Tobacco smoking and alcohol consumption are the main risk factors in the etiology of head and neck squamous cell carcinomas (HNSCC). In addition, over the last 15 years, infection with high-risk human papillomavirus (HPV) types has also been etiologically linked with a subset of HNSCCs1, 2, 3, 4, 5, 6, 7 and sexual transmission is suggested to cause the infection.5, 8, 9
High risk HPV type 16 (HPV16) has been identified in ∼90% of HPV-positive HNSCC, and HPV18, HPV 31 and HPV 33 in the remaining cases.10 These high risk viruses play a role in the carcinogenic process by producing 2 oncoproteins encoded by the viral E6 and E7 genes. These oncoproteins directly inactivate p53 and pRb, respectively, and promote cell cycle entry and DNA synthesis as well as blockade of apoptosis, cellular conditions favoring viral replication.11
At present, the causal role of HPV in cervical carcinomas, has been well established. High-risk HPV DNA is detected in almost all cervical carcinomas12, 13, 14 and this finding has instigated studies on the prevention or treatment of HPV infections using vaccines.15 As for etiology of HNSCC, the importance of HPV infection is more controversial, mainly because no unanimously accepted detection method exists. The reported prevalences of high-risk HPV DNA tremendously vary per study from 0 to 100%.10, 16 Part of the variation in HPV prevalence can be explained by differences with regard to the location of the tumor, i.e. it is relatively low in the oral cavity and high in the tonsil.1, 4, 7 More importantly, variations in the type of tissue material studied and the HPV detection methods used, likely have a major impact on the discrepancy in reported prevalence rates. The most widely applied detection methods are based on PCR amplification of viral DNA. However, these methods are extremely sensitive and can even detect a few DNA copies per sample, which might yield false-positive results.4, 10, 17 Moreover, HPV-DNA presence does not per se indicate viral involvement in the carcinogenic process and may reflect a transient infection that does not carry any risk of neoplastic transformation.18 It is our opinion that the lack of agreement of what to consider a HPV-involvement is for a large part the cause of confusion in the HNSCC literature. As an example, some studies reported an association between HPV infection and improved survival,4, 19, 20, 21 whereas other studies failed to show this.22, 23, 24
Recently, we demonstrated HPV DNA by PCR in 24/143 (16.7%) frozen oral and oropharyngeal tumors, but could confirm viral involvement by E6/E7 expression analysis in only 12/24 (50%) of these samples.22 This observation shows that presence of HPV DNA solely does not necessarily mean that the virus is biologically active in these tumors. In addition, it was found that tumors containing transcriptionally active HPV can be genetically distinguished as a specific subgroup,1 strongly supporting the hypothesis that HPV only plays a carcinogenic role when viral oncogenes are expressed. This group of tumors with transcriptionally active HPV showed absence of TP53 mutations and a limited number of genetic abnormalities, whereas their counterparts without transcriptionally active HPV showed TP53 mutations in 75% of cases and many genomic abnormalities.1, 2, 25, 26 Given this concept, measurement of E6/E7 expression seems to be the most reliable way to detect a biologically relevant association between HPV and malignancy, but a precondition to measure E6/E7 mRNA expression in a user-friendly reliable manner, is the availability of frozen material. A method feasible for more accessible routinely collected material, such as formalin-fixed paraffin-embedded (FFPE) tissue is still missing.
Commonly used HPV detection methods for FFPE tissue are known from studies on cervical cancer, and include, besides HPV consensus PCR methods, type-specific HPV-DNA detection by fluorescence in situ hybridization (FISH)27 and real-time PCR assays, the latter allowing viral load analysis.28, 29 Also the use of surrogate biomarkers such as p16 immunostaining can be considered for studying FFPE specimens of HNSCC.7, 21, 30 Alternatively, detection of serum antibodies31 directed against HPV epitopes might be an option when serum has been collected.
In this study we aimed to find a reliable and easily implementable test algorithm to assess a directly carcinogenesis-related HPV involvement in both frozen and FFPE tumor specimens of HNSCC patients.
Material and methods
Patients, tissue samples and group definitions
For this study we used tumor specimens and serum samples of 48 patients who underwent surgical treatment for HNSCC at the VU University Medical Center. The study was approved by the Institutional Review Board, and informed consent was obtained from all patients. Regarding the tumors a dual work-up was followed: a part of the tumorous tissue was snap-frozen in liquid nitrogen and stored for research purposes.32 The remaining part of each tumor was processed for routine histopathology and the tissue detection methods that are the subject of this study were performed on this routinely processed material.
