Head and neck squamous cell carcinoma (HNSCC) is the fifth to sixth most common malignancy in the world and has one of the lowest 5-year survival rates among all cancers.1 HNSCC is usually regarded as one tumor entity. This seems partially justified due to the fact that most squamous epithelia in the head and neck region, including the entire larynx, hypopharynx and oropharynx, are derived from the endoderm (only the anterior floor of mouth area and the dorsal and lateral part of the tongue are derived from the ectoderm). However, the prognosis of an individual tumor is actually strongly influenced by the tumor site. Tumors arising in the larynx, especially in the glottic region, have a favorable prognosis when compared to tumors in the oral cavity or in the hypopharynx, with the latter being associated with the worst prognosis among HNSCC.
Among the multitude of genetic alterations underlying human carcinogenesis including HNSCC, mutation in the p53 tumor suppressor gene seems to be one of the most important events.2 Most of these mutations are missense mutations within the core domain of the protein, and most but not all of these result in stabilization (overexpression) of the p53 protein. Conversely, p53 overexpression is frequently, but not always, associated with gene mutation.3 The spectrum of p53 missense mutations is very complex and is influenced by both exogenous carcinogenic agents and endogenous mutagenic mechanisms.4, 5 For instance, deamination of methylated CpG sites of p53 leads to C-to-T transition mutations in colon carcinomas, while exogenous carcinogenic agents induce G-to-T transversions at CpG sites.2, 4 The mutation spectrum in HNSCC is different from that in lung cancer, although both cancer types are strongly associated with tobacco carcinogens. It is possible that alcohol abuse and occupational exposures modulate the mutation spectrum in the different tumor sites, but detailed studies comparing the mutation spectrum between the different HNSCC sites are still missing.
There is some controversy regarding the timing of p53 mutational events in HNSCC. The analysis of histologically manifest premalignant lesions (leukoplakias) points to an increasing prevalence of mutations with increasing dysplasia, suggesting that these events are relatively late.5, 6 On the other hand, in apparently histologically normal tumor-distant biospies, p53 mutations have also been identified at high frequency,7, 8 which suggests that p53 mutations can be very early events. There is a consensus that the mutations occur prior to the conversion to the invasive state and that they are associated with the carcinogenic exposure of most HNSCC patients toward tobacco and alcohol.9
Regarding the prevalence of p53 mutations in HNSCC, the results from a large number of studies have indicated that around 50% of the tumors harbor mutations.4, 10, 11 Thus, the mutation prevalence in HNSCC seems to be very similar to that in other epithelial cancers.4 In most of these studies, exons 5–8, 5–9 or 4–9 have been analyzed, but in a limited number of studies all exons were included, and only about 2–5% of mutations were reported outside exons 5–8.4 Examination of all exons12 or reexamination of the exons 2–4 and 10–11 in those tumors in which the aforementioned exons did not reveal mutations11 has led to only a minor increase in the mutation rate. Recently, p53 sequencing by a p53GeneChip has been introduced by Affymetrix (Santa Clara, CA), which provides information on all protein coding exons 2–11 as well as intronic sequences. The GeneChip has been compared with the traditional dideoxy sequencing approach in squamous cell carcinomas of lung and bladder13, 14 and in ovarian carcinoma.15 The chip turned out to be more sensitive than dideoxy sequencing in mutation detection, although its current design does not allow reliable detection of deletions and insertions. Importantly, the increased mutation detection rate was largely due to a higher mutation rate within the conserved region of the p53 gene and by the identification of splice site mutations, whereas expansion of the mutation spectrum by including exons outside of the core region was minimal. Thus, the p53GeneChip did not change the overall picture, and importantly, a large fraction of tumors was still recognized as being p53 wild type. In 2 recent publications, a much higher prevalence (95% and 80%) of p53 mutations in HNSCC was reported.16, 17 It is notable that in both of these studies, the approach was sequencing of cDNA reverse-transcribed from p53 mRNA. However, whether this approach is really valid is not clear.
