Novel biomarker panel predicts prognosis in human papillomavirus-negative oropharyngeal cancer

An analysis of the TAX 324 trial




New treatment strategies for locally advanced head and neck squamous cell carcinoma combine induction chemotherapy and chemoradiation. Identifying the predictors of outcome in sequentially treated patients is critical for focusing therapeutic research. In this analysis, the authors evaluated human papillomavirus type 16 (HPV-16) status and the expression levels of a defined set of biomarkers to identify predictors of response to this treatment modality.


In total, 114 patients with oropharyngeal cancer (OPC) who were treated on the TAX 324 trial (cisplatin and fluorouracil with or without docetaxel in patients with locally advanced head and neck squamous cell carcinoma) had pretreatment biopsy specimens that were evaluable for HPV-16 DNA and immunohistochemical expression of the following biomarkers: beta-tubulin II (βT-II), glutathione S-transferase (GST-π), p53, and B-cell lymphoma 2 (Bcl-2). Patients were categorized into risk groups based on their HPV status and biomarker expression levels.


Patients with high-risk OPC were defined by HPV-negative status and either elevated expression of βT-II or levels of at least 2 of the other 3 adverse markers (elevated GST-π, elevated p53, or low Bcl-2). All other HPV-negative patients were categorized as moderate risk. In total, 55 patients were HPV-positive, and 59 patients were HPV-negative, with 34 were categorized as high risk and 25 categorized as moderate risk. The median survival for HPV-positive patients was not reached. The median survival was 44.2 months for moderate-risk patients (95% confidence interval, 20.9 months to not reached) and 12.1 months for high-risk patients (95% confidence interval, 7.5-19.7 months). The 24-month survival rate was 89% for HPV-positive patients, 67% for moderate-risk patients, and 29% for high-risk patients (P < .0001).


The molecular data set in this study readily differentiated between 2 distinct groups of patients with locally advanced, HPV-negative OPC. This risk-stratification strategy may serve as a guide for treatment selection. Cancer 2012;. © 2011 American Cancer Society.


Head and neck cancers account for 3% of all cancers in the United States. In 2009, approximately 48,010 new cases were diagnosed, and 11,260 patients died of the disease.1 Long-term survival for patients with head and neck squamous cell cancer (HNSCC) has improved over the past 30 years, particularly for those with cancers of the oropharynx. This improvement is caused by several factors, including the development of more effective treatment strategies, new understanding of the mechanisms of carcinogenesis, and a shift in the epidemiology of HNSCC.

Traditionally, concurrent chemoradiotherapy formed the cornerstone of management for the majority of patients with HNSCC with locally advanced, regionally confined (M0) disease. However, long-term survival for these patients remained poor at approximately 50%.2 Considerable efforts have been made to improve outcomes with novel chemotherapeutic agents, radiotherapeutic techniques, and multimodality strategies. The armamentarium of active chemotherapeutic agents in HNSCC was expanded beyond platinum compounds and fluoropyrimidines to include taxanes and biologic agents. More important, recent studies have demonstrated a significant benefit for induction chemotherapy followed by chemoradiation, with salvage surgical resection reserved for poor responders.3-6 One of the landmark studies was the TAX 324 trial, which was a randomized phase 3 trial in which patients received induction chemotherapy either with cisplatin and fluorouracil alone (PF) or with PF plus docetaxel (TPF) followed by radiotherapy and concomitant carboplatin. The results demonstrated a significant benefit for the induction chemotherapy strategy.5

Research efforts in the past 10 years also have brought forth new insights into the molecular changes underlying the development HNSCC. High-risk human papillomavirus (HPV), and specifically HPV type 16 (HPV-16), is implicated as the causative factor in a large proportion of HNSCCs,7 especially those of the oropharynx.8 HPV-related cancers have a phenotype9 and a genotype10 distinct from those of HNSCCs related to tobacco and alcohol. They also behave differently from the latter group of cancers. Although patients with HPV-related HNSCC tend to present with locally advanced disease, they respond well to treatment and have a better prognosis.11-13 Furthermore, epidemiologic studies indicate that the prevalence of HPV-related HNSCC is on the rise in Western countries.14-16

