Loss of p53 Expression Correlates with Metastatic Phenotype and Transcriptional Profile in a New Mouse Model of Head and Neck Cancer Free

Squamous cell carcinoma of the head and neck (HNSCC) is the sixth most frequent cancer worldwide (1). HNSCC is a major cause of morbidity and mortality in developing nations, comprising up to 50% of all malignancies. HNSCC is the most common malignant tumor of the oral cavity with nearly 30,000 new cases and 8,000 deaths reported in the U.S. each year (2). Tobacco carcinogens are the primary etiologic agents of the disease, with age and genetic background as contributory factors. The overall 5-year survival rate of ∼50% has not changed significantly in recent decades.

A recent report identified a shortage of suitable animal models with which to study different biological and clinical aspects of HNSCC (3). Because HNSCC is largely acquired by environmental carcinogen exposure rather than through germ line mutations, there are no known familial forms of the disease in humans nor are there inbred rodent strains prone to spontaneous head and neck tumors (4). A classical animal model of HNSCC was carcinogen exposure of the hamster buccal cheek pouch (5). A second model of HNSCC relied on human tumor cell xenografts in immunodeficient mice. Studies using this model s.c. injected cultured human HNSCC cells into the backs of nude mice (6). These models have been used in many types of cancer to determine in vivo tumorigenicity but typically fail to replicate local invasion and lymph node metastasis of HNSCC. Variations on this model have injected human tumor cells into other anatomic sites (7, 8). However, the xenograft models are limited to human cancer cell lines that can adapt to the murine environment and do not replicate the early stages of carcinogenesis. Recently, a transgenic mouse model expressing activated K-ras reportedly developed epithelial lesions ranging from oral papillomas (9) to squamous cell carcinomas of skin, esophagus, stomach, uterine cervix, oral mucosa, and salivary glands (10). However, K-ras is reported to be infrequently mutated in human HNSCC cases (11, 12).

Transgenic animals with inactivation of tumor suppressor genes commonly mutated in human cases of HNSCC provide attractive models for studying the pathogenesis of head and neck cancer. p53 is the most frequently inactivated tumor suppressor gene in HNSCC (13). High frequencies of p53 mutations have also been found in mucosal dysplasias, which are potential precursor lesions for HNSCC (14, 15). The p53 DNA binding domain is the mostly commonly mutated region of the gene in DNA isolated from cancer cells (16). p53 functions in a variety of ways to suppress tumor formation, including activation in the presence of DNA damage, thereby arresting the cell cycle and inducing apoptosis in damaged cells (17). p53 regulates the transcription of a number of genes involved in these cellular processes (18). In normal cells, DNA-damaging agents result in p53 induction and growth arrest. Loss of p53 function may allow cells to replicate damaged DNA and divide, creating the potential for genomic instability in the daughter cells.

Mice in which the p53 gene has been inactivated by homologous recombination develop normally but have a higher incidence of tumors, primarily lymphoma (19). The majority of mice with inactivation of both p53 alleles (−/−) develop tumors by 5 months of age, and heterozygous animals (+/−) are also more susceptible to tumor formation. Overexpression of human papillomavirus type 16 oncogenes E6 and E7 (which inactivate p53 and Rb tumor suppressors) in mouse skin resulted in hyperproliferation and susceptibility to chemical carcinogens (20). The spontaneous development of epithelial cancers in mice is rare; therefore, to evaluate how p53 inactivation contributed to head and neck carcinogenesis in this animal model, we used a chemical induction protocol in p53 heterozygous mice. Metastatic squamous cell carcinomas developed in 100% of the animals. Histopathologically, the tumors ranged from well to poorly differentiated and showed many molecular features of human HNSCC. Mice carrying one p53 allele developed tumors with significantly reduced latency compared with wild-type controls (average, 18 versus 22 weeks). Metastatic cancer cells showed complete loss of p53 expression when compared with primary tumors. Transcriptional profiling showed not only distinct genetic differences between primary and metastatic tumors, but also when cancers from heterozygous null and wild-type animals were compared. Our results provide novel insights into the molecular genetics of tumor progression in head and neck cancer.