The frozen tumor specimens were analyzed for the presence of HPV-DNA and E6/E7 mRNA as described previously.1, 2 All HPV DNA-positive carcinomas contained high risk HPV16. For this study we considered presence of HPV16 E6/E7 mRNA in the frozen specimens as “gold standard” reflecting direct viral involvement in carcinogenesis and this was used as selection criterion for the case group. Among 24 HPV16 DNA-positive tumors, 12 (50%) samples were positive for E6 and E7 transcripts and were considered true HPV-positive,1, 2 and further referred to as HPV D+/R+ (DNA+/RNA+). The group of twelve HPV16 DNA-positive, but E6/E7 mRNA-negative tumors was also included in the comparative analysis as an as yet undefined group regarding the HPV status, and was further referred to as the HPV D+/R− group. A third HPV-negative (DNA−/RNA−) control group of 24 tumors were selected in such a way that a similar distribution over the groups was ensured for clinical parameters that might in theory confound the analysis (i.e., age, gender, tumor site and stage, tobacco and alcohol consumption and histology). After analysis, the HPV D+/R− group showed in general negative test results and together with the negative E6-expression in frozen tissue, this group was scored as being HPV-negative and was included as such in the calculation of sensitivity and specificity.
Preparation of paraffin sections and isolation of nucleic acids
Paraffin sections were prepared according to the sandwich method: the first and last sections were stained by haematoxylin and eosin to check for tumor presence and to guide microdissection. One 4-μm section was used for FISH analysis and one 5-μm section for p16 immunohistochemical staining. Routinely 10–20 sections were used for microdissection of the neoplastic cells for viral load analysis. From 10–20 subsequent sections RNA and DNA was isolated simultaneously for GP5+/6+ PCR and HPV16 E6* RT-PCR (see later). To avoid cross-contamination, a new microtome blade was used each time a new case was sectioned, aerosyl tips were used for all pipetting steps and separate laboratories were used for pre- and post-PCR handling.
Detection of High Risk HPV DNA by GP5+/6+-PCR
Detection of high risk HPV was performed by general primer GP5+/6+-PCR on 50 ng of DNA, quantified by the Quant-It Picogreen dsDNA assay kit (Invitrogen, Breda, The Netherlands), followed by reverse line blot genotyping.33, 34 Serial dilutions of DNA isolated from cervical carcinoma cell line SiHa (ATCC; HTB35; 1–2 integrated HPV16 copies), and reactions without template were run in parallel as controls.
Viral load analysis
Quantification of HPV16 DNA copy numbers per cell was performed by real-time PCR using the LightCycler® technology (Roche Molecular Biochemicals, Mannheim, Germany) as described previously,35 except that for PCR detection of both HPV16 and β-globin other primers were employed to shorten the amplicon lengths to 114 and 111 base pairs, respectively (Table I). On the basis of the assumption that at least 1 viral copy is needed per cell for clonal expansion, tumors with >0.5 copies per cell were scored as positive, allowing some normal tissue contamination in the microdissected tumor samples.
Table I. Primer and Probe Sequences used For Viral Load and E6*I Expression Analysis
Assay (amplicon length in basepairs)
Sequence 5′ → 3′
Primer location in genome sequence
fw, forward primer; re, reverse primer; dp, donor probe (3′end labeled with fluorescein); ap, acceptor probe (3′end phosphorylated, for HPV16 5′end labeled with LightCycler-Red-640 and for β-globin 5′end labeled with LightCycler-Red-705); pr, Enzyme-Immuno-Assay (EIA)-probe; tp, TaqMan-probe (5′end labeled with FAM and 3′end labeled TAMRA).
Genbank accession number.
Length in basepairs.
This gene is located at chromosome 11.
HPV16 E6*I is spliced between genome position 226 and 409.
FISH was performed on paraffin-embedded tissue sections as described previously.7 Controls included hybridizations on FFPE sections of known HPV16- and 18-positive human cervical carcinoma cell lines (CaSki [ATCC; CRL1550; 500 integrated HPV16 copies], HeLa [ATCC; CCL2; 20–50 integrated HPV 18 copies] and SiHa [ATCC; HTB35; 1–2 integrated HPV16 copies]) as well as hybridizations on tissue sections of cervical lesions with proven integration or episomal presence (replication) of HPV genomic DNA to guarantee probe specificity, sensitivity and interpretation accuracy.36 Negative controls consisted of HPV PCR- and FISH-negative cell lines and tissue sections and hybridizations omitting the viral probe. Evaluation of nuclear hybridization signals was performed by 3 investigators according to the criteria described by Cooper et al.37 without information on the HPV-status of the samples. Both staining intensity (0–3) and punctate and/or diffuse signals throughout the nucleus indicating integrated and episomal HPV DNA, respectively, were evaluated. The level of inter-observer agreement was determined by calculating Cohen's kappa values.38 A definitive consensus score was determined by mutual agreement in a separate session.
