Stromal PDGFRβ Expression in Prostate Tumors and Non-Malignant Prostate Tissue Predicts Prostate Cancer Survival

Background The identification of new prognostic markers for prostate cancer is highly warranted, since it is difficult to identify patients requiring curative treatment. Data from both experimental models and clinical samples have identified important functions of PDGFRβ on pericytes and fibroblasts in the tumor stroma. Methodology/Principal Findings In this study the prognostic significance of PDGFRβ in prostate cancer stroma, and in matched non-malignant tissue, was evaluated with immunohistochemistry. PDGFRβ expression was analyzed in normal and tumor stroma from more than 300 prostate cancer patients. High PDGFRβ expression in tumor stroma was associated with large tumor size, advanced stage, high Gleason score and high vessel density. Perivascular PDGFRβ staining in tumors was also correlated with high Gleason score. Correlations were also observed between PDGFRβ status in tumor stroma and non-malignant stroma. Similarly, high PDGFRβ expression in adjacent non-malignant tissue stroma correlated with large tumor size, advanced stage, high Gleason score and proliferation in non-malignant epithelium. Interestingly, high levels of PDGFRβ in the stroma of tumor and non-malignant tissue were associated with shorter cancer specific survival in prostate cancer patients. Conclusions/Significance The study revealed a number of novel associations between stromal PDGFRβ expression in prostate tumors and several important clinical characteristics, including survival.


Introduction
Tumor behavior is, in general, largely governed by characteristics of the tumor stroma. In the case of prostate cancer this is well illustrated by a series of recent studies that have identified stromal characteristics and markers with prognostic and response-predicative significance.
Tumors with the most pronounced alteration in stroma morphology, characterized by a major loss of smooth muscle cells and increases in fibroblasts, myofibroblasts, and collagen fibres (reactive stroma grade 3), have a poor outcome compared to those with a stroma morphologically more similar to that in normal prostate tissue, i.e. reactive stroma grades 1 and 2 [1]. Recent studies have also shown that prostate cancers are characterized by loss of androgen receptors (AR) in the stroma [2]. A low frequency of stromal AR was coupled to increased Gleason score, metastasis, poor response to castration therapy and an unfavorable outcome. Furthermore, stromal characteristics, such as angiogenesis, accumulation of macrophages, lymphocytes and mast cells as well as changes in the extracellular matrix have been linked to variations in prostate tumor behaviour [3,4,5,6,7]. Notably, it has also been observed that cancer field effects and/or adaptive changes may change the stroma and glandular epithelium in the surrounding non-malignant tissue in a prognostically significant manner [2,8,9,10].
PDGF aand b-tyrosine kinase receptors exert important control functions in mesenchymal cells, such as pericytes, fibroblasts and vascular smooth muscle cells during development [11,12]. Experimental studies have identified different functional effects of stromal PDGF receptor signaling in various tumor models. It has been demonstrated that paracrine activation of PDGF receptors on fibroblasts acts as a potent signal for tumor stroma recruitment [13,14,15]. Other studies have demonstrated therapeutic benefits of targeting stromal PDGF receptors involving either direct anti-tumoral effects, as well as beneficial effects on tumor drug uptake [16,17,18,19,20]. Clinical significance of the findings from experimental models is indicated by numerous studies demonstrating stromal PDGF receptor expression in different human solid tumors [13]. Most recently these studies have been supplemented with analyses revealing that high stromal PDGF receptor expression is a marker of an unfavorable outcome in breast cancer patients [21].
The biological effects of PDGF receptors in tumor fibroblasts and pericytes together with the advent of drugs with PDGF receptor-inhibitory activity, such as imatinib, sorafenib and sunitinib, thus motivates a systemic characterization of the expression pattern of PDGF receptors in human solid tumors. To this end, this study describes the expression of PDGFRb in approximately 300 cases of prostate cancer and in matched surrounding non-malignant prostate tissue, and also reports on associations between PDGFRb expression and molecular, histopathological and clinical characteristics.