To determine the contribution of the p53 gene product to head and neck carcinogenesis, we evaluated three induction protocols on groups of p53 wild-type, heterozygous null, and homozygous null mice. Groups of mice were treated with 7,12-dimethylbenz(a)anthracene (DMBA) alone, a single dose of DMBA followed by 12-O-tetradecanoylphorbol-13-acetate (TPA), or TPA alone as described in Materials and Methods. DMBA is a well-characterized carcinogen in mouse models but is less abundant in tobacco smoke than other polycyclic aromatic hydrocarbons such as benzo(a)pyrene. Homozygous null animals died by 4 months of age from other tumors (primarily lymphoma) prior to the onset of HNSCC. Among p53 wild-type and heterozygous null mice, no tumors were observed in the group treated with TPA alone for up to 52 weeks (n = 20). In the group treated with an initiating dose of DMBA followed by twice weekly doses of TPA, 2 out of 30 mice developed one small papilloma each on the buccal mucosa which did not progress even after 52 weeks of treatment. In marked contrast, 100% of mice (n = 18) treated with twice weekly doses of DMBA developed papillomas on the labial and buccal mucosa which grew larger and progressed to large keratinizing tumors in the absence of additional induction. p53+/− mice developed tumors after a mean latency period of 18 weeks versus 22 weeks for wild-type animals (P < 0.02, t test; Fig. 1A). Both genotypes developed euthanasia criteria with similar time courses (mean of 13 weeks after onset of tumors; Fig. 1B). The most common criterion was weight loss followed by tumor size. There were no significant differences between genotypes with respect to euthanasia criteria. These results indicate that p53+/− mice developed head and neck tumors significantly earlier than wild-type animals.

The gross appearance of an early squamous cell carcinoma arising from the labial mucosa in the mouse model is shown in Fig. 2A. These tumors underwent progression in the absence of additional induction and formed large highly vascularized keratinizing carcinomas (Fig. 2B). Definitive cervical lymph node metastasis could be detected histopathologically as early as 3 weeks after the onset of the primary tumor. The representative gross appearance of the enlarged metastatic lymph nodes at necropsy 12 weeks after the onset of the primary tumor is shown in Fig. 2C. The average number of metastatic cervical lymph nodes in this model was 4.1 ± 0.5 per mouse, and the largest node found in this model was 6 mm in diameter (mean, 3.2 ± 0.3 mm; P < 0.00001). The size of representative metastatic lymph nodes was compared with that of a normal lymph node in Fig. 2D. At necropsy, there was no evidence of tumor extension through the lymph node capsule or distant metastasis to any organ. These results indicate that HNSCCs in the p53 mouse model rapidly metastasize to cervical lymph nodes after the onset of the primary tumor.

Histopathologic analysis of primary tumors and cervical lymph nodes from this model is shown in Fig. 3. Papillomatous lesions from the TPA control and DMBA treatment groups are shown in Fig. 3A and B, respectively. The two papillomas detected in the TPA control group appeared as hyperplastic lesions composed of histologically normal stratified squamous epithelium with submucosa consisting of dense fibrous connective tissue. In contrast, the papillomatous lesions from the DMBA treatment group showed evidence of well differentiated squamous cell carcinoma at initial presentation. Foci of well differentiated tumor cells had invaded the basement membrane and replaced normal submucosal tissues. At necropsy, the advanced stage HNSCCs showed clear evidence of tumor progression with all degrees of differentiation represented. Well differentiated squamous cell carcinoma was the predominant histologic type in 16 of 26 tumors examined by histopathology (62%; Fig. 3C). These tumors showed extensive basal layer and suprabasal differentiation with evidence of basement membrane formation and prominent keratinization. Seven tumors were histopathologically classified as moderately differentiated (27%; Fig. 3D). These tumors showed less evidence of stratification, basement membrane production, and keratin formation. These carcinomas were also characterized by the loss of intercellular junctions, increased nuclear/cytoplasmic ratio, nuclear pleomorphism, and occasional mitotic figures. Three tumors were classified as poorly differentiated or anaplastic (11%; Fig. 3E and F). These tumors were composed of sheets and bundles of spindle-shaped cells with elongated nuclei and complete loss of intercellular junctions. Eosinophilic inclusions were frequently observed in the cytoplasm of these cells which likely represented abnormal keratin production; mitotic figures were rarely observed in these tumors. There were no statistically significant differences in the distribution of these tumor types between p53 genotypes. These results indicate that the mouse HNSCC model can recapitulate the full histopathologic spectrum of tumors found in the human disease.

