Hypoxia-Inducible Factor-1α Suppresses Squamous Carcinogenic Progression and Epithelial-Mesenchymal Transition

The transcription factor hypoxia-inducible factor-1 (HIF-1) is a fundamental mediator of cellular adaptation to microenvironmental stresses, such as hypoxia, free radical exposure, activation of oncogenes, loss of tumor suppressor genes, or enhanced survival signaling (1, 2). HIF-1α is induced in multiple tissue biological contexts, including embryonic development, hematopoiesis, inflammation, cerebral and myocardial ischemia, and carcinogenesis (3).

HIF-1 can contribute to cancer progression in multiple, nonoverlapping mechanisms by transactivating suites of genes regulating the microvasculature, glycolysis, oxidative phosphorylation, cell motility, migration, tissue invasion, and metastasis (1, 4). Detectable and elevated levels of HIF-1α protein are present in premalignant lesions and in epithelial, soft tissue, central nervous system, and hematologic malignancies (1, 5). HIF-1α protein stabilization is also associated with poor prognosis in most of these malignancies (1).

However, the concept that HIF-1α invariably facilitates growth and spread of cancer is counterbalanced by its putative role as a tumor suppressor. Thus, HIF-1α up-regulation has been associated with increased patient survival in patients with squamous carcinomas of the head and neck or the lung (6, 7). Hypoxic cell cycle arrest mediated by p21 and p27 induction seems to require HIF-1α (8–10). Moreover, loss of HIF-1α function is also associated with enhanced outgrowth of embryonic stem cell carcinomas (10). HIF-1α has also been shown to bind and sequester p53 as a mechanism for enhanced genomic instability associated with chronic cellular hypoxia (11). Additionally, in certain cellular contexts such as renal cell or lung cancer, HIF-1α directly and indirectly inhibits c-Myc function, resulting in either p21-mediated cell cycle arrest in the former or apoptosis in each cancer type (12, 13). Induction of the proapoptotic target genes BNIP3 or NIX may also explain the tumor suppressor capability of HIF-1α (14). However, each of these facets of HIF-1α–mediated tumor suppression is controversial. The precise roles of cyclin-dependent kinases in hypoxic cell cycle arrest and the functional significance of either HIF-1α–p53 or HIF-1α–c-myc interactions remain complex and are dependent on cell type and levels of oncogene expression (15–17). The contribution of HIF-1α or hypoxia-induced BNIP3 to cancer cell apoptosis has also been questioned (18).

We designed our experiments to test HIF-1 gain of function throughout the entire process of epithelial carcinogenesis from cancer stem cell initiation and subsequent promotion to malignant conversion of high-risk premalignant precursors using a combination of two-stage chemical carcinogenesis and K14-HIF-1α^{Pro402A/P564G} (HIF-1αDPM) transgenic mice (19). Use of the keratin-14 promoter targeted transgene expression to basal keratinocytes (20), whereas mutation of prolines 402 and 564 rendered the HIF-1α protein resistant to binding by the ubiquitin ligase pVHL (see references within ref. 19). HIF-1 gain of function facilitated the outgrowth of initiated foci into papillomas; however, differentiation of these premalignant neoplasias was maintained and malignant conversion was potently inhibited. Expression of N-Myc downstream-regulated gene-1 (NDRG1), a known HIF-1 target (21), and a gene with tumor suppressor functions (22, 23), was up-regulated and redistributed to basal keratinocytes of transgenic skin and papillomas. We also discovered a novel HIF-1 target gene, selenium binding protein-1 (SELENBP1), whose loss of function in several common epithelial cancers (24–26) suggested that it too had tumor suppressor function.

