm³ without angiogenesis because of the limited diffusion of O₂, glucose, and other nutrients (3, 4). In many cancers, the degree of vascularization is inversely correlated with patient survival (5); (c) the probability of invasion, metastasis, and death are positively correlated with the degree of intratumoral hypoxia (6, 7), which is caused by an architecturally defective microcirculation such that even cells adjacent to neovessels may be hypoxic (8). Cancer cell proliferation may also outpace the rate of angiogenesis (3); and (d) tumor hypoxia is associated with resistance to chemotherapy, immunotherapy, and radiotherapy (9). Despite the critical importance of these observations, their molecular basis is not well understood. Transcription factors that regulate expression of angiogenic growth factors (such as VEGF³) or glycolytic enzymes involved in the Warburg effect are compelling targets for interrogation. HIF-1 performs both of these functions.

HIF-1 is a bHLH-PAS transcription factor that plays an essential role in O₂ homeostasis (10, 11, 12, 13). HIF-1 is a heterodimer composed of HIF-1α and HIF-1β subunits (10). Whereas HIF-1β (also known as the aryl hydrocarbon receptor nuclear translocator) is a common subunit of multiple bHLH-PAS proteins, HIF-1α is the unique, O₂-regulated subunit that determines HIF-1 activity (14, 15). HIF-1 transactivates genes whose protein products function either to increase O₂ availability or to allow metabolic adaptation to O₂ deprivation. Included among these are genes encoding erythropoietin, transferrin, endothelin-1, inducible nitric oxide synthase, heme oxygenase 1, VEGF, IGF-2, IGF-binding proteins -2 and -3, and 13 different glucose transporters and glycolytic enzymes (15, 16). Remarkably, most of these proteins are implicated in tumor progression (17). Analysis of isogenic tumor cell lines injected into nude mice revealed a dramatic correlation of HIF-1 expression levels with tumor growth and angiogenesis (18, 19).

Recently, we found that HIF-1α mRNA was overexpressed in six rat PCA cell lines compared with the normal prostate, and metastatic potential was correlated with HIF-1α mRNA levels in those cell lines (20). A human PCA cell line derived from a bone metastasis was found to overexpress HIF-1α protein under nonhypoxic culture conditions (20). Because HIF-1α expression was dysregulated in PCA cell lines, we tested the hypothesis that HIF-1α is generally overexpressed in solid tumors. In this study, we screened HIF-1α protein expression by immunohistochemistry in normal tissues and human cancers, including lung, prostate, breast, and colon carcinoma, which are the leading causes of U.S. cancer mortality.

A human HIF-1α cDNA fragment encoding amino acids 432–528 was cloned into pGEX2T. The GST/HIF-1α fusion protein was purified from bacteria (14) and used to immunize BALB/c mice. Spleen cells from immunized mice were fused with P3X63- Ag8-653 myeloma cells. Hybridoma supernatants were screened by ELISA against GST and GST/HIF-1α. Supernatant from clone 67 was affinity-purified using protein G-Sepharose (Pharmacia). The adsorbed protein was eluted with 0.1 m glycine-HCl (pH 2.7) and neutralized with 1 m Tris-HCl (pH 9.0). Nuclear extracts, prepared from human Hep3B and mouse ES cells (11), were subjected to immunoblot analysis as described previously (14) except that the primary MAb was H1α67 (1:500), and the secondary MAb was horseradish peroxidase-conjugated sheep antimouse immunoglobulin (1:2000).

Human embryonic kidney 293 cells, growing exponentially on 10-cm dishes, were transfected by calcium phosphate coprecipitation with 10 μg of pCEP4 (Invitrogen), pCEP4/HIF-1α (21, 22), or PL477 (23), a HIF-2α expression vector that was generously provided by Dr. Christopher Bradfield (University of Wisconsin, Madison, WI). For reporter gene assays, the cells were cotransfected with pSVβgal and 2xWT33-luciferase, which contains two copies of a 33-bp hypoxia-response element from the human erythropoietin gene cloned upstream of a basal SV40 promoter (22).