In contrast to our findings in the primary tumors, well differentiated squamous cell carcinoma represented only 22% of 59 metastases examined by histopathology. Moderately differentiated squamous cell carcinoma was the predominant histopathologic tumor type observed in cervical lymph node metastasis (78%). Spindle cell metastasis was not observed in any lymph nodes. Metastases were observed immediately adjacent to the lymph node capsule as early as 3 weeks following the onset of the primary tumor. These foci became larger and clusters of tumor cells disseminated throughout the lymph node (Fig. 3G and H), eventually replacing most of the normal lymph node architecture (Fig. 3I and J). In the most advanced lymph node stages, large areas of necrosis were noted in addition to complete replacement of lymphocytes by sheets and foci of tumor cells (Fig. 3K and L). These results indicate that HNSCC metastases occur soon after the onset of the primary tumor, was less differentiated than their corresponding primary tumors, and that these metastases progressively replaced the normal structure of cervical lymph nodes.

To begin to determine gene expression differences between HNSCC arising in p53 heterozygous and wild-type mice, we initially correlated p53 protein levels with those of the growth factor receptors and cell cycle regulatory proteins previously examined in human HNSCC. We detected p53 protein expression in 89% of primary tumors, and these levels correlated with the expression of the growth factor receptor met (P < 0.02; Fig. 4A-D). In marked contrast, p53 was detected in only 11% of cases of metastases in cervical lymph nodes. This loss of p53 expression correlated with significantly decreased epidermal growth factor receptor (EGFR; P < 0.04; Fig. 4E-H) and cyclin D1 protein levels (P < 0.05; Fig. 4I and J). These data suggested that proliferation of metastases in the lymph nodes may be less than that of those in the primary tumor. To test this hypothesis, we did immunohistochemistry using anti–proliferating cell nuclear antigen (PCNA) antibody on primary and metastatic tumor tissue. As shown in Fig. 4K-L, PCNA was highly expressed in >90% of the proliferative basal layer cells in the primary tumor. In contrast, <10% of metastases in cervical lymph nodes expressed PCNA (P < 0.006). To determine if p53 allelic loss contributed to the lack of expression of this gene in metastatic tumors, we extracted genomic DNA from these cells and did PCR using p53-specific primers. All metastatic tumors had lost one or both p53 alleles. Only two metastatic tumors in the wild-type background had one remaining p53 allele based on comparisons with amplified normal genomic DNA. PCR amplification of the p53 gene in these two, and nine other representative samples from wild-type and heterozygous backgrounds, are shown in Fig. 4M. These data indicate that loss of p53 expression is common during metastatic progression of HNSCC, which correlates with decreased levels of proliferation markers in the mouse model.

To determine if the mouse HNSCC model recapitulated additional molecular characteristics of the human disease, we did immunohistochemistry on sections of primary and metastatic tumors. We previously showed that human cases of HNSCC overexpressed EGFR, cyclin A, cyclin B, and cyclin E proteins (21). In the mouse model, 39% of HNSCC primary tumors overexpressed EGFR, 56% overexpressed cyclin A, 61% overexpressed cyclin B, 44% overexpressed cyclin D1, and 39% overexpressed cyclin E proteins. High levels of EGFR protein statistically correlated with increased cyclin A (P < 0.004), cyclin D1 (P < 0.01), and cyclin E (P < 0.05) expression. Representative immunohistochemical sections are shown in Fig. 5. These results indicate that EGFR expression is closely linked to that of downstream cell cycle regulatory proteins. Growth factor receptor and cell cycle regulatory proteins are overexpressed in the mouse HNSCC model, which correlated with the expression of those genes in the human disease.

We also examined the expression of growth factors important in human HNSCC in the mouse model using immunohistochemistry, i.e., epidermal growth factor, transforming growth factor α (TGFα), and hepatocyte growth factor (HGF; Fig. 6). Both epidermal growth factor and TGFα are ligands for EGFR (22), and HGF binds to the met receptor which promotes the invasion of reconstituted basement membrane by HNSCC cell lines in vitro (23, 24). Both epidermal growth factor and TGFα protein expression was detected in 89% of HNSCCs in the mouse model; high levels of TGFα correlated with the overexpression of cyclin E (P < 0.05), which further suggests that TGFα/EGFR autocrine signaling influences the expression of downstream cell cycle regulatory proteins. High levels of HGF expression correlated with cyclin B protein levels (P < 0.05), suggesting a link between met receptor signaling and expression of the G₂-M phase cell cycle regulatory protein in HNSCC.

