Most head and neck squamous cell carcinoma (HNSCC) patients present with late-stage cancers, which are difficult to treat. Therefore, early diagnosis of high-risk premalignant lesions and incipient cancers is important. HNSCC is currently perceived as a single progression mechanism, resulting in immortal invasive cancers. However, we have found that ∼40% of primary oral SCCs are mortal in culture, and these have a better prognosis. About 60% of oral premalignancies (dysplasias) are also mortal. The mortal and immortal tumors are generated in vivo as judged by p53 mutations and loss of p16INK4A expression being found only in the original tumors from which the immortal cultures were derived. To investigate the relationships of dysplasias to SCCs, we did microarray analysis of primary cultures of 4 normal oral mucosa biopsies, 19 dysplasias, and 16 SCCs. Spectral clustering using the singular value decomposition and other bioinformatic techniques showed that development of mortal and immortal SCCs involves distinct transcriptional changes. Both SCC classes share most of the transcriptional changes found in their respective dysplasias but have additional changes. Moreover, high-risk dysplasias that subsequently progress to SCCs more closely resemble SCCs than nonprogressing dysplasias. This indicates for the first time that there are divergent mortal and immortal pathways for oral SCC development via intermediate dysplasias. We believe that this new information may lead to new ways of classifying HNSCC in relation to prognosis. (Cancer Res 2006; 66(15): 7405-13)
- head and neck cancer
- cancer progression
Head and neck squamous cell carcinoma (HNSCC) patients often develop a series of premalignancies and SCCs over a number of years, many of which are genetically related, being derived by separate mutations within the same abnormal mucosal “field” altered by exposure to carcinogens or growth promoters in tobacco or alcohol ( 1, 2). Loss of heterozygosity (LOH) studies show that oral cancers develop from a field of altered mucosa that is polyclonal ( 1). However, most oral carcinomas are clonal ( 3), suggesting that carcinomas develop from earlier lesions by a succession of cumulative genetic changes ( 4). Second field or second primary cancers may subsequently develop, distinguishable from recurrences by their patterns of allele loss ( 1).
Although most HNSCC patients present without prior diagnosis of premalignancy, two main types of premalignancy are well documented: leukoplakias (white patches) and erythroplakias (red patches). Erythroplakias are much more likely to show histologic features of dysplasia and progress to SCC ( 5). However, a major clinical problem is that there is no clear correlation between histologic grade of dysplasia and prognosis in individual cases. Various genetic and gene expression changes are found in oral dysplasias but single molecular markers do not have sufficient predictive power to identify high-risk lesions ( 6). The only previous gene expression profiling study of oral dysplasias did not address the question of progression ( 7).
We have previously shown heterogeneity in HNSCCs and dysplasias with respect to their mortality/immortality status in culture, with ∼40% of HNSCCs ( 8, 9) and 60% of dysplasias being mortal ( 10, 11). This raises the question about whether there are separate routes for mortal and immortal HNSCC development via preceding mortal and immortal dysplasias. To test this idea, we have examined the gene expression relationships of mortal and immortal HNSCCs in relation to the various premalignant types, including high-risk dysplastic lesions that subsequently progressed to SCC.
Materials and Methods
Clinical samples. Ethical approval (with informed consent) was granted by the Glasgow Dental Hospital Area Ethics Committee (10MAR97/AGN4vi) and the Edinburgh Dental Hospital Area Ethics Committee (before 1995).
Cells and culture conditions. Details have previously been reported ( 8– 11). Cells were freshly thawed from early passage stocks in DMEM medium containing 10% FCS, 0.4 μg/mL hydrocortisone, 10 ng/mL cholera toxin, 5 μg/mL insulin, 8 μg/mL transferrin, 30 μg/mL adenine, and 10 ng/mL epidermal growth factor. Cells were used at subconfluent densities after removal of feeders immediately before analysis.
