In contrast to the progress seen in the molecular classification, pathogenesis elucidation and management of quite a few neoplasms, little has been achieved towards personalized approach in the NPC field so far. In contrast to the paradigm of other neoplasms, no activating mutations, such as EGFR or PI3K, are known to drive nasopharyngeal carcinogenesis [14, 15]. It seems that nasopharyngeal carcinogenesis is far more composite than the almost ubiquitous EBV presence depicts. As a result, survival rates of patients with advanced disease remain poor, especially for non-endemic populations . During the last few years, we have witnessed combined efforts to unravel NPC pathogenesis by utilization of multiple gene expression profile datasets, definition of novel gene signatures and molecular or cytokine markers, integration of miRNA technologies, and exploration of epigenetics.
The present study incorporates novel features which differentiate it from the rest utilizing IHC in NPC. In particular, 21 protein targets have been simultaneously profiled for their expression levels in nasopharyngeal tumors, in order to observe possible cross-talks between pathways. Biomarkers were carefully chosen, in order to include representatives from as many as possible oncogenic processes, as reviewed recently . In view of this, the clustering analysis performed is considered even more informative. More importantly, the study population was a Caucasian one, representing not only the entire Greek region but also the Balkan Peninsula, since HECOG-affiliated hospitals from both Greece and Romania have participated in this study. Although a considerable number of Caucasian NPC patients have been studied during the last decades [18–23], the respective series have either focused just on clinical parameters or have investigated only a limited number of biomarkers.
The high frequency of EBER CISH expression was not surprising, even for the Caucasian population of our study, as EBV infection is almost ubiquitous globally and plays a major role in the pathogenesis of NPC in both endemic and non-endemic areas . Generally, our study confirmed that the well-established characteristics of EBV-related NPC are also applicable to Caucasian NPC patients. Namely, the strong connection of EBV to WHO Type II and III histology as well as to the more favorable disease course . Importantly, the univariate analysis indicated EBER expression as a significant prognostic factor of improved OS.
Cyclin D1 is one of the key proteins involved in cell cycle control, a process considered as critical in the development of NPC . Although our study did not find as high uniform Cyclin D1 expression as reported in the literature , IHC expression levels were interestingly informative. Firstly, a statistically significant association between Cyclin D1 and Ki67 as well as between Cyclin D1 and p-AKT expression was indicated, pointing to an active cell proliferation axis, which has already been investigated as a therapeutic target in NPC . More importantly, univariate analysis pointed to a predictive role for Cyclin D1 protein levels, as differences in PFS and OS emerged only when the study population was adjusted for treatment. In particular, patients treated with induction chemotherapy prior to chemo-radiotherapy fared significantly better (in terms of PFS) in the case of positive Cyclin D1 expression, while in the absence of Cyclin D1 this finding was inverted (for OS). Consequently, it seems that tumors with enhanced Cyclin D1 function are vulnerable to more aggressive treatment. This result is in fine agreement with the reported radio-sensitivity of Cyclin-enriched nasopharyngeal , laryngeal , and breast  cancers.
The function of p63, a transcription factor and member of the p53 family, is rather complicated in NPC, since two major classes of p63 isoforms have been identified: the full-length TAp63 and the N-terminal truncated ΔNp63 . The two isoforms exert opposite purposes; TAp63 isoform has similar function to the wild-type p53 protein, while ΔNp63 is thought to antagonize TAp63 and p53 in target gene regulation. Each isoform can be targeted by a specific antibody for IHC evaluation, while a pan-p63 one binds to common epitopes. Consequently, reports of overexpression and oncogenic properties of p63 in NPC [30, 31] should be construed through this prism, also bearing in mind that the predominant isoform is the truncated ΔNp63 one. In our study, a pan-p63 antibody was utilized, thus limiting the importance of the borderline favorable prognostic value for OS, which emerged for p63 in the univariate analysis. Accordingly, the strong correlation of p63 expression with p-AKT and Ki67 as well as its association with early disease stage should be interpreted with caution.
