Molecular markers predict outcome in squamous cell carcinoma of the head and neck after concomitant cisplatin-based chemoradiation


  • Guido B. van den Broek,

    Corresponding author
    1. Department of Head and Neck Oncology and Surgery, The Netherlands Cancer Institute–Antoni van Leeuwenhoek Hospital, Amsterdam, The Netherlands
    2. Department of Otorhinolaryngology, Amsterdam Medical Center, Amsterdam, The Netherlands
    • Department of Head and Neck Oncology and Surgery, The Netherlands Cancer Institute-Antoni van Leeuwenhoek Hospital, Plesmanlaan 121, 1066 CX Amsterdam, The Netherlands
    Search for more papers by this author
    • Fax: +31-20-5122554.

  • Maarten Wildeman,

    1. Department of Head and Neck Oncology and Surgery, The Netherlands Cancer Institute–Antoni van Leeuwenhoek Hospital, Amsterdam, The Netherlands
    Search for more papers by this author
  • Coen R.N. Rasch,

    1. Department of Radiation Oncology, The Netherlands Cancer Institute–Antoni van Leeuwenhoek Hospital, Amsterdam, The Netherlands
    Search for more papers by this author
  • Nicola Armstrong,

    1. Department of Molecular Biology, The Netherlands Cancer Institute–Antoni van Leeuwenhoek Hospital, Amsterdam, The Netherlands
    Search for more papers by this author
  • Ed Schuuring,

    1. Department of Pathology, University Medical Center Groningen, Groningen, The Netherlands
    Search for more papers by this author
  • Adrian C. Begg,

    1. Department of Experimental Therapy, The Netherlands Cancer Institute–Antoni van Leeuwenhoek Hospital, Amsterdam, The Netherlands
    Search for more papers by this author
  • Leendert H.J. Looijenga,

    1. Department of Pathology, Erasmus Medical Center/Daniel den Hoed Cancer Center, Josephine Nefkens Institute, Rotterdam, The Netherlands
    Search for more papers by this author
  • Rik Scheper,

    1. Department of Pathology, Free University Medical Center, Amsterdam, The Netherlands
    Search for more papers by this author
  • Jacqueline E. van der Wal,

    1. Department of Pathology, University Medical Center Groningen, Groningen, The Netherlands
    Search for more papers by this author
  • Lorian Menkema,

    1. Department of Pathology, University Medical Center Groningen, Groningen, The Netherlands
    Search for more papers by this author
  • Paul J. van Diest,

    1. Department of Pathology, University Medical Center Utrecht, Utrecht, The Netherlands
    Search for more papers by this author
  • Alfons J.M. Balm,

    1. Department of Head and Neck Oncology and Surgery, The Netherlands Cancer Institute–Antoni van Leeuwenhoek Hospital, Amsterdam, The Netherlands
    2. Department of Otorhinolaryngology, Amsterdam Medical Center, Amsterdam, The Netherlands
    Search for more papers by this author
  • Marie-Louise F. van Velthuysen,

    1. Department of Pathology, The Netherlands Cancer Institute–Antoni van Leeuwenhoek Hospital, Amsterdam, The Netherlands
    Search for more papers by this author
  • Michiel W.M. van den Brekel

    1. Department of Head and Neck Oncology and Surgery, The Netherlands Cancer Institute–Antoni van Leeuwenhoek Hospital, Amsterdam, The Netherlands
    2. Department of Otorhinolaryngology, Amsterdam Medical Center, Amsterdam, The Netherlands
    Search for more papers by this author