With the knowledge that HNSCC is a biologically heterogeneous disease comes the recognition that traditional prognostic markers, such as tumor extent and size, may be inadequate for the risk stratification of patients with HNSCC. There is a growing need for clinically relevant and practical biomarkers that will allow the appropriate selection of high-risk patients for more aggressive treatment strategies. Previously, as part of the TAX 324 study, we evaluated and confirmed the prognostic significance of a panel of rationally selected biomarkers in HNSCC, which included beta-tubulin II (βT-II), glutathione S-transferase (GST-π), p53, and B-cell lymphoma 2 (Bcl-2).17 In the current study, we correlated HPV status and expression of the same biomarkers with clinical outcome data as of December 2008 of in patients with oropharyngeal cancer (OPC) from the TAX 324 trial.5 Then, we developed a risk model to stratify HPV-negative patients into subsets of relatively good and poor responders to sequential treatment. Such information will allow us to move toward a risk-based treatment paradigm in the future.


Patient Samples

Of the 539 intent-to-treat (ITT) patients who were enrolled on the TAX 324 trial, paraffin sections from pretherapy biopsy specimens were available in 265 patients (from both trial arms); and, of these, 114 patients with OPC had complete data available for the evaluation of HPV status and biomarker expression levels. Patients were selected prospectively and met predetermined selection criteria before entry, as previously described.

HPV Analysis

Sections of pretreatment paraffin-embedded biopsy specimens were deparaffinized in xylene. The tissue was scraped with a sterile scalpel into a microcentrifuge tube, rehydrated with graded ethanol washes, and air dried. DNA extraction was performed using the QIAamp DNA Micro Kit (Qiagen, Valencia, CA), following the instruction manual protocol for laser-microdissected tissue. DNA was quantified using the Quant-iT dsDNA Assay Kit, (Invitrogen, Eugene, OR) and stored at −80°C. Working 0.5-ng/μL dilutions were prepared from stored aliquots. Sufficient DNA for polymerase chain reaction (PCR) analysis was recovered in 265 of 270 specimens. PCR was performed for the E6 (forward, AAACTAAGGGCGTAACCGAAA; reverse, TAGTTG TTTGCAGCTCTGTGC) and E7 (forward, ACAAG CAGAACCGGACAGAG; reverse, GATGGGGCACA CAATTCCTA) genes of HPV-16. These primers are type-specific for HPV-16. Forty cycles of standard, 3-step PCR (annealing temperature, 55°C) were performed on 0.5 ng of DNA. Multiple negative controls (without DNA) were included in every PCR run. Products were observed on 1.7% agarose gels that were stained with GelStar nucleic acid stain (Cambrex, Rockland, MD). Only specimens that were positive for both genes were scored HPV-16 positive. Specimens that were positive for 1 gene but negative for the other gene were excluded. If there was ambiguity in interpreting the expression of either gene, then both genes were amplified again from a freshly diluted DNA aliquot using a different primer set (overlapping but not nested with the original primer set). Specimens that remained ambiguous were excluded. HPV-16 status could be scored for 237 of the 265 specimens with sufficient DNA. In a validation series of 49 repeat reactions, the error rate was 0% for the E6 gene and 1% for the E7 gene.


Immunohistochemical analysis of putative prognostic markers was performed using the commercially available antibodies GST-π (clone 353-10; Dako, Carpinteria, CA), Bcl-2 (clone 124; Dako), p53 (clone BP53-12; Zymed Laboratories, Inc., South San Francisco, CA), and β-tubulin-II (clone JDR 3B8; Biogenex, San Ramon, CA). The selection of the βT-II antibody was based on a pilot study in our laboratory comparing the prognostic significance of βT-II, III and IV expression in an unrelated sample set. The selection of markers for analysis was based on the state of the literature at the time the trial began in 1999, the availability of antibodies, and our preliminary testing of available β-tubulin antibodies. Immunohistochemistry for all antibodies was performed as previously described.18

Sample Scoring

Samples were scored by a pathologist who was blinded to all clinical and outcome data. Samples were scored for the percentage of tumor cells with positive staining as well as the intensity of staining according to the method described by Allred et al.19 For subsequent statistical analysis, staining intensity scores below the median were considered low, and scores above the median were considered high.

Data Collection

A minimum of 5 years of follow-up data as of December 1, 2008 were collected on patients who were treated on the TAX 324 trial. Information was obtained through a questionnaire that was approved by the Institutional Review Board. Endpoints for follow-up included overall survival (OS), progression-free survival (PFS), locoregional failure, and distant metastasis.