Although the literature is not entirely consistent, evidence has been reported that in some cancer types p53 mutations are associated with worsened disease outcome, e.g., in breast cancer,18 ovarian cancer15 and hematologic malignancies such as B-cell lymphoma.19 In colon and rectum cancers, p53 mutations appear to be more frequent in distal colon and in rectum, whereas some specific hot-spot mutations (e.g., at codon 245) seem to have stronger clinical impact in proximal colon and in rectal cancer.20, 21, 22 DNA contact (class I) mutations, which affect the L3 loop and the H2 helix of p53,23 and structural (class II) mutations in the L2 loop were shown to correlate with resistance to primary and adjuvant therapy and with poorer survival in breast cancers.24, 25, 26, 27 Some of these prognostic mutations, especially those affecting the L2 loop (R175H, C176P and H179L), appear to have the capacity to transactivate or transrepress genes other than those transactivated by wild-type p53 (gain-of-function phenotype28, 29). Based on these findings, it seems relevant to consider the type of p53 mutation when studying the possible prognostic value.
In HNSCC, the prognostic role of p53 mutations is unclear due to contradictory reports. Initial findings that p53 mutations were associated with locoregional treatment failure could not be confirmed later by the same group,30, 31 and the reported association of mutations with survival32 also could not be confirmed by others.10, 32 In most of the previous studies, the biochemical consequences of different mutations were not considered.29 Furthermore, most studies so far have examined mixed tumor cohorts from all head and neck sites. Since the tumor site is an established clinical factor of its own, the anatomical site might have confounded the role of p53 mutations. In laryngeal cancer, some studies have reported an inverse relationship between p53 alterations and outcome, i.e., p53 mutations was associated with decreased survival,33 whereas p53 overexpression was associated with improved larynx preservation but not better survival.34 We have previously reported that DNA contact mutations and those affecting the zinc-binding sites in L2 and L3, but not p53 mutations per se, were associated with a poorer survival.32 In that study, we have also analyzed a mixed cohort of small size. Regarding p53 overexpression, Shin et al.35 in their cohort of 69 patients with definitive local therapy (46% being stage I and II tumors) demonstrated a predictive value of p53 overexpression for shorter survival. However, other studies again failed to confirm these results or revealed contradictory results.36, 37, 38
In the present study, we have compared the genetic and protein expression status of p53 in squamous cell carcinomas from the larynx, hypopharynx, oropharynx, oral cavity and minor sites. We have also employed the p53GeneChip from Affymetrix to examine whether a higher mutation rate could be detected as compared with fluorochrome-based dideoxy sequencing.13 We analyzed whether the tumor sites differ in the prevalence and in the nature of p53 alterations and whether these alterations have a tumor site-specific effect on the clinical course of the disease. The results obtained indicate that p53 is indeed differentially targeted in the different anatomical sites. However, within each site, as in total, the p53 alterations observed had only limited prognostic value.
MATERIAL AND METHODS
Patients and biopsy material
Tumor biopsies from untreated patients were obtained at the Ear, Nose and Throat University Hospital in Heidelberg, Germany, and were divided into 2 parts. One part was snap-frozen in precooled isopentane/liquid nitrogen immediately after surgery and stored at −80°C. The second part was formalin-fixed and paraffin-embedded. Histopathologic assessment was performed on paraffin sections (by C.F.). Data on characteristics of the patients under study are accessible via the homepage of our institution (www.speculum-online.de).
Immunohistochemistry and immunohistochemistry with tyramide signal amplification
Paraffin sections of 5–6 μm thickness were deparaffinized in xylene for 2 × 5 min, rehydrated with graded ethanol, and antigen retrieval was performed by heating at 90°C for 10 min in 10 mM sodium citrate (pH 6.0). Endogenous peroxidase was blocked with 3% hydrogen peroxide in PBS for 10 min and nonspecific reactivity with 10% goat serum (Dianova, Hamburg, Germany) for 1 hr at room temperature. The sections were incubated with the primary monoclonal antibody Bp53-12 (at 1:8,000 dilution; Progen Biotechniques, Heidelberg, Germany) at 4°C overnight in the presence of 10% goat serum. As secondary antibody, biotinylated goat antimouse serum (Dianova) was added for 30 min at room temperature. For washing, PBS was used.