Not all patients with squamous cell carcinomas of the head and neck (HNSCC) benefit from concurrent cisplatin-based chemoradiation, but reliable predictive markers for outcome after chemoradiation are scarce. We have investigated potential prognostic biomarkers for outcome in a large group of patients. Ninety-one tumor biopsies taken from consecutive HNSCC patients were evaluated for protein expression on a tissue microarray. Using immunohistochemistry, 18 biomarkers, involved in various cellular pathways were investigated. Univariable and multivariable proportional hazard analyses were performed to investigate associations between each individual marker and outcome. In addition, the global test was used to test all variables simultaneously and selected combinations of markers for an overall association with local control. Univariable analysis showed statistically significant increased relative risks of RB, P16 and MRP2 for local control and MDR1 and HIF-1α for overall survival. MRP2, MDR1 and P16 levels were positively associated with outcome whereas RB and HIF-1α had a negative relationship. Using Goeman's global testing no combination of markers was identified that was associated with local control. Grouping the markers according to their function revealed an association between a combination of 3 markers (P16, P21 and P27) and outcome (p = 0.05) was found. In the multivariable analysis, MRP2 and RB remained significant independent predictive markers for local control. This study describes the prognostic value of biomarkers for the outcome in patients uniformly treated with concurrent chemoradiation. MRP2 and RB were found to be associated with outcome in patients treated with concurrent chemoradiation. © 2009 Wiley-Liss, Inc.

Head and neck squamous cell carcinoma (HNSCC) is the fifth most common cancer in men.1 Over 70% of head and neck cancer patients present with advanced stage III and IV disease. Concomitant chemoradiation (CCRT) leads to improved local control (LC) and overall survival (OS) in advanced head and neck cancer compared with conventional radiotherapy,2–7 making this modality the most suitable curative treatment option in these patients currently. However, CCRT is not effective in all patients, and when unsuccessful, patients suffer the potential side effects and toxicities of chemotherapy (e.g., swallowing problems, hearing loss) and radiation therapy (e.g., mucositis, late toxicity). Therefore, identification of reliable outcome predictors in this setting is of clinical interest and especially important if alternatives such as surgery and postoperative radiotherapy or cetuximab with radiotherapy are possible.

Clinical variables have been intensively studied for prognostic accuracy. TNM classification is universally used as a staging system. However, in patients with advanced head and neck cancer it has been demonstrated that T and N stage do not have sufficient predictive value.8, 9 Tumor volume has been shown to be the most predictive factor for LC.8–12 For OS, tumor volume, comorbidity, lowest involved neck level and pretreatment weight loss have been found to be prognostic. Although tumor volume has strong prognostic value, we have demonstrated that tumors with the same volume can have different outcomes.13 These and other findings14 suggest that other biological factors are important in determining tumor response and knowledge of these and other factors could be helpful in clinical decision making. These can be hypoxia, repopulation rate and intrinsic sensitivities to radiation and cisplatin (e.g., repair pathways and drug pumps), amongst others.

Recent studies have described many biological markers correlating with outcome.13, 15–27 However, only a few have been tested for predictive accuracy after chemoradiation treatment. Such markers include those associated with cell cycle control (cyclin D1, retinoblastoma gene product (RB), P16, P21 and P27), apoptosis (TP53, MDM2, TP73 and BCL-2), growth regulation (EGFR), cyclooxygenase-2 (COX-2 and ki-67), focal adhesion signaling (cortactin), hypoxia (CA9, HIF-1α) and sensitivity to chemotherapy (XPA, MRP2 and MDR1).

Tissue microarray (TMA) technology in combination with immunohistochemistry has been demonstrated to be an appropriate tool to analyze the prognostic value of genetic abnormalities in a large number of tumor samples simultaneously.28–31 In this study we investigated the role of several biomarkers in predicting LC and OS in patients with squamous cell carcinoma after treatment with chemoradiation. These markers were chosen as the most promising individual markers in previous studies. We also conducted a review of the published literature to assess and compare the expression and prognostic value of biomarkers in patients with advanced head and neck cancer treated with radiotherapy, chemotherapy or a combination of both modalities. Our data suggest that only a few molecular markers known from current head and neck literature might play a role in chemoradiation resistance.

Material and methods

Patients and tissue samples

From 1997 until 2000, 104 consecutive patients with advanced squamous cell carcinoma of the oral cavity, oropharynx, supraglottic larynx and hypopharynx were treated in 2 chemoradiation trials (RADPLAT). Paraffin embedded biopsies of the primary tumor were available for immunohistochemical analysis in 95 of these patients.

