Expression of ERBB Family Members as Predictive Markers of Prostate Cancer Progression and Mortality

Simple Summary Patients diagnosed with prostate cancer are usually offered a standard treatment plan based on their Gleason score, stage, and prostate-specific antigen (PSA) level. However, studies on other cancers have shown the importance of using biomarkers in addition to clinical and pathologic parameters to personalize therapeutic decisions. Given the important heterogeneity in the natural history of localized prostate cancer, novel prognostic biomarkers would aid in patient stratification and decision making. Here, our study shows that members of the ERBB family are markers that have high prognostic value for predicting biochemical relapse, metastasis development, and even prostate cancer-related mortality. The integration of these markers into clinical practice may eventually lead to more appropriate therapeutic decisions in newly diagnosed patients and potentially reduce prostate cancer morbidity and mortality. Abstract Background: EGFR, ERBB2, ERBB3, and ERBB4 are growth receptors of the ERBB family implicated in the development of epithelial cancers. Studies have suggested a role for EGFR and ERBB3 in the development of prostate cancer (PC), while the involvement of ERBB2 and ERBB4 remains unclear. In this study, we evaluated the expression of all members of the ERBB family in PC tissue from a large cohort and determined their contribution, alone or in combination, as prognostic markers. Methods: Using immunofluorescence coupled with digital image analyses, we quantified the expression of EGFR, ERBB2, ERBB3, and ERBB4 on radical prostatectomy specimens (n = 285) arrayed on six tissue microarrays. By combining EGFR, ERBB2, and ERBB3 protein expression in a decision tree model, we identified an association with biochemical recurrence (log rank = 25.295, p < 0.001), development of bone metastases (log rank = 23.228, p < 0.001), and cancer-specific mortality (log rank = 24.586, p < 0.001). Conclusions: Our study revealed that specific protein expression patterns of ERBB family members are associated with an increased risk of PC progression and mortality.


Introduction
Prostate cancer (PC) is one of the most commonly diagnosed and lethal cancers in men worldwide. PC is a heterogeneous disease encompassing low-(slow and nonaggressive progression) and high-risk (rapid progression) diseases. Approximately a quarter of patients will develop the latter, characterized by the development of metastasis and subsequent death [1]. Currently, clinicians use prostate-specific antigen (PSA) levels, Jurkat T lymphoma cells were kindly provided by Dr. Lapointe Réjean (CRCHUM), while MCF-7, SKOV3, and all PC cell lines (22Rv1, LNCaP, DU145, and PC3) were obtained from the American Type Culture Collection (ATCC, Manassas, VA, USA). Jurkat T lymphoma cell and PC cells were maintained in RPMI 1640 medium (Wisent Inc., St-Bruno, QC, Canada), MCF-7 was grown in DMEM medium (Wisent Inc.), and SKOV3 in OSE medium (Wisent Inc.). All culture media were supplemented with 10% fetal bovine serum (FBS) (Gibco ® , Thermo Fisher Scientific, Waltham, MA, USA), 0.454 µg/mL of amphotericin B (Wisent Inc.), and 90 µg/mL gentamycin sulfate (Wisent Inc.).

Creation of Cell Line Pellets
Cell line pellets were used as a control on tissue microarray (TMA) and prepared as previously described [31]. This method was developed to fix and embed cell suspensions in paraffin using HistoGel™ (Thermo Fisher Scientific) to ensure high cell density per core when arrayed on a TMA. The embedded cell suspensions in paraffin made it possible to reproduce the same conditions with which the patient samples were tested.

Patient Cohort
The TF123 cohort included 300 primary PC patients who underwent radical prostatectomy at the Centre hospitalier de l'Université de Montréal (CHUM, Montréal, QC, Canada) between 1993 and 2006. Each patient signed an informed consent form for their participation in the Centre de recherche du Centre hospitalier de l'Université de Montréal (CRCHUM) PC biobank. The CRCHUM ethics review committee approved the study. A total of 15 patients were excluded due to preoperative hormone therapy. The time to BCR was defined as the time interval between the date of surgery and an increase in PSA levels above 0.2 ng/mL and rising, or when a decision to institute additional therapy was made.

