Expression profiling of 21 biomolecules in locally advanced nasopharyngeal carcinomas of Caucasian patients

This is a translational study in LA-NPC patients performed as an extension of a randomized trial, conducted by HeCOG [5], which tested the addition of induction chemotherapy (cisplatin 75 mg/m², epirubicin 75 mg/m² and paclitaxel 175 mg/m² every 3 weeks) (patient Group A) to the standard approach of concurrent chemo-radiotherapy with cisplatin (patient Group B). The main eligibility criteria were as follows: (i) biopsy-proven, previously untreated WHO type I, II or III NPC; (ii) age >15 years; (iii) stage II–IVB according to the American Joint Committee on Staging of Cancer classification (AJCC 7th edition, 2009); (iv) measurable or evaluable disease; (v) no other primary tumors; (vi) performance status (PS) of 0–2 according to the Eastern Cooperative Oncology Group (ECOG) scale.

Tissue paraffin blocks and peripheral blood for DNA analysis were prospectively collected from each patient registered in the study. The clinical protocol and collateral translational research studies were approved by the HeCOG Protocol Review Committee, the Institutional Review Boards in participating institutions and the Bioethics Committee of the Aristotle University of Thessaloniki School of Medicine under the general title: “Investigation of putative predictive or prognostic biomarkers in patients with locally advanced nasopharyngeal carcinoma”. The study was also registered at the Australian New Zealand registry (ACTRN 12609000730202). Prior to randomization each patient provided a study-specific written informed consent and optionally a separate informed consent for the provision of biological material for future research studies.

One hundred and nineteen formalin-fixed paraffin-embedded (FFPE) tumor tissue samples from 118 patients were available for analysis (only in one case, 2 samples were examined from 1 patient). Eleven blocks were not further evaluated due to inappropriate material (inadequate material in 7 cases and no tumor in 4 cases). Thus, 108 samples from 107 patients were examined for EBV-related small RNA (EBER) detection with the use of chromogenic in situ hybridization (CISH) [6] and for a number of key regulatory proteins by utilizing IHC [7, 8]. Finally, 105 samples from 105 patients were successfully assessed with CISH and IHC (REMARK diagram, Figure 1). Representative slides (hematoxylin and eosin, H&E) from the tissue blocks were reviewed by an experienced pathologist (M.B.) for confirmation of the diagnosis and adequacy of material as well as calculation of the percentage of tumor cells in each case. In 97 cases, the tumor tissue in the paraffin block was adequate for the construction of tissue microarrays (TMA), while for 11 cases all assays were performed on whole tissue sections.

Tumor tissue specimens were arrayed (3 cores per case, 0.6 mm in diameter) into a recipient paraffin block using a manual arrayer (MTA-1, Beecher Instruments, Sun Prairie, WI, USA). Representative cores from normal (colon, tonsil, placenta, kidney, thyroid, breast, and prostate) and cancerous tissues (melanoma, testicular seminoma, colon carcinoma, squamous head and neck carcinoma, and endometrial carcinoma) were also loaded on the TMA block, serving as positive and negative assay controls.

In order to detect latent EBV infection [6], CISH assays were performed on serial 4 μm-thick sections from the original blocks and the TMA block, using a Bond Max™ autostainer (Leica Microsystems, Wetzlar, Germany), according to the provider’s specifications as described in detail elsewhere [5]. The poly-A probe, used as an indicator of the preservation of mRNA in the cells (positive control), resulted mainly in dark brown nuclear staining and less in a cytoplasmic one.

Immunohistochemical labelling was performed on serial 3 μm-thick sections from the original blocks and the TMA block. Proteins involved in angiogenesis, cell adhesion, assembly, proliferation and differentiation, cell cycle and transcription regulation, DNA repair, microtubule assembly, and inflammation mediation were investigated. A list of the studied proteins, the sources of antigens and the staining procedures for IHC and the cut-offs used [6‐11] are quoted in Table 1. The Multi-Cytokeratin (AE1/AE3) was used as a control stain for the identification of tumor cells.