Formalin-fixed, paraffin-embedded tissue specimens were obtained and handled by standard surgical oncology procedures. Serial 4-μm sections were prepared, and one was stained with H&E. Flanking sections were stained for HIF-1α using Catalyzed Signal Amplification System (DAKO) which is based on streptavidin-biotin-horseradish peroxidase complex formation. In brief, after deparaffinization and rehydration, slides were treated with target retrieval solution (DAKO) at 97°C for 45 min, and the manufacturer’s instructions were followed. MAb H1α67 (1 mg/ml) was used at a dilution of 1:1000. Nuclei were lightly counterstained with hematoxylin. Negative controls were performed using nonimmune serum or PBS instead of the MAb. A preadsorption test was also performed using GST/HIF-1α protein. Twenty-four-well plates were coated with GST/HIF-1α protein (2.9 mg/ml), air-dried, and incubated with H1α67 (1:1,000 dilution), followed by immunohistochemistry. Automated immunohistochemistry was performed using a BioTek-Tech Mate 100 Automated Stainer (Ventana-BioTek Solutions, Inc., Tucson, AZ) with the following MAbs: (a) anti-Ki67 (MAb MIB-1, Immunotech, 1:100); (b) antihuman p53 protein (MAb DO-7, DAKO, 1:250); (c) antihuman bcl-2 (MAb 124, DAKO, 1:25); (d) anticytokeratin (AE1/AE3, Boerhinger Mannheim, 1:2000); and (e) anti-prostate-specific antigen (MAb 5126, Immunotech, dilution 1:50).

Three investigators (H. Z., A. M. D., and J. W. S) independently evaluated the immunohistochemistry. All of the PCA bone metastases were verified by cytokeratin and prostate-specific antigen staining. The immunohistochemical results for HIF-1α protein were classified as follows: −, no staining; +, nuclear staining in less than 1% of cells; ++, nuclear staining in 1–10% of cells and/or with weak cytoplasmic staining; +++, nuclear staining in 10–50% of cells and/or with distinct cytoplasmic staining; ++++, nuclear staining in more than 50% of cells and/or with strong cytoplasmic staining. When independent scoring of a case differed, the case was rechecked, and the final score was determined by recounting HIF-1α positive cells using a multiheaded microscope with all of the three reviewers simultaneously viewing the slide. For Ki67 analysis, nuclei from approximately 1000 tumor cells from 10 randomly selected fields were counted, and the LI was determined as the percentage of positive nuclei. Bcl-2 reactivity was scored positive if >10% of tumor cells showed distinct cytoplasmic staining. Aberrant p53 accumulation was scored positive if nuclear staining was present in >10% of tumor cells.

Nonparametric statistical analyses were conducted by Dr. Steven Piantadosi, Johns Hopkins Oncology Center Biostatistics Center, using Microsoft Excel (Microsoft Corporation, Redmond, WA) and STAT-XACT, Version 4 for Windows (Cytel Software, Berkeley CA, 1998). As a singly ordered table, the Kruskal-Wallis test was used to evaluate the correlation between HIF-1α and aberrant p53 or bcl-2 expression. As a doubly ordered table, the correlation between HIF-1α protein expression and Ki67 LI was analyzed by Jonkchere-Terpstra test (24).

A GST fusion protein containing amino acids 432–528 of human HIF-1α was used as immunogen for MAb production. Five hybridoma clones were identified that reacted with GST/HIF-1α but not with GST. Clone 67 was chosen for further characterization. MAb H1α67 was identified as IgG2b/κ subtype and purified from hybridoma supernatants by protein-G affinity chromatography. Immunoblot assays demonstrated that MAb H1α67 recognized a hypoxia-induced protein of approximately M_r 120,000 that was identical in size to HIF-1α, in Hep3B cells and wild-type ES cells, but not in HIF-1α-null (11) ES cells (Fig. 1 A). MAb H1α67 showed reactivity against human, monkey, sheep, mouse, bovine, rat, and ferret HIF-1α (data not shown).

MAb H1α67 also recognized human HIF-1α purified 11,250-fold by anion-exchange and DNA-affinity chromatography (25) at concentrations too low to allow protein quantitation (data not shown). As a final test of its specificity, cells were transfected with expression vectors encoding no protein, HIF-1α, or HIF-2α (Fig. 1 B). MAb H1α67 detected overexpressed HIF-1α (Lanes 3–4), whereas cells overexpressing HIF-2α (Lanes 5–6) gave the same pattern as cells transfected with the empty vector (Lanes 1–2). HIF-2α expression in the transfected cells was confirmed by cotransfection of a reporter gene containing a hypoxia response element, which was activated 9- to 13-fold over background in cells transfected with HIF-1α expression vector and 33- to 106-fold over background in cells transfected with the HIF-2α expression vector (data not shown). These highly stringent tests provide convincing evidence that MAb H1α67 specifically recognizes HIF-1α.

HIF-1α expression was extensively screened in normal tissues and human cancers resected during routine surgical oncology procedures. Twenty-one normal human tissues (174 specimens), 19 primary malignant cancers (131 specimens), and 36 metastases from 6 tumor types were interrogated (Tables 1 and 2). Most normal human tissues (14 types) showed no HIF-1α immunoreactivity (153 of 174 clinical specimens, 88% negative). In some autopsy specimens, weak staining was detected in adrenal cortical cells (3 of 8), renal distal tubular epithelium (3 of 9), pancreatic acinar cells (4 of 11), fetal hepatocytes (1 of 1), proliferating B cells from tonsil (2 of 3) and spleen (1 of 9), and seminiferous tubules of testis (7 of 7; Table 1).