Construction of TMA
A specialized genitourinary CHUM pathologist identified and traced out regions of cancer (Tumor: T), as well as adjacent non-cancerous areas (adjacent benign: BA) on fresh hematoxylin and eosin-stained slides obtained from formalin-fixed paraffin-embedded (FFPE) specimens. Two or three cores (0.6 mm) of BA and T were arrayed on two separate recipient blocks using a TMA array (Pathology Devices, Inc., Westminster, MD, USA). This TMA series (TF123) was composed of a total of six TMA blocks.
To properly identify basal cells, a cocktail containing antibodies against p63 (1:650, 4A4, Ab-1, Neomarkers, Fremont, CA, USA) and high molecular weight cytokeratin (1:50, 34bE12, CLSG36689-05, Cedarlane, Fremont, CA, USA) was applied for 45 min to the section, and this was followed by the secondary fluorescent antibody Alexa Fluor ® 488 goat anti-mouse IgG (1:250, Thermo Fisher Scientific). Following a DAPI staining, to identify nuclei, each slide was incubated for 15 min at room temperature with a 0.1% solution of Sudan Black B (Research Organics, Cleveland, OH, USA) in 70% ethanol to quench tissue autofluorescence.
Finally, slides were mounted using Fluoromount™ Aqueous Mounting Medium (F4680, Millipore Sigma, Burlington, MA, USA). A negative control slide was performed in parallel (one for each biomarker) and incubated with PBS instead of the primary antibodies, then processed with the appropriate secondary antibodies.

Digital Image Analyses and Pre-Processing of Scoring Data
All slides were scanned within 24 h with a 20× Olympus Optical microscope BX61VSF (Olympus, Shinjuku, Tokyo, Japan) and visualized with OlyVIA software (Olympus). Scanned images were imported to VisiomorphDP software (Visiopharm, Hoersholm, Denmark). This software allows the development of semi-automated analysis protocol packages (APPs) to determine the expression levels of each biomarker by the mean fluorescence intensity (MFI) in each compartment (i.e., stroma and epithelium cytoplasm) [6].
We performed quality control of the tissue cores to exclude those that were damaged during the processing or cores containing less than 5% of epithelial cells. Duplicate cores presenting with significant differences were identified with scatter plots and Mann-Whitney test using GraphPad Prism software V6 (GraphPad, La Jolla, CA, USA). We then reviewed the images to determine if the difference observed was due to a technical issue. In such cases, the core was excluded from the analysis. However, data were kept if no unspecific staining anomaly was noted. To properly compare all TMAs together, the mean fluorescence intensity values of each core were normalized according to a calculated ratio. This ratio results from the mean fluorescence intensity across all TMA sections for a given biomarker divided by the mean fluorescence intensity (biomarker) for a given slide.

Statistical Analysis
Statistical analyses were performed with SPSS Statistics 25.0 software package (SPSS Inc., Chicago, IL, USA). To compare biomarker expression between tissue compartments (epithelial versus stroma), a Mann-Whitney test was used. To identify the appropriate threshold for survival analyses, data were displayed as quartiles to explore data trends and identified the percentile providing the best dichotomization for each biomarker. Survival analyses were performed using the Kaplan-Meier method coupled with a log-rank test. Univariate and multivariate Cox regression analyses were used to estimate the hazard ratios (HR) for each biomarker. A two-sided p-value < 0.05 was considered statistically significant. The construction of the decision tree model was done using R software version 3.4.3 with RPART package (R Core Team, R Foundation for Statistical Computing, Vienna, Austria).