Variable PDGFRb expression in normal and tumor prostate tissue
In order to evaluate the significance of PDGFRb in prostate cancer, a TMA containing matched non-malignant and tumor tissue from 377 prostate cancer patients with up to 25 year of follow-up was analyzed by PDGFRb immunohistochemistry.
In agreement with previous studies, PDGFRb expression was predominantly found in the fibromuscular stroma and in perivascular cells ( Figure 1 and Table 1). In the non-malignant prostate 27% and 18% were scored as having positive PDGFRb expression in the fibromuscular stroma and perivascular cells, respectively. Perivascular and stroma staining were not correlated. In the tumor areas 34% were positive in the fibroblast-like stroma and 17% in the perivascular cells. Staining in tumor stroma and around tumor blood vessels were correlated (Rs = 0.59). Significant associations were also observed between the PDGFRb status in the malignant and non-malignant stroma (Table 1).
In some cases, the PDGFRb expression also varied within the prostate tissue of the same patient ( Figure S1).

Stromal PDGFRb expression occur predominantly in aSMA-positive cells
To further characterize the stromal PDGFRb expression, a double staining was performed with antibodies against aSMA and PDGFRb.
These analyses confirmed that the majority of the PDGFRb expression in the non-malignant fibromusclar and the fibroblastlike tumor stroma occurred in aSMA-positive cells ( Figure 2). Furthermore, the analyses also demonstrated the presence These findings thus confirm the existence of subsets of stromal aSMA-positive cells, which differ with regard to PDGFRb expression.

Stromal and perivascular PDGFRb staining in prostate tumors is correlated with prognostic markers
Analyses were performed to investigate possible associations between PDGFRb expression in tumors and clinical parameters or histological characteristics of the tumors.  Presence of PDGFRb staining in the fibroblast-like tumor stroma was significantly positively correlated with large tumor size, advanced stage, high Gleason score and high vessel density (Table 1). In patients with a Gleason score 4-6 or Gleason score 7 tumor, only about 20% of the tumors showed stroma PDGFRb staining, but it was considerably more common in cases with Gleason score 8-10 tumors (Figure 3 A).
Perivascular PDGFRb expression in the tumor area was also positively correlated with advanced stage, tumor vessel density and high Gleason score (data not shown).
Together these findings demonstrate previously un-recognized associations between stromal and perivascular PDGFRb expression in prostate tumors and characteristics associated with a more aggressive cancer phenotype.
Stromal PDGFRb staining in the adjacent non-malignant tissue are correlated with prognostic markers PDGFRb expression in adjacent non-malignant tissue was also analyzed with regard to associations with tumor characteristics (Table 1).
High PDGFRb in non-malignant, fibromuscular stroma was correlated with large tumor size, advanced stage, epithelial cell proliferation and high Gleason score (Table 1). It was considerably more common to find PDGFRb expression in the normal prostate tissue stroma when a Gleason score 8-10 tumor was present elsewhere in the organ compared to if the tumor was Gleason score 4-6 (Figure 3 B).
Perivascular PDGFRb expression in the surrounding nonmalignant tissue was also positively correlated with epithelial cell proliferation (data not shown).
These analyses thus suggest novel relationships between prostate cancer properties and the phenotype of the stroma of the nonmalignant prostate tissue adjacent to prostate tumors.
Stromal PDGFRb expression in tumors and in the nonmalignant prostate tissue surrounding tumors predicts cancer specific survival The results from the analyses of PDGFRb expression were finally combined with survival data to investigate possible prognostic significance.
The cut-offs for tumor stromal PDGFRb and non-malignant stromal PDGFRb in the analyses was set to the third quartile, corresponding to 1.0 and 0.5 for tumour and non-malignant stromal PDGFRb staining, i.e. high PDGFRb immunoreactivity was $1.0 for tumour and $0.5 for non-malignant stroma.
Kaplan-Meier analysis showed that the watchful waiting patients with the highest quartile of PDGFRb expression in tumor stroma had a significantly shorter cancer specific survival compared to the rest (15-year probability of event-free survival (P-EFS) was 6765% and 3469% in the two groups) (Figure 4 A).
Interestingly, high expression of PDGFRb in the stroma of adjacent normal prostate tissue was also associated with a significantly reduced survival in patients managed by watchful waiting (15-year P-EFS was 6865% and 4169% in the two groups) (Figure 4 B).
High tumor and non-malignant stroma PDGFRb immunostaining was associated with an increased relative risk for prostate cancer specific death in a univariate Cox regression analysis ( Table 2). In multivariate Cox regression analysis including the known prognostic marker GS and local tumour stage, high stromal PDGFRb in tumor and non-malignant tissue was not an independent prognostic marker (data not shown).