Animal studies. K14-HIF-1αDPM transgenic mice, created and used in the FVB/n inbred strain, were described previously (19). Mice were housed in pathogen-free conditions, and the Washington University Animal Studies Committee approved all experiments described in this study. For mouse skin carcinogenesis, the back skin of each mouse was shaved 2 d before topical treatment, with 7,12-dimethylbenz(a)anthracene (DMBA), 25 μg in 200 μL of acetone. One week later, mice were topically treated with 12.5 μg of 12-O-tetradecanoylphorbol-13-actate (TPA; Sigma) in 200 μL of acetone, once a week for 20 wk. Mice were sacrificed either when lesions with an appearance consistent with cancers reached 1 cm in diameter or arbitrarily at 25 wk postinitiation.

Tissue harvest and histology. Tissues for routine histopathology and selected immunohistochemical analysis were obtained from formalin-perfused mice, fixed using a microwave processing technique (19), and following a PBS wash and processing through graded alcohols and xylenes were embedded in paraffin. Five-micrometer tissue sections were stained with H&E or underwent immunohistochemical analysis.

Histopathology and immunohistochemistry. All sections were deparaffinized, rehydrated, washed in PBS, and blocked with Dako protein block for 30 min at room temperature. Antigen retrieval was performed in a pressure cooker (Decloaking chamber, Biocare Medical) in citrate buffer (pH 6.0) and used for anti–cytokeratin 10, biotinylated anti–cytokeratin 14, anti-desmin, and anti–E-cadherin immunostaining. Sections immunostained for bromodeoxyuridine (BrdUrd) were pretreated with 2 N HCl for 1 h at room temperature and 0.01% protease (type XXIV, Sigma) for 15 s. Antibodies/dilutions for the following markers were used in Dako antibody diluent and applied overnight at 4°C: rabbit anti–cytokeratin 10 and 14 (1:50, 1:2,000, Covance), biotinylated anti–cytokeratin 14 (1:1250, Lab Vision), goat anti–N-myc downstream-regulated gene 1 (NDRG1; 1:200, Santa Cruz Biotechnology), rabbit anti-desmin (1:250), rabbit E-cadherin (1:200, Cell Signaling), biotinylated anti-BrdUrd (1:200, Caltag), and rat anti–panendothelial cell antigen (MECA-32; 1:20 BD Pharmingen) cytokeratin 8 antibody (TROMA-1; 1:1,000, Developmental Studies Hybridoma Bank). Secondary antibodies labeled with Alexa Fluor 488, Alexa Fluor 594, or Alexa Fluor 594 streptavidin conjugate were placed on tissue sections for 1 h at room temperature (1:400, Molecular Probes). Nuclei were counterstained using SlowFade Gold Anti-fade reagent with 4′,6-diamidino-2-phenylindole (DAPI; Vector).

BrdUrd incorporation. DMBA/TPA–treated mice were injected i.p. with 100 mg/kg BrdUrd (Sigma). Tissue was collected after 3 h after BrdUrd injection for measurement of cell replication. Incorporated BrdUrd was detected as described above.

Cell lines. PDV cells originated from a DMBA-treated C57BL/6 neonatal keratinocyte culture (27). HEK 293 and PDV cells were cultured in DMEM (Life Technologies) and supplemented with 10% fetal bovine serum (Sigma) and 1,000 units/mL penicillin-streptomycin (Sigma) at 37°C in 21% O₂/5% CO₂. Where indicated, PDV cells were treated with 100 μmol/L CoCl₂ (Sigma) for 24 h.

Keratinocyte culture. Primary mouse keratinocytes were isolated from newborn transgenic and wild-type littermate epidermis as described previously (19) and seeded at 5 × 10⁶ cells per 60-mm dish (or equivalent concentrations) in Ca²⁺- and Mg²⁺-free MEM (Invitrogen Life Technologies) supplemented with 8% chelexed (Bio-Rad Laboratories) fetal bovine serum (Gemini Bioproducts) and 0.2 mmol/L Ca²⁺. After 24 h, cultures were switched to the same medium with 0.05 mmol/L Ca²⁺ to select for basal cells.