The intra-arterial chemoradiation treatment has been described earlier.32, 33 Briefly, treatment consisted of 4 consecutive weekly selective intra-arterial infusions of cisplatin (150 mg/m2) simultaneous with intravenous sodium thiosulfate rescue combined with radiotherapy (2 Gy per day, 5/week × 7 to a total dose of 70 Gy) according to the RADPLAT protocol.34 Before the start of treatment, all patients signed an informed consent form approved by our institutional protocol review committee. Treatment response was evaluated 6 to 8 weeks after completion of radiotherapy by magnetic resonance imaging, followed by examination under general anesthesia. Thereafter, patients were subjected to regular outpatient follow-up and an annual chest X-ray. To decrease the chance of a late recurrence, all living patients had a minimum follow-up period of 2 years. Patient characteristics are detailed in Table I.

Table I. Patient Population
VariableNumberPercentage (%)
 Stage III78
 Stage IV8492
 Oral cavity1516


The markers have been chosen to reflect different biological processes that might be involved in the response to concurrent chemoradiation. We performed a literature search to compare expression and prognostic value of several molecular markers. The following search terms were used: squamous cell carcinoma, head and neck neoplasm, immunohistochemistry and outcome for each molecular marker. Studies describing less than 25 cases, any surgical treatment, esophageal or nasopharyngeal cancer were excluded. Based on these criteria, we selected the following markers: BCL-2,17, 26, 35–40 CA9,18–20, 41 COX-2,42 cyclin D1,17, 23, 35 EGFR,43, 44 HIF-1α,20, 45 ki-67,17, 21–23, 35, 38, 39, 42, 46, 47 MDM2,48, 49 MRP,24 P16,23 P21,23, 49–51 P27,23, 51 TP5316, 17, 21–25, 35, 36, 38, 39, 41, 42, 45, 46, 48–53 and RB.23 In addition, Cortactin, XPA, MDR1 and TP73 were chosen, because these genes were of interest to the authors.


TMAs were constructed as described by Chen et al.30 Briefly, an H & E slide from the tumor-embedded paraffin block was used to guide the sampling of morphologically representative regions of the tumor. To construct the TMA, 3 core tissue biopsy specimens (diameter: 0.6 mm) from the selected regions of the donor block were taken and precisely arrayed into a new recipient paraffin block. Each microarray block contained a maximum of 168 punches and 2 paraffin blocks were produced. After the construction of the array block, 5 μm paraffin serial sections were cut with a microtome using an adhesive-coated tape sectioning system (Instramedics Hackensack, NJ) of which one H & E staining was performed to verify histology.

Staining with antibodies was performed using standard methodologies previously described.30, 54, 55 In short, paraffin-embedded, formalin-fixed sections were deparaffinized and antigen retrieval was performed (Table II). After blocking endogenous peroxidases with 0.3% H2O2, the sections were stained for primary antibody diluted in 1% BSA-PBS. Secondary rabbit anti-mouse peroxidase or goat anti-rabbit peroxidase were precipitated using 3.3′ diaminobenzidine tetrachloride as a substrate and slides were counterstained using routine hematoxylin.

Table II. Details on Primary Antibodies Used for Immunohistochemical Stainings
AntigenAntibodySpeciesSourceLocalization of staining
  • 1

    Neomarkers, Union City, CA, USA.

  • 2

    Dako, Glostrup, Denmark.

  • 3

    Immunotech, Marseille, France.

Cyclin D1SP4RabbitNeoMarkersNuclear
Cortactin30/cortactinMouseBD Transduction LaboratoriesMembranous/cytoplasmic
COX233MouseBD Transduction LaboratoriesCytoplasmic
HIF-1α54/HIF-1αMouseBD Transduction LaboratoriesNuclear/membranous
BCL-2Bcl-2α AB-3MouseNeomarkers1Cytoplasmic
XPAXPA AB1MouseNeomarkersNuclear
MRP2M2III5MouseRik ScheperMembranous
MDR1JSB1MouseRik ScheperMembranous
MDM2Clone sample 14MouseNeomarkers1Nuclear
TP73P73 (H-79)RabbitSanta CruzNuclear

Immunohistochemistry scoring

TMAs contained 3 cores from each tumor biopsy taken from every patient before therapy. Immunohistochemical stainings were scored by 2 independent observers (G.B.vd.B, M.W.). In case of disagreement, after discussion with a pathologist (M.v.V.) the final score was determined by general agreement. Staining of all antibodies was evaluated for both positivity percentage and intensity independently. The score for positivity was the percentage of positive cells in increments of 5% and based on visual assessment. The final score was the percentage of positive cells averaged over 3 cores. Intensity scoring was performed using 4 categories (−; +; ++; +++). In case of heterogeneity, the most intense cores were scored.

