Expression of cell cycle markers and human papillomavirus infection in oral squamous cell carcinoma: Use of fuzzy neural networks

Authors


Abstract

Our aim was to evaluate in oral squamous cell carcinoma (OSCC) the relationship between some cell cycle markers and HPV infection, conditionally to age, gender and certain habits of patients, and to assess the ability of fuzzy neural networks (FNNs) in building up an adequate predictive model based on logic inference rules. Eighteen cases of OSCC were examined by immunohistochemistry for MIB-1, PCNA and survivin expression; presence of HPV DNA was investigated in exfoliated oral mucosa cells by nested PCR (nPCR, MY09-MY11/GP5-GP6), and HPV genotype was determined by direct DNA sequencing. Data were analyzed by traditional statistics (TS) and FNNs. HPV DNA was found in 9/18 OSCCs (50.0 %) without any significant higher risk of HPV infection with respect to the sociodemographic variables considered (p > 0.2), apart from tobacco smoking, reported in 44.4% of OSCC HPV-positive vs. 100% HPV-negative subjects (p = 0.029). Regarding cell cycle markers, TS and FNN revealed that survivin was expressed significantly more in HPV-negative than in HPV-positive OSCC [root mean-square error (RMSE) = 5.89 × 10–6, % predicted 100.0]; furthermore, smoking played a protective role for survivin expression in HPV-positive cases (OR = 0.019, 95%CI 0.001–0.723, RMSE = 0.20, % of prevision 94.4). FNN, although on a small sample size, allowed us to confirm data by TS and to hypothesize a different cell cycle pattern for HPV-positive vs. HPV-negative OSCC. In the latter cases, the relevance of apoptotic vs. proliferative markers suggested that they may be related to the different supposed outcome of HPV-negative OSCC and that HPV may have a protective role in the expression level of survivin, especially in tobacco smokers. © 2005 Wiley-Liss, Inc.

OSCC is rapidly increasing in incidence and represents the most frequent malignant oral tumor. The incidence of metastasis depends on the degree of cellular differentiation, deep invasion and site of the primary tumor; however, outcome is difficult to predict if only standard clinicopathologic parameters are taken into account. Biologic markers that can help to identify lesions with an aggressive phenotype and worse prognosis need to be identified.

Since the majority of human neoplasms are characterized by an imbalance of the regulatory cell cycle control processes, the study of OSCC using the expression of proteins involved in the critical checkpoints of cell apoptosis and growth starts by elucidating the processes of carcinogenesis and appears to have good prognostic value.1 Indeed, we know that apoptosis, or programmed cell death, and its suppression are involved in the carcinogenesis of several tumors; this process, mediated by caspases and cysteine proteinases present as proenzymes, is inhibited by several proteins, such as those of the Bcl-2 and IAP families. Survivin is an IAP protein, which is abundantly expressed in most solid and hematologic malignancies but undetectable in normal adult tissues.2 However, the growth fraction, i.e., the proportion of cells committed to the cycle, may be easily assessed by Ki-67 or MIB-1 antibodies, which identify antigens expressed in the G1, S and G2 phases of the cell cycle.3, 4 In particular, SPF, besides being the gold standard technique,5 can be assessed by immunohistochemical detection of PCNA, a 36 kDa auxiliary protein of DNA polymerase-delta nuclear protein involved in DNA synthesis, with a relevant correlation to the proliferative state of the cell.6

Besides the relevance of studies on several markers, it was mandatory to investigate the risk factors of continued use of tobacco, alcohol misuse, diet and viral infections.7 In particular, it has been generally hypothesized that HPV infection may play a role in the occurrence of potentially malignant oral lesions and/or in their malignant transformation,8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23 even if the detection rate of HPV DNA reported in carcinomas of the oral cavity is very discordant.19 A previous study, carried out in our geographic area (southern Italy), determined that HPV prevalence was 61.5% in carcinomas, 27.1% in potentially malignant lesions, 26.5% in erosive ulcerative lesions and 5.5% in controls.20

In a framework in which several markers and risk factors operate and need to be analyzed simultaneously, ANN analysis could be a powerful tool for accurately detecting causal relationships. The FNN is one of the most advanced ANN models, and its most attractive feature is that relationships between input and output variables can be described accurately from the acquired model.