Overexpression of HIF-1α protein was found in 69 (53%) of 131 primary malignant tumors representing 13 of 19 tumor types screened (Table 2). Cases of human prostate, breast, lung, colon, pancreas, brain, gastric, ovarian, and renal cell carcinomas, mesothelioma, and melanoma were positive. Immunohistochemistry was performed on adjacent sections of vena caval invasion from a renal cell carcinoma using MAb H1α67 that was preadsorbed with GST/HIF-1α protein. Whereas nonadsorbed MAb resulted in strong (++++) staining, preadsorbed MAb resulted in no (−) staining (data not shown).

Two-thirds of all of the regional lymph node and bone metastases were also positive for HIF-1α overexpression. HIF-1α was overexpressed in only 29% of primary breast cancers, whereas 69% of breast metastases were positive. All four of the preneoplastic and premalignant lesions found incidentally within biopsy specimens were positive for HIF-1α immunoreactivity, including two cases of breast comedo-type ductal carcinoma in situ, one case of prostatic intraepithelial neoplasia, and one case of colonic adenoma (Fig. 3, a and b). In contrast, all 12 of the benign tumors (breast fibroadenoma and uterine leiomyoma) were negative (Table 2).

HIF-1α immunostaining was heterogeneous with signal concentrated primarily within the nucleus (Figs. 2 and 3). Cytoplasmic staining was also detected in colon (Fig. 2,e), breast, pancreas, and prostate adenocarcinomas. The results were reproducible, and cytoplasmic staining was not observed in flanking normal tissue. Within tumors, clusters of HIF-1α positive cells were most dense at the invading edge of tumor margins, the periphery of necrotic regions, and surrounding areas of neovascularization (Fig. 2, b and f). Some lymphocytes in lymph nodes containing metastatic cancer cells were positive for HIF-1α immunostaining (Fig. 3 d), but expression was not detected in nonmalignant stromal cells under the assay conditions used for this study.

To investigate whether HIF-1α expression levels correlated with the degree of tumor angiogenesis and/or disease progression, we evaluated nine brain tumors of different grades and degrees of neovascularization. HIF-1α expression was strongest in glioblastomas multiforme and hemangioblastomas (Fig. 2, f and g), which are respectively the most malignant and most highly vascularized primary tumors arising in the central nervous system. In glioblastomas, the staining was especially intense in pseudopalisading tumor cells surrounding areas of necrosis.

On the basis of tissue availability, most tumor samples used for HIF-1α staining were also stained with anti-Ki67 MAb; some tumor samples, the majority of which were colon and breast cancers, were also stained with anti-p53 and/or anti-bcl-2 MAbs. These markers were scored in a blinded manner relative to the HIF-1α staining. Expression of HIF-1α protein was positively correlated with aberrant p53 accumulation (P < 0.01), but the correlation with bcl-2 expression was of marginal statistical significance (P = 0.05; Table 3). Nonparametric statistical analyses demonstrated a highly significant correlation of HIF-1α protein expression with Ki67 LI as a marker of cellular proliferation (P < 0.001; Table 3). HIF-1α expression also correlated with Ki67 LI in some normal cell types. Fetal hepatocytes, proliferating B cells in tonsil and spleen, and seminiferous tubules of testis demonstrated weak HIF-1α expression, and these cell types manifested high Ki67 LI relative to other normal tissue types.

HIF-1α protein was overexpressed in multiple types of human cancer and in regional and distant metastases. This study has identified increased HIF-1α expression (relative to adjacent normal tissue) in 13 tumor types including lung, prostate, breast, and colon carcinoma, which are the leading causes of U.S. cancer mortality. HIF-1α protein was also overexpressed in preneoplastic and premalignant lesions such as colonic adenoma, breast ductal carcinoma in situ, and prostate intraepithelial neoplasia. These data suggest that overexpression of HIF-1α can occur very early in carcinogenesis, before histological evidence of angiogenesis or invasion. Additional studies are under way to assess whether HIF-1α may represent a novel biomarker for precancerous lesions that warrant clinical surveillance or therapeutic intervention. It is provocative that every benign noninvasive tumor analyzed was negative for HIF-1α overexpression.

