Validation of Androgen Receptor loss as a risk factor for the development of brain metastases from ovarian cancers

Background Central nervous system (CNS) spreading from epithelial ovarian carcinoma (EOC) is an uncommon but increasing phenomenon. We previously reported in a small series of 11 patients a correlation between Androgen Receptor (AR) loss and localization to CNS. Aims of this study were: to confirm a predictive role of AR loss in an independent validation cohort; to evaluate if AR status impacts on EOC survival. Results We collected an additional 29 cases and 19 controls as validation cohort. In this independent cohort at univariate analysis, cases exhibited lower expression of AR, considered both as continuous (p <  0.001) and as discrete variable (10% cut-off: p <  0.003; Immunoreactive score: p <  0.001). AR negative EOC showed an odds ratio (OR) = 8.33 for CNS dissemination compared with AR positive EOC. Kaplan-Meier curves of the combined dataset, combining data of new validation cohort with the previously published cohort, showed that AR <  10% significantly correlates with worse outcomes (p = 0.005 for Progression Free Survival (PFS) and p = 0.002 for brain PFS (bPFS) respectively). Comparison of AR expression between primary tissue and paired brain metastases in the combined dataset did not show any statistically significant difference. Conclusions We confirmed AR loss as predictive role for CNS involvement from EOC in an independent cohort of cases and controls. Early assessment of AR status could improve clinical management and patients’ prognosis.


Background
Epithelial ovarian cancer (EOC) is the third cause of death among gynaecological malignancies worldwide [1].
Unfortunately, no predictive biomarkers of CNS involvement have been validated in EOC at present.
Considering its already known negative prognostic value, we focused our interest on Androgen Receptor (AR) [12,13]. Interestingly, in our previously published small series of 11 patients, AR-negativity predicted a 9.5 times higher propensity of CNS involvement [14].
Aims of the present study were: i) to confirm AR's predictive role for CNS dissemination from EOC in an independent validation cohort, and ii) to assess the role of AR expression in terms of progression free survival (PFS) and brain progression free survival (bPFS) in the combined dataset.

Patients' validation cohort collection
In order to validate AR's predictive role of CNS involvement form EOC, we extended our previously published series of eleven patients [14]. Fourteen new cases with paired brain lesions and 15 new cases with CNS involvement not surgically removed, diagnosed from April 2000 to December 2015, were added. These cases were obtained from the Pathology archives of Hospitals with recognized expertise in treatment of EOC, members of the Neuro-Oncological Network of the Piedmont Region (Italy) (AOU Città della Salute e della Scienza of Turin, AO San Giovanni Bosco Hospital of Turin, AOU Maggiore della Carità of Novara, AO S. Croce e Carle Hospital of Cuneo, AO SS. Antonio, Biagio e Cesare Arrigo of Alessandria, ASL CN2 Hospitals of Alba and Bra, AO Martini, Hospital of Turin, AO Maria Vittoria, Hospital of Turin, AO Mauriziano, Hospital of Turin),and from Azienda Socio Sanitaria Territoriale (ASST) Spedali Civili of Brescia, Lombardy Region (Italy).
Inclusion criteria were: i) clinical evidence of CNS progression by radiological imaging; ii) access to follow-up data; iii) availability of histological blocks of at least the primary ovarian tumor.
For each case the following clinical and pathological parameters were collected: i) age at diagnosis; ii) date of diagnosis of both primary ovarian cancer and CNS metastasis; iii) morphological and histological features of both ovarian tumor and brain metastasis (histotype and grade); iv) FIGO stage at diagnosis and residual tumor after surgery; v) date and site of first relapse; vi) treatments received at diagnosis and relapse; vii) date of death or last follow-up; viii) overall survival (OS), counted as the time from the date of EOC diagnosis to the date of death or last follow-up; ix) progression free survival (PFS) estimated as the time from EOC diagnosis to the date of first clinical relapse; x) progression brain metastasis free survival (bPFS) calculated as the time from primary tumor diagnosis to brain metastasis development; xi) brain metastasis overall survival (bOS) determined as the time from the date of brain metastasis diagnosis to death or last follow-up. Nineteen new controls, diagnosed from June 2011 to June 2016, without CNS progression but with similar clinical and histopathological features and follow up were selected from the clinical records of Candiolo Cancer Institute (FPO-IRCCS).
The study was submitted to and approved by the Ethic Institutional Review Board for "Biobanking and use of human tissues for experimental studies" of the Pathology Service of the AOU Città della Salute e della Scienza (Turin, Italy). The project provided a verbal and not written informed consent from the patients due to the retrospective approach of the study, which did not impact on their treatment. All the cases were anonymously recorded. The Institutional Review Board approved this consent procedure.

