Ambivalent role of pFAK-Y397 in serous ovarian cancer-a study of the OVCAD consortium

In the frame of an FP6 EU-project named OVCAD (http//http://​www.​ovcad.​eu) we coordinated the development of a comprehensive and well described tumor bank and database of EOC patients. Within this project, samples from primary EOC were collected at the University clinics of Berlin, Hamburg, Innsbruck, Leuven, and Vienna (OVCAD-consortium) according to standardized operation procedures. Clinical and histopathological data as well as follow-up data were collected and cured by experienced clinicians. Patients presenting with benign ovarian diseases, low malignant potential ovarian cancer, clear-cell ovarian cancer, FIGO I stage EOC and patients with secondary malignant diseases were excluded [21]. The study protocol was approved by the Ethics Committees of the participating OVCAD partners (EK207/2003, ML2524, HEK190504, EK366, and EK260). All patients gave pre-operative written informed consent before enrollment in the study. Only patients undergoing debulking surgery and platinum–based chemotherapy were included to the OVCAD patient cohort. Overall survival (OS) was defined as the time interval between diagnosis and tumor associated death and progression free survival (PFS) as the time between cytoreductive surgery and disease progression or death. Overall observation time was the time interval between diagnosis and last contact, defined as death from the disease or last follow-up. Therapy response to chemotherapy was defined according to Chekerov et al. [21]. Patients were classified as non-responder if progression was diagnosed during treatment or recurrence within six months after end of first-line chemotherapy. Patients without recurrence, disease progression or death were censored at the time of last follow-up. Experienced gynecological oncologists and pathologists from the OVCAD consortium performed the clinical and histopathological evaluation and the evaluation of response to first-line treatment. As the impact of biomarkers on patients’ survival has been described to vary between histological subtypes [22] only patients with serous histology were included, resulting in a total of 179 serous EOC patients. Tumors were graded as well (1), moderately (2), or poorly differentiated (3).

Tissue micro arrays (TMAs) were constructed, whereby two 1 mm-diameter cores were obtained from each tumor sample. The immunohistochemistry procedure was performed essentially as described previously [23]. Antigen heat retrieval was performed using citrate buffer (Citra-BioGenex no. HK 087-5 K). The sections were incubated at 4°C overnight with primary antibodies (FAK, 1:200, monoclonal mouse IgG1, Millipore, catalog no. 05–182; pFAK, 1:150, polyclonal rabbit, abcam, catalog no. ab4803). As a positive control, breast cancer tissue sections and kidney tissue sections were used. For negative control breast and kidney sections were incubated in absence of the primary antibody. The specificity of the pFAK staining was shown by a peptide competition experiment, i.e. blocking of the primary antibody with the phosphorylated peptide mouse FAK (phospho Y397, abcam, catalog no. ab40145) in a molar excess ratio of 200-fold over night (Additional file 1: Figure S1). Samples were examined by two independent observers blinded for the clinical data, including a pathologist specialized in gynecology. FAK expression levels were determined using a scoring system based on staining intensity (0–3) and percentage of positive cells. pFAK scoring was performed as previously described (Fan and Shi, [19]), including staining intensity and percentage of positive cells. As both, FAK and pFAK staining were homogenous within the EOC tissue (Figure 1) percentage of positive cells did not differ significantly within the analyzed samples. The staining was grouped into FAK low and pFAK low (0–1; not stained at all and low expression) as well as FAK high and pFAK high (2–3; moderate to high expression).

Colocalisation of FAK and pFAK was determined by immunofluorescence staining of MCF7 and CaOV3 cells. The fluorescence labeled secondary antibodies, goat anti-rabbit (1:1,000; Invitrogen, AlexaFluor® 488 fragment of goat anti-rabbit IgG (H + L)) and goat anti-mouse (1:500; Invitrogen, AlexaFluor® 568 goat anti-mouse IgG1) were used besides DAPI for nuclear counterstaining. FAK staining intensity was designated as FAK expression and pFAK staining intensity as pFAK abundance throughout the paper, since phosphorylation is not a result of expression but a result of posttranslational modification.