Microarray analysis. Primary keratinocytes (n = 3 for nontransgenic and transgenic) were homogenized in TRIzol (Invitrogen); total RNA was isolated using the manufacturer's guidelines and treated with RQ1 DNase (Promega) for 30 min at 37°C. Microarray probes were synthesized using RNA samples and hybridized with Affymetrix MU430Av2 GeneChips in the Siteman Cancer Center Multiplexed Gene Analysis Core, and differentially expressed transcripts were identified using unpaired t tests.

Plasmid construction, transfections, and luciferase reporter assays. A 479-bp (−635 to −156) fragment from the Selenbp1 (NC_000069) promoter was PCR amplified from mouse tail DNA and cloned into pCR2.1-TOPO (Invitrogen). The insert was sequenced and subcloned into pRL-null (Promega; pSELENBP1-Luc). The full-length HIF-1α^{P402A/P564A/N803A} cDNA, a kind gift from R. Bruick (Department of Biochemistry, University of Texas Southwestern Medical Center, Dallas, TX), was cloned into pIRES-hrGFP2a (Stratagene). Human embryonic kidney 293 cells (HEK293), passaged in 48-well plates, were cotransfected with DNA-liposome complexes Lipofectamine 2000 (Invitrogen Life Technology), containing 15 ng of either pRL-null reporter alone or pSELENBP1-Luc, with either 200 ng of pIRES-hrGFP2a or increasing amounts (50, 100, and 200 ng) of CMV-HIF-1α-PPN overnight. We determined luciferase activity in quadruplicate transfections using a Synergy HT luminometer (Bio-Tek).

Quantitative reverse transcription-PCR. Back skin, papillomas, and cancers were snap frozen in liquid nitrogen; homogenized in TRIzol (Invitrogen); and total RNA was treated in RQ1 DNase (Promega) for 30 min at 37°C. Reverse transcription-PCR (RT-PCR) was performed as described previously (19) using a Stratagene MX3000P. Primer Express software (version 2.0 Applied Biosystems) was used to design primer/probe sets (Supplementary Table S1) to amplify the genes described below. All target cDNAs were normalized to histone 3.3A (28).

Western blotting. Cells or tissues were lysed in radioimmunoprecipitation buffer [10 mmol/L Tris (pH 7.4), 150 mmol/L NaCl, 1 mmol/L EDTA, 0.1% SDS, 1% Triton X-100, 1% sodium deoxycholate, 0.25 mmol/L phenylmethylsulfonylfluoride]; 20 μg of total lysate were loaded on polyacrylamide gels; transferred to polyvinylidene difluoride membranes (Amersham Biosciences); blocked in Blotto-Tween solution (5% nonfat dry milk, 0.1% Tween in PBS) for 1 h; and incubated overnight in PBS with a rabbit polyclonal anti-SELENBP1 (1:20,000, a gift from D. Medina, Baylor College of Medicine), goat polyclonal anti-NDRG1 (1:1,000, Santa Cruz Biotechnology), rabbit anti–cytokeratin 10 and 14 (1:500, Covance), rabbit polyclonal anti-Snail and E-cadherin (1:1,000, Cell Signaling), or β-tubulin (1:5,000, Sigma). Peroxidase-coupled AffiniPure goat anti-rabbit or donkey anti-goat IgG (1:200, Jackson Immunoresearch Laboratories) was used in secondary incubations for 1 h followed by chemiluminescence detection (ECL Plus, Amersham Biosciences).

Microscopy. All microscopy images were obtained with a BX61 microscope (Olympus America) using the following objectives: UPlan Apochromatic 4×/NA 0.16, UPlan Apochromatic 10×/NA 0.40, UPlan Apochromatic 20×/NA 0.70, and UPlan Apochromatic 40×/NA 0.85. Tissue sections stained with H&E or diaminobenzadine immunoperoxidase were mounted with Permount (SP15-500, Fisher Scientific), and coverslipped microscopy images were obtained with a DP70 color Bayer mosaic digital camera, Peltier device cooled to −10°C (Olympus). Tissue sections for fluorescence microscopy images were mounted with SlowFade Gold antifade reagent with DAPI (Invitrogen, Molecular Probes), coverslipped, and images were obtained with a Soft Imaging Solutions FVII cooled monochrome digital camera, Peltier cooled to −10°C (Olympus). All images were captured with MicroSuite Biological Suite version 5 software (Olympus Soft Imaging Solutions) and resized and formatted with Adobe Photoshop CS3 software (Adobe Systems Incorporated).