Microarray experiments and analysis. RNA was obtained with the RNeasy Total RNA Isolation Kit (Qiagen, Gatwick, Sussex, United Kingdom). Double-stranded cDNA made using the Superscript Double Stranded cDNA Synthesis Kit (Invitrogen, Inchinnan, Renfrewshire, United Kingdom) was labeled using the Enzo Bioarray High Yield Transcript Labelling Kit (Affymetrix, Abbington, Oxon, United Kingdom). Fragmented cRNAs were then hybridized to Affymetrix U133A&B chips by the Cancer Research UK Microarray Facility at the Paterson Institute for Cancer Research in Manchester, United Kingdom. Data were normalized using the RMAexpress method ( 12). Genes were analyzed further if classified as expressed in >5 of 80 samples by the Affymetrix MAS 5.0 software. Significant differences in gene expression were determined using the statistical analysis of microarrays software (SAM; TIGR open access software TMEV), or by ANOVA, carried out in R-package. 5 Rank Products ( 13) and Iterative Group Analysis were also used to identify differentially expressed gene classes ( 14) by the Sir Henry Wellcome Functional Genomics Facility (http://www.gla.ac.uk/functionalgenomics/rp/affy_analysis.html). Spectral clustering using the singular decomposition value was done as previously described ( 15, 16). If the expression data for M genes and N samples are arranged in an M × N matrix, with the element on the ith row and jth column representing gene i in sample j, then the singular decomposition value expresses this matrix as the product of three matrices: an M × M orthogonal matrix, U; an M × N diagonal matrix, S (with nonnegative elements on the diagonal ordered from high to low); and the transpose of an N × N orthogonal matrix, V. The columns of U can be used to assign numerical values to the genes and the columns of V can be used to assign numerical values to the samples. The first columns, U1 and V1, represent the overall background conditions and are not useful for analysis. The second columns, U2 and V2, may be viewed as partitioning the genes and samples, respectively, into clusters. If components r and s of V2 are relatively close, then this means that sample r and sample s are close. Similarly, V3 is another (orthogonal) assignment of real values to samples that can be used to cluster. Hence, samples that are close in this two-dimensional ordering can be viewed as similar.
Quantitative reverse transcription-PCR. The Finnzymes DyNAmo SYBR Green 2-Step qRT-PCR Kit (F-430 S/L) was used with the MJ Research Chromo4 Detection System. cDNA was prepared using 1 μg mRNA, random hexamer primers, and M-MuLV RNase H+ reverse transcriptase: primer extension, 10 minutes 25°C; cDNA synthesis, 30 minutes 37°C; termination, 5 minutes 85°C. One microliter of product was subjected to 35 cycles of PCR using Thermus brockianus DNA polymerase and SYBR Green: initial denaturation, 15 minutes 95°C; denaturation, 10 seconds 94°C; annealing, 30 seconds 58°C; extension, 30 seconds 72°C (in some cases followed by heating to 75°C for 12 seconds each cycle before plate reading). For characterization of the PCR product, a final extension at 72°C was done for 10 minutes followed by a melting curve over 70°C to 95°C (1 second, 0.3°C steps). The size of the reverse transcription-PCR (RT-PCR) product was determined after reannealing for 10 minutes at 72°C. Triplicate samples, including minus-reverse transcriptase and minus-template controls, were analyzed using the MJ Research Opticon 3.1 software. RNA levels were calculated relative to glyceraldehyde-3-phosphate dehydrogenase (GAPDH), β-actin, and RPLP0 RNAs using the Relative Expression Software Tool (REST; ref. 17). Primers spanning intron-exon