In the present study, we applied TS and FNN methodology to evaluate, in an OSCC population grouped by HPV infection, the relationship of immunohistochemical (survivin, MIB-1 and PCNA) expression and sociodemographic findings with HPV status and to define or predict any proliferative/apoptotic behavior related to HPV.

Abbreviations:

AEC, 3-amino-9-ethylcarbazole; ANFIS, adaptive neuro-fuzzy inferences system; ANN, artificial neural network; CI, confidence interval; FL, fuzzy logic; FNN, fuzzy neural network; H&E, hematoxylin and eosin; HPV, human papillomavirus; HR, high-risk; HRP-LSAB, horseradish peroxidase–labeled streptavidin-biotin; IAP, inhibitor of apoptosis; MAb, monoclonal antibody; nPCR, nested PCR; OR, odds ratio; OSCC, oral squamous cell carcinoma; PCNA, proliferating cell nuclear antigen; RMSE, root mean square error; SPF, S-phase fraction; TS, traditional statistics.

Material and methods

The study group was composed of 18 immunocompetent adult outpatients with OSCC (11 female, 7 male) consecutively diagnosed at the Section of Oral Medicine (University of Palermo); mean age was 63.06 years (range 44–84). Informed consent was obtained from all participants. Microscopic evaluation was performed by one oral pathologist (A.M.F.), who confirmed lesion diagnosis and determined the degree of differentiation. Histologic grading was assessed on paraffin H&E-stained sections. Tumor extent was classified according to the TNM system,24 and tumors were divided into grades 1, 2 and 3 using the WHO classification of histologic differentiation. Historical and clinical data of each subject were recorded on a clinical report form and collected by means of a data entry program. Information regarding age, smoking and alcohol use was obtained by personal interviews; patients were divided into 2 categories for each habit: yes or no.

Virologic evaluation

HPV DNA presence was researched by nPCR (MY/GP primers) in oral mucosa brushed cells, and HPV genotype was determined by direct sequencing of PCR fragments.

Sample collection and processing

Oral cytologic specimens were obtained from the site of the lesion by means of a cytobrush (RAM, Mirandola, Italy).

PCR analysis

DNA extraction was performed as previously described.20 All clinical samples were checked for DNA by amplification of the human β-globin gene and tested in duplicate. Three types of control were included in each reaction series: blank control, HPV-negative Wi cells as negative control and HPV-18 DNA-positive HeLa cells, in dilutions from 20,000–50,000 down to 2–5 HPV DNA copies, as positive control. All controls were prepared and analyzed in parallel with clinical specimens to ensure that proper reaction conditions were maintained. Special care was taken to prevent contamination; standard precautions for molecular biology reactions were observed,25 and individual procedural steps (reaction mix preparation, sample preparation, amplification, electrophoresis) were carried out in separate rooms. HPV DNA was amplified by nPCR assay (MY09–MY11 primer pair in combination with GP5–GP6 primer pair) as previously described,20 and amplifications were performed in a DNA thermal cycler (Mastercycler gradient; Eppendorf, Hamburg, Germany); amplification products were analyzed in 8% polyacrylamide gel.

Sequencing analysis

HPV genotyping was based on direct sequencing of MY or MY/GP PCR fragments.26

Amplification products were purified by Microcon YM-100 (Amicon-Millipore, Billerica, MA); the sequence of both DNA strands was determined by the BigDye Ready Reaction Kit in the automatic sequencer ABI Prism 310 Analyzer (both from Perkin-Elmer Applied Biosystems, Foster City, CA). Alignments were obtained from the GenBank on-line BLAST server and HPV sequences downloaded from the HPV database (http://hpv-web.lanl.gov).

Immunohistochemical analysis

Immunohistochemistry was performed on the remaining sections mounted on poly-L-lysine-coated glass slides. Deparaffinized and rehydrated sections were incubated for 30 min in 3% H2O2/methanol to quench endogenous peroxidase activity, then rinsed for 20 min with PBS (M107; Bio-Optica, Milan Italy). Nonspecific protein binding was attenuated by incubation for 30 min with 5% horse serum in PBS.