Statistical methods

Reproducibility of the TMA scores was determined from the differences between the 3 cores and primarily expressed using the intraclass correlation coefficient (ICC). The ICC expresses the percentage of overall variance. Positivity scores, taken as continuous percentages, were used for the statistical analysis. In other words, no cut off values were used in scoring or to analyze expression. Time was calculated from start of treatment until local failure, end of follow-up or death. Correlations between variables were calculated with Pearson's correlation coefficient.

Because of the large number of variables (N = 18) compared to the number of events (N = 16), methods developed for cDNA microarrays were used. The global test of Goeman56 was used to test all 18 variables simultaneously or a combination of some related variables for an overall association with LC. Combinations of related variables were chosen according to known specific pathways: cell cycle control (cyclin D1, cortactin, RB and P27, P16, P21); apoptosis (TP53, TP73, MDM2, P21); hypoxia (HIF-1α, CA9) and chemotherapy sensitivity (XPA, MRP2, MDR1). Ten thousand permutations were used to calculate the p-value. The statistical package R (version 2.2.1 and 2.4.1) was used to perform the global test. Reproducibility, univariable and multivariable analyses of outcome and additional analyses were performed using the statistical package S+ (version 6.2 for Windows). Outcome curves were calculated using the Kaplan-Meier method.57 To visualize the significant outcome differences using Kaplan-Meier curves, the continuous positivity scores were converted to categorical scores (low/high). Cox proportional hazard analyses, both univariable and multivariable, were performed using the continuous positivity scores.58 Clinical factors, such as T-classification, N-classification and site were included in the multivariable analysis.


In 4 patients, the tumor tissue appeared to be of insufficient quality to be used in the analysis (defined as more than 20 missing values among the 36 final intensity and final percentage scores). These subjects were excluded from all analyses; the remaining 91 tumors were deemed to be of sufficient quality to be included in the analysis. Median follow-up for OS was 18 months (range, 1–58 months). The median age of the patients was 54 years (mean = 56; range, 29–78). During follow-up, 16 patients had a local recurrence and 53 patients died. Two-year LC and OS rates were 78% and 48%, respectively. Other patient and tumor characteristics are detailed in Table I.

Immunohistochemical scoring by intensity had poor reproducibility in general (ICC < 0.5). The reproducibility of cores scored with the positivity-method was moderate (ICC 0.5–0.8) to good (ICC > 0.8). Data for expression of the markers are shown in Table III. Several markers were expressed in many tumors (e.g., EGFR, HIF-1α, cortactin and RB), whereas others demonstrated immunoreactivity in only a limited number of tumors (e.g., BCL-2, MRP2 and P16). The correlation between the positivity scores for all pairs of markers was then calculated revealing no strong associations. The strongest association was between P21 and P27, where the Pearson's correlation coefficient was 0.54.

Table III. Marker Characteristics (Positivity Percentage)
VariableNumber of patientsMean (%)Standard deviation (SD)Coefficient of variation (CV)
Cyclin D19033260.79

The reproducibility of markers scored by the positivity-method was much better than the intensity-scoring method. However, since intensity scoring has been described for some markers,17, 42, 44 we performed univariable analyses for these markers: COX-2, P53, MDR1 and EGFR. No significant association was found between these markers and outcome (data not shown). For further analyses, we used the positivity scores only. From the univariable analysis, 3 markers were found to have a significant predictive value for LC: RB (p = 0.036), P16 (p = 0.008) and MRP2 (p = 0.007; Table IV). MRP2 and P16 expression had a positive association with LC (Figs. 1 and 2), whereas RB expression had a negative relationship with LC. According to the global test of Goeman, a combination of all markers did not show an association with LC. Several related groups of genes were then tested for prognostic value. The combination of P27, P21 and P16 was found to be significantly related to LC (permutation p-value = 0.03). MRP2, RB, P16, P21, P27 and clinical factors (T-classification, N-classification and site) were included in the multivariable analysis for LC. TNM stage was not included, because all patients with a local recurrence had stage IV disease. After the multivariable analysis, MRP2 and RB remained as independent significant variables and P16 was borderline significant (p = 0.06) for LC (Table V).