Immunohistochemical staining of p16
For p16INK4a or short p16 (the protein encoded by CDKN2A) immunohistochemistry the CINtec™ Histology Kit (DakoCytomation B.V., Heverlee, Belgium) was used. For every case analyzed, an extra tissue section was stained with a mouse IgG, as a negative control. Staining intensity as a result of this mouse antibody was considered background and all samples with staining intensity above that background were scored as positive. An extra tissue section of an HPV-positive tumor with high p16 expression was included as positive control. Both the staining intensity (graded 0–3 proportional to staining intensity) and the percentage of the tumor cells positively stained per slide were assessed independently by 3 investigators. The level of interobserver agreement was determined by calculating Cohen's kappa values. A definitive consensus score was determined by mutual agreement in a separate session. The eventual decision if tumor was analyzed was made after consultation of an experienced pathologist (CJLMM).
Detection of antibodies against the proteins HPV16 L1, E6 and E7
Antibodies against L1, E6 and E7 of HPV16 were measured in a sandwich ELISA using a glutathione S-transferase capture method with native recombinant L1-tag, E6-tag and E7-tag proteins.39, 40
All sera were measured at least twice and the median of the absorbance values was taken as the final read out.
Detection of HPV16 E6*I mRNA on paraffin embedded tissue
We developed an RT-PCR assay to detect the most abundant splice variant within the HPV16 E6 open reading frame, namely E6*I, in FFPE specimen.41, 42 For further details about the primer and assay design: see Supplementary Information. Primer sequences are listed in Table I.
PCR products were detected using an enzyme immunoassay (EIA) as described previously43 with a HPV16 E6*I specific probe (Table I) according to a method that was described previously.33 Samples were scored positive when the EIA signal was above the threshold value of 3 times the average of the EIA signals of 4 negative PCR controls. RNA from cells of the HPV16 containing SiHa cell line that were formalin-fixed and embedded in paraffin, was run in parallel as positive control. For each clinical sample a parallel RT-PCR without reverse transcriptase was used as a control for possible amplification of contaminating HPV16 DNA. RNA integrity as assayed by detection of β-glucuronidase transcripts.
All results are summarized in Table II. Three groups of carcinomas have been tested, divided according to the HPV DNA and E6/E7 mRNA status as determined in frozen tissue.22 In general, all tests scored positive in the carcinomas of the D+/R+ group and negative in the D−/R− group. The D+/R− group showed in most cases negative test results and was also negative for E6/E7-expression in frozen tissue. For these reasons this group was scored as being HPV-negative and was included as such in the calculation of sensitivity and specificity.
Table II. Different Methods for Detecting Clinically Relevant HPV Infections in Paraffinembedded Head and Neck Carcinomas and Patient Sera
OC, oral cavity; OP, oropharynx; NE, not evaluable (some tumors could not be analyzed reliably, as the tissue section did not contain enough neoplastic cells); NA, not available for serologic detection of HPV16 L1, E6 and E7 antibodies.
p16 IHC staining intensity: 0–3 and the percentage of tumor cells positively stained.
The cut-off value to define HPV-antibody-positive sera was calculated separately for each antigen as the median of the specific absorbance values of control sera from individuals who did not have cancer plus three standard deviations excluding positive outliers, as described elsewhere.44 The samples were scored positive when the optical density (OD)-value was above the following levels: 16L1: 100; 16E6: 150 and 16E7: 60 and are in bold.
(+) = Positive or (−) negative respectively for E6*I expressing according to the criteria as described in the material and methods.
Viral DNA copy number per cell.
FISH staining intensity 0–3 was scored according to Cooper et al.37 Tumors are grouped on basis of the presence of HPV DNA and HPV E6/E7 transcripts as determined in the frozen samples (gold standard).
HPV DNA- and E6/E7 RNA-positive (D+/R+).
HPV DNA-positive, but E6/E7 RNA-negative (D+/R−).
Sensitivity: Relative number of positive samples detected in the HPV D+/R+ group.
Specificity: Relative number of negative samples detected in the HPV DNA+/R− and HPV-negative group together. Sera results: positivity was scored when any of the three markers was scored positive.