An anti-Ki-67 MAb (clone MIB-1; DakoCytomation, Copenhagen, Denmark) was used at a dilution of 1:75. The antibody was applied on the sections after 20 min heat-induced epitope retrieval with Target Retrieval Solution, Citrate (pH 6.0, DakoCytomation) for 15 min at room temperature. After rinsing in PBS and application of a biotinylated secondary antibody for 10 min, an HRP-LSAB visualization system (DakoCytomation) was used and applied for 10 min. To visualize the reaction, sections were incubated with an AEC substrate chromogen (DakoCytomation) for 20 min at room temperature and counterstained with Mayer's hematoxylin. Slides were mounted with Glycergel (DAKO Corp., Milan, Italy) mounting medium and evaluated under a conventional light microscope. Human tonsil tissue was used as a positive control, and a negative control was performed by omitting incubation with primary antibody replaced by mouse IgG (DakoCytomation) diluted at the same concentration. Proliferating (MIB-1+) cells were visually counted. Slides were evaluated at low magnification, and 10 areas with different percentages of positive cells were randomly selected. Cells were counted at ×400 magnification; the average number of positive cells was calculated, with mean proliferation index determined and expressed as a percentage of counted cells. The score was graded in 3 classes as follows: grade 1, <5% positive cells; grade 2, 5–10%; grade 3, >10%.

PCNA staining was performed using a conventional indirect immunoperoxidase technique on routinely fixed tissues. Sections were incubated with PCNA MAb (PC10; Dakopatts, Hamburg, Germany), diluted 1:100, for 60 min at 24°C; the conventional streptavidin-biotin method was performed. Reactivity was evidenced with diaminobenzidine and H2O2 at 0.03%. Sections were counterstained with hematoxylin. Random fields were sampled with the aid of a random table, and mean percentage of positive tumor cells was determined by examining 300 cells in at least 5 areas at ×400 magnification and assigned to one of the following categories: 0, <5%; 1, 5–25%; 2, 26–50%; 3, >50%.

For survivin, immunohistochemical staining was carried out with a polyclonal antibody (NB 500-201; Novus Biologicals, Littleton, CO) after antigen retrieval by pressure cooking and detection by streptavidin-biotin-peroxidase as previously reported.27 Negative control slides without primary antibody were included for each staining. Survivin expression was quantified by the following scoring method: mean percentage of positive tumor cells was determined examining 300 cells in at least 5 areas at ×400 magnification and assigned to one of the following categories: 0, <5%; 1, 5–25%; 2, 26–50%; 3, >51%.

Data analysis

Data were analyzed by means of the computer packages S-Plus 4.0 (Insightful, Seattle, WA) and SPSS 9.0 (SPSS, Chicago, IL). Comparison of baseline characteristics was made for continuous and categorical variables by nonparameter tests, typical in TS. Fisher's exact test was used when the frequency observed was <5; in all other evaluations, p ≤ 0.05 was considered statistically significant.

In particular, to measure the association level, crude ORs and the 95% corresponding test-based CIs were calculated.

By FNN, an fuzzy logic (FL) system was found from the same data; a fuzzy inference system (Fig. 1) is a rule-based system using FL. FNN is one of the most advanced ANN models (computational methodologies for multivariate analyses) and is considered to have an important role in the future of medicine and, in particular, oncology.28 FNN is founded on FL; FL is a superset of conventional logic that has been developed to handle the concept of partial truth. It was introduced by Zadeh29 as a tool to model the uncertainty within natural language. The key notion of fuzzy theory is “graded membership”, according to which a set could have members belonging only partly to it. So, if we assume X is a set serving as the universe of discourse, a fuzzy subset A of X is associated with the function μA: X→[0, 1], which is generally called “membership function”. The idea is that for each x, μA(x) indicates the degree to which x is a member of the fuzzy set A. The classic set theoretical operations can thus be extended to fuzzy sets, which have membership grades that are in the interval 0–1.30 The basic structure of FL includes 4 main components: (i) a fuzzifier, which translates crisp (real value) inputs into fuzzy values; (ii) an inference engine, which applies fuzzy reasoning, a mechanism to obtain a fuzzy output; (iii) a defuzzifier, which translates this latter output into a crisp value; and (iv) a knowledge base, which contains both an ensemble of fuzzy rules, known as the rule base, and an ensemble of membership functions, known as the database. In the present study, primary selection of entry variables (membership functions) was used as input and later the ANFIS methodology (Fig. 2), with Sugeno's model of the first order. The FL system takes into account the following input variables: age (<57, 57–71, >71); gender; smoking status (0, nonsmoker; 1, smoker); alcohol use (0, nondrinker; 1, drinker); survivin (0, negative; 1; 2; 3); MIB-1 (0, negative; 1; 2; 3); PCNA (0, negative; 1; 2; 3). The output variable was HPV (0, negative; 1, positive).