Figure 1.

Kaplan-Meier curves of 88 patients stratified for MRP2. Cut off value: 0%, expression (——) and no expression (········). Log-rank test: p = 0.03.

Figure 2.

Kaplan-Meier curves of 91 patients stratified for P16. Cut off value: 25%, high expression (——) and low expression (········). Log-rank test: p = 0.02.

Table IV. Univariable Analysis for Local Control
Variablep-ValueHazard ratio95% Hazard ratio
Table V. Multivariable Analysis for Local Control1
Variablep-ValueHazard ratio95% Hazard ratio
  • 1

    T-classification, N-classification, site, P21 and P27 were included in the model as well, but were eliminated in a backward manner.


To study interaction between markers, we split the patients into positive and negative subgroups for either P16 (marker for HPV) or P53 and looked at the Cox proportional hazard models for both LC and OS, including the interaction between the dichotomized P16 or P53 score and all other markers. In the P16 subgroup with zero or less than 1% expression, 14 of the 16 patients with a local recurrence were found, indicating that indeed, in this HPV negative subpopulation, most recurrences occur. For LC, Cycline D1 was borderline significant (p = 0.054) with a HR of 3. After correcting for multiple testing, this significance disappeared. For OS, no marker had a significant interaction with P16. For P53, the samples with local recurrence had positivity scores of either 0 or more than 85%. In these populations with either a very low or very high expression, most patients have a P53 mutation. For P53, there was a significant interaction with XPA (p = 0.05, HR = 5.1) for LC while MDR1 (p = 0.012, HR = 0.5) and COX2 (p = 0.03, HR = 1.8) were significant for OS, again, significance disappeared after adjusting for multiple testing. We note that the subgroup numbers are not large enough to give reliable results.

When OS was considered as the outcome, 2 markers as well as gender (p = 0.03; Table VI) were found to be significantly prognostic in univariable analysis: HIF-1α (p = 0.03), MDR1 (p = 0.04). MDR1 expression had a positive association with OS, whereas HIF-1α expression had a negative relationship with OS. HIF-1α, MDR1 and clinical variables were included in a multivariable analysis for OS. This analysis demonstrated that HIF-1α remained borderline significant as a prognostic marker for OS (p = 0.053, Table VII).

Table VI. Univariable Analysis for Overall Survival
Variablep-ValueHazard ratio95% Hazard ratio
Gender (male vs. female)0.030.480.250.92
Table VII. Multivariable Analysis for Overall Survival
Variablep-ValueHazard ratio95% Hazard ratio
N-classification (N0-1 vs. N2–3)0.031.391.041.85
T-classification (T3 vs. T4)0.091.300.961.77
 Oral cavity 1.00  


Several biomarkers have been described to be prognostic for outcome in squamous cell carcinoma of the head and neck,17, 35, 36, 59–62 although most reported studies included HNSCC carcinomas that were treated surgically and the findings in many of these studies are contradictory.63 In HNSCC cancer patients, chemoradiation is increasingly being used to treat locally advanced tumors. Therefore in this study, the prognostic value of 18 of the more promising biomarkers was investigated in tumors from HNSCC patients treated with concurrent chemoradiation. The markers were chosen because of their known role in chemosensitivity or radiosensitivity. Genes included those for cell cycle control, apoptosis, hypoxia, DNA repair, drug transport and growth signaling, all of which have been associated with drug or radiation response. Since there are a multitude of genes involved in these different pathways and processes, we chose representatives that had shown their value as predictors in previous studies, but had never been studied as a group, or in patients treated with one of the emerging current treatment standards of concurrent high dose cisplatin and radiation. A multivariable analysis showed that 2 biomarkers (RB and MRP2) were significantly and independently predictive for LC and 1 biomarker (HIF-1α) had prognostic value for OS. We also found that a combination of P16, P21 and P27 (cell cycle control) was significantly associated with LC, although P16 alone showed the strongest association.