boundaries were designed using Perlprimer software with the following constraints: GC content of 40% to 60%; exclusion of primers containing known genomic repeat sequences or runs of four identical nucleotides; low probability of primer-dimer formation; screening for primer specificity using the National Center for Biotechnology Information Blast server. The primers used were GAPDH F 5′-AAGGCTGAGAACGGGAAGCTTGTC-3′, R 5′-AGCCCCAGCCTTCTCCATGGTGGT-3′; β-actin F 5′-AAGATCAAGATCATTGCTCCTCCT-3′, R 5′-TCATAGTCCGCCTAGAAGCA-3′; CES2 F 5′-TTAACAAGCCTTTCAAGATGATCCC-3′, R 5′-CCAAATGTAGGAGGCAACATCAG-3′; HES2 F 5′-CTCATTTCGGACCTCGGTT-3′, R 5′-TTCGAGCAGTTGGAGTTCT-3′; ITGAV F 5′-TAACCAATTAGCAACTCGGAC-3′, R 5′-ACACATCTGCATAATCATCTCC-3′; ITGB6 F 5′-CGGCTTTCCAAAGAGATGTC-3′, R 5′-GAAGTATGGAATACTACTGCAAGG-3′; KLK10 F 5′-AGCTGATCCAGATGTTATGCTC-3′, R 5′-AAACCTTTGCCACCACTTCC-3′; MMP9 F 5′-TCACTTTCCTGGGTAAGGAG-3′, R 5′-GAACAAACTGTATCCTTGGTC-3′; DTL F 5′-TGGCCTAGTAACAGTAACGA-3′, R 5′-AGTATTTGAAGGAAGAGGGAG-3′; RPLP0 F 5′-GAAGGCTGTGGTGCTGATGG-3′; R 5′-CCGGATATGAGGCAGCAGTT-3′;SAT F 5′-GATTATAGAGGCTTTGGCATAGG-3′, R 5′-CTTGTCGATCTTGAACAGTCTC-3′.TGFB2 F 5′-ACTTTCTACAGACCCTACTTCAG-3′, R 5′-AGGTTCCTGTCTTTATGGTG-3′; and ZIC2 F 5′-GTTCCAGTGTGAGTTTGAGG-3′, R 5′-GACTCATGGACCTTCATGTG-3′.
Antibodies. Western blotting was done as described ( 11, 18). Immunohistochemistry was done by standard procedures. Primary antibodies used were BIRC5 (Abcam, Cambridge, United Kingdom), CCNB1 (NeoMarkers, Lab Vision, Newmarket, Middlesex, United Kingdom), CDC2 (Santa Cruz Biotechnology, Insight Biotechnology, Wembley, United Kingdom), ECT2 (Santa Cruz Biotechnology), HMMR (Dr. Volker Assmann, University Hospital, Hamburg, Germany), IGFBP2 (Santa Cruz Biotechnology), IVL (Sigma, Poole, Dorset, United Kingdom), p16INK4A (Santa Cruz Biotechnology), LAP2/TMPO (Abcam), p38 (Cell Signaling, New England Biolabs, Hitchin, Hertfordshire, United Kingdom), p53 (Santa Cruz Biotechnology), S100A9 (Santa Cruz Biotechnology), S100P (BD Transduction Laboratories, Becton and Dixon, Cowley, Oxford, United Kingdom), SMC4L1 (Upstate, MSC/ALPLAB, Newton Abbey, County Antrim, Northern Ireland), STK6 (Abcam), and UBE2C (Abcam). Sodium citrate was used for antigen retrieval; Vectastain ABC kit (p53) or Immpress (p16INK4A; Vector Laboratories, Burlingame, CA) kit was used for staining.
Clinical characteristics of tumors. Virtually all the biopsies were from oral cavity sites (Supplementary Table S1). Erythroplakias E2, E4, and E5 either had adjacent malignancy present at diagnosis or developed malignancy at the same site subsequently ( 19); the outcome of erythroplakia E1 is unavailable. Two “mixed” erythroleukoplakias (D19 and D35) also developed SCCs at the same sites whereas only 1 of 13 of the leukoplakias progressed (D20). None of the 15 SCC cultures or the 16 cultures from dysplastic lesions (“dysplasia cultures”) contained human papillomavirus (HPV)-16 or HPV18 E6/E7 DNA (by PCR techniques sensitive down to ∼0.5 copies per cell) ( 8, 19). 6 This probably reflects the fact that only one biopsy was from pharynx, which is most commonly associated with HPV infection ( 20).
SCC and dysplasia cultures. All cultures were isolated and maintained using the 3T3 feeder layer method, which supports growth of epithelial cells at all stages of cancer progression and is believed to maintain in vivo characteristics of tumors ( 21). Overall, 9 of 16 primary SCCs and 7 of 19 dysplasias were immortal in culture ( 9– 11). All three cultures of SCC recurrences were immortal, as were two of three cultures from metastases. The single mortal metastasis failed to generate an immortal culture despite multiple attempts ( 9).