Figure 1.

Description of steps involving the fuzzification/defuzzification processes.

Figure 2.

Network scheme of ANFIS: input/output flow.

Results

HPV DNA was found in 9/18 (50.0%) OSCCs; in all cases, HPV genotypes were HR: HPV-18 in 77.7% (7/9), HPV-16 in the remaining 2 cases (Fig. 3). Prevalence of oral habits (tobacco smoking and alcohol drinking) and immunohistochemical expression (PCNA, MIB-1 and survivin; Fig. 4) related to HPV status are detailed in Table I.

Figure 3.

Gel showing representative MY-PCR (450 bp) and MY/GP-PCR (140 bp) products of HPV DNA from oral mucosa samples. Lanes 1 and 9, ϕχ 174 restriction fragment (RF) DNA cleaved with HaeIII molecular size standard; lanes 2–5, HeLa cells (positive control), dilutions corresponding to 2 × 105, 2 × 104, 2 × 103 and 2 × 102 HPV DNA copies, respectively; lanes 6 and 7, positive samples; lane 8, Wi cells (negative control); lanes 10 and 11, HeLa cells (positi3ve control), dilutions corresponding to 20 and 2 HPV DNA copies, respectively; lanes 12 and 13, positive samples.

Figure 4.

(a) MIB-1 expression in a case of OSCC (×10). (b) High magnification (×25). (c) PCNA expression in a case of OSCC (×10). (d) High magnification (×40). (e) Survivin expression in a case of OSCC (×20). (f) High magnification (×40).

Table I. Habits and Histologic Markers in OSCC (n = 18) Grouped by HPV Status
 HPV+n = 9 (%)HPVn = 9 (%)Fisher's exact p value
  • 1

    M.S., mean score.

  • 2

    χ2 = 6.923.

  • 3

    χ2 = 6.923.

Smoking   
 Yes4 (30.8)9 (69.2)0.0292
 No5 (100)0 (0)
Alcohol   
 Yes6 (66.7)3 (33.3)>0.2
 No3 (33.3)6 (66.7)
Markers   
 MIB-1 (11/18, 61.1%)4 (44.4), MS1 1.12 ± 0.777 (77.7), MS 1.22 ± 0.97>0.2
 PCNA (14/18, 77.8%)6 (66.7), MS 1.67 ± 0.478 (88.8), MS 1.67 ± 0.33>0.2
 survivin (13/18, 72.2%)4 (44.4), MS 1.1 ± 1.3649 (100), MS 2.22 ± 0.830.033

Our further main finding is the following: all HPV-negative OSCCs (100.0%) were smokers and showed expression of survivin (p = 0.03 and 0.029, respectively). In univariate analysis, smoking showed a protective role of survivin expression in HPV-positive cases (OR = 0.019, 95% CI 0.001–0.723).

A higher prevalence of alcohol drinkers (66.7%) was registered among OSCC HPV-positive patients; but the same group presented slight expression of MIB-1 (44.4%) and PCNA (66.7%) compared to HPV-negative patients (77.7% and 88.8%, respectively).

Expression and intensity of the markers considered were evaluated in relation to HPV status and the main sociodemographic variables using an FNN system (i.e., FL). Different from TS, by means of the FNN system it was possible to extrapolate the variables with the best system behavior, such as expression of PCNA, expression and intensity of survivin and smoking (RMSE = 5.89 × 10–6, % predicted 100.0) (Fig. 5). We observed a protective role of smoking for HPV status (all nonsmokers were HPV-positive), independent on any PCNA and survivin expression; among smokers, HPV-negative cases were significantly related to PCNA expression and to the highest values of survivin (Fig. 6). Finally, a new FNN system was developed, taking into account only 2 variables, expression and intensity of survivin and smoking (RMSE = 0.20, % of prevision 94.4), confirming the finding of TS as to the protective role of HPV infection in nonsmokers.

Figure 5.

RMSE = 5.89 × 10–6; % of prevision 100.0 calculated by ANFIS.

Figure 6.

System control surfaces for variables PCNA, survivin and output HPV in nonsmokers.