The evaluation of CISH and IHC stained sections was done simultaneously by two pathologists (G.K. and M.B.) blinded as to the patients’ clinical characteristics and survival data, according to the criteria quoted in Table 1. Tumor cells containing EBER transcripts were evaluated as positive when an intense, brown, predominantly nuclear staining was present in >1% of tumor cells [12]. Only poly-A probe positive cases were evaluable for EBER status. IHC cut-off values were defined prospectively, based on distributional analyses in frequency histograms and the medical literature. In order not to miss cut-off values with prognostic utility, a Receiver-Operating Curve analysis of IHC staining scores was followed with disease progression as the indicator parameter; this analysis failed to show any cut-off with significant sensitivity and specificity (Table 1). Based on the selected IHC cut-offs, protein expression status was defined for each tumor as either negative or positive, except for Ki67 which was graded as low (≤13%) or high (>13%). Representative cases of CISH and IHC staining are presented in Additional file 1: Figure S1.

Stained sections for Cyclin D1 and Ki67 were evaluated using an image analysis system as described [5]. The cut-offs used for the evaluation of Cyclin D1 and Ki67 staining are quoted in Table 1.

Categorical data are displayed as frequencies and corresponding percentages, while continuous data by median and range. For the comparison of continuous data the non-parametric Kruskall Wallis test was used, while the comparison of categorical data between groups was performed by Fisher’s exact or Pearson chi-square tests, where appropriate. Overall survival (OS) was measured from the date of treatment randomization to patient’s death or last contact. Progression-free survival (PFS) was measured from the date of treatment randomization to documented disease progression, death without prior documented progression or last contact. Time-to-event distributions were presented using Kaplan-Meier curves and compared using the log-rank test. Unsupervised hierarchical clustering analysis using the majority of the examined biomarkers was conducted. Univariate Cox regression analyses were performed for OS and PFS, to assess the prognostic or predictive significance of biomarkers adjusted for treatment. A backward selection procedure with a removal criterion of p>0.10 was performed in the multivariate Cox regression analysis in order to identify significant factors among examined biomarkers, basic clinicopathological characteristics (categorized as given in Additional file 2: Table S1) and treatment group. For all comparisons the level of significance was set at α=0.05. All results are presented according to reporting recommendations for tumor marker prognostic studies [13]. Analyses were performed with the use of the SPSS v. 17.0 (SPSS, Inc., Chicago, IL), JMP version 8 and SAS version 9.3 (SAS, Institute Inc., Cary, NC, USA).

Selected patient and tumor characteristics are quoted in Table 2. The male/female ratio was 2.24/1. The majority of patients were of PS 1 or 2, early T stage and advanced lymphnode involvement. Findings deserving closer attention are the normal pattern of age distribution (Figure 2) and the relatively low incidence of WHO Type I disease, given that the population was Caucasian (9.3%).

The patients have been followed for an additional period of 21.8 months compared to the respective clinical study [5]. After a median follow-up time of 76.8 months (range 42.3-99.2 months), 3-year PFS and OS rate for all patients in the study were 60.7% and 66.4%, respectively. Mean PFS and OS for all patients were 49.7 months and 52.9 months, respectively. No statistically significant differences emerged when time-to-event times were analyzed with respect to the treatment group (mean PFS: 51.1 months for Group A vs. 47.4 months for Group B, p = 0.65; mean OS: 53.4 months for Group A vs. 51.7 months for Group B, p = 0.938).

Biomarkers with the most frequent expression (>90% positive cases) were: p53, COX-2, P-cadherin, EBER, phospho-GSK-3β (p-GSK-3β), and Fascin-1. On the contrary, the least IHC-expressed proteins (<50% positive cases) were as follows: VEGFR-2, phospho-mTOR^Ser2448 (p-mTOR), VEGF-A, and p16 (Table 3).

The statistically significant correlations of patients’ clinicopathological characteristics with the IHC proteins expression are presented in Additional file 2: Table S1. Notably, PTEN and p63 expression patterns were clearly linked with favorable characteristics: patient PS and T stage (for the former), T and AJCC TNM stage (for the latter). Positive phospho-AKT (p-AKT) and phospho-p44/42MAPK (p-MAPK) expression were related to male gender, positive p-GSK-3β and ERCC1 expression depicted a more advanced age at diagnosis, and deficiency of ERCC1 and p-mTOR expression were correlated to poor patient PS. As expected, positive EBER expression status was linked to WHO Type III histology. Unexpectedly, Ki67-positive tumors were more frequently associated with earlier N stage and good PS; likewise, Cyclin D1 expression was more often positive in tumors of earlier AJCC TNM stage.