To identify additional genetic targets of p53 in the mouse HNSCC model, we compared relative mRNA expression in primary and metastatic tumors in both wild-type and p53 heterozygous animals by microarray analysis. Hierarchical clustering analysis revealed that primary tumors from p53+/+ and p53+/− mice were highly related, as were metastases. When comparing gene expression signatures of HNSCC from p53+/+ and p53+/− mice, we detected 41 and 43 differentially expressed nonoverlapping transcripts in primary and metastatic tumors, respectively. Gene expression differences in primary tumors are shown in Table 1 and between metastases in Table 2. Up-regulated genes include those related to cell survival (bcl2-like 1, 4-fold; phosphatidylinositol-3-kinase catalytic δ polypeptide, 2.1-fold) and carcinogen detoxification (glutathione S-transferase, 3.5-fold) in p53+/− mice. Interestingly, the expression of TGFβ2, which inhibits the growth of stratified squamous epithelia was down-regulated by 2.1-fold in heterozygous animals, which correlated with the earlier onset of tumors in this group. These results indicate that primary and metastatic HNSCCs from p53 wild-type and heterozygous null mice exhibit distinct patterns of gene expression.

In contrast, hierarchical clustering analysis revealed dramatic differences between primary and metastatic tumors. Comparison of primary to metastatic tumors in p53 wild-type mice revealed 1,511 differentially expressed genes. This analysis showed dramatic down-regulation of epithelial differentiation genes (filaggrin, −181.6-fold; keratin complex 2 basic gene 5, −165.8-fold; epithelial membrane protein 2, −4.5-fold), which correlated with the less differentiated histopathology of the metastases (Table 3; Fig. 3). Expression of genes encoding intercellular adhesion proteins were also down-regulated (plakophilin 1, −458.7-fold; cadherin 3, −16.7-fold; protocadherin 7, −4.7-fold), which correlated with loss of cellular adhesion evident in histopathologic sections from the metastases. Growth factors and receptors were notably down-regulated in metastases: Jagged 1 (−9.5-fold), TGFβ2 (−8.7-fold), fibroblast growth factor receptor 3 (FGFR3; −8.3-fold), EGFR (−8.0-fold), Wnt3a (−5.2-fold), FGFR2 (−4.6-fold), and growth hormone receptor (−3.1-fold). In addition to decreased expression in metastases, TGFβ2 expression was also down-regulated in p53+/− primary tumors (Table 1). Decreased EGFR expression observed by microarray analysis in metastases was confirmed by our immunohistochemical results (Fig. 4). A number of gene products involved in signaling cascades was up-regulated: signal transducer and activator of transcription 4 (STAT4; 23.4-fold), c-src tyrosine kinase (4.4-fold), and STAT5A (2.7-fold). However, the expression of several transcription factors was down-regulated in metastases: vitamin D receptor (−8.4-fold), c-jun (−3.8-fold), hypoxia-inducible factor 1α (−3.7-fold), retinoic acid receptor γ (−2.8-fold), and Jun B (−2.2-fold). Taken together, these results indicate that changes in metastatic gene expression are reflected in the observed phenotype and histopathology.