Immortal SCCs and dysplasias are generated in vivo. The p53 mutations found in our immortal SCC cell lines were detected in the original tumors in vivo ( 8); this is also the case for all five immortal dysplasia biopsies tested using laser capture microdissected material (data not shown). Immortal dysplasias show expression of p53 protein in the suprabasal layers in vivo, whereas in mortal dysplasias p53 expression is restricted to the basal layer ( Fig. 1 ). Seven of eight of our mortal dysplasias express p16INK4A whereas six of seven immortal dysplasia cultures do not ( 10, 11). We have now shown that these differences also occur in vivo in the biopsies from which the cultures were derived ( Fig. 1). Two dysplasia cultures with atypical p16INK4A expression patterns ( 10, 11) also have the same phenotypes in vivo; the extended-life-span D17 dysplasia lacks expression of p16INK4A whereas the immortal dysplasia D38 continues to express p16INK4A ( Fig. 1B).
Gene expression profiling of HNSCCs and dysplasias. The gene expression data for 16 SCC, 19 dysplasia, and 4 normal mucosa cultures were analyzed by “spectral clustering,” which is a useful and sensitive tool for analyzing complex microarray data. For example, it correctly classifies the three subtypes of leukemia ( 15, 16) and the known lymphoma, breast, and brain cancer subtypes ( 15, 22). This method resolves the gene expression data for all the genes in each of the samples as a series of component vectors, of which V2 and V3 are the most useful in the present context (see Materials and Methods). The principle is that samples that are close in this two-dimensional ordering can be viewed as similar in overall gene expression profile.
Because the differences between duplicate cultures were minimal in comparison with differences between cultures from different biopsies, for clarity, Fig. 2 gives spectral clustering analysis using the mean expression levels of the duplicate cultures. This shows major distinctions between mortal and immortal SCCs and normal mucosa, principally by vector 2. However, within the immortal SCC group, the three recurrences and one of the two metastases cluster somewhat separately on vector 3 from the main group of primary SCCs. The SCC BICR7 culture (P7) that is in “M2/crisis” ( 23) differs substantially from all the other SCCs.
The mortal and immortal dysplasias cluster as heterogeneous groups overlapping the mortal and immortal SCC groups, respectively ( Fig. 2). The only samples that cluster between the main mortal and immortal groups are the two dysplasias, D17EL and D38, with atypical p53 status and p16INK4A expression patterns referred to previously ( Fig. 1). Within the mortal dysplasia group, three of the four erythroplakias (especially E1 and E2) cluster separately from the nonprogressing mortal dysplasias and closest to the mortal SCCs ( Fig. 2). Similarly, the three immortal dysplasias that progressed, D19, D20, and D35, cluster closest to immortal SCCs on vector 2 ( Fig. 2). Heterogeneity in tumor site locations is unlikely to significantly complicate the analysis because all samples were from oral sites, except for three from larynx (B30, B80, and M18). Moreover, spectral clustering relationships of normal mucosa, dysplasia, and SCC cultures are little affected when the three non-oral (laryngeal) samples are removed (Supplementary Fig. S1).
Furthermore, the mortal/immortal gene expression clustering patterns do not correlate with smoking status. The only normal mucosa sample from a smoker (N9) clusters within the other normal mucosa cultures from nonsmokers. Virtually all of the mortal and immortal SCCs were obtained from smokers. Of the dysplasia patients for which information is available, one of eight mortal dysplasias was from a nonsmoker (D41) and another was from a light smoker (D48) whereas four of seven immortal dysplasias were from nonsmokers (D9, D34, D38, and D20). Moreover, mortal dysplasia D41 and immortal dysplasia D38 arose concurrently in the same patient (a nonsmoker), both in the lateral tongue.