The virtual absence of p16 IHC expression in our study (only 5% positive tumors) has been constitutively documented in nasopharyngeal cancers , attributable mainly to aberrant methylation or even to gene deletion. In contrast to reports supporting a predictive role of p16 to both chemotherapy (5-flouoruracil and cisplatin)  and radiotherapy , our study did not find any connection of p16 protein expression with disease outcome. Likewise, despite the documented unfavorable value of p16 expression in NPC [7, 35], such a role did not emerge in our study. The inevitable small number of p16-positive cases undoubtedly calls for a cautious reading of these results.
Activation of the PI3K/AKT/mTOR pathway has been reported in NPC several years ago . AKT phosphorylation can be either a result of LMP1  or EGFR  action on PI3K or a compensatory consequence of decreased PTEN levels . However, EBER/p-AKT or EGFR/p-AKT protein interactions did not emerge in our study. Moreover, the comparable expression levels of p-AKT and PTEN argue in favor of a downward activation of AKT instead of a compensatory one. p-AKT association with the Ki67 status may reflect the known propensity of AKT-activated NPC to metastasize ; however, we could not confirm the contribution of decreased E-cadherin levels to this phenomenon .
Cadherin IHC expression did not acquire any significance in our study with regard to disease prognosis or association with the rest biomolecules examined. Despite the reported depletion of E-Cadherin expression in NPC tissues in comparison to the normal ones [32, 41], a high level of IHC expression was observed in our study, both for P- and for E-Cadherin. One possible explanation is the low IHC cut-off that was selected, compared to other reports . On the other hand, the enhanced Cadherin expression might not merely be an artifact but could reflect a lack of its contribution to NPC pathogenesis in Caucasian patients. Such a scenario would justify the absence of any unfavorable prognostic value, contrary to reports of low tumor Cadherin levels in NPC patients with advanced disease stage and decreased survival .
The predictive role of high ERRC1 protein levels to cisplatin-containing regimens has been firstly described in ovarian cancer 20 years ago . However, it was only recently that similar reports were published in NPC . In line with this evidence, multivariate analysis in our study showed a significant unfavorable association of increased ERCC1 expression with both PFS and OS. Since cisplatin was the chemotherapy backbone in both patient groups, this outcome could be explained as a negative predictive effect of high ERCC1 protein levels. In addition, a poor prognostic value of ERCC1 could also be contemplated, as increased ERCC1 protein levels were significantly associated with aggressive disease characteristics, such as positive Ki67 and p-AKT IHC expression.
Loss of PTEN expression is a frequent event in NPC , accounting for the extensive deregulation of cellular signaling pathways (e.g. PI3K/AKT/mTOR and Wnt) and the metastatic propensity via induction of the epithelial-mesenchymal transition phenotype. Our study, in line with the literature evidence, has found a relatively low rate of PTEN expression. Moreover, PTEN presence was significantly associated with early tumor stage, a finding which has been also described in Chinese NPC patients .
Increased COX-2 IHC expression in NPC has been linked to the presence of lymph node metastasis  and worse survival, as well to enhanced sensitivity to the radiotherapy effects . In contrast to such well-established evidence, no prognostic or predictive role for COX-2 was revealed in our study, despite the higher COX-2 expression frequency compared to previous reports . However, in line with our observations are a few studies  which indicate no prognostic significance of COX-2 or even a positive one , thus perplexing COX-2 contribution to NPC pathogenesis. Commonly reported co-expression patterns, such as COX-2/EGFR/VEGF , COX-2/EGFR , and COX-2/LMP1  were not confirmed in the present study. The axis LMP1/COX-2/VEGF , in which COX-2 is promoted by LMP1 and subsequently induces VEGF, was also not manifested in our Caucasian patient series.