Discussion

Several factors are involved in oral carcinogenesis, such as age, gender, ethnicity, lifestyle, genetic background, status of health and exposure to one or more oncogenic factors.31 In several epidemiologic studies, tobacco smoking and alcohol consumption have been well documented as major risk factors for oral cancer, with attributable fractions of approximately 90%.32 However, 15–20% of head-and-neck squamous cell carcinomas have no known tobacco or alcohol exposure.33 Thus, other agents, such as viruses, are being investigated.34

In particular, with regard to viral involvement, it is still highly controversial whether HR HPV can be considered an etiologic or a malignant risk factor in oral carcinogenesis:14 some research groups14, 35, 36 have identified HR HPV antigens and viral DNA in potentially malignant and malignant oral lesions, and others have defined HR HPV as playing an important role in OSCC, especially in the absence of common oral habits.37 Historically, evidence of the carcinogenic role of HR HPVs is based on 2 oncoproteins: HPV-E6, which promotes degradation of the p53 tumor-suppressor gene product,38 and HPV-E7, which modifies pRb tumor-suppressor gene product function, leading to increased cell proliferation and contributing to carcinogenesis.39, 40 In a review on epidemiologic and molecular bases, Ha and Califano41 attributed to HR HPV a role in oral carcinogenesis but only in a small subset of cases, with differences reported in clinical outcome, response to radiotherapy and prognosis.21, 22, 33, 42, 43, 44, 45 In this context, it becomes important to identify this cluster principally on the basis of key immunohistochemical cell cycle markers that are inexpensive and reliable, such as PCNA, MIB-1 and survivin. Several reports have focused on the identification of useful markers such as molecules involved in the proliferation pathway or in the regulation of apoptosis, which may influence the cell death/cell viability balance toward cell proliferation or cancer.46, 47, 48, 49, 50, 51

In our study, different immunohistochemical markers of proliferation and apoptosis (PCNA, MIB-1, survivin) were evaluated in relationship to HR HPV infection by means of TS and FNN system analyses. The finding of an association of HPV infection with any of the markers considered could contribute to our understanding of the immunohistochemical cluster of OSCCs found more likely to be HPV-positive.

Data evaluation indicated that, in the present series of OSCCs, the prevalence of HPV DNA was 50.0%, in agreement with a previous report.41 Indeed, HPV DNA detection rate was not considered a main finding but as the output of the analysis system; the issue of HPV infection in OSCC has been formerly debated elsewhere by the same research group.20, 52 Also, tobacco smoking represented the most common habit in HPV-negative patients, whereas in HPV-positive ones alcohol drinking was the most reported.

In the present study, TS and FNN (an FL model) revealed some interesting results, similar as to association values but with an advantage for FL, which was able to realize a predictive model.

Firstly, PCNA and MIB-1 expression, both indexes of proliferative activity, was higher in HPV-negative cases (respectively, 88.0% and 77.0%) than in HPV-positive ones (respectively, 66.0% and 44.0%). These detection rates are lower than some reports in the literature, with PCNA expression up to 94.7% in HPV-positive OSCC;53, 54, 55, 56 furthermore, OSCCs with HPV-positive genotype 16/1853, 54, 55 or 389 have been associated with more intense expression of PCNA. PCNA is an accessory protein of DNA polymerase-delta, associated with cell cycle progression and detectable in the replicating cells of normal tissues, reaching an expression peak during the S phase of the cell cycle. PCNA tethers the DNA polymerase catalytic unit to the DNA template and is therefore essential for DNA replication. If we consider that higher proliferative activity is related to a worse prognosis, in our cohort of HPV-positive OSCCs, we could forecast a better prognosis, as pointed out by Gillison et al.,57 with a 59% reduction in risk of cancer death compared to HPV-negative tumors.

Secondly, the most important finding of our research is that survivin expression, an index of apoptosis blockage, is higher in HPV-negative cases (100.0%), independent of all the variables considered, than in HPV-positive cases (44.0%), with the maximum protective effect in HPV-positive smokers. This is in agreement with previously reported links between high levels of survivin expression and worse prognosis of the neoplastic lesions along with better prognosis in HPV-positive OSCC.57 In retrospective trials, survivin expression has been correlated with reduced overall survival in several solid tumors,58, 59, 60, 61, 62, 63 especially in OSCC;27, 64, 65 similarly, HPV infection could influence oral carcinogenesis, consistently with results suggesting that HPV-positive patients have significantly reduced overall and disease-specific mortality.22, 66 Besides Gillison et al.,57 Herrero et al.34 suggested HPV-positive OSCC as a distinct small subgroup of tumors with different biomolecular and clinical behaviors. Ringstrom et al.22 reported no disease recurrence and 5% mortality in HPV-16-positive oral cancers, while 31% of an HPV-negative group developed recurrence and 46% died.