Of all PCR-based detection methods in FFPE specimen, only E6* mRNA detection showed both a sensitivity and specificity of 100%. Typical examples are shown in Figure 1. FISH analysis showed to be very specific, but was less sensitive: In 10 out of 12 HPV D+/R+ tumors HPV16 DNA was detected. In all positive tumors a punctuated hybridization pattern was observed consistent with viral DNA integration, although some tumors in addition showed patterns suggestive of the presence of episomal viral DNA as well (Fig. 2). Three samples could not be analyzed, because of a high background. Between the 3 investigators, there was total agreement on FISH scoring in 43 of the 45 cases (96%). When the interobserver agreement was analyzed, an average κ score of 0.85 was observed. This can be interpreted as a very good strength of agreement.38
All HPV D+/R+ tumors showed expression of p16 by immunohistochemistry. One sample could not be analyzed as the tissue section appeared not to contain neoplastic cells anymore. Between the 3 investigators, there was full agreement on p16-IHC scoring, with respect to positive staining intensity (graded 0–3 proportional to staining intensity) in 34 of the 44 cases (77%), with an average κ score of 0.81. This can be interpreted as a very good strength of interobserver agreement.38 Figure 3 shows typical examples of tumors with different staining intensities.
We measured HPV16 L1, E6 and E7 antibody levels above the cut-off values in the serum of patients in the HPV D+/R+ group, the HPV D+/R− and the HPV-negative group (Table II). Regarding the E6 antibodies, a statistically significant difference (p < 0.05, Fisher's exact test) was observed between the HPV D+/R+ group and the HPV D+/R− group as well as the HPV-negative group. The significant difference is based on the number of samples that were above the cut-off level, defined as specified in the legend of Table II.
Since all single methods showed limitations with respect to sensitivity and specificity, and E6* mRNA detection is not yet available for HPV-types other than 16, combinations of high throughput methods were considered. Our results showed that when p16 immunochemistry is followed by GP5+/6+-PCR on the p16-positive cases, 100% sensitivity and specificity might be reached (Fig. 4).
The reported large variation of HPV prevalence in HNSCC has complicated understanding of the role of HPV in this tumor type.4, 10, 16, 17 The high sensitivity of the widely used PCR methods for HPV-DNA amplification is a likely explanation for this variation, and a HPV DNA-positive assay may not always reflect biologically meaningful viral involvement.18 Recent findings point out that the only way to support a conclusive viral involvement is measuring levels of E6/E7 RNA, thus far necessitating frozen tissue material.1, 2, 26, 32 We aimed to find a sensitive and specific high throughput HPV detection algorithm for FFPE specimen.
The D+/R− group was included with an unknown definitive HPV status and the present results were interpreted to be additional arguments to consider this group of tumors HPV negative and to include this as such in the test performance assessment. This decision was motivated initially by the argument that E6 RT-PCR was negative in frozen tissue. All other data supported this decision: most tests showed negative results, these tumors showed genetic profiles indistinguishable from HPV negative tumors,1, 2 and the GP5+/6+-PCR signal intensities in the frozen tissues, the starting situation, were just above background. Regarding this last item, details on how this relatively low level was detected and interpreted has been mentioned in a previous report.22 Nevertheless, we consider samples that are positive for the DNA assay but negative for the E6/E7 assay as false positive. The presence of HPV DNA in these cases is not relevant for the carcinogenic process. This is supported by all other assays that we performed. How the presence of HPV DNA should be explained is not easy to answer. HPV DNA could be present in the saliva or mucosa without influencing the physiology of the host. It has to be stressed that no matter how this virus ended up in these samples, no viral oncogene expression was detected, making it unlikely that the virus is biologically active in carcinogenesis. The E6*I mRNA detection designed for FFPE material showed a hundred percent sensitivity and specificity and seems very well applicable for large retrospective studies, i.e. the readout is performed by PCR-EIA resulting in a basically dichotomous output. This assay also does not require laborious microdissection. A limitation is that the HPV-type should be known, and similar assays for types other than 16 needs to be developed. Nevertheless, this novel assay may already solve current clinical controversies about HPV involvement.45, 46
All methods based on PCR amplification of viral DNA showed limitations with respect to sensitivity and specificity. As expected, the GP5+/6+ PCR assay detected HPV in all tumors of the D+/R+ group. However, only a single case was GP5+/6+-PCR positive in the D+/R− group, while the corresponding frozen samples were all positive, although with relative low values,22 demonstrating that HPV is more readily detectable in frozen tumor tissue than in paraffin embedded tissue, which is not unexpected. PCR amplification is more efficient on frozen than on FFPE material, since it is known that the fixation procedure leads to DNA fragments that are often shorter than 200 bp.47 In the group of 24 tumors that were HPV-negative on basis of the frozen biopsy, 3 corresponding FFPE tumor samples were GP5+/6+ PCR positive. An explanation for this phenomenon might be related for instance to tumor heterogeneity and sampling error. The pieces that were selected for cryopreservation were often small biopsies, while the FFPE samples encompassed transsectional cuts through the whole tumor. It has to be emphasized that the E6/E7 mRNA expression remained negative in these samples, indicating that the virus was not biologically active and it demonstrates that PCR-based HPV DNA detection methods can lead to falsely positive results in terms of biologically active HPV detection.