Statistical Analysis

Among the 539 ITT patients, 114 patients (21.1%) with OPC had identified HPV status as well as complete data on all 4 tumor markers. These 114 patients represent the tumor marker study population. The remaining 425 patients were excluded from the tumor marker analysis. Comparison of patients' baseline demographic and clinical characteristics demonstrated that the tumor marker population was representative of the ITT population (Table 1). There were no statistically significant differences in OS or PFS between the tumor marker population and the ITT population. The Kaplan-Meier survival distribution functions were estimated for different categories of patients. The estimated time-to-event functions (OS, PFS) were compared using the log-rank test at the .05 level of significance. A Cox proportional hazards model was used to analyze survival data (OS, PFS) and to assess the effect of the plausible risk factors on the hazard function. The levels of biomarkers identified as positive or elevated versus negative were specified by the plan in place before data analysis. Elevated levels of βT-II, GST-π, and p53 and a zero level of Bcl-2 were defined as positive and in an adverse state. The number of plausible predictors in the regression model was limited by the number of actual events (ie, death, progression). The adequacy of the model was assessed with Martingale residuals. Predictive accuracy of the assigned risk category based on HPV status and biomarkers level was characterized using the receiver operating characteristic (ROC) curve associated with fitting the Cox model. The area was integrated using the distribution of survival time to obtain a concordance summary.20

Table 1. Baseline Characteristics of the TAX 324 Intent-To Treat Population and Subsets of All Patients With Oropharyngeal Cancer and Those With Complete Biomarker Data Available
CharacteristicAll, n = 539OPC/Markers, n = 114All OPC, n = 287
  1. Abbreviations: OPC, oropharyngeal cancer.

 Men451 (83.7)92 (80.7)239 (83.3)
 Women88 (16.3)22 (19.3)48 (16.7)
 Black52 (9.7)8 (7)21 (7.4)
 White456 (85.1)102 (89.5)250 (87.7)
 Other28 (5.2)4 (3.5)14 (4.9)
Age, y   
Anatomic site   
 Hypopharynx80 (14.8)  
 Larynx98 (18.2)  
 Oral cavity74 (13.7)  
 Oropharynx286 (53.1)114 (100)287 (100)
Clinical stage   
 III96 (17.8)16 (14)38 (13.3)
 IV366 (67.9)82 (71.9)203 (73.1)
 IVA61 (11.3)13 (11.4)29 (10.1)
 IVB15 (2.8)3 (2.6)10 (3.5)
 Missing1 (0.2)  


Our previous study demonstrated that elevated expression levels of GST-π and βT-II were associated with adverse OS. Elevated expression of p53 was associated with a trend toward diminished OS, and expression of Bcl-2 was a favorable prognostic indicator with borderline significance.17

Given the recent information on the prognostic importance of HPV, we performed a retrospective analysis of the TAX 324 trial samples and demonstrated that patients with HPV-positive HNSCC had a markedly superior prognosis compared with HPV-negative patients. The median OS for patients with HPV-positive HNSCC was not been reached, and the 25th quartile was 77.4 months; whereas the median OS for patients with HPV-negative HNSCC was only 19.7 months (95% confidence interval [CI], 12.4-34.9 months), and the hazard ratio (HR) was 5.3 (95% CI, 2.8-9.9; P < .0001). The 48-month survival rates were 81% and 36%, respectively.21, 22

Because HPV-positive patients clearly had an excellent prognosis, we investigated whether these biomarkers could stratify patients in the HPV-negative group according to risk. The molecular markers β-TII, GST-π, p53, and Bcl-2 were used to define high-risk and moderate-risk, HPV-negative patients. We defined high-risk patients as those who had elevated expression of βT-II or adverse findings for 2 of the other 3 markers (elevated GST-π, elevated p53, and negative Bcl-2). Conversely, we defined moderate-risk patients as those who demonstrated low expression of βT-II and had no more than 1 other marker with an adverse finding. For patients with HNSCC who were at moderate risk according to this definition, the median survival was 44.2 months, and the lower boundary of the 95% CI was 20.9 months. In contrast, for patients in the high-risk group, the median survival was 12.1 months (95% CI, 7.5-19.7 months). The 3-year survival rate for the high-risk group was only 26% compared with 54% for the moderate-risk group and 85% for the HPV-positive group (Fig. 1, Table 2). The HR for HPV-positive patients was 0.30 (95% CI, 0.14-0.62) compared with the moderate-risk patients and 0.16 (95% CI, 0.08-0.30) compared with the high-risk patients. The HR for moderate-risk patients compared with high-risk patients was 0.47 (95% CI, 0.25-0.88). All HRs differed significantly between the 3 risk groups in pairwise comparisons (Table 3). These results clearly differentiate patients with HPV-negative HNSCC who will fare reasonably well with a sequential treatment plan combining induction chemotherapy followed by chemoradiation as well as a group that will fare very poorly with that approach.