Antibody Validation in PC Cell Lines
Although all antibodies have already been reported in the literature, we validated their specificity in a Western blot assay. We observed that all antibodies showed specific bands ( Figures S1A and S2). We noted that the EGFR protein levels were higher in DU145, 22Rv1, and PC3 cells when compared to the LNCaP cell line. PC cell lines only weakly express ERBB2 with greater expression in LNCaP and 22Rv1 cells. No ERBB3 expression was detected in LNCaP or PC3 cells, while 22Rv1 and DU145 cells presented high expression. Finally, only the 22Rv1 cell line expressed ERBB4. Jurkat T lymphoma cells were used as control (negative) for ERBB receptors along with the well characterized MCF-7 (EGFR-, ERBB2-, ERBB3+, and ERBB4+) and SKOV3 (EGFR+, ERBB2+, ERBB3 weak, and ERBB4+) cell lines ( Figures S1A and S2). Since in this study, we used these antibodies in formalinfixed paraffin-embedded tissue, we created, fixed, and embedded cell pellets from these cell lines and performed an immunofluorescence (IF) assay. We noted that the expression of the ERBB family members was similar to those observed in the Western blot ( Figure S1B).

Patient Characteristics and Clinical Parameters
The analyzed TF123 TMA series was composed of 285 PC patients who did not receive neoadjuvant androgen deprivation therapy before radical prostatectomy. This was a mature cohort with a median follow-up of 129 months. Their demographic, histopathological, and clinical parameters are detailed in Table 1. The incidence of BCR at five years was 33% (94 patients), the incidence of bone metastasis at 10 years and death specific mortality were 6.3% (18 patients).

ERBB Family Member's Expression in Human PC Specimens
To assess the usefulness of the ERBB family members as PC prognostic markers, we performed a multiplex IF assay incorporating one receptor (red: EGFR, ERBB2, ERBB3, or ERBB4) with specific masks to define the epithelium (yellow: CK8/18), the basal cells (green: p63/CKHMW, present in benign/normal prostate glands), and the nucleus (blue: DAPI) on the TF123 TMA series.
As expected, EGFR, ERBB2, ERBB3, and ERBB4 presented a membrane and cytoplasmic localization in the epithelium of both T ( Figure 1A) and BA tissue cores. EGFR expression was significantly higher in BA compared to T tissue (p < 0.0001, MFI = 705 vs. 654, Figure 1B) as opposed to ERBB2 (p = 0.0230, MFI = 112 vs. 115, Figure 1C) and ERBB3 (p < 0.0001, MFI = 1126 vs. 1163, Figure 1D). Since we observed a low level of expression of ERBB4 in the PC cell lines and their derivates, and to avoid any waste of material, we decided to stain only one TMA slide with ERBB4. Despite the specificity of the antibody used, ERBB4 did not show a clear signal compare to background staining and had an MFI similar to the negative control ( Figure 1E). Therefore, we did not include the analyses with ERBB4. We also noted that the expression of all receptors was significantly weaker in the stroma when compared to the epithelium ( Figure 1B

EGFR, ERBB2, and ERBB3 Expression Is Associated with an Increased Risk of BCR at 5 Years
To quantitate the expression of each biomarker, we performed digital image analysis of each core using an algorithm that detected only the epithelial compartment. This algorithm targets the region of epithelial cells stained by the cytokeratins (CKs) cocktail used for the detection of epithelial cells then measures the fluorescence intensity in the channel corresponding to the marker of interest.
To assess the prognostic capacity of the ERBB, we first evaluated if they were associated with BCR (less than five years). To determine the appropriate threshold for each biomarker to dichotomize their expression levels, we used the quartiles methods. Using Kaplan-Meier curves coupled with a log-rank test, we observed an increased risk of BCR with the high expression of EGFR (75th percentile defined as EGFR high ; log rank = 5.861, p = 0.015) (Figure 2A), while ERBB2 did not show such significance (under 50th percentile defined ERBB2 low ; log rank = 2.441, p = 0.118) ( Figure 2B). Finally, the low expression of ERBB3 was also an indicator of BCR (25th percentile defined as ERBB3 low ; log rank = 3.768, p = 0.052) ( Figure 2C). rs 2021, 13, x FOR PEER REVIEW 8 of 18   Univariate Cox regression analyses also demonstrated that EGFR high in continuous (HR = 1.006, CI = 1.003-1.009, p = 0.001) or dichotomized (HR = 1.703, CI = 1.0097-2.644, p = 0.018) values, as well as dichotomized ERBB3 low values (HR = 0.650, CI = 0.418-1.011, p = 0.056), showed an association with an increased risk of BCR (less than five years) ( Table 2). Continuous ERBB2 and ERBB3 or dichotomized ERBB2 expression values failed to show significance. However, the ERBB family members were not independent of known prognostic clinical parameters (Table 2).