Comparison of gene expression signatures of primary and metastatic HNSCC from p53+/− mice revealed 2,503 differentially expressed genes. A large number of genes regulating cell migration and cytoskeletal organization was up-regulated in metastases: rhophilin (18.6-fold), Rac2 (15.9-fold), Rho GTPase-activating protein 15 (15.6-fold), Wiskott-Aldrich syndrome human homologue (15.1-fold), Rac/cdc42 guanine nucleotide exchange factor 6 (14.2-fold), Rho GTPase-activating protein 25 (5.4-fold), Rho GTPase-activating protein 9 (5.3-fold), Rho guanine nucleotide exchange factor 1 (3.9-fold), Rho GTPase-activating protein 6 (3.8-fold), and Rho/Rac guanine nucleotide exchange factor 18 (2.0-fold). Some genes in this group were down-regulated: Rho GTPase-activating protein (−7.3-fold), rhotekin (−5.5-fold), Rho guanine nucleotide exchange factor 12 (−4.0-fold), cdc42 effector protein 4 (−3.5-fold), cdc42 effector protein 5 (−3.0-fold), and Rho GTPase-activating protein 12 (−2.6-fold). Expression of genes encoding intercellular adhesion molecules was significantly down-regulated in metastases: cadherin 1 (−53.5-fold), desmocollin 2 (−21.3-fold), and cell adhesion molecule with homology to L1CAM (−6.2-fold). Growth factors and receptors were notably down-regulated in metastases from p53+/− mice: epiregulin (−37.7-fold), FGFR3 (−11.0-fold), Wnt10a (−10.3-fold), Wnt9a (−10.2-fold), EGFR (−6.5-fold), Jagged 1 (−3.6-fold), and the met proto-oncogene (−3.4-fold). Expression of a number of cell cycle regulatory genes also was down-regulated: cyclin-dependent kinase inhibitor p21 (−5.3-fold), cyclin-dependent kinase 6 (−4.5-fold), cyclin-dependent kinase 4 (−2.2-fold). However, cyclin D2 was up-regulated by 2.9-fold in metastases from p53+/− mice, which contrasts with decreased cyclin D1 protein expression observed by immunohistochemistry. (25, 26) Expression of gene products involved in signal transduction pathways was up-regulated in metastases from p53+/− mice: Janus kinase 3 (6.8-fold), STAT1 (6.2-fold), c-src tyrosine kinase (4.2-fold), STAT5B (3.2-fold), mitogen-activated protein kinase–interacting serine/threonine kinase 1 (3.2-fold), and Janus kinase 2 (2.0-fold). Interestingly, the expression of genes regulating apoptosis was up-regulated: BH3-interacting domain death agonist (2.7-fold), caspase 8 (2.5-fold), and caspase 9 (2.4-fold). These data revealed distinct differences between metastases arising in p53 wild-type and heterozygous null mice.

p53 is the most frequently inactivated tumor suppressor gene in HNSCC (13). Previous studies have shown concomitant p53 protein accumulation and genomic instability during head and neck tumorigenesis (27). p53 has been evaluated as a predictive marker for lymph node metastasis and response to ionizing radiation and chemotherapy (28-30). Reintroduction of wild-type p53 into HNSCC lines and xenografts results in growth inhibition and sensitization to chemotherapy drugs (31, 32). Patients with Li-Fraumeni syndrome have decreased p53 function and higher risk of multiple cancers, including those of the head and neck (33). However, in human studies, it is difficult to prospectively evaluate the contribution of loss of p53 function to head and neck tumorigenesis in the microenvironment in which these tumors most frequently arise (i.e., chronic carcinogen exposure). We report a new mouse model of HNSCC that produced metastatic squamous cell carcinomas in 100% of animals. The lack of one p53 allele correlated with a statistically significant earlier onset of primary tumors. Microarray analysis showed that expression of TGFβ2 was down-regulated in HNSCCs arising from p53+/− mice (Table 1). TGFβ2 is an obvious candidate gene for mediating shortened tumor latency in these animals given the growth-inhibitory effects of this family of ligands on stratified squamous epithelium (for a review, see ref. 34). Decreased expression of TGFβ ligands might also contribute indirectly to carcinogenesis by promoting inflammation in the local microenvironment (for a review, see ref. 35). Paradoxically, TGFβ expression may promote epithelial to mesenchymal transdifferentiation in later stage epithelial cancers (36, 37). In this regard, none of the anaplastic spindle cell tumors in our mouse model arose from the p53+/− genotype in which TGFβ2 expression was down-regulated. Conversely, overexpression of growth factor receptors such as EGFR has been linked to anaplastic tumor progression in some types of cancer (38, 39). In line with these findings, we did not observe the progression to anaplastic tumors in cervical lymph node metastases which correlated with reduced expression of growth factor receptors including EGFR, c-met, and FGFR3 (Fig. 4; Tables 3 and 4). These results suggest important roles for growth factors in regulating the onset, progression, and metastasis of HNSCC in the mouse model.