Gene expression differences between mortal and immortal SCCs. Using various statistical methods, SAM, ANOVA, and rank products, there are surprisingly few gene expression changes shared during the development of both mortal and immortal SCCs ( Fig. 3A ; Supplementary Table S2). Iterative Group Analysis shows that the Gene Ontology–based signatures of mortal and immortal SCC cultures are significantly different, with increases in mitosis and DNA replication categories in immortal SCCs, in contrast to increases in extracellular matrix and collagen catabolism categories in mortal SCCs. Gene expression differences >5-fold between mortal and immortal SCCs by SAM analysis (false discovery rate <1%) are illustrated in Fig. 3B and Supplementary Table S3. Immortal SCCs show increased expression of genes encoding several cell cycle regulators, compared with mortal SCCs (e.g., CCNA2, CCNB1, CCNB2; CDC2, CDC6, CDC20, CDCA7; CDKN3; several MCMs; PCNA; BIRC5), but reduced expression of genes encoding terminal differentiation markers [e.g., several keratins (KRT), involucrin (IVL), and the family of small proline-rich proteins (SPRR)], as well as the S100 family of calcium-binding motility proteins and various kallikreins (KLK; Supplementary Table S3). In contrast, a variety of adhesion, invasion, and extracellular matrix proteins are up-regulated in mortal SCCs [e.g., CEACAMs, fibronectin (FN1), integrins, metalloproteinases (MMP), periostin (POSTN)], as well as various chemokines (CXCLs, CCL20, and IL-8) and TGF-β (TGFB2; Supplementary Table S3). Some gene expression changes distinguish all mortal and immortal cultures (whether dysplasias or SCCs; Supplementary Table S4).
We have confirmed several of these distinctive gene expression differences between mortal and immortal SCCs or dysplasias by protein or quantitative RT-PCR studies: the increased expression of the cell cycle regulators, STK6, BIRC5, CCNB1, SMC4L1, and CDC2, in immortal compared with mortal SCCs ( Fig. 4A ); the reduced expression of S100P, S100 A9, and IVL in immortal SCCs ( Fig. 4A); and increased expression of MMP9, integrins ITGAV and ITGB6, and TGF-β RNAs ( Fig. 4B) in mortal SCCS. Up-regulation of KRT7 in most mortal SCCs and of KRT18 in immortal SCCs has been confirmed at the protein level (data not shown).
In some respects, mortal SCCs resemble wound keratinocytes (e.g., up-regulation of PLAU, PAI-1/SERPINE1, MMP10, ITGB4, CRABPII, and CDH1 genes; reviewed in ref. 24 and references therein) but differ in other respects (e.g., lack of up-regulation of S100A8 and A9, GALS7, and connexin 26/GJB2 genes; Supplementary Table S3; Fig. 4; and data not shown). We have also analyzed the extent to which the mortal-immortal SCC differences might be senescence related because mortal SCCs, like normal mucosa cultures, always contain a small proportion of terminally differentiating and senescent cells, even under optimal growth conditions soon after isolation ( 25). However, only ∼10% of the ∼200 gene expression changes found in senescent oral epithelial cells in other studies ( 26, 27) show a significant difference in expression between mortal and immortal SCCs (Supplementary Fig. S2).Four microarray studies have identified putative gene expression signatures in primary HNSCCs in vivo that correlate with subsequent lymph node metastasis ( 28– 31). Although these signatures show little overlap at the individual gene level (e.g., due to HNSCC subsite heterogeneity, differences in sample numbers or array platforms), there are more consistent changes in certain gene families (Supplementary Table S6) and these are generally characteristic of immortal SCC cultures ( Fig. 5 ).
p53 target genes. Because mortal and immortal SCCs differ in p53 status (Supplementary Table S1), we have analyzed which of the known p53 target genes ( 32) differ in expression between mortal and immortal SCCs; those that are differentially expressed are involved in cell cycle/apoptosis, signal transduction, DNA damage responses, adhesion or protein catabolism, and transcription ( Fig. 6A ; Supplementary Table S5). Most are most significantly down-regulated in immortal SCCs compared with normal mucosa or mortal SCCs (CCNG2, CDKN1A, TGFA, TRIM22, RRAD, CALD1, RPS27L, and HIG1) but MSH6 and MYBL1 are up-regulated. Thus, these changes correlate with the wild-type or mutant p53 status of the mortal and immortal SCCs, respectively (Supplementary Table S1). However, some genes (e.g., BCL2AP1, COL4A1, and VIM) are up-regulated mainly in mortal SCCs and must therefore be due to other transcription factor changes.