Combining the Expression Levels of EGFR, ERBB2, and ERBB3 Predicts BCR at 5 Years
Since all ERBB family members can dimerize with one another, we evaluated how combinations could be informative of BCR. Therefore, we developed a decision tree model, including dichotomized EGFR, ERBB2, and ERBB3 expression values. Four ERBB status groups were defined ( Figure 2D). The decision tree indicated, within a receptor status group, the number of patients with a BCR event. Thereby, we obtained a percentage reflecting the probability to experience the event within the receptor status group. To better represent differences among these newly identified receptor status groups we performed a Kaplan-Meier analysis and observed a significant overall difference (log rank = 25.295, p < 0.001) ( Figure 2E). Despite significant overall results, some groups are only distinct from one another when compared two-by-two and not from all of the other groups ( Figure 2F).
We performed univariate Cox regression analysis using the four ERBB groups and we noted a significant risk of BCR (HR = 1.609, CI = 1.276-2.027, p < 0.001) ( Table 3). More importantly, patients with PC tissue expressing EGFR high , ERBB3 high , and ERBB2 low present a higher 2.189-fold risk of experiencing a BCR that increased to a 5.455-fold higher risk when high ERGF and low ERBB3 expressions were present. In the univariate analyses, the hazard ratio was greater than all clinical parameters. However, in the multivariate analyses, these status groups were not shown to be independent of the clinical parameters (HR = 1.204, CI = 0.930-1.559, p = 0.158) ( Table 3).

Expression of EGFR, ERBB2, and ERBB3 Can Predict Bone Metastasis Development at 10 Years
Another important endpoint in PC is the development of bone metastasis, which is also recognized as a surrogate for PC mortality. We observed that both EGFR high (log rank = 8.103, p = 0.004) and ERBB2 low (log rank = 4.539, p = 0.033) were significantly associated with an increased risk of developing bone metastasis ( Figure 3A,B). However, ERBB3 expression did not confer risk for bone metastasis development (log rank = 0.889, p = 0.346) ( Figure 3C). Patients with EGFR high in their PC tissue (continuous or dichotomized values) showed an increased risk of developing bone metastasis when performing Cox regression analyses (Table 4); a risk that reached 3.462-fold (CI = 1.404-1.016, p = 0.008) when EGFR expression was dichotomized. In contrast, ERBB2 high in PC tissue was associated with a lower risk for bone metastasis (HR = 0.312, CI = 0.101-0.969, p = 0.044) and, to a lesser extent, ERBB3 high (continuous values) with a protective effect (HR = 0.994, CI = 0.989-1.00, p = 0.037) ( Table 4).  Patients with EGFR high in their PC tissue (continuous or dichotomized values) showed an increased risk of developing bone metastasis when performing Cox regression analyses (Table 4); a risk that reached 3.462-fold (CI = 1.404-1.016, p = 0.008) when EGFR expression was dichotomized. In contrast, ERBB2 high in PC tissue was associated with a lower risk for bone metastasis (HR = 0.312, CI = 0.101-0.969, p = 0.044) and, to a lesser extent, ERBB3 high (continuous values) with a protective effect (HR = 0.994, CI = 0.989-1.00, p = 0.037) ( Table 4).