HIF-1α activates the transcription of genes encoding transferrin, VEGF, endothelin-1, and inducible nitric oxide synthase, which are implicated in vasodilation, neovascularization, and tumor metastasis (15, 17). Particularly strong HIF-1α expression was observed in glioblastoma multiforme and hemangioblastoma. High VEGF mRNA expression has been reported in these highly malignant and vascularized brain tumors (26). HIF-1α-positive cells were prominent at tumor margins and surrounding areas of neovascularization. In colonic adenocarcinoma, cancer cells at the leading edge of infiltrating carcinoma manifested the most intense HIF-1α immunostaining. Comparison of tumor and flanking normal tissue allows the patient to serve as his own control and supports the hypothesis that HIF-1α overexpression is associated with angiogenesis, invasion, and metastasis. Experimentally, xenografts of mutant mouse hepatoma cells lacking HIF-1 expression manifested significantly reduced growth rates and vascularization compared with parental and revertant cells that expressed HIF-1 (18, 19). Conversely, human colon carcinoma cells transfected with a HIF-1α expression vector manifested significantly increased growth rates in nude mice as compared with parental cells.⁴

The patterns of immunohistochemical staining in different human cancers suggest that HIF-1α overexpression may result from both physiological (hypoxia) and nonphysiological mechanisms. It is clear from previous studies that many human tumors have regions of significant hypoxia (6, 7, 8). This pattern was most obvious in glioblastoma multiforme in which HIF-1α was detected in viable tumor cells that were closest to areas of necrosis and farthest from a blood vessel, as previously demonstrated for the expression of VEGF mRNA in these tumors (26, 27). In contrast, expression of HIF-1α in hemangioblastoma could not be attributed to hypoxia because tumor cells immediately adjacent to patent blood vessels stained intensely, which indicated that factors other than hypoxia may contribute to HIF-1α expression in human cancers.

A growing number of observations indicate that genetic alterations also affect HIF-1α expression in cancer cells:

(a) we have correlated HIF-1α expression with cell proliferation, both in cultured PCA cells (20) and in vivo (Table 3). Treatment of cultured cells with insulin, IGF-1, or IGF-2 induced expression of HIF-1α protein, which was in turn required for expression of IGF-2 mRNA (16), suggesting the involvement of HIF-1α in an autocrine growth factor loop. Remarkably, all of the 22 primary colon cancers analyzed overexpressed HIF-1α, and the most highly up-regulated gene in colon cancer encodes IGF-2 (28);

(b) cells transfected with the v-Src oncogene overexpressed HIF-1α, HIF-1 DNA-binding and transcriptional activity, and downstream genes including VEGF (18);

(c) HIF-1α overexpression was associated with aberrant p53 accumulation in human tumors (Table 3). The anti-p53 MAb used in this study recognizes an epitope in the NH₂ terminus of the wild-type and mutant forms of human p53 protein. Point mutations in the TP53 gene occur frequently in human cancers, leading to increased expression of a nonfunctional p53 protein with a prolonged half-life that is detectable by immunohistochemistry. Thus, the presence of strong nuclear staining in the majority of cancer cells is frequently observed (29). Expression of HIF-1α protein, HIF-1 DNA-binding activity, and VEGF mRNA are increased in p53−/− knockout colon carcinoma cells as compared with the parental p53+/+ cells⁴; and

(d) in renal clear cell carcinoma cell lines, the loss of von Hippel-Lindau tumor suppressor function results in constitutive high-level expression of HIF-1α (30). The primary (Fig. 2,i) and metastatic (Fig. 3 f) renal clear cell carcinomas analyzed here represent the first demonstration of this overexpression in vivo. Thus, in addition to hypoxia, both oncogene activation and tumor suppressor gene inactivation are associated with increased HIF-1α expression.

Some tumors did not stain positive for HIF-1α in this study. These tumors may overexpress HIF-1α but at levels that were below the limits of detection by immunohistochemistry using current methodology. Alternatively, other bHLH-PAS transcription factors that may have similar biological properties to HIF-1α, such as HIF-2α (also known as EPAS1, HLF, HRF, and MOP2) or HIF-3α (23, 31, 32, 33), may also mediate hypoxic adaptation. Nevertheless, HIF-1α was overexpressed in the majority of preneoplastic, malignant, and metastatic cancers analyzed. Given the overexpression of HIF-1α in common human cancers relative to normal tissues and its vital importance in mediating hypoxic adaptation, additional investigations of HIF-1α as a biomarker of metastatic potential and as a novel target for therapeutics are warranted.

The authors acknowledge Dr. Mary Ann Accavitti and the Hybridoma Core Facility (University of Alabama at Birmingham, Birmingham, AL) for MAb production; the important contributions of Drs. Josef Prchal, Christopher Bradfield, Donald S. Coffey, Frank Kujhada, William G. Nelson, and Steven Piantadosi; and the technical assistance of Jurga Sauvengot, Natalia Rioseco-Camacho, Bahar Mikhak, Colleen Hanrahan, and Kimberly Heaney.