Immunohistochemistry procedures
Haematoxylin and eosin (H&E) slides were assessed and the most representative paraffin block of each lesion was selected for immunohistochemistry (IHC). A threemicrometer-thick section was collected on a SuperFrost Plus slide and, IHC reaction against AR (monoclonal antibody, clone SP107, pre-diluted in Tris Buffer, pH 7.3-7.7, with 1% BSA and < 0.1% Sodium Azide, Ventana, Roche) was performed on an automated immunostainer (VentanaBenchMark XT AutoStainer, Ventana Medical Systems, Tucson, AZ, USA). Nuclear staining was considered positive. Prostate tissue was used as positive control, while a triple negative basal-like breast cancer sample was used as negative control.

Statistical analysis
All statistical analyses were performed using SPSS software for Windows (version 22.0; SPSS Inc., Chicago, IL). Categorical variables were initially compared with Pearson Chi-square test, but when results were not reliable, Fisher exact test has been considered for further statistical analyses. Continuous variables were compared using analysis of variance (ANOVA) or dependent T test for paired samples. Kaplan-Meier curves were drawn to analyse survival outcome using the log-rank test method. P values < 0.05 were considered significant, and all tests were two-tailed.
All cases relapsed, with a median PFS of 15 months (range 0-62). Median age at diagnosis of CNS involvement was 59 years, with a median bPFS of 25 months (range 0-87 months). Fourteen out of 29 (48%) patients developed brain metastases as the first site of relapse.
For more detailed features of CNS involvement in the combined dataset see Table 2.
Cases and controls of the validation cohort were comparable: no statistical difference was observed for age, histotype, tumor grade, FIGO stage, and first line chemotherapy (see Table 1). Significant differences between two group were observed for upfront surgery vs interval cytoreductive surgery, absence/presence of residual tumor after surgery, incidence of relapse and number of surviving patients.

AR expression in cases and controls of the validation cohort
Immune-histochemical stainings were performed on both cases and controls of the validation cohort. Table 3 shows comparisons between cases vs controls, considered as continuous variables.
For AR protein expression in cases and controls of the validation cohort as dichotomized variable see Table 4.

AR expression is significantly reduced in cases vs controls' validation cohort
Case dataset vs. control dataset AR shows a statistically significant difference between the subgroups when considered as continuous variable (mean case dataset: 13.21%; mean control dataset: 43.21%, p < 0.001, see Table 3).  Moreover, considered as dichotomized variable, different expression of AR among the two populations emerges (p = 0.003 for10% cut-off and p < 0.001 for IRS). Thus, odds-ratio (OR) was evaluated for 10% AR's cut-off and the risk to develop a brain metastasis was 8, 33 times higher in women with AR-negative primary EOCs (CI 95%: 2.15-32.29). (See Table 4).

Combined analysis
We analyzed the combined dataset including previous published cohort and new validation cohort, for a total of 40 cases and 40 controls.
All cases relapsed, with a median PFS of 18 months (range 0-62). Median age at diagnosis of CNS involvement was 59.5 years, with a median bPFS of 25 months (range 0-87 months). Twenty one out of 40 (52.5%) patients developed brain metastases as the first site of relapse. Further sites of first relapse were lymph nodes and/or peritoneum (14/40, 35%) lung or liver (5/40, 12.5%).
For more details about features of CNS involving see Table 2.

Control dataset
This group included 40 women with a median age of primary EOC diagnosis of 63.5 years (range 36-78). All of our controls were identified as high grade (G3) Table 5 shows the most relevant clinico-pathological parameters of control subgroup of combined dataset.
Cases and Controls proved to be comparable populations: no statistical difference was observed for age (p = 0.960), histotype (p = 0.262), tumor grade (p = 1), FIGO stage (p = 1), residual tumor after surgery (p = 0.611) and first line chemotherapy (p = 1) ( Table 5). There were significant differences between cases and controls, in particular upfront surgery vs interval cytoreductive surgery, incidence of relapse and number of surviving patients.
For AR protein expression of combined cohort as continuous variable of primary vs metastatic lesions and of cases vs controls see Table 6. AR expression in cases and controls dataset of combined cohort considered as dichotomized variable is showed in Table 7.
AR expression is significantly reduced in cases vs controls' combined cohort Case dataset vs. control dataset AR shows a statistically significant difference between cases and controls when considered as continuous variable (mean case dataset: 14.15%; mean control dataset: 42.05%, p < 0.001, Table 6).
Considered as dichotomized variable, the two populations showed different expression of AR (p < 0.001 for both 10% cut-off and IRS). Odds-ratio (OR) evaluated for 10% AR's cut-off was 9.43 (CI 95%: 3.290-27.020) (See Table 7). Figure 1 shows the immune-histochemical expression of AR with the corresponding H&E staining in two representative cases and two controls.
AR expression is not differentially expressed in primary ovarian cancers vs paired brain metastases Primary ovarian cancers vs paired brain metastases IHC expression of AR doesn't exhibit statically significant difference of expression between the two groups in the combined population, considering continuous variable only (See Table 6).