To enhance the sensitivity of immunohistochemistry (IHC), the tyramide signal amplification (TSA) system (NEN-Life Science) was employed following the manufacturer's instructions. Biotinylated tyramide was applied to the sections after they had been consecutively incubated with primary antibody, biotinylated secondary antibody, the horseradish streptavidin-peroxidase complex (diluted 1:100 in blocking buffer, 30 min) and washing steps in PBS in between. For staining, reagents from the ABC kit (Vectastain, Vector Laboratories, Burlingham, CA) and DAB (Diaminobenzidine, Vector Laboratories) were added. Precipitate development of the DAB substrate solution was monitored under the microscope. The sections were counterstained with methyl green or hematoxylin.
PCR cycle sequencing of p53 exons 4–9 by dideoxy chain termination method
Tumor DNA was prepared from carefully microdissected tissues using the Qiamp tissue kit from Qiagen (Hilden, Germany). PCR was performed using nested primers after initial amplification of PCR fragments containing exons 4–9. Primers were designated E4/C4, E9//C1 for exons 4–9, E4/C3, 2997 and E7-C4 and E9//C1 for exons 4–6 and exons 7–9, respectively. PCR was performed as described previously.32 Sequencing was performed using the BigDyeDeoxy Terminator Cycle Sequencing kit (Applied Biosystems, Weiterstadt, Germany) and the sequences were analyzed on the ABI-Prism 310 Genetic Analyzer. Primers used for PCR were E4/C4, 5′-GCTAGAGACCTGGTCCTCT–3′; E9//C1, 5′-CGGCATTTTGAGTGTTAGAC-3′; E4/C3, 5′-GACTGCTCTTTTCACCCATC-3′; 2997, 5′-CCACTGACAACCACCCTTAAC–3′; E7/C4, 5′-CTGGCCTCATCTTGGGCCTG-3′. Sequences were always obtained from both directions, and manual reading of all electropherograms without automatic assignment of mutations was mandatory.
Mutational analysis by p53 GeneChip from Affymetrix
Although the p53 GeneChip (Affymetrix) should tolerate contaminating normal cells in tissues, stringout microdissection was performed to allow direct comparison of the 2 sequencing approaches. The manufacturer's protocol was strictly followed; the hybridization step was performed for 90 min to obtain a more even hybridization intensity over all positions on the chip.13 Detailed description of the chip and the mixture detection algorithm used for mutation detection has been published.13, 14, 15
The chi-square test and Fisher's exact test were used for assessing independence in contingency tables. Survival was analyzed by means of standard methods, i.e., Kaplan-Meier estimates, the log-rank test and the Cox proportional hazards model with backward selection of relevant variables. If patients died for reasons not related to cancer, they were censored at the time of their death or at their last follow-up. In view of the large number of tests performed on the available data, the analysis has primarily a descriptive character; no adjustments of p-values for the multiplicity of testing were done.
Analysis of p53 protein expression in tumors of different head and neck sites by IHC
Using the monoclonal antibody Bp53-12, paraffin sections and tissue microarrays (TMAs) from a total of 514 untreated primary HNSCC were examined for p53 protein expression. A hundred twenty tumors were from the hypopharynx, 138 from the larynx, 183 from the oropharynx, 57 from the oral cavity and 16 from minor sites such as nasal sinus and nasopharynx. We used conventional IHC as well as TSA-IHC. In IHC, 198 tumors were negative; a further 42 tumors revealed up to 5% positive cells. Of the 274 p53-positive tumors, 99 received score 2 (between 5% and 20% positive cells), 139 tumors had score 3 (between 20% and 50% positive cells) and 36 tumors had score 4 (over 50% positive cells). For scoring of p53 protein levels, only the proliferating cells of the tumors were regarded, since p53 expression is turned off upon cell differentiation. Since exact determination of the histologic grade is not possible in many cases due to intratumor heterogeneity, we used parallel staining for the proliferation marker Ki67 (with the MIB antibody; not shown) as a guideline. Thus, a well-differentiated tumor with 20% of the tumor cells positive for Ki67 and also 20% positive for p53 also received score 4.