Associations between the expression status of the tested biomarkers and response to treatment did not reveal significant predictive values for any of them (Data not shown).

Additional file 3: Table S2 presents all statistically significant paired associations of protein IHC expression. The most interesting paired interactions were as follows: p63/p-AKT, p63/Ki67, p-AKT/Ki67, p-AKT/ERCC1, Ki67/ERCC1, Ki67/PTEN, Ki67/Cyclin D1, p-mTOR/ERCC1, and p-mTOR/VEGFR3.

Consequently, the most informative biomarkers in NPC with regards to IHC expression were: p63, p-AKT, Ki67, ERCC1, CyclinD1, p53, COX-2, and p-mTOR.

In an effort to study the complex interactions among the examined biomarkers, cluster analysis of the IHC protein expression was performed.

Firstly, all protein targets were included in the analysis and gradually increasing numbers of clusters were tested for their prognostic significance with respect to PFS and OS as well as their association with T and N status, AJCC TNM stage, and overall response rate (ORR) to treatment. The only model with prognostic significance was a 2-cluster one, with a borderline association to favorable OS (p=0.048) (Figures 3 and 4) and earlier TNM stage (p=0.027 for stage IIB-III vs. IV). Cluster analysis for each of the two treatment groups (A and B) did not result in the emergence of any meaningful prognostic significance.

A following step was to look at a more compact set of protein targets (such as p-AKT, p-mTOR, PTEN, VEGF-A, ERCC1, and p53) based on their relative significance in driving carcinogenesis or their possible synergistic action. In this case, no cluster appeared to confer prognostic information for any of the endpoints tested.

In univariate Cox regression analysis adjusted for treatment, the only biomarkers with positive prognostic value were EBER and p63, the latter being important also in the premature translational analysis of the relevant clinical study [7]. Particularly, EBER expression was linked to a favorable OS in a statistically significant way (hazard ratio HR = 0.38, 95% CI = 0.15-0.99, p = 0.048), while its association to improved PFS was only of borderline significance (HR = 0.44, 95% CI = 0.17-1.14, p = 0.09). The favorable prognostic value of positive p63 expression was borderline only for OS (HR = 0.48, 95% CI = 0.21-1.09, p=0.08) (Table 4 and Additional file 4: Figures S2).

Of special interest was the predictive significance of Cyclin D1 expression, which emerged only when the patient population was adjusted for the administered treatment (Group A vs. Group B). In this case, immunopositivity for Cyclin D1 resulted in a statistically significant worse prognosis in terms of PFS for patients in Group B (HR = 2.64, 95% CI = 1.08-6.48, p=0.034) (Table 4), while the absence of Cyclin D1 expression conferred a trend of better OS for the same Group (HR = 0.44, 95% CI = 0.18-1.04, p=0.062) (Table 4). Log-Rank test was subsequently performed, which confirmed the results of Cox regression analysis mentioned above (Figures 5A and 5B).

The following parameters were pointed out by multivariate analysis as the significant prognostic indicators for both PFS and OS: age at diagnosis, treatment group, PS, AJCC TNM stage, p-mTOR and ERCC1 expression. Notably, not only the same variables were indicated as important for both PFS and OS, but also the magnitude of significance (denoted by HR value) was similar in both conditions (Tables 5 and 6).

Particularly, advanced age, the addition of induction chemotherapy, poor PS, AJCC stage IV, and positive ERCC1 expression were associated with a higher probability of disease progression and death. On the contrary, positive p-mTOR expression was related to a decreased risk of progression and death (Additional file 4: Figures S2).

Compared to the preliminary analyses performed in the context of the respective clinical study [7], Ki67 and p63 lost their independent prognostic value in the longer follow-up, in contrast to age, PS, and AJCC TNM stage which maintained their significance.