The present study demonstrates the possible association between 2 markers (MRP2 and RB) and response to concurrent chemoradiation. MRP2 is a member of the ABC transporter family64, 65 which is able to export cisplatin out of the cell.66–68 It would therefore be expected that tumors showing overexpression of MRP2 might be clinically chemoradiation resistant. However, we found increased expression of MRP2 in the sensitive tumors. Correlations between MRP2 expression and resistance to cisplatin have been demonstrated in in vitro studies.68, 69 In addition, downregulation of MRP2 using siRNA resulted in increased sensitivity to cisplatin.70 In HNSCC patients treated with chemoradiotherapy, MRP2 expression was not associated with clinical outcome.24 However, few studies in HNSCC have been reported on the role of MRP2 in cisplatin resistance. In this study, we found increased rather than decreased MRP2 expression in sensitive tumors of patients treated with concurrent chemoradiation. Recently, Guminski et al.71 demonstrated that MRP2-mediated efflux is a determinant of cisplatin sensitivity in normal cells. However, a higher rate of MRP2 expression was described in platinum-sensitive ovarian carcinomas (55%) than in resistant cases (23%) in agreement with our findings. These findings contrast the in vitro observations.

Since tumor volume is a strong predictor of outcome8, 9 in advanced head and neck squamous cell carcinomas, we looked for a possible association between MRP2 levels and volume. However, mean tumor volumes were 43 and 37 cm3 for the MRP2 positive and MRP2 negative cases, respectively, which are not significantly different. Another reason for the association of MRP2 expression and sensitivity to chemoradiation treatment could be the granular cytoplasmatic staining in MRP2 positive tumor cells. It is possible that cells not showing membrane localization do not have the active efflux mechanism due to incorrect protein localization. A third explanation is that MRP2 is known to pump reduced glutathione and conjugates out of the cell.72 Tumors overexpressing MRP2 may therefore have lower glutathione levels, and since glutathione is the major radical scavenger protecting cells from oxidative DNA damage of the sort produced by ionizing radiation, this could render them more sensitive to radiation.73 Although we do not know exactly how high levels of MRP2 trigger resistance to cisplatin, our study indicates that MRP2 expression is correlated significantly with sensitivity to cisplatin-based treatment in head and neck carcinomas.

The role of RB in predicting outcome in head and neck cancer has been investigated mostly in patients treated with surgery.74, 75 To our knowledge, only one study has been performed in chemoradiation patients and showed no firm association with treatment response.23 Here we observed an inverse association with LC. Although this association was not as strong as for MRP2, patients with high RB expression had an increased probability of having a recurrence. A predictive value of P16 has been found in surgically treated patients,76–78 but we describe here an association between P16 and outcome in chemoradiation-treated patients. In our study, P16 was borderline significant in a multivariable analysis, independent of MRP2 expression and other clinicopathological factors. Patients with high P16 expression had an increased probability of LC compared to patients with low P16 expression. P16 is a negative regulator of the cell cycle by acting as an inhibitor of cyclin-dependent kinase 4 and 6-cyclin D complexes.79, 80 It has been hypothesized that cisplatin affects nuclear transport and stabilization of P16, which might partly explain the association with LC.81 It is also an indirect marker of HPV, being expressed in HPV induced cancers but not in most other SCC, and HPV expressing SCC tumors have been found to be more sensitive to treatment.82

HIF-1α was the only independent marker showing a strong trend related to OS in the multivariable analysis (marginally significance, p = 0.053). HIF-1α positive patients had worse survival than HIF-1α negative patients. This is in agreement with earlier published data in HNSCC. Aebersold et al.15 found a significantly higher OS in oropharyngeal cancer patients showing increased HIF-1α expression. HIF-1α is induced by hypoxia, subsequently regulating erythropoiesis, glycolysis and angiogenesis that may promote survival, invasion and metastasis.83–85 The stimulus for angiogenesis which promotes distant metastasis might explain why HIF-1α is associated with OS and not with LC.