Combining the Expression Levels of EGFR, ERBB2, and ERBB3 Can Predict Bone Metastasis Development
By incorporating EGFR, ERBB2, and ERBB3 expression in a decision tree model ( Figure 3D), four ERBB receptor status groups could be developed to segregate patients based on the risk of developing bone metastasis. Kaplan-Meier analyses performed using these groups revealed an overall log rank of 23.228 with p < 0.001 ( Figure 3E). The Kaplan-Meier highlights two distinct patient profiles (ERBB2 high /ERBB2 low /EGFR low ) at risk of the development of bone metastases compared to two groups (ERBB2 low /EGFR high /ERBB3 high and ERBB2 low /EGFR high /ERBB3 low ) ( Figure 3F). Moreover, an overall Cox regression analysis, taking into account the combination of all ERBB receptors, showed an increased risk of developing bone metastasis (HR = 2.036, p < 0.001) ( Table 4). More specifically, our results suggest that patients with tumors expressing ERBB2 low and both EGFR high and ERBB3 high present a risk that increased by 9.273-fold (CI = 2.311-37.197, p = 0.002). The risk of developing bone metastasis reached 14.774 (CI = 2.387-81.237, p = 0.002) for patients expressing ERBB2 low and ERBB3 low coupled with EGFR high and overperformed all clinical parameters.

Combining the Expression Levels of EGFR, ERBB2, and ERBB3 Can Predict PC Mortality
The decision tree model ( Figure 4D) revealed five groups with differential risk of PC-specific mortality as demonstrated by the Kaplan-Meier analyses (log rank = 24.586, p < 0.001) ( Figure 4E). More precisely, the patients presenting ERBB3 high /ERBB2 high are the group at the least risk of PC-specific mortality, and this group was significantly distinct from all other groups, except those with the ERBB3 high /ERBB2 low /EGFR low . The group ERBB3 low /EGFR high was the group with the highest risk of specific PC mortality, then this group was significantly different from all of the other groups, except one (ERBB3 high /ERBB2 low /EGFR high ) ( Figure 4F). Cox regression analysis, including the five groups, recapitulated the overall Kaplan-Meier analyses (HR = 1.865, CI = 1.256-2.768, p = 0.002) ( Table 5). With a hazard ratio greater than all clinical parameters assessed, three groups (orange, grey, and purple) were indicators of patient prognosis. The greatest risk of PC-specific mortality was observed for patients expressing ERBB3 low coupled with EGFR high (orange: HR = 36.732, CI = 4.901-275.271, p < 0.001), followed by a combination of EGFR high /ERBB2 low /ERBB3 high (HR = 11.755, CI = 1.958-70.566, p = 0.007), and finally by EGFR low /ERBB3 low (HR = 5.671, CI = 1.143-28.129, p = 0.034). Patients with ERBB2 high and ERBB3 high were those with a better prognosis. These were not significantly different from patients with EGFR low /ERBB2 low /ERBB3 high .