Kaplan-Meier's curves of combined population
Survival curves were obtained using PFS (progression free survival) and bPFS (progression brain metastasis free survival) as events of interest in the combined population. AR results associated with prognosis in this population as show the Kaplan-Meier's (KM) curves, in particular: KM curve by AR 10% and PFS (Fig. 2, curve a, p = 0.002) and KM curve by AR 10% and bPFS (Fig. 2, curve b, p = 0.005).

Discussion
Little evidence from small series of patients exists about predictors of CNS involvement from EOC. Few clinical and molecular factors have been studied, such as CD133 overexpression and platinum resistance [17], overexpression of MDR-1 (multi drug resistance 1) [18], having suffered from a previous breast cancer [19], loss of BRCA function [20] and, presence of mutations in BRCA1 and BRCA2 genes [21]. In our work we explored the predictive and prognostic role of AR. The main conclusion of the current study is that decreased AR levels predict CNS involvement from EOC. In an independent validation cohort, we showed a 8,33 times higher risk of disease's spreading to CNS if AR's amount on primary tumor is lower than 10%, statistically significant both for AR considered as dichotomized or continuous variable (see Table 4). Moreover, in our complete dataset, including cases and controls of previous work [14], we confirmed these data, with AR expression significant reduction in cases vs controls (see Tables 6 and 7) and a more than 9 times greater risk of developing brain metastases if AR showed less than 10% of expression (see Table 7).
In combined dataset AR levels also have a prognostic value, as reported in KM curve (Fig. 2, curve a). Indeed, a reduced AR, is associated with shorter PFS, regardless of site relapse.
Our data are consistent with other studies performed in EOC [16,22,23] breast [24,25] and endometrial cancer [26] that suggest a positive prognostic role of AR expression.
Most importantly we show for the first time that besides causing earlier involvement of any other site, AR loss also has a negative prognostic relevance if bPFS is considered: lower AR levels correlate with an earlier CNS spreading (Fig. 2, Curve b).
Moreover, we also first compare AR expression in primary and paired CNS metastatic site (See Table 6).
Data on this topic are limited [27,28]. When this receptor is considered as continuous variable, no difference of expression is observed between ovarian and brain lesions, differently from the previous study [14] (see Table 6); it is clear that a correlation to dedifferentiation exists, since metastatic lesions disclose lower percentages of the evaluated protein compared with primary ovarian tumors.
It is worthwhile mentioning that 21 cases (52.5%) of the combined population had the CNS as the first site of relapse. Such tendency may be explained with the improvement of chemotherapeutic agents, able to contrast typical routes of metastatic spread, though unable to cross the blood-brain barrier (BBB) [29].
Our study has some limitations. First, sample size was increased, but is still relatively small. Second, BRCA mutational status was known only in 4/40 (10%) of cases (3 were mutated, one was not). This lack of information may be important since BRCA mutation is a known independent positive predictive and prognostic factor [30]. Third, in 22.5% of patients residual disease after surgery is unknown. This is a limit because macroscopic residual tumor after surgery is an independent prognostic factor in ovarian cancer [31], although it is unlikely to impact specifically on the development of brain metastases. Fourth, nearly all patients in our study have serous cancers, a good prognostic factor for patients with CNS involvement [11]. We could have selected a population with better prognosis, but in any case it appears to be well balanced to the control group as regards histology (see Table 5).
In conclusion, we confirmed in an independent validation cohort that AR reduction is predictive of CNS involvement from EOC. Assessment of AR levels might help in the identification of high risk patients who may benefit from dedicated follow up procedures to anticipate diagnosis and treatment.
However, before applying this biomarker to clinical practice, further evidence on larger series of EOC patients is needed.