However extensive our investigation was, a bunch of biomolecules and processes were not shown to contribute in any way to NPC pathogenesis, despite the strong literature evidence to the contrary. In particular, p-MAPK was neither evident in our tumor series nor was clinically significant, counter to reports of an activated and tumor-promoting MAPK in NPC . Likewise, despite p-GSK-3β being reported as a vital member of the PI3K/AKT/GSK-3β/Cyclin D1 pathway in NPC promotion [26, 36, 46], the present study did not result in such evidence. Similarly, p53 expression was not informative in terms of disease prognosis or interaction with any other biomarker. Several reports state the overexpression or accumulation of p53 protein in the vast majority of NPC  and a resulting tumor-promoting role ; however, p53 expression was not detected to that extent in our study. Anyway, the functional status of p53 is disputable in NPC , which could explain the absence of its prognostic value , as indicated also in our study. The antagonizing effect of ΔNp63 protein on p53, as quoted above, renders the conclusion-making procedure even more insecure. Fascin, an actin cross-linking protein, has been implicated in the progression of various neoplasms, including NPC, mainly through the promotion of cell migration and adhesion . In spite of Fascin’s high expression frequency in our study, any significant correlation of Fascin-1 to NPC pathogenesis cannot be implied.
The absence of any prognostic role for EGFR in our Caucasian population, despite its high expression frequency, is a result which needs particular consideration. There is a substantial bulk of evidence concerning the unfavorable prognostic value of EGFR both for disease stage and for outcome , even in Caucasian populations , the only exception being a recent Korean study . Additionally, several co-expression patterns have been described, such as EGFR/COX-2 [53, 54], EGFR/VEGF , and EGFR/AKT , not being manifested in our series. Although EBV presence is believed to induce EGFR , such a paired expression model was also not found. The paucity of EGFR-expression information in the present study could be interpreted in the context of the evolving understanding of EGFR role in NPC pathogenesis; EGFR downstream signaling molecules are numerous in NPC and are still being defined . The relative inefficacy of EGFR-targeting attempts in NPC , in contrast to HNSCC , serves as a reminder of the underlying complexity.
VEGF-A and -C expression status were also not informative with regard to prognosis or any association with clinicopathological variables. In spite of the abundant literature references of an adverse prognostic role of VEGF expression in NPC [67, 68] and VEGF-C in particular , no such effect was shown. Furthermore, several co-expression pairs have been repeatedly described, mainly between VEGF, COX-2, EGFR, and LMP1 [56, 70], which were not prominent in our tumor series. Anyway, the exact setting in which angiogenesis exerts its tumor-promoting action needs to be clarified, as depicted by the modest results of anti-angiogenic therapy so far .
Phosphorylated-mTOR expression emerged as an independent favorable prognostic factor in multivariate analysis, although with borderline significance. mTOR is linked with a promoting role in nasopharyngeal carcinogenesis  and a unanimous adverse prognostic value in NPC , which is alleviated by mTOR inhibition ; thus, the aforementioned result is considered as unexpected. Possible hypotheses for this discrepancy are: a) the presence of p53 expression in all p-mTOR tumors could account for a favorable outcome upon treatment and b) multiple comparisons in a sample underpowered for such exhausting statistical analyses could lead to an artifact. Similarly, the association of increased Ki67 with an earlier N stage and good PS is rather surprising and could be attributed either to a statistical artifact, as mentioned above, or to the IHC cut-off level which have been used.
Clustering analysis did not reveal any discriminator group of genes/biomolecules which could portray a specific driving scenario of carcinogenesis in NPC. Moreover, no NPC subtype emerged on the basis of differential protein expression, which could point towards potential therapeutic targets. Following the breast cancer paradigm , the reported gene-profiling approaches in NPC end up to different gene sets [75, 76], calling once again for the need of bioinformatics implementation to data interpretation.
It should be noted that the reported high frequency of WHO Type I NPC histology in Caucasian populations (approximately 25%)  was not represented in our study (only 9%). Possibly, Balkan NPC patients represent an intermediate population in terms of epidemiology, as Balkan Peninsula shares the Mediterranean Basin and neighbors to North Africa, which is a known endemic NPC area.
The results of the present study should be interpreted on the notion of the inherent limitations of IHC. Protein expression levels often mirror the mechanisms leading to cellular growth deregulation; however, they have to be translated along with the coexisting genetic and epigenetic alterations. Moreover, clustering analysis is unable to provide a perfect outline of the complex interactions between the tumor-promoting networks; small-size samples and multiple comparisons undermine the validity of the results. Last but not least, since several phospho-antibodies were utilized, consideration should be paid to the disputable reliability of FFPE IHC for phosphorylated epitopes .