p53 expression correlated with that of the growth factor receptor c-met in primary HNSCC, and with levels of EGFR, cyclin D, and PCNA in metastases (Fig. 4). It is likely that mutant rather than wild-type p53 protein was expressed by these cells following carcinogen exposure, and these data suggest that p53 is an important regulator of growth factor expression and cellular proliferation. p53 protein expression was not detected in 89% of metastases, in some cases, due to allelic loss. Loss of p53 expression correlated with the reduced growth rate of metastases and the decreased expression of proliferation markers. This decreased proliferation might be due to intrinsic differences in cancer cells expressing mutant p53 versus those null for the tumor suppressor, or the result of changes in the microenvironments of the primary and metastatic tumors. In agreement with these findings, loss of p53 expression was shown to have tumor-inhibitory effects in transgenic mice expressing viral oncogenes in the skin (40). The EGFR promoter contains a putative binding site for p53 (41), and tumor-derived p53 mutants have been shown to induce promoter activity (42). We recently showed that overexpression of wild-type p53 in the mutant background of human HNSCC lines reduced EGFR protein expression (31). However, loss of p53 protein due to degradation induced by human papillomavirus type 16 E6 protein had mixed effects on EGFR expression in normal human keratinocytes (43). These results suggest that regulation of EGFR expression by p53 likely depends on the type of mutation, the contribution of wild-type p53, loss of heterozygosity, and differences in experimental models. The gene expression data also revealed other known p53 targets that were regulated in mouse HNSCC. p21, which is induced by p53 activation (44), is down-regulated 5.3-fold in metastases which do not express detectable p53 protein. Similarly, a second p53 target gene, the Werner syndrome homologue, which is repressed by p53 activation (45), was up-regulated by 2.8-fold in the metastases from p53 heterozygous null mice. These gene expression results clearly indicate a metastatic and invasive phenotype that is regulated in part by loss of p53 expression in head and neck cancer cells.

Using the well characterized mouse skin carcinogenesis protocol, Kemp et al. showed that epidermal tumors in p53 null animals underwent more rapid progression to poorly differentiated carcinomas (46). However, this DMBA/TPA protocol was completely ineffective in initiating HNSCC in our mouse model, resulting only in benign papilloma formation in ∼6% of treated animals. These differences may be due to the genetic background of the animals or to intrinsic differences between the epidermis and mucosa. Additionally, we observed poorly differentiated carcinomas only in wild-type mice, which raises the above question of whether growth factor receptor activation is required to drive tumor progression. DMBA is less abundant in tobacco smoke than other polycyclic aromatic hydrocarbons, but produced tumors with similar histopathologic and molecular characteristics of the human disease in our mouse model. Additional molecular characterization of the distinct histopathologic types of HNSCC in this model should provide important information about the biology of tumor progression.

In keeping with our previously published findings using human HNSCC cases (21), overexpression of growth factors and their receptors correlated with high levels of downstream cell cycle regulatory proteins which regulate G₁ to S phase progression. In particular, the expression of TGFα and its receptor EGFR independently correlated with cyclin E protein levels by immunohistochemistry (Figs. 5 and 6). High EGFR expression also correlated with increased cyclin A and cyclin D protein levels (Fig. 5). HGF treatment was reported to increase cyclin B expression in the liver (47), but this correlation had not been observed in human cancer before this report. HGF overexpression correlated with high levels of cyclin B protein in primary HNSCC in the mouse model (Fig. 6) and we also linked low-level HGF expression to decreased cyclin B levels in metastases (P < 0.07).⁵