Two other transcription factors that consistently differ in expression between mortal and immortal SCCs in the microarray data are ZIC2 and HES2 (Supplementary Table S3), and this has been confirmed by qRT-PCR ( Fig. 4B). ZIC2 is a member of the ZIC family of Cys2-His2 zinc finger transcription factors whereas HES2 is a member of the family of bHLH transcriptional repressors. This association of ZIC2 and HES2 with immortality is novel.
Mortal and immortal SCCs share most of the transcriptional changes found in their respective dysplasias but have additional changes. SAM (false discovery rate <1%) shows that 80% of the >2-fold gene expression differences between immortal dysplasias and normal mucosa are retained in immortal SCCs; similarly, 60% of the changes in mortal dysplasias are retained in mortal SCCs. This is also illustrated in Fig. 3B where many of the largest gene expression differences found between mortal and immortal SCCs (>5-fold) are already established at the dysplasia stage (e.g., groups marked A). Several of these changes have been confirmed at the protein level or by quantitative RT-PCR: UBE2C, STK6, S100P, BIRC5, HMMR, ECT2, CCNB1, SMCL1, and CDC2 ( Fig. 4A) and Zic2, TGFB, ITGAV, and ITGB6 ( Fig. 4B).
However, other gene expression changes found in SCCs are less marked in dysplasias (e.g., group B, Fig. 3B). Examples of this type that have been confirmed at the protein or RNA levels include the down-regulation of IGFBP2, IVL, and S100A9 ( Fig. 4A) and HES2 ( Fig. 4B) in immortal SCCs; and the up-regulation of spermidine acetyltransferase (SAT) or the down-regulation of the nuclear matrix-associated protein, DTL/RAMP, in mortal SCCs ( Fig. 6). Other changes occur principally only in high-risk dysplasias and the corresponding SCCs [e.g., the down-regulation of kallikrein 10 (KLK10) and carboxylesterase 2 (CES2) in immortal SCCs and high-risk dysplasias D19, D20, and D35; Fig. 6B]. Interestingly, some of the gene expression changes characteristic of poor-prognosis HNSCCs in vivo and immortal SCC cultures are also found in the three immortal dysplasias that subsequently progressed (D19, D20, and D35; e.g., group A, Fig. 5). This strongly suggests that oral SCCs can develop via low-risk and high-risk dysplasias within distinct mortal and immortal pathways.
The existence of mortal and immortal dysplasias and SCCs in vivo. Our work indicates that ∼40% of oral SCCs can develop without acquiring immortality. Whereas our immortal SCC cell lines frequently show aneuploidy and LOH at several sites also found in SCCs in vivo, such allele loss is rare in mortal SCC cultures, although they may have abnormal modal chromosome numbers ( 9). Our data for p53 status and p16INK4A expression in cultures and biopsies of origin show that the genetic differences responsible for immortality are established in vivo before culture. This adds considerable weight to recent evidence that senescence operates in vivo as well as in vitro (reviewed in ref. 33).
Because mortal neoplastic cells show no LOH using conventional techniques ( 9, 34), it is unlikely that they would have been detected in the in vivo allele loss studies previously published. Therefore, it remains to be clarified whether mortal and immortal SCCs coexist within the same tumor in vivo. Our previous evidence shows that neoplastic mortal and immortal cells are genetically very distinct ( 9– 11) with only limited evidence of late conversion from the mortal to the immortal state ( 23). For example, we were never able to generate an immortal culture from one “mortal” SCC (M37) despite multiple attempts ( 9) and cultures of two regions of an “immortal” SCC (P56) were both immortal. Moreover, mortal dysplasia D41 and immortal dysplasia D38 both arose concurrently in the lateral tongue of same patient. This is all consistent with our current data supporting the idea that mortal and immortal tumors progress along independent pathways.