TSA-IHC on TMAs: identification of tumors with reduced or lost p53 protein expression
Due to its very short half-life, the wild-type p53 protein is undetectable in normal cells by standard IHC. The tiny amounts of this protein present in normal squamous epithelium can, however, be visualized by TSA-IHC, as shown in Figure 1. Whereas the normal uvula mucosa is p53-negative in IHC (top), at a 1:200 dilution of TSA, some parabasal proliferating cells turn positive (bottom). The fraction of positive cells further increases at a higher TSA concentration (1:100), and the cells in the connective tissue also become positive (not shown; Fig. 2). Using a smaller series of approximately 100 tumor paraffin sections with known p53 expression scores, we then established that the TSA-IHC indeed had strongly enhanced sensitivity and enabled us to discriminate between normal low expression and reduced or completely lost expression (which could result, e.g., from nonsense mutations or the inability to activate the wild-type protein, respectively). We then applied this technique on a TMA containing most tumors of this study. All tumors with scattered (up to 5%) positive cells in IHC turned positive in TSA-IHC. Of the 198 tumors negative in IHC, 108 turned positive in TSA-IHC and received score 1, but 90 tumors remained negative (score 0). The value of TSA-IHC in conjunction with TMA is illustrated by the staining patterns shown in Figure 2. In the top panel, tumors staining positive (left) and negative (right) in IHC are shown. In the bottom panel, the tumor on the left was negative in IHC and remained negative in TSA-IHC (at TSA 1:100; note strongly positive stromal cells), whereas the tumor shown at the top right turned positive in TSA-IHC (bottom right; note similar intensity of staining in some tumor cells as in stromal cells). Thus, tumors as shown in Figure 2 at the bottom left had an expression level below that of normal mucosa, and in a fraction of them lost p53 expression could be substantiated by the identification of nonsense mutations (deletions/insertions with frameshifts, or termination mutations).
Summary of IHC and TSA-IHC
By aggregating the score values obtained in IHC and TSA-IHC, we have defined 3 levels of p53 protein expression: reduced expression, defined by score 0; normal expression, defined by score 1 plus score 2; and overexpression, defined by score 3 plus score 4. In Table I, these expression levels are summarized in relation to the different tumor sites. Both reduced expression and overexpression were distinctly and significantly more prevalent in hypopharynx as compared to the other sites, whereas in the larynx, normal level expression was most prevalent (and aberrant expression lowest). Because the oropharynx, oral cavity and minor sites were intermediate between hypopharynx and larynx (Table I), they have been combined in Table I (see www.speculum-online.de for detailed information).
Table I. Definition Of p53 Expression Levels In HNSCC; Comparison Between Sites
Oropharynx + Oral Cavity + Minor Sites %
Protein levels were defined as reduced, score o; normal, scores 1 and 2; high (overexpressed), scores 3 and 4.
P-value of hypopharynx vs. the other sites combined.
P-value of larynx vs. the other sites combined. Note that the global null hypothesis of equality of proportions for different sites in Table II could be rejected (chi-square test, p < 0.0001).
Sites of highest and lowest prevalence are denoted as 3 and 4, respectively.
Sites of highest and lowest prevalence are denoted as 3 and 4, respectively.
Table II. Relationship Between p53 Expression Levels And Mutation Types
Predictive value (%) with 95% CI (%)
Two of these tumors also contained a missense mutation, and the nonsense mutation was likely only present in a subpopulation of tumor cells. Predictive value of the expression levels were calculated for the presence of
(figures denoted as 5).
Each mutation type was compared with the other mutation types combined.