Gene expression profiling of primary and metastatic tumor cells revealed important new information about the signaling pathways that regulate the invasive phenotype in HNSCC. In primary tumors, expression of glutathione S-transferase which is involved in carcinogen detoxification and is a predictive marker for HNSCC (48, 49) was up-regulated 3.5-fold. The phosphatidylinositol-3-kinase catalytic subunit, an important survival factor for HNSCC (24), was up-regulated 2.1-fold in primary tumors arising from p53+/− animals. In addition to the striking down-regulation of genes controlling differentiation, intercellular adhesion, and proteinase activity, the gene expression profile of HNSCC metastases in this model was notable for the large number of differentially regulated genes involved in cytoskeletal reorganization, cell migration, and invasion. Many gene products which regulate actin cytoskeletal dynamics by activating or inhibiting Rac and Rho GTPases were differentially expressed. Previous reports have shown that Rho family members are overexpressed in gastric carcinoma (50), and expression of Rac2 and its effector Vav2 was increased in the mouse HNSCC model. The non–receptor tyrosine kinase c-src (up-regulated in the present model) is overexpressed in human HNSCC and regulates both proliferation of and invasion by these cells (51, 52). The Stat signaling pathway was also dysregulated in human HNSCC (53, 54), and the expressions of STAT1, STAT4, STAT5A, and STAT5B were up-regulated in metastases from our mouse model (Tables 3 and 4). Notch signaling was recently shown to suppress invasive growth (55); in this regard, Jagged 1 (a Notch ligand) was significantly down-regulated in metastatic HNSCC cells. Up-regulation of serine proteinases (serine protease 2, 3.1-fold) and down-regulation of their inhibitors (serine or cysteine proteinase inhibitor clade B member 5, −177.6-fold) have been associated with cancer invasion and metastasis (56). The Fyn proto-oncogene, which was up-regulated in metastases from the mouse model, has been associated with extracellular membrane signaling via integrins in head and neck cancer cell lines (57). These data reveal a highly metastatic gene expression profile for HNSCC in the mouse model.

Down-regulation of several members of the Wnt gene family was also apparent in metastatic HNSCC cells. Wnt signaling has been shown to regulate the proliferation of many cancer cell types (58, 59), but its role in human HNSCC is largely uncharacterized. It will be interesting to determine if the down-regulation of Wnt expression in the mouse model of HNSCC contributes to tumor spread or is reflected in the reduced growth rate of metastatic cells in cervical lymph nodes. Members of the steroid receptor superfamily were also down-regulated, including the vitamin D receptor, which promotes keratinocyte differentiation (60) and retinoic acid receptor γ, which confirmed our earlier results using human HNSCC lines (61). The proapoptotic genes bid, caspase 8, and caspase 9 were also up-regulated in metastases, although whether these cells are more sensitive to apoptotic stimuli remains to be determined.

In summary, we report a new p53 knockout-induction model of metastatic HNSCC. The early onset of metastasis in this model sheds important new light on the clonal selection theory of tumor progression. These new results suggest that metastatic cells arise early during tumorigenesis and do not need to evolve in an advanced primary tumor. In future experiments, the use of this model should provide important information on initiation, progression, metastasis, and experimental therapy of head and neck cancer.

This study was approved by the institutional animal care and use committee before any experiments were done. The p53 heterozygous mouse strain B6;129S2-Trp53^tm1Tyj was purchased from The Jackson Laboratory (Bar Harbor, ME) and bred to create p53 wild-type, heterozygous, and homozygous null mutant animals. Mice were housed in approved environmentally controlled facilities on 12 h light-dark cycles and unlimited access to food and water. Genotyping was done according to The Jackson Laboratory protocol using extracted tail DNA as the PCR template. Male and female mice from all genotypes were dosed orally twice weekly using one of three induction protocols: 25 μg of DMBA dissolved in 20 μL of ethanol, 1.2 μg of TPA dissolved in 20 μL of ethanol following an initiating dose of DMBA, or TPA alone. The time course and number of tumors were recorded for each animal. Mice were euthanized when any institutional criterion for experimental neoplasia in rodents was met. Euthanized mice were photographed and complete necropsies were done. A portion of each tumor specimen was flash-frozen in liquid nitrogen or fixed in 4% buffered formaldehyde for 16 h at room temperature.