Prognosis of mortal and immortal SCCs. We have 10- to 12-year follow-up data for 18 of our SCC patients. Six of 12 patients with immortal SCCs died during this period from recurrence of cancer and two from other causes. In contrast, none of the six patients with mortal SCCs died from cancer during the same period but two died from other causes; this survival difference is likely to be significant (P = 0.054, Fisher's exact test). Furthermore, we have shown that the in vivo gene expression profiles of poor-prognosis HNSCCs that subsequently metastasize tend to resemble those of immortal, rather than mortal, SCC cultures ( Fig. 5). Esophageal SCCs that are immortal in culture also have poorer survival ( 35). Many HNSCC studies have reported an association between p53 mutations and poor prognosis (reviewed in ref. 36). Thus, mortal SCCs that develop without p53 mutations may have a better prognosis because their proliferative life span has been almost “exhausted” during the course of their development into clinically detectable tumors, so that they cannot reestablish after surgery/treatment unless they become immortal. However, the possibility that mortal SCCs possess a small stem cell population that is immortal in vivo cannot be ruled out.
Chemokine differences between mortal and immortal SCCs. Mortal SCCs and some mortal dysplasias overexpress TGF-β RNA, in contrast to immortal SCCs and dysplasias ( Fig. 4B). TGF-β inhibits growth of normal epithelial cells whereas it stimulates invasion of SCCs ( 37). The majority of immortal HNSCC lines are sensitive to growth inhibition by TGF-β1 ( 38) but resistant lines behave more aggressively when transplanted orthotopically into athymic mice ( 39). Mortal SCC cultures also express a variety of chemokines in culture, such as IL-8, CXCL5, CXCL6, CXCL7, and CCL20, in contrast to immortal SCCs. These CXCL family members are involved in leukocyte infiltration, metastasis, and neovascularization of tumors ( 40). Peritumoral lymphocytic infiltration is associated with a better prognosis in other HNSCC studies ( 41).
Transcriptional regulation of the immortal phenotype. We have shown previously that, with very few exceptions, loss of expression of RAR-β and p53 gene mutations is consistently associated with dysplasia/SCC immortality in culture ( 9– 11, 18). p53 mediates the senescence of cells in response to DNA strand breaks created by dysfunctional telomeres and mitotic stress ( 42). We have now identified differences in the expression of a subgroup of p53 targets between mortal and immortal SCCs which correlate with p53 mutation status ( Fig. 6A). However, a few p53 target genes are more highly expressed in mortal SCCs than in normal mucosa; because mortal SCCs have normal levels of p53, we presume that this is due to differences in other transcription factors.
Many of the cell cycle regulators that are more highly expressed in immortal dysplasias and SCCs ( Fig. 4; Tables S2 and S3) are known E2F target genes (e.g., CDC2, CDC6, CKS1, CDC25B, CCNA, TTK, TK, BUB1, PLK, PCNA, SMC4L1, TOP2A, BIRC5/survivin; ref. 43). Their overexpression in immortal cells may therefore be due to release of inhibition of RB-mediated-E2F family members by loss or inactivation of p16INK4A in immortal dysplasias and SCCs ( 10, 11, 44).
We have now identified other transcription factors that are associated with immortality, such as down-regulation of HES2 mRNA and up-regulation of ZIC2 RNA ( Fig. 4B). HES family members are well established targets of Notch signaling involved in intestinal epithelium homeostasis in humans. HES family members per se have not yet been implicated in keratinocyte biology, but Notch 1 has been implicated as a tumor suppressor in mouse skin ( 45) and keratinocyte-specific deletion of Notch1 in mice results in epidermal hyperplasia ( 46). Thus, it may be significant that Notch3 is down-regulated in immortal SCCs in our microarray experiments (data not shown). ZIC family members are implicated in control of development but ZIC2 also seems to have a role in tumorigenesis because it has been identified as a tumor antigen in small-cell lung cancer ( 47) and esophageal SCC ( 48). We are currently investigating whether manipulating the expression of HES2 and ZIC2 in mortal or immortal dysplasias and SCCs affects their life span.
