Phosphorylation of Akt (Ser⁴⁷³) Predicts Poor Clinical Outcome in Oropharyngeal Squamous Cell Cancer

The prognosis of patients presenting with locally advanced head and neck squamous cell cancer (HNSCC) is still poor, with 5-year actuarial survival rates between 30% and 40% in most studies, despite decades of intensive clinical investigations (1). These poor survival rates combined with the severe functional impairment associated with surgery and radiation underscore the need to develop novel therapeutic strategies in the management of patients with advanced HNSCC. Recently, research efforts have focused on developing molecular targeted therapies or searching for molecular markers that are prognostic or predictive for treatment outcome. The advances in molecular biology with the availability of genomic and proteomic approaches have opened new research directions. Receptor tyrosine kinases have emerged as promising targets for cancer therapy (2-8).

The serine/threonine protein kinase Akt, a downstream target of phosphatidylinositol 3-kinase (PI3K), has been shown to be intimately involved in various cellular processes linked to tumorigenesis, including cell cycle progression, cell survival, cell motility, angiogenesis, as well as regulation of the mammalian target of rapamycin signaling pathway (9). PI3K/Akt pathway is activated by signaling inputs transmitted to the cytoplasm after growth factor ligand binding to receptor tyrosine kinases spanning the plasma membrane, such as epidermal growth factor receptor (EGFR). Specifically, ligand binding to receptors induces activation of the PI3K holoenzyme, leading to its recruitment to the plasma membrane and the formation of phosphoinositol triphosphate. Phosphoinositol triphosphate subsequently activates phosphoinositide-dependent kinase 1, leading to phosphorylation and activation of Akt. PTEN protein, the product of PTEN tumor-suppressor gene, is a lipid phosphatase that, by removing phosphate groups from phosphoinositide signaling molecules, limits the activity of PI3K pathway.

Although persistent activation of Akt pathway in HNSCC has been reported (10), the studies examining the prognostic significance of Akt activation in this tumor type are limited by small sample size and the use of conventional immunohistochemistry for evaluation of p-Akt expression (11, 12).

AQUA is a fully quantitative in situ proteomic method of analysis for tissue microarrays that allows calculation of expression ratios (13). This method allows measurements of protein expression within subcellular compartments that results in a number directly proportional to the number of molecules expressed per unit area (the concentration). This novel technology uses molecular methods to define subcellular compartments. It then quantifies the amount of protein expressed within the compartment by colocalization. Therefore, this technology permits preservation of tissue morphology while quantifying protein expression in paraffin-embedded tissue. In this study, we sought to evaluate the prognostic role of Akt activation by studying the levels of phosphorylated (activated) Akt (Ser⁴⁷³) expression using a cohort oropharyngeal squamous cell cancer (OSCC) tissue microarray and associating expression with clinical and pathologic data. We also sought to examine the correlation between PTEN and p-Akt levels. Our results indicate that increased levels of p-Akt represent an independent adverse prognostic indicator in HNSCC and that PTEN loss is a likely mechanism of Akt activation.

Following Yale institutional review board approval, a cohort of patients with OSCC was assembled. Inclusion criteria were histologically confirmed primary squamous cell carcinoma of the oropharynx treated at Yale-New Haven hospital between 1980 and 1999, and therapy with either external beam radiotherapy or gross total surgical resection and postoperative radiotherapy. Exclusion criteria included presentation with metastatic or recurrent disease, non-oropharyngeal site, and failure to receive a full course of radiation therapy. A tissue microarray was constructed as previously described (14), including 79 cases that met inclusion criteria and had complete p-Akt expression data. We excluded from the analysis 15 cases that had missing p-Akt expression information. Tissue cores were obtained from paraffin-embedded, formalin-fixed tissue blocks in the Yale University Department of Pathology archives. Slides from all blocks were reviewed by a pathologist to select representative areas of invasive tumor to be cored. The cores were placed in the recipient microarray block using a Tissue Microarrayer (Beecher Instrument, Silver Spring, MD). All tumors were represented with 2-fold redundancy. Previous studies have shown that the use of tissue microarrays containing one to two histospots provides a sufficiently representative sample for analysis by immunohistochemistry. Addition of a duplicate histospot, although not necessary, provides marginally improved reliability (15). Cores from HPV16-positive SiHa cell lines fixed in formalin and embedded in paraffin were selected for positive controls and included in the array. Additionally, 10 histologically confirmed normal squamous epithelium samples from formalin-fixed and paraffin-embedded skin were included for comparison of p-Akt expression in normal tissue. The tissue microarray was then cut to yield 5-μm sections and placed on glass slides using an adhesive tape transfer system (Instrumedics, Inc., Hackensack, NJ) with UV cross-linking.