Relationship between p53 protein expression levels and mutational status: overexpression but also loss of expression correlates with mutation
We then examined the relationship between the different p53 protein expression levels and the mutational status in 253 cases, as shown in Table II. As expected, the number of mutations in cases with normal level of p53 protein expression was low (25 of 116; 21.5%). In contrast, the rate of mutations increased to 75.5% in tumors with p53 overexpression. As is highlighted in Table II, lack of detectable expression even in TSA-IHC was also associated with the presence of mutations (58.3% mutations). None of these mutations were missense mutations; they comprised 18 nonsense mutations and 3 mutations in introns affecting splicing. Thus, overexpression was associated with missense mutations in 72.5% (95% CI = 62.5–81.0) of the cases; reduced expression was associated with nonsense mutations in 60.0% (95% CI = 40.6–74.5); normal expression was strongly associated with the absence of mutations in 78.5% (95% CI = 69.8–85.5). In the different tumor sites, similar relationships between p53 expression levels and the mutational status were obtained (data not shown).
Analysis of p53 mutations in tumors of different head and neck sites, with application of p53 GeneChip hybridization (Affymetrix)
Of the 253 tumors subjected to mutational analysis, 79 were derived from the larynx, 59 from the oropharynx, 30 from the oral cavity and 8 from minor sites. These tumors were subjected to PCR cycle sequencing by the dideoxy chain termination method and analyzed using the ABI-310 Genetic Analyzer. Dideoxy sequencing comprised exons 5–8 in 45 cases; in the others, exons 4–9 were analyzed. Of 77 tumors from the hypopharynx, 33 were sequenced by the dideoxy method, 44 were analyzed by the p53GeneChip. Thirty-six tumors, 24 from the hypopharynx and 12 from other sites, were analyzed by both methods. As shown in Table III, 60.6% of the hypopharngeal tumors revealed mutations by dideoxy sequencing (20/33). In contrast, only 35.4% of the tumors of the larynx (28/79), 39.0% of the tumors of the oropharynx (23/59), 36.7% of the tumors of the oral cavity (11/30) and 50% in the other sites (4/8) revealed mutations (in Table III, oropharynx, oral cavity and other sites were combined). Silent mutations (found in 21 tumors) were not included. The difference in the prevalence of mutations between the hypopharynx and the other sites, based on the results obtained by dideoxy sequencing, was significant (p = 0.01, chi-square test), whereas the distribution of the mutation classes did not differ significantly between the sites (not shown).
Table III. p53 Genotypes In Untreated Primary Carcinomas Of Different Tumor Sites
Silent mutations and polymorphisms were not registered as mutations; only one mutation/patient was recorded.
Class I, DNA contact mutations; class II, structural mutations; class III, nonsense mutations, including 4 intronic splice site mutations in hypopharynx. For simplicity, the site hypopharynx was compared with the total of the other sites by cross-tabulation and the p-value was derived from chi-square test.
It is assumed that dideoxy sequencing would have recognized mutations with confidence scores > 30 on the GeneChip.
In comparison with dideoxy sequencing, the p53GeneChip hybridization technique produced a insignificantly higher yield of mutations in hypopharyngeal tumors (65.9% compared with 60.6% by the dideoxy method, resulting in a total of 66.2% mutations by employing both sequencing methods). This difference was mainly due to the detection of 4 splice site mutations (from exon 4 to intron 4, from intron 4 to exon 5, from intron 5 to exon 6, from exon 7 to intron 7). In contrast, a single nonsense mutation in exon 10 was detected by chip analysis, changing CGA (R) to TGA (stop) at codon 342, whereas no changes were seen in exons 2–4 and in exon 9. In agreement with previous reports,13 single-base-pair deletions identified in 2 cases by dideoxy sequencing were not clearly displayed by the GeneChip. However, we have also experienced a few cases in which the sequence changes had p53GeneChip confidence scores above the threshold, which we were unable to detect by dideoxy sequencing of the same genomic DNA samples (data not shown).
Mutation spectrum in different tumor sites
On the basis of the data described so far, it was of interest to examine whether the tumor sites also differed with regard to the p53 mutation spectrum. Examination of the sequence data in the different sites revealed a distinctive mutation spectrum in the larynx. As shown in Table IV, in laryngeal tumors the mutations clustered in exon 5, specifically in the S2′ protein domain, and in codon 248, which is the hot spot most often affected by mutation. Although not significant, there were also more transversions than transition mutations in laryngeal tumors as compared to the other sites. The most frequent sequence changes observed were G:C→A:T (17.4%) and A:T→G:C (13.7%) transitions and C:G→T:A and G:C→T:A transversions (14.4% each). Deletions and insertions amounted to 14.4% as well (see www.speculum-online.de for detailed information).