Discussion
The present analyses uncovered a set of novel associations between the tumor stroma PDGFRb status and histopathological characteristics, including positive correlations with Gleason score, tumor stage and tumor size (Table 1). These findings are reminiscent of the recently described situation in breast cancer where stromal PDGFRb expression was positively correlated with high grade [21]. These findings raise the question of the underlying mechanism(s). As of now it is not possible to conclude whether the associations are caused by an epithelial-induced stromal phenotype or by a stroma-induced epithelial phenotype.
The analyses of the present study also revealed significant associations between clinical characteristics, including survival, and the stromal PDGFRb expression in the non-malignant present tissue (Table 1). This observation of prognostically significant properties of the non-malignant tissue adds to a series of similar observations that together have led to proposing the term ''tumor indicating non-malignant tissue'' (TINT), which implies that analyses of non-malignant prostate tissue can yield prognostic information [2,10]. Other recent studies have presented similar findings following analyses of e.g. pAKT, pEGFR and AR [2,10,22]. From a practical perspective this is important since many diagnostic biopsies only sample non-malignant tissue.
The underlying biology of the variations in stromal PDGFRb expression should be better clarified. From the present analyses it is unclear whether the alterations in the non-malignant tissue are secondary to the presence of near-by malignant tissue as suggested in animal models [8], or if it rather reflects a tumor-permissive environment which causally contributes to formation of more malignant tumors.
The analyses including double staining with aSMA and PDGFRb antibodies revealed that the PDGFRb expression occurred in aSMA -positive cells (Figure 2). However, it was also noted that these cells, in tumors scored as having a PDGFRbpositive stroma, occurred together with aSMA -positive/ PDGFRb-negative cells. These findings thus clearly exemplify the molecular heterogeneity among the cells of the non-malignant fibromusclar stroma and of the fibroblast-like tumor stroma. Similar findings have been made in experimental tumors [23]. As also suggested by others [24], the findings should stimulate to an improved classification of these subsets and continued analyses of their functional significance.
The associations between high stromal PDGFRb expression and shorter survival indicate that this receptor is causally linked to the clinical course of the disease and thus also suggests targeting of PDGF receptors for therapeutic purposes. Some smaller studies have been reported where imatinib have been used in prostate cancer without any obvious effects [25,26]. The findings from the present study, demonstrating large variations in stromal PDGFRb expression, suggest that future studies evaluating this approach will require more stringent procedures for patient selection. In this context it can also be noted that epithelial PDGFRb expression was only rarely detected in this study (data not shown). Furthermore, no associations between this expression and clinical or histopathological characteristics were observed.
A series of topics for future studies are suggested by the present findings. Experimental studies involving co-cultures of epithelial and stromal cells should help to clarify if the stroma-epithelial associations have their predominant mechanistic basis in the stromal or epithelial compartment. Another interesting topic for future studies would be to investigate to what extent the stromal and epithelial phenotypes are linked to individual prostate cancer risk genes. Finally, continued analyses are required to understand if the tumor-associated variations in the non-malignant tissue are a cause or a consequence of the tumor formation. Careful tissue analyses in prospective studies might be required to clarify this issue.