Tissue microarray slides were deparaffinized and stained as previously described. In brief, slides were deparrafinized with xylene followed by ethanol. Following rehydration in distilled water, antigen retrieval was accomplished by pressure cooking in 0.1 mol/L citrate buffer (pH 6.0). Endogenous peroxidase activity was blocked by incubating in 0.3% hydrogen peroxide in methanol for 30 min. Nonspecific antibody binding was then blocked with 0.3% bovine serum albumin for 30 min at room temperature. Following these steps, slides were incubated with primary antibody to p-Akt (Ser⁴⁷³), obtained from Cell Signaling Technology (Beverly, MA), at 4°C overnight. The antibody to p-Akt was used at 1:300 dilution. This antibody has been extensively validated in previous studies using immunohistochemistry and Western blot analysis of normal and neoplastic tissue and tumor cell lines (10, 16). Primary antibody to PTEN (Cascade BioScience, clone 6H2.1) was used at 1: 50 dilution in 0.3% bovine serum albumin/TBS. This antibody has been validated in previous studies using immunohistochemistry and Western blot analysis of neoplastic tissue (17-19). Subsequently, slides were incubated with goat anti-mouse secondary antibody conjugated to a horseradish peroxidase–decorated dextran polymer backbone (Envision, DAKO Corp., Carpinteria, CA) for 1 h at room temperature. Tumor cells were identified by use of anticytokeratin antibody cocktail (rabbit anti-pancytokeratin antibody z0622; DAKO) with subsequent goat anti-rabbit antibody conjugated to Alexa 546 fluourophore (A11035, Molecular Probes, Eugene, OR). We added 4′,6-diamidino-2-phenylindole to visualize nuclei. Target (p-Akt, PTEN) molecules were visualized with a fluorescent chromogen (Cy-5-tyramide, Perkin-Elmer Corp., Wellesley, MA). Cy-5 (red) was used because its emission peak is well outside the green-orange spectrum of tissue autofluorescence. Slides were mounted with a polyvinyl alcohol–containing aqueous mounting medium with antifade reagent (n-propyl gallate, Acros Organics, Geel, Belgium).

Automated image acquisition and analysis using AQUA has been described previously (20). In brief, monochromatic, high-resolution (1,024 × 1,024 pixel; 0.5 μm) images were obtained of each histospot. We distinguished areas of tumor from stromal elements by creating a mask from the cytokeratin signal. 4′,6-Diamidino-2-phenylindole signal was used to identify nuclei, and the cytokeratin signal was used to define cytoplasm. Overlapping pixels (to a 99% confidence interval) were excluded from both compartments. The signal (AQUA score) was scored on a normalized scale of 1 to 255 expressed as pixel intensity divided by the target area. AQUA scores for each subcellular compartment (nuclear and cytoplasmic p-Akt) as well as the tumor mask were recorded. AQUA scores for duplicate tissue cores were averaged to obtain a mean AQUA score for each tumor.

Histospots containing <10% tumor, as assessed by mask area (automated), were excluded from further analysis. Based on a left skewed distribution, automated AQUA scores were dichotomized for some analyses. The lowest quartile (“low expressers”) was compared with the rest of the cohort (“all others”). Comparison of prognostic, treatment, and demographic variables between high/low p-Akt expressers was made using Fisher's exact test for categorical variables and confirmed by χ² tests. Comparison of age between expression groups was made using t tests.

For cumulative incidence analyses, local recurrence was defined as time from day of diagnosis to development of locally recurrent disease. Death without recurrence was defined as the time from day of diagnosis to death without observation of local recurrence. Overall survival was defined as time from day of diagnosis to death from any cause. Analyses of the cumulative incidence of local recurrence and death without recurrence between the dichotomized p-Akt groups were done using the methods of Gray (21). Overall survival between the dichotomized p-Akt groups was assessed by Kaplan-Meier estimation of the survival function. We used Gray's (21) method for hypothesis testing of the null hypothesis of no population-level difference between high and low p-Akt expression groups in the subdistribution hazards generating the cumulative incidence curves. We used a log-rank test to test a similar hypothesis concerning the Kaplan-Meier survival curves.