Table IV. Larynx Carcinoma With A Distinct Mutation Spectrum
S2′ domain of the p53 protein is coded for by codons 132–135.23 Codon 248 was the most frequently mutated codon in this cohort.
Exon 5, n/%
S2′ domain, n/%
Codon 248, n/%
Transversion vs. transition, n/%
Comparison of p53 alterations with clinical parameters
The p53 expression levels as well as the presence and class of p53 mutation were analyzed for possible correlations to clinical parameters in Table V. Cross-tabulation of the p53 expression levels with the TNM stages revealed that both reduced (or lost) expression as well as overexpression were significantly more frequent in stage IV tumors as compared to tumor stages I–III (p = 0.001), whereas the normal expression level was underrepresented in stage IV tumors. Regarding mutations, cross-tabulation analysis showed that the prevalence of mutations of any class only marginally increased from stage I to stage IV tumors.
Table V. Differential Relationship Between p53 Alterations And Tumor Stage
Tumor stage n/%
p-values of Fisher's exact test of the 2 × 2 table were obtained by comparing stage IV with stages I–III combined. In case of the protein expression levels, both reduced expression and overexpression were separately compared with normal expression.
Figures representing highest levels and lowest levels are denoted a and b, respectively.
Figures representing highest levels and lowest levels are denoted a and b, respectively.
Survival analyses substantiated these differential findings, revealing a shorter survival of patients whose tumors showed aberrant p53 expression in comparison with patients with normal p53 expression (p = 0.044, log-rank test). However, when the tumors were stratified for stage, p53 expression was no longer an independent prognostic factor (data not shown). As can be seen from Figure 3, hypopharyngeal cancer patients with DNA contact mutations (class I) in their tumors only showed insignificant shorter survival than the other hypopharynx cancer patients (top), and very similar results were obtained in the other sites. In hypopharynx again, patients with class III mutations even had a statistically insignificant tendency for longer survival (middle), and in the total cohort, the difference was small (bottom).
Thus, p53 protein levels changed from normal to reduced as well as to increased expression with tumor stage progression, but were not independent predictors of clinical outcome. The different mutation classes that were present in all stages at a similar frequency did not correlate with the clinical course. In multivariate analysis, taking into account the TNM stage of the tumors, the p53 alterations examined, including DNA contact mutations, failed to achieve prognostic significance in the total cohort as well as in the individual tumor sites.
Although many studies have been performed to assess the possibly pivotal role of p53 in HNSCC progression, the results have been largely contradictory regarding both the prevalence of p53 alterations and their biologic and clinical influence on tumor behavior.39 The reasons for the failure to reach a consensus are attributable to small sample size and differences in the methodologic approach, e.g., for mutational analysis, for which none of the methods available is infallible.9, 13, 15 Perhaps most importantly, in the majority of investigations, mixed cohorts of tumors from all sites in the head and neck region had been analyzed. However, the tumor site is one of the strongest prognostic factors in HNSCC, which probably contributes to the heterogeneous clinical behavior of the tumor entity HNSCC. Therefore, it appeared important to study a number of tumors large enough to allow the assessment of the p53 status in the total HNSCC cohort and in the different sites: hypopharynx, oropharynx, larynx and oral cavity. The study here presented meets the criteria of a large-scale study. A total of 514 tumors were subjected to immunohistochemical analysis of p53 expression. As a novel and very informative method to examine the level of protein expression, we introduced TSA-IHC. This allowed the visualization of the small amounts of normal p53 protein present in normal squamous mucosa. Hence, in tumors we could discriminate between a normal, possibly tolerated p53 protein expression level and reduced or lost expression (Figs. 1 and 2). In addition, 253 tumors were subjected to mutational analysis using in part 2 different sequencing strategies. In the following, we discuss the most important results obtained.