Figure 1.

Kaplan-Meier overall survival (OS) functions are illustrated. Patients are grouped based on their human papillomavirus (HPV) status and a biomarker-defined risk category. The median survival for HPV-positive (HPV+) patients has not been reached, and the 25th OS quartile is 77.4 months. The median OS for patients in the high-risk category (ie, either with elevated β-tubulin II expression or with elevation in 2 of 3 the other markers) is 12.1 months (95% confidence interval, 7.5-19.7 months). The median OS for patients in the moderate-risk group is 44.2 months (the lower bound of the 95% confidence interval is 20.9 months). Patients' survival experiences differed across these 3 groups (P < .0001; log-rank test).

Table 2. Estimated Overall Survival Parameters
 Median OS (95% CI), moOS Rate
Group25th Quartile50th QuartileAt 24 MonthsAt 36 Months
  1. Abbreviations: CI, confidence interval; HPV, human papilloma virus; NR, not reached; OS, overall survival.

HPV-positive77.4 (32.1 to NR)NR0.890.85
Moderate risk20.2 (5.2-34.8)44.2 (20.9 to NR)0.670.54
High risk 6.3 (2.7-8.2)12.1 (7.5-19.7)0.290.26
Table 3. Comparison of Hazard Ratios of Death Between the 3 Risk Groups
Groups ComparedaHR95% CIP
  • Abbreviations: CI, confidence interval; HPV, human papilloma virus; HR, hazard ratio.

  • a

    For each group comparison of the hazard of death (overall survival), the supposedly “worst” category of patients is used as the reference category. For instance, the hazard of death for HPV-positive patients is approximately 30% of the hazard for the moderate-risk subgroup; or the chance of dying is >3.3 times greater for a patient in the moderate-risk group than for a patient in the HPV-positive group.

HPV-positive vs moderate risk0.300.14-0.62.001
HPV-positive vs high risk0.160.08-0.30<.0001
Moderate risk vs high risk0.470.25-0.88.02


Prior studies by our group and others have demonstrated that HPV-positive and HPV-negative OPC have distinct biologies and that positive HPV status confers an enormous prognostic benefit to patients. The dismal prognosis for the HPV-negative population warrants a refocusing of research on this distinct disease to better understand its biology, to develop appropriate biomarkers for directed therapy and research, and to improve its treatment. In the current study, we investigated the prognostic value of 4 molecular markers (β-TII, GST-π, p53, and Bcl-2) in patients with locally advanced, HPV-negative HNSCC. The outcome data were obtained from a uniformly, prospectively selected and prospectively treated cohort of patients in the TAX 324 trial, in which patients received induction chemotherapy with either PF alone or TPF followed by radiotherapy with concomitant carboplatin.5 Our results indicated that HPV-negative patients can be broken into 2 distinct prognostic groups based on expression levels of defined biomarkers. This allowed us to define a group of patients with high-risk, HPV-negative HNSCC as those with high expression of β-TII or with adverse findings for at least 2 of the 3 other prognostic markers.