Univariable (with p-AKT as a continuous measure as the sole covariate) and multivariable analyses of time until local recurrence and death without local recurrence were done using the proportional hazards competing risk method of Fine and Gray (22). Similar analyses for overall survival were done using a Cox proportional hazards regression. Fine and Gray's (22) model is similar to the traditional Cox proportional hazards model, but the model provides estimates of hazard ratios (HR) associated with the subdistribution hazard of the outcome of interest. The subdistribution hazard is defined to account for the fact that those who have a competing outcome event can never have the outcome event of interest. This is different from the Cox model in which those with competing outcome events are censored and such censoring is assumed to provide no further information about the outcome of interest. More details about Fine and Gray's method can be obtained from Resche-Rigon et al. (23). In multivariable regression analyses, we added as covariates p-Akt expression level and all variables listed in Table 1; we believed that the variables in Table 1 could be possible confounders of the relationship between p-Akt expression (in and of itself) and local recurrence or survival. We excluded those with missing covariate information from the multivariable analyses. We also excluded from multivariable analyses those with cancers located in other locations of the oropharynx because there were too few cases (four with high p-Akt expression) to reliably estimate a variable associated with this subgroup, and hence adequately adjust for potential confounding. We entered the discrete variables into the multivariable models as categorical variables. We entered age into the models using a restricted cubic spline with one interior knot to allow for a nonlinear relationship between age and recurrence or survival.

We used the R software package to implement Gray's (21) and Fine and Gray's (22) methods. We used STATA for all other analyses. All statistical tests were two-tailed where applicable.

There were 79 patients with primary oropharyngeal carcinoma in this cohort who met inclusion criteria. Sixty were male, 29 female with age ranging from 41 to 79 years (median 61 years). Nine patients were tumor-node-metastasis (TNM) stage II, 23 stage III, and 47 stage IV. Oropharyngeal subsites included 38 tonsillar fossae, 37 base of the tongue, and 5 other oropharynx. For histologic grade, 5 tumors were well differentiated, 38 were moderately differentiated, 26 were poorly differentiated, and 11 were not recorded. Forty-seven patients were treated by primary radiation therapy alone, and 30 by surgery with postoperative radiation (two patients were not recorded). Twenty-one patients received chemotherapy in addition to their primary treatment modality. These data are summarized in Table 1.

As visualized by fluorescent immunohistochemistry, p-Akt displayed mixed cytoplasmic and nuclear staining (Fig. 1). Normalized AQUA scores for p-Akt expression in tumors ranged from 1 to 255. Positive controls (SiHa cell lines) were in the top quartile (data not shown). Histogram analysis of p-AKT expression shows a left-skewed distribution where a number of cases (18%) exhibit p-AKT levels 1 SD below the mean (Fig. 2). Table 1 compares characteristics between low and high p-Akt expression groups. There were no statistically significant differences between the two groups.

Seventy-one percent (67 of 79) of tissue spots had sufficient tissue for analysis of PTEN expression by AQUA. PTEN displayed a primarily nuclear expression pattern by immunofluorescence analysis, and normalized AQUA scores were reported on a 1 to 255 scale.

The associations between tumor and nuclear PTEN and p-Akt expression (as determined by AQUA) were analyzed by Spearman correlation. There was no relationship between tumor PTEN and tumor p-Akt expression. For nuclear expression, a significant inverse relationship was found where tumors with high p-Akt in the nucleus had low PTEN in the nucleus and vice versa (Spearman ρ = −0.249, P = 0.045).

Tumor p-Akt as determined by AQUA was examined for association with local recurrence, survival without recurrence, and overall survival. To assess the prognostic value of p-Akt expression on a continuous scale, we used the proportional hazards competing risk regression method of Fine and Gray (22). Tumor p-Akt as a continuous variable was a significant predictor for local recurrence (P = 0.01), but not for death without recurrence (see Table 2). Tumor p-Akt showed a marginally statistically significant association with overall survival (P = 0.08). The magnitude of these effects was clinically relevant; although the point estimates were small, the range of p-AKT expression in the data set was large (1-242), and hence a sizable difference in p-Akt expression would also be associated with a sizable difference in the relative risk of developing local recurrence or dying.

Based on the presence of a low-expressing p-Akt group, we did additional univariable analyses comparing incidence and survival curves between low- and high-expressing p-AKT groups (Figs. 3 and 4). This analysis showed that cases with low p-AKT expression exhibit a decreased risk of 5-year local recurrence (5.3% versus 38.2%, P = 0.014), and a higher 5-year survival probability (44.9% versus 27.9%, P = 0.05). Table 3 presents 5-year local recurrence and mortality outcomes based on the Gray cumulative incidence and Kaplan-Meier estimates.