Statistical analysis. Data were analyzed using GraphPad PRISM; results were expressed as the mean ± SE; and statistical significance was determined using either the Student's t test or the Mann-Whitney U test or by using a contingency table analysis.

K14-HIF1-αDPM mice exhibit propensity for benign neoplasms with resistance to cancer formation. To study the effects of HIF-1 gain of function at initiation and then progression of epithelial carcinogenesis, we performed the classic two-stage carcinogenesis protocol on K14-HIF1-αDPM transgenic and nontransgenic mice. However, twice a week TPA application caused an exacerbated back skin inflammation in transgenic mice, consistent with our recent work (19). Therefore, we began a new study with an unchanged DMBA dose, but at half the dose and frequency of TPA application that was well tolerated by transgenic mice for 20 weeks of promotion. The nontransgenic papilloma frequency, 2 ± 1.5 per mouse (Fig. 1A), was consistent with a previous work with a similar chemical carcinogenesis protocol in FVN/n mice (29). Moreover, 40% of all papillomas, n = 20 for the entire group of nontransgenic mice, converted to cancer (Fig. 1B). Our low nontransgenic papilloma frequency with enhanced malignant conversion rate was consistent with induction of “high-risk” papillomas (30). Transgenic mice developed 3-fold more papillomas (6 ± 3 per mouse) than nontransgenic counterparts (Fig. 1A), and the total number of papillomas in the entire transgenic cohort (n = 57) was also nearly 3-fold higher than the 20 papillomas in the nontransgenic study group (Fig. 1B). These data were consistent with an increase in initiation and growth facilitation of transgenic initiated foci by HIF-1 gain of function. In contrast to our previous work (19), there was no differential increase in CD45-positive cells in transgenic papillomas, likely due to dose and interval reduction of TPA application (data not shown). Surprisingly, the malignant conversion rate of each individual transgenic papilloma (7%) was 7-fold lower than nontransgenic counterparts (Fig. 1B). Previously, we discovered that HIF-1α overexpression induced angiogenesis (28) and this was present in both transgenic papillomas and cancers (Supplementary Fig. S1A and C). Increased blood vessel density was previously documented to be an early event required for papilloma development (31, 32) and it could explain the enhanced papillomagenesis in transgenic mice.

The resistance of transgenic papillomas to malignant conversion resistance led us to further determine the differences in their tissue biology compared with the nontransgenic counterparts (Fig. 1C and D). Nontransgenic papillomas (Fig. 1E) evidenced loss of the differentiation-associated keratin-10 (Fig. 1E), whereas transgenic papillomas retained keratin-10 expression (Fig. 1F). Moreover, epidermal proliferation was enhanced in nontransgenic papillomas; BrdUrd-positive keratinocytes were detectable in both the basal and suprabasal layers (Fig. 1G, and inset; see also Supplementary Fig. S2A), whereas BrdUrd-positive transgenic keratinocytes were confined to the papilloma basal layer (Fig. 1H , inset; Supplementary Fig. S2B). Diminution of mRNA levels of Ki67 and cyclin D1 (Supplementary Fig. S3A and B) further supported the conclusion that transgenic papilloma proliferation was inhibited.