Discussion
In multiple cancers, including PC, a personalized therapeutic approach based on individual tumor characteristics has become an ongoing objective for both treating physi-cians and patients. Since PC is a heterogeneous disease, and that clinical parameters are insufficient to accurately predict disease outcomes, it is important to identify new tools to aid in clinical decisions and patient management.
In this study, we showed that ERBB family members are associated with a greater risk of BCR. These findings are in line with the literature, where a high expression of EGFR is associated with PC progression [22,32,33], and a low ERBB3 expression, located in the nucleus, is associated with a worse prognosis [27] and an increased risk of BCR. Moreover, this study confirmed the absence or the very low expression of ERBB4 in PC cell lines and primary cancer tissue [34,35]. However, ERBB2 expression in PC progression is more controversial, with one study showing that a high level of ERBB2 is associated with poor prognosis [36], as measured by BCR, while a previous study from our group failed to identify any correlation with BCR [37]. However, in our current study, we identified the importance of ERBB2 as a predictor of eventual bone metastases. These results are in line with those reported in breast cancer studies, where patients presenting with an ERBB2-positive tumor are more likely to metastasize to the bone when compared to the ERBB2-negative group [38]. In addition, several biological studies in pre-clinical prostate models have shown that ERBB2 signaling plays an essential role in the progression from a castration-sensitive to a castration-resistant state associated with bone metastases [39,40]. For example, ERBB2 was shown to play a role in the progression of PC through an increase in angiogenesis, thereby facilitating the dissemination of tumor cells. ERBB2 also confers an androgen independence state, leading to cell survival and proliferation when antiandrogen therapy is used [39]. The contradictory studies looking at ERBB2 expression and the correlation to outcomes in patient tumor tissue could reflect differences in results both in the composition and size of cohorts studied or by differing sources of antibodies used in the studies.
Most studies have highlighted a link between a single member of the ERBB family with PC progression using BCR as an endpoint [22,23,36,37]. When two markers were assessed in the same publication, it was discussed that EGFR is a better predictor of BCR when used alone than in combination [22]. Another study used an heterogenous set of specimens from patients treated (n = 29) or not treated with androgen deprivation therapie (n = 29) to perform their survival analyses in combining EGFR with ERBB2 [23]. They showed that patients presenting EGFR high and ERBB2 high were more likely to experience BCR. In this present study, we combined three out of the four ERBB family members and evaluated their prognostic value against three clinically relevant PC endpoints. We excluded ERBB4 from all analyses in reason of the low expression of this biomarker in prostate cell lines by Western blot and the non-specificity of IF staining for ERBB4 in patient samples. We found that the combination of ERBB3 low coupled with EGFR high was associated with the worst prognosis across all endpoints (BCR, development of bone metastasis, and PC-specific survival). We also noted a major role of ERBB2 in the development of bone metastasis and this marker was found to be the first marker used to stratify patients in the decision tree model. The important role of ERBBs in the development of metastases has previously been reported in the literature [40][41][42]. Indeed, in a pre-clinical model using cell line in vivo assays, it was demonstrated that EGFR promotes the survival of PC-circulating tumor cells, while ERBB2 supports cancer cell growth in bones by promoting the RANK signaling pathway [40]. These results support our findings of the association of the ERBB members with worse prognosis.
Since ERBB receptors are differentially expressed in radical prostatectomy specimens, they should form different combinations of dimers (homo-/heterodimer) to activate or inhibit diverse cellular pathways. These different dimerizations, in addition to the fact that the trials did not measure the levels of ERBB receptors, could explain why previous clinical trials studying ERBB family members failed to demonstrate a predictive effect on patient outcomes [43][44][45]. Moreover, it would be interesting to investigate how members of the ERBB family affect anti-androgen therapy. Our results also highlight that the different receptor combinations in the primary tumor are associated with very different outcomes much later in the post treatment setting in these patients. These molecular markers may provide early indicators of patients with worse prognosis requiring more intense follow-up strategies and possibly earlier and more aggressive therapeutic strategies.

Conclusions
Our results suggest that a different combination of ERBB could be useful to stratify patients following local therapy for PC. We demonstrated that patients presenting with EGFR high coupled with ERBB3 low were at a 5-fold increased risk of BCR. Patients expressing ERBB2 low had a 14-fold increased risk of developing bone metastasis, and were more than 36 times at higher risk of PC mortality. These biomarkers may become useful in the clinic if they are further validated on larger cohorts.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/10 .3390/cancers13071688/s1, Figure S1: Evaluation of antibody specific for all ERBB family members. Figure S2. Whole Western blots of ERBB expression in prostate cancer (PC) cell lines. Informed Consent Statement: All subjects gave their informed consent in the PC biobank of the CHUM, affiliated to the Réseau de la recherche sur le cancer (RRCancer), for inclusion before they participated in the study.