The level of expression is potentially an important parameter for prognosis. Unfortunately, a search of the literature revealed a large variation in cut off values used for expression for most markers (summary in Table VIII): BCL-2 (13–32%), CA9 (27–58%), COX-2 (46%), ki-67 (24–67%), P21 (34–60%), P27 (37–45%) and TP53 (36–67%). This makes it difficult to compare studies. Ideally a cutoff value should separate resistant cases from sensitive cases, but the ideal cutoff value is usually not known. Most investigators chose cutoff values to separate the study into comparably sized groups. However, because resistant cases are often a small portion of the study group, choosing cutoff values that differentiates a small subpopulation from a larger group might also be acceptable. Furthermore, it might well be that a small subpopulation of tumor cells determines resistance to a therapy. To partially circumvent these problems, we analyzed expression as a continuous rather than dichotomous variable.

Table VIII. Literature Review: Prognostic Value of Markers for Outcome
MarkerAuthor  N  SiteTreatmentCut off (%)Expression (%)LCu/mOSu/m
  • 1

    LC defined as >50% tumor response.

  • 2

    Case-control study.

  • 3

    Intensity was scored.

  • 4

    Intermediate group.

  • LC, local control; OS, overall survival; nm, not mentioned; u, univariable analysis; m, multivariable analysis.

bcl2Trask et al. (Ref.26)47LarynxCT>5015No1 No 
Homma et al. (Ref.39)59LarynxCCRT>3012No Yesm
Gallo et al. (Ref.37)85AllRT>3024YesmYesm
Homma et al. (Ref.38)111AllCCRT>3013YesmNo 
Nix et al. (Ref.40)2124LarynxRT>532Yesmnm 
Fouret et al. (Ref.36)139AllCT>518Yes1mnm 
Ataman et al. (Ref.35)309AllRT>513Yesmnm 
Buffa et al. (Ref.17)402AllRT>513YesmYesm
CA9Kaanders et al. (Ref.19)38AllRTnmnmNo No 
Schutter et al. (Ref.18)67AllRT>1750No No 
Koukourakis et al. (Ref.20)75AllCCRTStrong staining27YesuYesu
Koukourakis et al. (Ref.86)198AllRT>1058YesmYesm
COX-2Cho et al. (Ref.42)123LarynxRTint(3)346No Yesm
Cyclin D1Rodriguez et al. (Ref.23)122AllCCRT>564No No 
Ataman et al. (Ref.35)309AllRTContinuous100No nm 
Buffa et al. (Ref.17)402AllRTContinuous80No Yesm
EGFRDemiral et al. (Ref.43)31LarynxRT>516Yesunm 
Gupta et al. (Ref.44)38OropharynxCCRTint(4)379No No 
HIF-1αKoukourakis et al. (Ref.20)75AllCCRT>3652No No 
Aebersold et al. (Ref.15)98OropharynxRT>094YesmYesm
ki-67Valente et al. (Ref.47)31Oral cavityRT>50nmNo nm 
Raybaud et al. (Ref.22)56AllRT>2032Yesmnm 
Homma et al. (Ref.39)59LarynxCCRT>5049No No 
Lavertu et al. (Ref.21)105AllCCRT>024No Yesm
Homma et al. (Ref.38)111AllCCRT>4067No No 
Rodriguez et al. (Ref.23)122AllCCRT>2053No No 
Cho et al. (Ref.42)123LarynxRT>1028No Yesm
Couture et al. (Ref.46)304AllRT>2059No No 
Ataman et al. (Ref.35)309AllRT>2053No nm 
Buffa et al. (Ref.17)402AllRT<20; 20–40; >4046; 32; 22No No 
MDM2Osman et al. (Ref.49)71AllCCRT>2074No Yesm
Friesland et al. (Ref.48)70TonsilRTint(4)3nmNo No 
MRPShiga et al. (Ref.24)68AllCT + RT>543No No 
p16Rodriguez et al. (Ref.23)122AllCCRT>5068No No 
p21Jeannon et al. (Ref.50)60LarynxRT>5058nm Yesu
Korkmaz et al. (Ref.51)68LarynxRT>1060No No 
Osman et al. (Ref.49)71AllCCRT>2054No No 
Rodriguez et al. (Ref.23)122AllCCRT>1034No No 
p27Korkmaz et al. (Ref.51)68LarynxRT>1037YesmNo 
Rodriguez et al. (Ref.23)122AllCCRT>2545No No 
TP53Raybaud et al. (Ref.22)56AllRT>1041Yesmnm 
Homma et al. (Ref.39)59LarynxCCRT>1059No No 
Jeannon et al. (Ref.50)60LarynxRT>2548nm No 
Narayana et al. (Ref.52)267LarynxRT>1046Yesunm 
Korkmaz et al. (Ref.51)68LarynxRT>1061No No 
Shiga et al. (Ref.24)68AllCT + RT>537No Yesm
Osman et al. (Ref.49)71AllCCRT>2049No Yesm
Friesland et al. (Ref.48)75TonsilRT>1055No No 
Bradford et al. (Ref.16)94LarynxCT + RT>057No No 
Koukourakis et al. (Ref.41)95All(C)CRT>2039No No 
Aebersold et al. (Ref.45)100OropharynxRT α CT>1067No No 
Narayana et al. (Ref.53)102LarynxRT>1037YesmNo 
Lavertu et al. (Ref.21)105AllCCRT>255No No 
Temam et al. (Ref.25)105AllCT>561No1 nm 
Homma et al. (Ref.38)111AllCCRT>1055No No 
Rodriguez et al. (Ref.23)122AllCCRT>1055No No 
Cho et al. (Ref.42)123LarynxRT>2036No No 
Fouret et al. (Ref.36)139AllCT>559No1 nm 
Couture et al. (Ref.46)304AllRT>1044YesmNo 
Ataman et al. (Ref.35)309AllRT<5; 5–75; >7550; 25; 25No nm 
Buffa et al. (Ref.17)402AllRTint(3)342No Yes4m
RBRodriguez et al. (Ref.23)122AllCCRT>1081No No 