Suppression of epithelial-mesenchymal transition in transgenic squamous cancers. To determine whether resistance of transgenic papillomas to malignant conversion also produced a different cancer phenotype compared with transgenic counterparts, we histopathologically classified all squamous cancers in the study (31) into three histotype groups: well-differentiated squamous cell carcinoma, in which most of the tumor mass composed of epithelial cells, with eosinophilic cytoplasmic staining consistent with keratin expression (Fig. 2A); transition lesions, consisting of epitheliod cells with a paucity of keratinization and apparent gaps between adjacent malignant cells consistent with decreased homotypic cellular adhesion (Fig. 2B); and poorly differentiated spindloid tumors with no keratinization (Fig. 2C). This histopathologic differentiation classification was bolstered by immunohistochemical analysis of patterns of K14 and K8 expression (20) wherein well-differentiated cancers displayed strong immunofluorescence for K14, with only sporadic K8 detection (Fig. 2D). Transition cancers displayed an overall diminished signal for K14 with a sporadic desmin signal localized in linear tube–like structures consistent with microvessels (Fig. 2E). Spindloid cancers evidenced strong but scattered K8 immunofluorescence, undetectable K14 (Fig. 2F), and a ubiquitous desmin signal (Fig. 2G), a hallmark of epithelial-mesenchymal transition (EMT). Using these criteria, all three transgenic cancers were well differentiated (Fig. 2H). In contrast, only 37.5% of nontransgenic cancers were well differentiated, another 25% were transition lesions, and 37.5% were spindloid/EMT cancers (Fig. 2H). Thus, HIF-1 gain of function not only inhibited conversion of premalignant lesions to cancer but it also suppressed development of poorly differentiated malignancies.

Induction of the tumor suppressor genes Ndrg1 and Selenbp1 by basal keratinocyte HIF-1α gain of function. As our tissue-derived data suggested that HIF-1 was mediating repression of squamous carcinogenesis via cell autonomous mechanisms, we established primary keratinocyte cell cultures from 2- to 3-day-old neonates and performed microarray analysis (Supplementary Table S2) focusing on differential expression of genes known to be involved in regulation of differentiation or proliferation. Surprisingly, and again consistent with the carcinogenesis data, we did not detect a significant differential down-regulation of E-cadherin mRNA in transgenic keratinocytes (data not shown). However, we discovered a 7-fold increase of N-Myc downstream regulated gene-1 (NDRG1) mRNA expression, a previously characterized HIF-1 target gene (21). We also found a 80-fold induction in Selenbp1 and Selenbp2 mRNAs (33). Mouse Selenbp1 and Selenbp2 are highly related, with only 20 nucleotides and 14 amino residues differentiating the two genes in the coding sequence (34), making it impossible to separate their expression by real time RT-PCR or even protein expression. Despite their similarity, the two genes are regulated differently and likely have different functions in that Selenbp2 seems to be involved in acetaminophen metabolism (34). Our subsequent analysis determined that alterations in protein expression during carcinogenesis were likely due to SELENBP1 expression (see below); thus, we used this terminology collectively when unable to determine the genes individually. As antibodies for immunohistochemical detection of SELENBP1 in mouse tissues were not available, in contrast to NDRG1, we restricted our analysis of expression of SELENBP1 protein or its mRNA to tissue or cell extracts. We validated the transgenic induction of Ndrg1 and Selenbp1 in independent sets of keratinocyte cultures and found a 10-fold elevation of Selenbp1 mRNA and protein and a 7-fold increase in Ndrg1 mRNA and protein in transgenic compared with nontransgenic cultures (Fig. 3A and B). Transgenic NDRG1 overexpression was further validated by immunofluorescence, revealing markedly enhanced punctate cytoplasmic NDRG1 protein expression (Fig. 3C and D) consistent with previously reported localization to recycling/sorting vesicles (35).

Enhanced Ndrg1 and Selenbp1 mRNA (data not shown) and protein expression was also present in untreated transgenic back skin that contained a 2-fold increase in the former and a 4-fold elevation of the latter protein (Fig. 4), respectively. HIF-1 gain of function also redirected NDRG1 expression from the terminally differentiated suprabasal layer and inner root sheath of nontransgenic skin (Fig. 5A; Supplementary Fig. S4A and C; ref. 36), to the proliferative basal cell compartment of transgenic epidermis (Fig. 5B; Supplementary Fig. S4B and D). The paranuclear localization of NDRG1 expression in transgenic skin (arrowheads, Supplementary Fig. S4B) was consistent with localization in trans Golgi recycling/sorting vesicles (35).