First, in this study, the analysis of p53 protein expression has proven to be an excellent supplement to mutational analysis. We have established 3 levels: reduced (lost), normal and overexpression. Reduced or lost expression correlated very well with and was predictive for the presence of nonsense mutations; overexpression of p53 correlated very well with and was predictive for the presence of missense mutations; and normal expression correlated with the absence of mutations.
Second, the prevalence and nature of p53 alterations differed significantly between the tumor sites. The hypopharynx showed the highest rate of alterations, i.e., mutations, overexpression and loss of expression, whereas the larynx showed the lowest rate of alterations (Tables I–III).
Third, the distribution of the mutation classes did not differ between the sites. However, the larynx revealed a distinct mutation spectrum. Fourteen of the 28 mutations occurred in exon 5, and the S2′ domain (codons 132–135) was almost exclusively affected. The S2′ domain is part of the loop-sheet-helix motif critical for sequence-specific DNA binding of p53.23 The prevalence of transversion mutations was also higher in the larynx tumors (Table IV). It is thus noteworthy that the distinctive mutation spectrum seen in the larynx is more similar to that in lung cancer than, e.g., that in hypopharynx, consistent with the surface mucosa of the larynx being exposed only to tobacco, whereas in the other head and neck tumor sites (oropharynx, oral cavity, hypopharynx), tobacco carcinogens act in conjunction with alcohol abuse in most cases.4 This site-specific difference has not been reported in previous studies.10, 11
Fourth, neither in the total cohort nor in individual sites did the rate of p53 mutations increase from low to advanced tumor stages. This finding strongly supports the occurrence of mutations prior to the invasive phenotype, in agreement with the previous identification of mutations in premalignant leukoplakias5, 6 and in histologically normal-appearing tumor-distant mucosa of cancer patients.7, 8
Fifth, in contrast to mutations, there was a significant increase in alteration of protein expression from low- to high-stage tumors. Both overexpression and loss of expression predominated in stage IV tumors, whereas normal expression predominated in stage I tumors. Whether these changes in expression levels reflect a general alteration in p53 metabolism in higher tumor stages or changes in p53-protein interactions is unclear.
Sixth, none of the p53 alterations observed had significant prognostic power in multivariate analysis (Fig. 3), although low and high protein expression levels correlated with high tumor stage.
It is likely that the application of the p53GeneChip to tumors of the other head and neck sites would also marginally increase the mutation rate in these sites. It is unlikely, however, that the mutation rate in HNSCC is as high as reported in recent publications relying on the analysis of mRNA instead of genomic DNA.8, 16, 17 This conclusion is based on the fact that genomic sequencing of all exons and exon/intron borders by others and by us has only produced a moderate increase in the mutation rate,9, 12 suggesting the possibility that the analysis of cDNA reverse-transcribed from mRNA may yield artifacts (e.g., due to error-prone RT activity). Indeed, in the study by Kropveld et al.,16 which involved only a very small number of cases, there were discrepant mutations, i.e., seen in the mRNA but not in the DNA analysis, which are difficult to explain (they were not confirmed by mutation-specific oligonucleotide hybridization), and a very unusual mutation was reported to have occurred in 2 different patients, suggesting the possibility of sequencing artifact or of carryover contamination. In the second study, neither DNA sequencing nor mutation-specific oligonucleotide hybridization has been performed to confirm the presence of mutations, nor was immunohistochemistry used to confirm the absence of protein in cases with frameshift, nonsense or aberrant splicing mutations.17 Perhaps even more important, we have found in this study an excellent agreement between the p53 protein expression levels and the genetic status detected by sequencing. In particular, the predictive value of normal p53 protein expression for the absence of mutation (78.5%) with its narrow 95% confidence interval between 69.8 and 85.5 makes it highly unlikely that the p53 mutation prevalence will greatly differ from the range reported here (46.5%; 95% CI = 41.7–54.4) and that reported by others.4, 9, 10, 11, 12