Relationships of SCCs to dysplasias: evidence of divergent routes to oral malignancy. Our work shows that mortal and immortal SCCs share most of the transcriptional changes found in mortal and immortal dysplasias but have additional changes. Moreover, high-risk dysplasias that subsequently progress to SCCs more closely resemble SCCs than nonprogressing dysplasias. The fact that we can identify these gene expression relationships in unselected dysplasias and SCCs adds to their significance. Thus, mortal and immortal SCCs seem to follow divergent routes of progression, one of which is linked to multiple sites of LOH and the other not ( 9). This suggests that not all oral SCCs develop as a sequential accumulation of genetic changes as is commonly perceived ( 4). Although we have noticed rare examples of dysplasias (D17 and D38) and SCCs (BICR7) that are apparently intermediate between the mortal and immortal state ( 10, 11, 18, 23), most dysplasias and carcinomas seem to belong to either the mortal or immortal class at diagnosis. The subclassification of oral SCCs described here may have clinical relevance in the future because the different tumor classes may need different therapeutic approaches.
Our spectral clustering analysis seems to be able to distinguish the high-risk dysplasias within the subsets of mortal and immortal dysplasias, at least in primary culture. We have been able to identify some gene expression changes that seem to be characteristic of the high-risk mortal or immortal dysplasias available in our study ( Figs. 5 and 6B). KLK10 is a member of the kallikrein family of trypsin-like serine proteases that is down-regulated in a subset of breast, prostate, and testicular cancers, although up-regulated in ovarian, colon, and pancreatic cancers (reviewed in ref. 49). CES2 is commonly expressed in tumor tissue and is involved in the activation of irinotecan, a chemotherapy agent ( 50). These studies make a strong case for further larger-scale in vivo studies to identify the genetic and gene expression changes associated with dysplasia progression by the mortal and immortal pathways. The only previous in vivo gene expression profiling study found that, by hierarchical clustering or principal component analysis, six moderate to severe dysplasias clustered close to the group of seven SCCs, whereas a hyperplasia and a mild focal dysplasia clustered with the contralateral unaffected mucosa samples ( 7). However, no erythroplakias were included in this study nor was there information about whether any of the leukoplakias progressed. Lack of frozen material from the biopsies from which our dysplasia cultures were originally derived precluded an in vivo gene expression profiling in the present study, but long-term prospective studies are in progress.
Grant support: Cancer Research UK (J.K. Thurlow, J. Fleming, P.J.H. Drake, J.K. Vass, E.K. Parkinson, and P.R. Harrison); University of Glasgow (K.D. Hunter, D.G. MacDonald, and P. Herzyk); and Strathclyde University and Engineering and Physical Science Research Council grant gr/s62383/01 (G. Kalna and D.J. Higham).
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 our clinical colleagues, Dr. D. Felix (Glasgow Dental School and Hospital), D. Soutar [Plastic Surgery (Canniesburn) Unit, Glasgow Royal Infirmary], L. Clark (Victoria Hospital, Glasgow, United Kingdom), and Roy Mitchell (while at City Hospital, Edinburgh, United Kingdom), for provision of biopsies; Dr. V. Assmann (University of Hamburg, Hamburg, Germany) for the anti-HMMR antibody; Prof. Brad Ozanne for his critical interest throughout the project; and Prof. Dave Gillespie, Prof. Margaret Frame, Prof. Robert Brown, Dr. Kevin Ryan, and Dr. Karin Oien (all at the Beatson Laboratories in Glasgow) and Prof. Nalin Thakker (University of Manchester) for valuable comments on the manuscript.
Note: Figures 1-6, Supplementary Figures S1-S2 and Tables S1-S6 are available at http://www.beatson.gla.ac.uk/supplement/harrison.
K.D. Hunter and J.K. Thurlow contributed equally to this work.
Current address for E. Ken Parkinson: Centre for Clinical and Diagnostic Oral Sciences, Institute for Cell and Molecular Sciences, 4, Newark Street, London E1 2AT, United Kingdom.
D.G. MacDonald, deceased December 12, 2005.
↵5 E. Wit and J.K. Vass, unpublished data.
↵6 K.D. Hunter, J.K. Thurlow, J. Fleming, P.J.H. Drake, J.K. Vass, G. Kalna, D.J. Higham, P. Herzyk, D.G. MacDonald, E.K. Parkinson, P.R. Harrison, J. Brown, E. Wagner, unpublished data.
- Received January 18, 2006.
- Revision received March 29, 2006.
- Accepted May 31, 2006.
- ©2006 American Association for Cancer Research.