The 4 biomarkers that we studied comprise is a novel panel of molecular biomarkers to be used in risk stratification of patients with HNSCC who receive sequential therapy. It has been demonstrated that 3 of these markers have predictive value in terms of response to chemotherapy and/or radiotherapy. Previous work from our laboratory identified β-TII as a negative prognostic factor in HNSCC. An earlier analysis of the TAX 324 study suggested a role for βT-II in both cisplatin resistance and taxane resistance.17 We also previously demonstrated that overexpression of GST-π had strong negative prognostic value in patients who were receiving cisplatin-based induction chemotherapy,18, 23 and our subsequent work demonstrated that phase 2 enzymes in the glutathione pathway mediate resistance to cisplatin.24 The p53 tumor suppressor gene is mutated in approximately 50% of patients with HNSCC.25 Its mutation confers resistance to both chemotherapy26-28 and radiotherapy.29 The Bcl-2 family of genes regulates apoptosis. Previous studies from our laboratory indicated that the expression of Bcl-2 was associated with a favorable prognosis, consistent with other studies published in the literature.30

HPV-related HNSCCs consistently have been associated with a favorable prognosis in multiple studies.11-13, 31-37 In HPV-related HNSCC, p53 protein typically is wild type, but its presence is low because of the activity of the HPV E6 protein, which targets p53 for ubiquitination and degradation.38 This characteristic of HPV infection may lead to an HPV-related tumor's greater sensitivity to radiotherapy,39 because functional p53, even at a low expression level, allows radiation-induced apoptosis.40 Two prospective clinical trials also have indicated that HPV-positive tumors have an improved response to chemotherapy compared with HPV-negative tumors.12, 35 Current studies are evaluating whether such patients are served better by less aggressive, organ-sparing approaches.32, 41 In contrast to the multiple lines of data indicating a divergent prognosis based on HPV status in patients with OPC who receive chemoradiotherapy, the impact of HPV status on patients who undergo with primary surgery is less clear. Two studies indicated that HPV status also conferred a favorable prognosis in patients with OPC who underwent surgery.33, 42 However, a recent study from the Mayo Clinic demonstrated that HPV status did not impact the prognosis of patients who presented with surgically resectable cancers of the oropharynx.43 In the latter study, the association between HPV status and OS had an HR of 1.02 (95% CI, 0.53-1.99). Survival in these resectable patients was high and was not altered significantly by HPV status.

In contrast, the impact of HPV on survival in the current study was profound. The HR for survival in our HPV-negative, high-risk cohort was an impressive 0.16 (95% CI, 0.08-0.30) compared with the HPV-positive group. We also used a Cox model and area under the ROC curve (AUC) analyses to assess the predictive performance of HPV status based on risk categories in 114 patients (“HPV-positive,” “high-risk,” and “moderate-risk”). The estimated AUC of 0.68 was between a value of 0.5 for random prediction and 1.0 for perfect separation. The AUC for the Gail model, as a reference in invasive breast cancer, was 0.60.44

Our current results clearly stratify patients with locally advanced OPC into 3 risk groups based on biomarkers and not on clinical factors, such as smoking status. Patients with HPV-related HNSCC fare extremely well. Patients with moderate-risk, HPV-negative disease fare reasonably well with sequential therapy, as described in the TAX 324 trial. However, high-risk, HPV-negative tumors appear to be resistant to chemoradiotherapy, and patients with such tumors are not served well by this approach. Surgery may be an alternative to consider for selected patients in this group. For example, patients who have tonsil cancer with low tumor classification and high lymph node status who fall into the poor-risk category according to our current definition may be good candidates for tonsillectomy and neck dissection followed by chemoradiotherapy. Although the treatment of head and neck cancer has moved away from the primary surgical management of locally advanced OPC, the very poor survival in the high-risk population warrants a new look at that important issue.

This risk-stratification model should be validated in an independent data set. If the findings presented here are duplicated, then they could form the basis for a prospective clinical trial. Our data also support the examination of less intense therapy for favorable-risk, HPV-positive patients in a prospective trial setting.

We believe that this straightforward panel of biomarkers can more precisely and accurately predict clinical outcomes and guide therapeutic choices. The process of tumorigenesis is complex and involves multiple molecular pathways. Their joint effects must be assessed to evaluate risk. We have selected biomarkers that represent independent and unrelated pathways based on the current available knowledge. We believe that this molecular marker set is an example of a test that can be both practical and clinically useful. Testing is easily accomplished on standard paraffin-embedded biopsy samples in a few days. This analysis is far less costly and simpler than deep sequencing genomic profiling. Our hope is that this will allow us to move toward a more individualized treatment paradigm in HNSCC.


This work was supported by Sanofi-Aventis US (the parent trial TAX324 was sponsored by Sanofi-Aventis Groupe), the State of Maryland Cigarette Restitution Fund Program, and the Orokawa Foundation.


The authors made no disclosures.