To further investigate potential NDRG1 and SELENBP1 tumor suppressor functions, we determined the levels and patterns of expression of these molecules in papillomas, the premalignant precursor lesion of chemical carcinogenesis. NDRG1 protein expression was markedly down-regulated in nontransgenic papillomas (Fig. 5C), in contrast to its strong expression in the basal and suprabasal neoplastic epidermal layers of transgenic papillomas (Fig. 5D). Western blotting revealed a 2-fold increase of NDRG1 protein in transgenic compared with nontransgenic papillomas (Fig. 4). SELENBP1 protein was also elevated 3-fold in transgenic papilloma extracts (Fig. 4). Elevated NDRG1 and SELENBP1 protein expression also correlated with a 3-fold elevation of expression of keratin-10, a differentiation marker, in transgenic, compared with nontransgenic, papilloma extracts (data not shown). NDRG1 was known to induce differentiation in colon cancer cell lines and was also elevated by several different differentiation-inducing agents (22). As commitment to terminal differentiation is initiated in the epidermal basal layer (20), the expression of a molecule such as NDRG1 in stem-like/transient amplifying cells could have been responsible for differentiation maintenance and lower proliferation of transgenic papillomas.

As both NDRG1 and SELENBP1 expression were lost and correlated with poor prognosis in human cancers (22, 24–26, 37), we determined the expression of these molecules in nontransgenic and transgenic cancers (Fig. 4). Nontransgenic protein extracts were correlated with the histologic presence or absence of EMT. Snail, an EMT marker and initiator (38), was not detectably expressed in transgenic cancers despite gain of HIF-1α function (Fig. 4; refs. 39, 40), whereas it was consistently expressed at either a low or high level in either histologically well-differentiated or spindloid nontransgenic malignancies, respectively (Fig. 4). NDRG1 protein was 3-fold lower in Snail-high compared with Snail-low nontransgenic protein extracts (Fig. 4). SELENBP1 protein was expressed at a low level in nontransgenic cancers, independent of Snail expression, compared with elevated SELENBP1 levels in transgenic cancers (Fig. 4). These data were bolstered by immunofluorescence analysis of malignant tissue sections from transgenic (Fig. 5E and F) and nontransgenic mice (data not shown). Strong NDRG1 expression was detectable in well-differentiated regions of transgenic squamous cancers (Fig. 5E). Adjacent sections revealed that these malignant cells retained membrane-bound E-cadherin (Fig. 5F). NDRG1 expression was lost in malignant squamous cells of both transition and EMT cancers along with E-cadherin in the latter lesions (data not shown). The association of NDRG1 and E-cadherin expression patterns mirror NDRG1 small interfering RNA knockout studies in prostate cancer cell lines wherein NDRG1 was shown to control E-cadherin recycling and E-cadherin plasma membrane localization (35).

Mouse selenbp1 is a HIF-1α target gene. We focused further investigation on the regulation and function of Selenbp1 in keratinocyte and epithelial cell lines. Direct regulation of Selenbp1 by HIF-1α was determined by the hypoxia mimetic CoCl₂, which produced a 2-fold elevation of Selenbp1 mRNA (Fig. 6A) PDV cells (27). Transient overexpression of the triple HIF-1α point mutant, HIF-1α^{P402A/P564A/N802A} (HIF-1αPPN), increased Selenbp1 mRNA 4-fold (Fig. 6B).