Using Fine and Gray's method, we did multivariable analyses to assess the independent predictive value of p-Akt tumor expression as a continuous variable for local recurrence and death without recurrence. The following prognostic and potentially confounding variables were also included: sex, age, chemotherapy status, subsite within oropharynx, TNM stage, management (external beam radiotherapy versus primary surgical excision plus external beam radiotherapy), and histologic grade. In these models, p-Akt as a continuous variable was a strong predictor of local recurrence (P = 0.03; see Table 4) after controlling for the potential confounding covariates, but not death without local recurrence (P = 0.78; results not shown). Again, because the range of p-Akt was so large, the magnitude of the effect in the regression model presented in Table 4 was substantial. A large increase in p-Akt is associated in the sample with a negative effect on recurrence even after controlling for prognostic and potentially confounding covariates. As a sensitivity analysis to examine whether our assumption of proportional hazards over time held, we also included an interaction between time and p-AKT status in the competing risk regressions to allow for a time-varying HR. The interaction term in the competing risk regressions was not statistically significant, indicating that the data do not provide evidence of a time-varying p-Akt effect. The effect of p-Akt on overall survival in the multivariable Cox model was not statistically significant (P = 0.28; results not shown).

In this study, we show that increased p-Akt level in OSCC is a strong predictor for poor patient outcome. In multivariable analysis, adjusted for well-recognized prognostic indicators, p-Akt status remained a strong predictor, whereas TNM stage did not. Our data suggest that p-Akt status may be a powerful prognostic marker in OSCC. Further study is needed to determine whether it may also predict benefit from or resistance to therapy or constitute an important therapeutic target.

Two other groups have shown that p-Akt is associated with poor prognosis in oral cancer (11, 24). However, both of these studies used conventional immunohistochemistry to assess p-Akt status. Measurement of protein levels with conventional immunohistochemistry often fails to provide accurate results. Specifically, the pathologist tends to group things as positive or negative, whereas the automated device yields results in continuous full-scale scores. Two prior studies used two different immunohistochemical cutoffs (15% and 20%) to differentiate between “high” versus “low” expressers. The present study is the only one of its kind that uses a quantitative method of protein analysis to determine the prognostic value of Akt activation in oropharyngeal cancer. This method allows measurements of protein expression within subcellular compartments and generates a numerical result directly proportional to the number of molecules expressed per unit area (25). Thus, we avoid biases introduced from the arbitrary cutoff points used in conventional immunohistochemistry studies while preserving spatial and morphologic information that techniques such as Western blotting lose.

Activation of Akt in tumors is mediated via several mechanisms, including activation of cell membrane receptor tyrosine kinases such as EGFR and loss of phosphatase PTEN with dephosphorylation of phosphoinositol triphosphate. In HNSCC, amplification of the 3q26 chromosomal region has also been reported (26). The gene PIK3CA, encoding the catalytic subunit of PI3K (p110), resides at this area (27). Amplification of PIK3CA leads to Akt activation. Whether the remarkable activation of Akt in OSCC results from a unique event or from multiple unrelated molecular mechanisms converging on PI3K activation is presently unknown. Akt activity was found to increase in parallel with the promotion stages of mouse skin carcinogenesis and remained high during the malignant conversion of papillomas to squamous cell carcinoma (28). In one study, simultaneous abnormalities in Akt and p53 pathways in upper aerodigestive tract tumors were rare, indicating that amplification of PIK3CA and mutation of p53 are mutually exclusive events and either event is able to promote a malignant phenotype (26).

Taken together, these studies suggest that Akt activation is an early event in head and neck carcinogenesis, sufficient to induce malignant conversion independent of p53 tumor-suppressor pathway inactivation. Experiments in our laboratory are under way to determine the association between p-Akt expression and expression of related proteins, such as p53, EGFR, PTEN, and PIK3CA in OSCC.

In addition to providing useful prognostic information, our results have important clinical implications. Akt inhibitors, such as UCN-01, and inhibitors of the p-Akt target mammalian target of rapamycin are currently in clinical development. In addition, constitutively activated Akt has been implicated in resistance to EGFR-targeted therapies and EGFR is an established molecular target in HNSCC. So, the combination of EGFR and Akt inhibitors may prove to be a promising therapeutic strategy for OSCC. In the future, p-Akt AQUA score may be useful in predicting response to Akt-targeted therapies.