Microarray analysis of this patient cohort has been previously published by Pils et al. [4]. In the present study we tried to further elucidate differentially expressed genes between pFAK positive and pFAK negative samples. Microarray data were available from 141 samples [24]. The parameters pFAK (high, n = 51 vs. low, n = 90) and FAK (high, n = 132 vs. low, n = 9) were used for shrinkage and a non-parametric prior was calculated for the pFAK parameter (allowing non-unimodality). Bayesian False Discovery Rate (BFDR) values below 10% were considered as statistically significant [24]. Functional analysis of differentially expressed genes was performed with Database for Annotation, Visualization and Integrated Discovery (DAVID) v6.7 [25]. In addition a gene set enrichment analysis [26] using the Gene Set Database MSigDB v3.1 (http://​www.​broadinstitute.​org/​gsea/​) was performed with the romer function [27] from the R-package limma v3.14.4 [28].

Statistical analyses were performed using SPSS software version 19 (IBM Corporation, Armonk, New York, United States). Associations between FAK expression and pFAK abundance, and between these two factors and clinicopathological parameters were assessed by T-tests (age), Chi-square tests, and Fisher’s exact tests as appropriate. Results were adjusted for multiple testing by the Bonferroni-Holm method [29]. Impact on progression free survival (PFS) and overall survival (OS) was determined by univariate and multiple Cox proportional-Hazards regression model analyses. To assess the independent impact of factors not significant in the univariate Cox regression analyses, all factors were included in the multiple models according to suggestions from Harrell [30] and Sun et al. [31]. Impact on chemotherapy response was determined by univariate and multiple logistic regression models. In addition, the estimates of the impact of pFAK on overall survival (i.e. the multiple Cox regression model) corrected for the clinicopathologic parameters age, stage, grade, residual tumor load, and peritoneal carcinomatosis on overall survival was illustrated by survival curves. For this task all parameters were averaged and pFAK was used as stratifying variable.

The characteristics of the 179 patients included in this study show a typical heterogeneous serous EOC population (Table 1 and 2). The clinicopathological characteristics are presented in Table 1 and 2 separately for the cohorts of pFAK-positive and -negative as well as FAK-positive and -negative patients, respectively. Mean age of the EOC patients at time of cytoreductive surgery was 57.6 years (SD ±12.6 years). The median observation period was 49 months (range: 1–69 months). Within the observation period, 82 patients died (45.8%) and 138 patients (77.1%) experienced tumor progression. A total of 43 patients (25%) did not respond to first-line chemotherapy.

For validation of the specificities of the FAK and the pFAK-Y397 antibodies immunofluorescence co-staining experiments were performed on MCF7 and CaOV3 cancer cell lines. As presented in Figure 2, both antibodies clearly stained focal adhesions, which proved the specificities of the used antibodies. A nuclear expression of pFAK (compared to only cytoplasmic expression of FAK) has been described previously by Murata et al. in colon cancer and breast cancer tissue. [32] In accordance with this data, staining of MCF7 cells also showed a nuclear staining of pFAK (Figure 2). However, in the ovarian cancer cell line CaOV3 and the immunohistochemistry stainings nuclear abundance could not be observed, as also described by Fan and Shi [19].

Immunohistochemistry of FAK and pFAK revealed a granular cytoplasmic immunostaining in serous EOC tumor tissue, typical for focal adhesion staining (Figure 1). High FAK staining was detected in a total of 92.2% (n = 165). pFAK was positive in a total of 87.2% and highly abundant in 36.9% (n = 66) of the samples. As expected, all the 66 samples showing high pFAK abundance also expressed FAK (p = 0.002). Comparing expression and abundance levels, a positive association between FAK and pFAK could be observed (p < 0.001).