We searched the promoter and first intron regions of the Selenbp1 and Selenbp2 genes for hypoxia response elements (HRE); in particular, matches to a consensus sequence containing the core HIF-1–binding site 5′-CGTG-3′ and a 1-8 nucleotide spacer followed by a CAC sequence (41). A total of eight HREs were detected in the first 5,000 bp of the mouse Selenbp1 gene, four in the promoter and four in the first intron (data not shown). A 0.479-kb DNA fragment containing the three upstream Selenbp1 HREs (Fig. 6C) produced a titratable 4-fold induction of Selenbp1 promoter activity when cotransfected with CMV-HIF-1α^{P402A/P564A/N802A} in HEK293 cells (Fig. 6D). In marked contrast, expression from a fragment spanning the homologous promoter region in Selenbp2 (data not shown) was unaffected by HIF-1 overexpression. Collectively, the data showed that Selenbp1 was a bona fide new HIF-1 target gene and further emphasized the distinction and differential regulation of the murine Selenbp1 versus Selenbp-2 genes. Notably, the homologous region in the human SELENBP1 promoter contained four HREs within 1,400 bp of the transcriptional start site (data not shown).

In this study, K14-HIF-1αDPM mice subjected to two-stage carcinogenesis were sensitive to the outgrowth of initiated foci to form benign neoplasms with an inherent resistance to malignant progression. Rapid emergence and frequency of transgenic papillomas was likely facilitated by angiogenesis, a prominent component of this model of HIF-1 gain of function (19, 28). A robust angiogenesis was previously shown in squamous cell hyperplastic foci, the immediate antecedents of nascent papillomas (31). Moreover, epidermal vascular endothelial growth factor overexpression itself accelerated papilloma development and enhancement of papilloma number, further supporting a functional role for angiogenesis in papillomagenesis (32).

A striking finding in our study was that despite enhanced papillomagenesis, proliferation was diminished and differentiation was maintained in transgenic versus nontransgenic papilloma precursor lesions. Moreover, the resistance to EMT of transgenic malignancies was also surprising, as HIFα proteins were known to decrease E-cadherin expression via induction of the transcriptional repressors Snail, TCF3, ZFHX1A, and SIP1 (39, 40). However, one explanation for our findings could be the redirection and induction of NDRG1 in the basal keratinocytes of transgenic epidermis and maintenance of expression of this molecule in transgenic papillomas and cancers compared with the loss of its expression in nontransgenic counterpart lesions. Ndrg1 was a known HIF-1 target with two HREs identified upstream of its promoter (21). Several studies confirmed the role of Ndrg1 in epithelial cell differentiation. NDRG1 increased during epidermal keratinocyte differentiation in culture and localized to the differentiated layers of intact skin (36). However, in contrast to our work, NDRG1 was overexpressed in squamous cancers, resulting from two-stage chemical carcinogenesis. The discrepancies between that study and our work was likely due to differences in promotion protocols with our once weekly TPA administration favoring emergence of EMT and poorly differentiated malignancies, in which we showed loss of NDGR1 expression. NDRG1 was also functionally linked to differentiation of colorectal cancer cells (22) and inhibited tumor cell matrix invasion in colon and prostate cancers (22, 23). Regulation of differentiation and invasion could be linked to the requirement of intact NDRG1 function for E-cadherin recycling and stabilization (35). One comprehensive explanation of cancer resistance in our work could be that NDRG1 up-regulation in basal keratinocytes enhanced their fate choice from continual proliferation to commitment to terminal differentiation (20). Maintenance of NDRG1 expression mediated by HIF-1 gain of function could have facilitated continual E-cadherin expression and EMT inhibition in transgenic squamous cancers. The fact that NDRG1 expression is lost in many epithelial malignancies in which HIF-1α is overexpressed (22, 23, 42) could be due in part to negative selection similar to that seen for p53 in hypoxic cell cultures (43). However, repression of NDRG1 expression by c-myc, or N-myc, PTEN, and p53, and DNA methylation (44, 45) could also contribute to loss of function of this gene in human carcinomas.