Tumor tissue was dehydrated in an ethanol series, cleared in xylene, and embedded in paraffin. Five-micrometer sections were prepared and mounted on poly-l-lysine–coated slides. Representative sections were stained with H&E and histologically evaluated by a pathologist. Immunohistochemical analysis was done using a commercially available kit (Invitrogen, Carlsbad, CA). Sections were incubated at 60°C for 30 min and deparaffinized in xylene. Endogenous peroxidase activity was quenched by incubation in a 9:1 methanol/30% hydrogen peroxide solution for 10 min at room temperature. Sections were rehydrated in PBS (pH 7.4) for 10 min at room temperature. Sections were blocked with 10% normal serum for 10 min at room temperature followed by incubation with anti-p53, met, EGFR, cyclin A, cyclin B, cyclin D, cyclin E, TGFα, HGF, and PCNA antibodies (Santa Cruz Biotechnology, Santa Cruz, CA) for 16 h at room temperature. After washing thrice in PBS, the sections were incubated with secondary antibody conjugated to biotin for 10 min at room temperature. After additional washing in PBS, the sections were incubated with streptavidin-conjugated horseradish peroxidase enzyme for 10 min at room temperature. Following final washes in PBS, antigen-antibody complexes were detected by incubation with a hydrogen peroxide substrate solution containing aminoethylcarbazole chromogen reagent. Slides were rinsed in distilled water, coverslipped using aqueous mounting medium, and allowed to dry at room temperature. The relative intensities of the completed immunohistochemical reactions were evaluated using light microscopy by independent trained observers who were unaware of the mouse genotypes. A scale of 0 to 3 was used to score relative intensity, with 0 corresponding to no detectable immunoreactivity and 1, 2, and 3 equivalent to low, moderate, and high staining, respectively. Nonparametric data was analyzed by Fisher exact test.

Genomic DNA was extracted from microdissected metastatic tumor cells using proteinase K digestion and phenol-chloroform extraction. Following ethanol precipitation, washing, and dissolving the DNA pellet in 10 mmol/L of Tris-HCl, 1 mmol/L of EDTA (pH 8), PCR reactions were done using 40 ng of genomic DNA template in 20 mmol/L of Tris-HCl (pH 8.3), 1.5 mmol/L of MgCl₂, 63 mmol/L of KCl, 0.05% Tween 20, 1 mmol/L of EGTA, 50 μmol/L of each deoxynucleotide triphosphate, and 2.5 units Taq DNA polymerase (Roche Molecular Biochemicals, Indianapolis, IN) using p53 exon 7 primers (5′-CCCGAGTATCTGGAAGACAG-3′ and 5′-ATAGGTCGGCGGTTCAT-3′). The cycling variables were 1 cycle of 94°C for 3 min, 12 cycles of 94°C for 35 s, 64°C for 45 s (decreasing 0.5°C per cycle), 72°C for 45 s, 25 cycles of 94°C for 35 s, 58°C for 30 s, 72°C for 45 s, and 1 cycle of 72°C for 2 min. Mouse normal genomic DNA was used as the positive control. Reactions with no template, no primers, or treated with DNase I were used as negative controls. PCR products were separated by agarose gel electrophoresis followed by ethidium bromide staining.

Total RNA was extracted from microdissected primary and metastatic tumor tissue from p53 wild-type and heterozygous mice using a commercially available kit (RNeasy, Qiagen, Valencia, CA). Individually matched well differentiated primary and metastatic tumor tissue was used in microarray analysis. Three independent samples from each group were used in this gene expression analysis. The integrity of rRNA bands was confirmed by Northern gel electrophoresis. Total RNA (10 μg), with a spike in controls, was first reverse-transcribed using a T7-Oligo(dT) promoter primer in the first-strand cDNA synthesis reaction. Following RNase H–mediated second-strand cDNA synthesis, the double-stranded cDNA was purified and served as a template in the subsequent in vitro transcription reaction. The in vitro transcription reaction was carried out in the presence of T7 RNA polymerase and a biotinylated nucleotide analogue/ribonucleotide mix for complementary RNA amplification and biotin labeling. The biotinylated complementary RNA targets were then purified, fragmented, and hybridized to Affymetrix GeneChip expression arrays (Santa Clara, CA). The murine genome 430 2.0 microarray was used to investigate 39,000 possible transcripts in each sample (Genomics Core Facility, Children's Hospital, Los Angeles, CA). After washing, hybridization signals were detected using streptavidin-conjugated phycoerythrin. Affymetrix GCOS software was used to generate raw gene expression scores and normalized to the relative hybridization signal from each experiment. All gene expression scores were set to a minimum value of 2 times the background determined by GCOS software in order to minimize noise associated with less robust measurements of rare transcripts. Normalized gene expression data was imported into dChip software⁶

We thank Dr. Timothy Triche, Betty Schaub, and Sitara Waidyaratne (Children's Hospital) for assistance with microarray analysis.