pFAK abundance in grade 3 tumors was significantly higher than in grade 1&2 tumors (adjusted p = 0.014; Table 1). High pFAK abundance was also associated with the presence of either nodal positivity (N1) and/or distant metastasis (M1) (adjusted p = 0.030; Table 1). No significant differences were found for the clinicopathological parameters age, presence of residual tumor after debulking surgery, peritoneal carcinomatosis, and ascites (≤ vs > 500 ml). There was no association between FAK expression and the described clinicopathological parameters (adjusted p-values > 0.05; Table 2). The hypothesis, that pFAK abundance might be different between tumors of the previously described [4] molecular subclass 1 and molecular subclass 2 proved wrong. The subclasses were classified using a published 112 gene set [4]. Neither the FAK expression status correlated significantly with the molecular subclasses (p = 0.910), which is in accordance with previous data [4], nor the pFAK-Y397 abundance (p = 0.108), which was the initial hypothesis of this study.

Table 3 shows the impact of pFAK abundance and FAK expression on OS and PFS in the subgroup of FIGO III/IV serous EOC (n = 172), together with various clinicopathological parameters considered as potential prognostic factors. In univariate analyses, age (in decades) (hazard ratio [HR] 1.48; p < 0.001), FIGO stage (HR 2.48; p < 0.001), residual tumor (HR 1.77; p < 0.001), peritoneal carcinomatosis (HR 3.11; p < 0.001) and the molecular subclassification (HR 2.51; p = 0.001) affected OS. Likewise, PFS was negatively affected by higher age, FIGO IV stage, residual tumor, peritoneal carcinomatosis, and molecular subclass.

In addition, the impact of FAK and pFAK together with the molecular subclassification on FIGO III and IV serous tumors was analyzed by multiple Cox regression. In the final model, pFAK (HR 0.54; p = 0.034) and molecular subclass (HR 2.23; p = 0.004) showed a significant independent impact on OS besides age, FIGO stage, and the presence of peritoneal carcinomatosis. Although residual disease is known as an important clinical prognostic factor, it was not independently prognostic in the multiple analysis, because it was outperformed by the other factors in the model - probably by the very closely linked factor peritoneal carcinomatosis. For PFS, the parameters age, stage, and peritoneal carcinomatosis showed a significant independent impact but neither FAK, pFAK nor the molecular subclass did (Table 3). Figure 3 shows the univariate impact of pFAK on PFS and OS and the impact of pFAK, corrected for the relevant clinicopathologic parameters, on prognostication of PFS and OS as estimated survival curves (Figure 3C).

The predictive value of FAK and pFAK abundance, i.e. influencing response to first-line chemotherapy, was estimated by multiple logistic regression analyses. The risk to be a non-responder was neither significantly influenced by FAK expression, nor by pFAK abundance (HR 1.13, 95% CI [0.29-4.31], p = 0.858; HR 0.73, 95% CI [0.35-1.51], p = 0.729, respectively).

Using an empirical Bayesian approach for differential gene expression analysis between pFAK high and pFAK low abundant samples, 780 ProbeIDs were found to be statistically (BFDR < 10%) significantly differentially expressed. An over-representation analysis (602 of the 780 ProbeIDs were annotated in DAVID) revealed the annotation clusters “ribosome” and “RNA processing” as over-represented (Additional file 2: Figure S2; Enrichment Scores 5.26 and 2.34, respectively) from the differentially expressed genes and the KEGG pathway “Ribosome” as the most over-represented single pathway (Additional file 3: Figure S3; Benjamini-Hochberg corrected, p = 0.00022). A sensitive Gene Set Enrichment Analysis using 6,769 gene sets and the romer-function from the limma package revealed 227 down-regulated, 300 mixed (partly down- and partly up-regulated), and 414 up-regulated gene sets as significantly deregulated in pFAK high tumors (FDR < 5%; totaling in 862 gene sets; Additional file 4: Table S1; overlap shown in Additional file 5: Figure S4). Among the up-regulated gene sets were the “Creighton AKT1 signaling via mTOR” [33], the “mRNA-binding” and the “microtubule associated complex” gene set and among the mixed gene set was the “cell adhesion molecules” gene set. For FAK no such analysis was performed mainly because there were only nine FAK negative samples compared to 132 FAK positive samples, which renders a very unbalanced, thus underpowered, design.