Patients and tissue microarray
Tissue specimens were collected from patients who underwent transurethral resection of the prostate (TURP) at the hospital in Vä sterås, Sweden, between 1975 and 1991. Histological analysis showed presence of prostate cancer. Median age at TURP was 74 years (range 51-95 years). Information concerning absence or presence of benign prostate hyperplasia was not available. Tissue specimens were formalin-fixed and paraffin-embedded followed by regrading according to the Gleason system. The specimens were used to construct a tissue micro array (TMA) using a Beecher Instrument (Sun Prairie, WI, USA). The TMA:s contained 5-8 samples of tumor tissue representing both the primary and secondary Gleason grade and 4 samples of non-malignant tissue from each patient. The patients had not received any anti-cancer  therapy before TURP. Radio nuclide bone scan was performed shortly following diagnosis for detection of metastases. There were 377 patients included in the study, of which 293 patients were followed with watchful waiting after TURP. At symptoms from metastases patients received palliative treatment with androgen ablation and in a few cases radiation therapy or oestrogen therapy, according to therapy traditions in Sweden during that time. Also, 84 patients that were treated with palliative treatment immediately after diagnosis were analyzed. The median overall survival for the patient group followed with watchful waiting was 5.6 years. Ninety-two of the TURP specimens were graded as Gleason score (GS) 4-5, 107 patients as GS  In this material we have already analyzed factors of potential prognostic significance such as Gleason score, tumor volume, tumour stage, tumor cell proliferation and vascular density and the data obtained were now related to the current PDGFRb findings [2,27].

PDGFRb immunohistochemistry
PDGFRb immunohistochemistry was performed as described in Nupponen et al 2008 [28]. Anti-PDGFRb (rabbit monoclonal, #3169, Cell Signaling Technology, Danvers, MA, USA) was used at a concentration of 2 mg/ml. The staining intensity was scored separately in stroma and around vessels as negative (0), weak (1), moderate (2) or strong (3). The PDGFRb staining score are the median values of five to eight scored samples of tumor tissue or four scored samples of nonmalignant tissue. For correlation analysis the samples were scored as positive for PDGFRb if staining was detected in at least one of the TMA cores.

Immunofluorescence
Immunofluorescent stainings of the TMAs was performed as above (in the immunohistochemistry section) until the incubation with the anti-PDGFRb antibody. Sections were then washed in PBS with 0.1% Tween-20 (PBT) 365 min. This was followed by a 1 h incubation with the second primary antibody, mouse monoclonal aSMA (M0851, clone 1A4, Dako Cytomation, Glostrup, Denmark) diluted 1:100 in 20% goat serum in PBT. After washing in PBT for 365 min, slides were incubated with biotinylated goat anti-mouse antibody (E0432, Dako Cytomations 1:500) for 45 min at RT. This was followed by another 365 min washing in PBT. The slides were then incubated with Alexa-Flour 488 goat anti-rabbit secondary antibody (A11008, Invitrogen, Carlsbad, CA, USA) diluted 1:100 in PBT for 45 min at RT. Following washing; the slides were incubated for 45 min with Cy3-Streptavidin conjugate (Sigma-Aldrich) diluted 1:500 in PBT for 45 min at RT. After washing the slides were dried and mounted with Vectashield mounting medium with DAPI (Vector Laboratories, Burlingame, CA, USA).

Statistics
Bivariate correlations were calculated with the Pearson's correlation test. Correlations between categorical variables and continuous variables were analyzed using the Spearman's rank correlation test. Data was collected at the time of prostate cancer diagnosis.
Patients included in survival analyses with the Kaplan-Meier and Cox regression were followed with watchful waiting. The duration of event-free survival (EFS) is defined as the time from TURP until the date of prostate cancer death, death of other cause, or until the date of last follow-up. Event in the survival analysis was defined as prostate cancer death, thereby showing cancer-specific survival. Differences in outcome were tested with the log-rank test. The prognostic significance of PDGFRb immunoreactivity was evaluated with Cox regression analysis alone and combined with GS and local tumour stage. Probability of event-free survival (P-EFS) is presented6standard error (SE). The level of statistical significance was defined as P ,0.05 (twosided). Statistical analysis was performed using the SPSS 17.0.0 software for OS X (SPSS Inc., Chicago, IL, USA).