Despite variations in cutoff values, less than 40% of published studies reported an association between a predictive molecular marker and outcome after chemoradiation (Table VIII). BCL-2 has been reported to be predictive for LC in more than half of all published studies. In 4 studies17, 35, 36, 40 the same cut off value (>5%) was used and a significant association between BCL-2 positivity and LC was observed. Using a cut off value of 30%, a significant association was also seen in 2 other studies.37, 38 The different cutoff values were not related to the different antibodies used. In our current study, if the expression of BCL-2 was categorized as positive or negative, with 5% taken as the cutoff value, the difference in LC rates was not statistically significant (78% vs. 61%; p = 0.39). Two other studies26, 38 also did not find an association with LC. Because of different reported expression and cutoff values, the value of BCL-2 as a tool for clinical decision making therefore remains weak, and the lack of predictive value found in our study cannot be regarded as surprising.

As there is no single biomarker that appears to have repeatedly strong predictive value, several investigators have tried to find a combination of markers with a stronger predictive value. Gallo et al.37 found a worse outcome in patients with mutated TP53 and positive BCL-2 expression in head and neck squamous cell carcinomas. De Schutter et al.18 demonstrated an association between LC and a combination of positive expression of CA9 and glucose transporter-I. We performed the global test of Goeman on selected groups of markers. One combination was found to be significantly associated with LC: P16, P21 and P27. However, after correction for multiple testing this association became weaker. A grouping of several biomarkers did not result in a better correlation with outcome compared to some biomarkers separately. It seems likely that multiple factors like DNA repair, apoptosis, cell cycle control, hypoxia and transmembrane drug pumps can contribute to chemoradiation-resistance, and the relative importance of these factors is likely to vary in different tumors, which makes identification of one strong prognostic molecular marker difficult. Prospective analysis of candidate markers and further investigation of their function is needed to help explain the role of chemoradiation resistance in head and neck squamous cell carcinoma.

In conclusion, this is the first study describing the possible prognostic value of 18 biomarkers for outcome in patients uniformly treated with concurrent chemoradiation. Although the exact role of MRP2 and P16 in the chemoradiation response in HNSCC has to be elucidated, both biomarkers were associated with clinical outcome. These markers need further validation in a similar patient group. In addition, it is unlikely that these are the best or the only possible markers for predicting outcome in this patient group, and it can be anticipated that present genome-wide screens (comparative genome hybridization, expression microarrays, epigenetics, etc) will lead to more and robust clinically useful markers in the future.