From the present findings, the relationship between survivin expression and HPV presence suggests that HPV may have a direct or indirect effect on regulating the levels of survivin expression. Survivin may be recruited to participate in HPV-mediated deregulation of the normal patterns of squamous differentiation; and the different distribution of survivin in OSCC would indicate that survivin participates in or influences oral carcinogenesis. HPV-negative OSCC characterized by high levels of survivin expression could have a poor clinical response compared to HPV-positive OSCC characterized by lower levels of survivin expression since in retrospective trials survivin expression correlated with reduced overall survival in OSCC.27, 67 Conversely, survivin expression in HPV-positive OSCC may be useful to detect lesions with unfavorable behavior, similar to that of HPV-negative oral cancers, which almost always have high survivin expression. HPV-positive OSCC lesions with higher levels of survivin expression could have a worse clinical outcome: survivin could stop apoptosis and promote cell transformation.64

Thirdly, all HPV-negative cases were smokers (100%), while only 44% of HPV-positive cases were smokers. Tobacco smoking appears to protect subjects from HPV infection. A previous study reported that HPV DNA was detected less frequently in specimens from ex-smokers (OR for HPV detection = 0.6, 95% CI 0.2–1.5) and current smokers (OR for HPV detection = 0.4, 95% CI 0.2–0.9) than in specimens from nonsmokers.34 In India, HPV DNA was detected less frequently in tumor specimens from tobacco chewers (OR for HPV detection = 0.5, 95% CI 0.1–2.0) than in those from nonchewers.34 Nonsmokers are more likely than smokers to have HPV-related tumors.

Fourthly, alcohol assumption is higher in HPV-positive patients with OSCC. Probably, alcohol helps viral penetration in oral mucosa; alcohol could act as a permeability enhancer of human oral mucosa, altering the mucosal structure and thereby aiding penetration of HPV through the epithelial layers. Furthermore, to date, only in tonsillar squamous cell carcinoma has the morphology of HPV-positive tumors has suggested that HPV may have a predilection for a population of nonkeratinized squamous cells, easier to penetrated especially in the presence of alcohol, or that the virally transformed cells inhibit keratinization of the tumor cells.68 This latter hypothesis is consistent with the results of Smith et al.,69 who consider HR HPV infection of oral exfoliated cells to be a risk factor for head-and-neck squamous cell carcinoma, independent of alcohol and tobacco use. Furthermore, Smith et al.70 observed a significant interaction effect between younger age and heavy alcohol use associated with HR HPV cancer status.

As previously stated, our purpose in the present study was to investigate by FNN the likelihood of an HPV infection association in OSCC with PCNA, MIB-1 and survivin; the eventual finding of any association could improve our protocols, especially in HPV-positive cases. Whereas we failed to recognize markers unequivocally related to HPV infection, we could confirm that these well-known markers are related to OSCC, principally in HPV-negative cases.

An additional finding was the high detection rate of HPV-18 genotype among HPV-positive cases. The same viral HR genotype had been previously reported as prevalent in oral samples from our population.70 These different genotype frequencies could indicate a peculiar distribution pattern of HPV genotypes, probably reflecting the geographic origins of samples, as also previously noted for cervical HPV infections.71, 72 Interestingly, in another geographic area (Shanghai, People's Republic of China), HPV-18 was found, alone or mixed with genotype 16, in OSCC and normal mucosa.73

The present study, although limited by a small sample size, confirms findings on HPV-positive OSCC, indicating a different behavior of HPV-negative with respect to HPV-positive OSCC, consistent with the hypothesis of a complex relationship between HR HPV infection and either apoptosis or proliferation in malignant oral lesions. Also, FNN appears to be an effective tool in analyzing the relationships between HPV infection and the markers considered; in comparison with TS, FNN allowed us to find nonlinear relationships between input and output variables, overcoming the bias sometimes generated by TS and linear regression.74 Hence, longitudinal studies supported by FNN systems on larger sample size are recommended.

Finally, taking into account the oncogenic potential of HR HPV and the differences in clinical outcome, response to radiotherapy and prognosis for HPV-positive OSCCs,21, 22, 33, 42, 43, 44, 45 our data suggest that the presence of oral HPV infection should be determined in any OSCC along with the concomitant expression of survivin.

Ancillary