Our disagreement with the above-mentioned recent reports8, 16, 17 only concerns the p53 mutation rate in HNSCC. However, we do share the view expressed in these studies as well as in others that it is reasonable to assume that the p53 pathway is impaired in every single tumor, since many proteins normally under p53 control show deregulated expression in HNSCC seemingly wild type for p53 (such as p21WAF1, Bax, GADD45, or angiogenic factors). The involvement of high-risk types of human papilloma viruses in HNSCC clearly provides an alternative mechanism to inactivate functionally the p53 protein and the retinoblastoma protein pRb as well.11, 40, 41, 42, 43 However, the prevalence of HPV oncogene expression in HNSCC17, 40, 42, 43 is not high enough to fill the gap in the tumor group in which we could not find evidence for mutational inactivation of p53. Even in the tonsils and the base of tongue, which are the oropharyngeal subsites showing the highest prevalence of HPV,40 p53 mutations and HPV involvement account for only about 70% of the tumors, because the high HPV prevalence is counterbalanced by a lower p53 mutation prevalence in these subsites. If a reduced or absent expression of pRb, concomitant with overexpression of p16INK4a, is taken as an indicator of HPV involvement,40 the prevalence of p53 inactivation by mutation or HPV in the entire oropharynx can be estimated to be approximately 56% (17% by HPV), in the oral cavity 45% (8.6% by HPV), in the larynx 47% (12% by HPV) and in the hypopharynx 74% (7% by HPV).40 Thus, in all sites, a fraction of tumors with apparently wild-type p53 remains, in which it is unclear as to whether and how p53 is inactivated.
There is a number of candidate proteins that are known to interact with p53 either in a positive or in a negative regulatory fashion, the loss or increased activity of which might result in the inhibition of p53 activity or might result in toleration of wild-type p53 expression. Besides already established candidates such as p14ARF or hdm-2, new candidates have been found. For instance, the loss of hSIR2 that functions as a p53 deacetylase might inhibit the proapoptotic activity of p53,44, 45 whereas the loss of ASPP proteins that are specific activators of the apoptotic function of p53 might constitute a mechanism to tolerate wild-type p53 expression in malignant tumors.46 Similarly, lack of promyelocytic leukemia protein (PML) might impair p53-dependent apoptosis.47 Furthermore, inhibitory forms of ASPP proteins have been identified. These not only inhibit p53 activity but in addition are oncogenic by cooperation with Ras, E1A, or E7 (but not mutant p53).48 A possible role of some of these proteins in HNSCC has not been studied as yet. Another alternative to p53 mutation is the amplification and consequent overexpression of oncogenes contained in the distal part of the short arm of chromosome 3, such as cyclin L or PIK3CA. PIK3CA, which is negatively regulated by p53 via PTEN in normal cells, has been shown to be preferentially overexpressed in p53 wild-type HNSCC and derived cell lines.49
If the assumption were correct that p53 is at least in part inactivated in every tumor and p53 inactivation occurs prior to and is a prerequisite for the invasive phenotype (as suggested from studies on the transformation process of cells in culture), then the mode of inactivation, either by mutation or by other mechanisms, might only be of secondary importance for tumor progression. This might explain the contradictory results reported in previous studies, including our own, in which we had obtained evidence for a prognostic role of class I mutations (DNA contact) and mutations affecting the Zn-binding residues.32 Considering the scale of the present study, p53 does not seem to be a major driving force once the invasive stage has been reached.
It should be stressed that even if the prognostic value in tumor progression is limited, the state of p53 retains an important role in HNSCC carcinogenesis as a very useful marker for risk assessment for disease recurrence, in particular with regard to the identification of minimal residual cancer.9, 49, 50 Equally important, p53 alterations remain very important markers for cancer risk (development of a primary tumor or second primary tumor), especially in conjunction with functionally related changes such as allelic imbalances and/or numerical chromosomal aberrations.51, 52, 53
The authors are greatly indebted to the medical staff of the Department of Otolaryngology, Head and Neck Surgery, for collecting and processing the tissue specimens. They also thank S. Huntgeburth, A. Schuhmann, W. Klein-Kühne and F. Devens for their excellent technical assistance and P. Benjamin for photographic work.