Quantitative assessment of the viral copy number per cell might improve specificity, but resulted in 1 false negative observation, and it is very labor-intensive as the neoplastic cells need to be enriched by microdissection.17
FISH allows for the direct visualization of down to 1 copy of viral DNA per cell and in addition the discrimination between integrated and replicative (episomal) HPV as a punctuated or diffuse hybridization signal, respectively.7 It is unclear why FISH was unable to identify the virus in the D+/R+ cases 2 and 11. The viral load values were relatively low in these cases, suggesting a low viral copy number. In a second instance, sample 11 was repeated on a different section and came out positive. This outcome was not included in the results as we decided to present the first unbiased results. Thus, to increase sensitivity of FISH one could think of testing multiple sections to minimize a possible sampling error. The sensitivity of FISH was 85% (10/12), but the specificity was optimal (100%) in our data set.
Another potentially interesting method, exploiting a surrogate marker to identify clinically relevant HPV infections, is detection of antibodies to E6 and E7 in sera, chosen for its easy applicability and used in several large clinical studies with cervical cancer patients.48, 49, 50 We exploited the potential value of serology and were able to analyze serum samples of most of the patients. There was indeed a significant difference between the HPV D+/R+-group and the HPV D+/R− group, in particular for antibodies against HPV16 E6. Highest sensitivity was reached with positive serology with any of the 3 antibodies (91%), but the specificity was then limited (74%).
A second method using a surrogate marker is p16 over-expression as it is strongly related to active HPV infection based on the concept that functional inactivation of Rb by E7 induces p16 up-regulation. P16 immunohistochemistry is a relatively standardized technique and easy applicable on FFPE samples. Analysis of p16 over-expression can have clinical value21 and although we and other investigators found an association with viral oncogene activity,7, 51 p16 overexpression is certainly not limited to HPV D+/R+ samples only. Some cases in the HPV D+/R− and the HPV-negative group showed very high p16-expression and these were further analyzed in detail for the possibility that we failed to detect the virus by GP5+/6+ PCR due to viral integration into the host genome in the L1 gene. This was performed by type-specific HPV DNA PCR using primers in the E6/E7 region that also enables detection of all high-risk HPV types when integrated. However, these assays did not yield positive results, indicating that these cases should be considered as truly false-positive cases for p16 immunostaining.
As we found that each single assay seems to meet limitations, which is in line with observations of others,52 we decided to investigate algorithms based on the combination of assays. We could extract an algorithm with a satisfactory performance and allowing high-throughput analysis (Fig. 4). P16-immunstaining is upfront and when positive staining is observed GP5+/6+ PCR is used for confirmation. In our series the sensitivity and specificity of this approach is 100%. P16 immunohistochemistry can easily be combined with standard histology when a hematoxylin/eosin (HE)-stained tissue section is prepared for examination by a pathologist. Preselection by p16-staining reduces the workload and the combination gives a dramatic decrease of the number of false-positive observations by either of these assays. On the basis of the test outcome and the easy applicability we judge this as a reliable algorithm for HPV detection in FFPE specimen. One possible limitation of this algorithm is that both assays may yield a false-positive result in the same sample. However, based on the presently available data the chance that this occurs is around 2%. It should be emphasized that the proposed algorithm should be validated in subsequent studies, particularly in multi-center designs. Our data clearly show that HPV DNA detection by PCR only overestimates the number of samples with a clinically relevant infection.
In conclusion, we propose an algorithm for detecting a clinically relevant HPV infection, applicable for high-throughput analysis of paraffin embedded material, using a combination of p16 immunohistochemistry and GP5+/6+ PCR. This algorithm may provide the opportunity to assess the role of HPV in large retrospective studies, and could be helpful to select patients who may benefit from immunotherapeutic strategies.
Authors thank Prof. Dr J.J. Manni for continuous support, and Dr. A.H.N. Hopman for support and evaluation of the FISH experiments. Authors also thank Ms. J. de Vries for help with the p16-immunostaining and Mr. H. Verdurmen and Ms. N. Fransen-Daalmeijer for their assistance in the PCR-based detection methods.