The resistance for malignant conversion in this experiment was also prima facie surprising given our previous work demonstrating enhanced inflammation in the same K14-HIF-1αDPM transgenic model. Inflammation is known to facilitate cancer development; however, the nature and intensity of this process govern recruitment of distinctive inflammatory cell subsets, helper, Th1, versus suppressor, Th2, T-cells, myeloid suppressor cells, or tumor-associated macrophages, that either inhibit or stimulate carcinogenesis (46). In our study, biweekly TPA application produced an extremely robust inflammatory reaction and transgenic mice were actually resistant to both papilloma formation and cancers, which mirrored our previous work (data not shown). In contrast, reduction of TPA dosage to once a week, although sufficient for induction of “high-risk” papillomas, did not produce a differential inflammatory reaction in transgenic versus nontransgenic mice as determined by density of CD45 cells. As such, we are confident that although K14-HIF-1αDPM mice can possess an inflammatory hyperresponsiveness to TPA, we found a dose of this tumor promoter that rendered this variable irrelevant and supported the potential primacy of the NDRG1 and SELENBP1 HIF-1 targets in carcinogenic inhibition in this model.

We also identified Selenbp1 as a novel HIF-1 target gene, the homologue of SELENBP1, in humans (47). As the name suggests, SELENBP1 belongs to the selenium-containing protein family. Selenium is an important micronutrient with novel anticancer, “nutraceutical” activities. There is a statistically significant inverse relationship between selenium levels and cancer risk (48) and mechanisms proposed for its nutraceutical activity include antioxidant protection, altered carcinogen metabolism, inhibition of proliferation and tumor cell invasion, and induction of apoptosis (49). SELENBP1 is the only human selenium-associated protein in which selenium is covalently bound, in contrast to the more common incorporation of selenocysteine into the amino acid sequence via the TGA codon (33). Most importantly, the cellular levels of SELENBP1 protein are directly regulated by selenium, increasing with elevated media selenium concentration (50).

Inferences on SELENBP1 function in normal epithelium and carcinomas were derived from several different types of studies. SELENBP1 expression was widespread in epithelial tissues, including kidney, liver, lung, colon, prostate, pancreas, kidney, and ovary (47). SELENBP1 was commonly lost in many types of human epithelial cancers and its loss correlated with poor prognosis (25, 26, 51). The molecular mechanisms of Selenbp1 down-regulation in cancer cells have remained a mystery; neither promoter hypermethylation nor gene deletion was responsible (26). Previous work suggested that Selenbp1 exerted tumor suppressor function by inhibition of proliferation. Androgen-stimulated LNCaP cells were shown to down-regulate SELENBP1 expression (51), and SELENBP1 protein levels correlated with DNA synthesis inhibition (50). However, we were unable to delineate an affect on proliferation or apoptosis in transfection rescue of malignant cells with absent SELENBP1 expression (data not shown). As such, the precise function of this protein remains to be determined.

In summary, we have shown that HIF-1 could definitely function as a tumor suppressor gene potentially by up-regulation of a known, Ndrg1, and a novel, Selenbp1, gene, both of which are provocatively lost and associated with poor prognosis in human epithelial malignancies. One inference of our data is that coordinate negative selection for both of these HIF-1 targets may be present in regions of human cancer with stabilization of HIF-1α protein expression. An outstanding corollary inference is that HIF-1 functions to protect cells against malignant conversion by direct or indirect HIF-1 regulation of cell cycle and apoptosis regulators (8, 9, 14–17) and the more recent demonstration of HIF-1α up-regulation in response to keratinocyte UVB irradiation (52). As such, profligate application of HIF inhibitors as novel chemopreventive agents should be carefully evaluated.

No potential conflicts of interest were disclosed.

Grant support: National Cancer Institute grant R01 CA90722.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We thank Dan Medina for the gift of the rabbit anti-mouse SELENBP1 antibody; Greg Longmore for provision of the PDV cells; Stuart Yuspa, Loren Michel, and Greg Longmore for critical input on the manuscript; and Karline Bringe for technical assistance.