Fibroblast as a critical stromal cell type determining prognosis in prostate cancer

Abstract Background Tumor stroma associates with prostate cancer (PCa) progression, but its specific cellular composition and association to patient survival outcome have not been characterized. Methods We analyzed stromal composition in human PCa using multiplex immunohistochemistry and quantitative, high‐resolution image analysis in two retrospective, formalin‐fixed paraffin embedded observational clinical cohorts (Cohort I, n = 117; Cohort II, n = 340) using PCa‐specific mortality as outcome measurement. Results A high proportion of fibroblasts associated with aggressive disease and castration‐resistant prostate cancer (CRPC). In a multivariate analysis, increase in fibroblast proportion predicted poor cancer‐specific outcome independently in the two clinical cohorts studied. Conclusions Fibroblasts were the most important cell type in determining prognosis in PCa and associated with CRPC. Thus, the stromal composition could be critically important in developing diagnostic and therapeutic approaches to aggressive prostate cancer.


| INTRODUCTION
Prostate cancer (PCa) is the most common noncutaneous cancer in males globally. 1 Normal prostate tissue consists of epithelium and stroma, which interact to maintain the physiological homeostasis of the organ. The stromal compartment comprises fibroblasts, smooth muscle cells, immune cells, and a collagen-rich extracellular matrix that is located between the secretory prostatic acini. In tumorigenesis, the normal organ architecture is disrupted, which activates various feedback responses both to the epithelial and stromal compartments.
In 1986, Dvorak 2 described tumors as "wounds that do not heal" and suggested that the stromal cells actively interact with the cancer cells of epithelial origin. Indeed, reactive stromal cells often form a fibrotic reaction around tumors mimicking the wound healing process. 3 with aggressive disease features 8,9 and shorter time to biochemical recurrence using needle biopsy 10 as well as prostatectomy samples. 11,12 Moreover, stromal gene expression analysis in prostatectomy samples revealed signatures specific for metastatic potential of PCa. 13 Vimentin (VIM, VIM) and alpha-smooth muscle actin (aSMA, ACTA2) have been widely used for the classification of the different stromal cell types. VIM is expressed in fibroblasts and myofibroblasts, whereas aSMA is expressed in smooth muscle and in myofibroblasts. 14  Here, we performed a quantitative analysis of PCa stroma using multiplex immunohistochemistry (mIHC) and unbiased automatic image analysis to determine the different stromal cells including fibroblasts, myofibroblasts, and smooth muscle cells in situ in human PCa tissue samples. We show that a stromal phenotype with an increased number of fibroblasts associates with aggressive disease and castration resistance. The proportion of fibroblasts independently predicts poor cancer-specific outcome in two independent patient cohorts.

| Patient material
Study cohorts were formalin-fixed paraffin embedded (FFPE) samples from transurethral resection of prostate (TURP; Cohort I, n = 159 patients) and retrospective radical prostatectomy (Cohort II, n = 350 patients). Tissue microarrays (TMA) were constructed for Cohort I by taking 1 to 5 cores per patient and for Cohort II as described earlier. 15 Prostate cancer-specific mortality (PCSM) was recorded as end-point in both cohorts. Cohort II included only patients who had non-metastatic (M0) primary PCa at diagnosis and who received prostatectomy as the first treatment (excluding also neoadjuvant therapy). Phosphatase and tensin homolog (PTEN) and androgen receptor protein expression data for Cohort II were obtained from Lahdensuo et al 16

| Immunohistochemistry
Antibodies used in the study are listed in Table S1. Antibodies for vimentin (VIM), alpha-smooth muscle actin (aSMA), and CAV2 (Caveolin-2, CAV2) were optimized for mIHC using immunohistochemistry in prostatectomy samples as described earlier. 17  were optimized earlier for mIHC. 17 The amount of VIM-positive immune cells per patient was visually scored by SB in Cohort II as "low", "medium", and "high" if <1%, 1-5%, or >5% of the stromal area was occupied by VIM-positive immune cells, respectively, which morphologically resembled either lymphocytes or macrophages. The scoring scheme follows previously published data reporting varying immune cell content PCa tissues. 18 2.3 | Multiplex immunohistochemistry mIHC was performed as described by Blom et al. 17 See full details in Supplementary Information. Briefly, for a 5-plex staining, paraffin was removed from the FFPE and heat-induced epitope retrieval (HIER) was performed. After HIER, endogenous peroxide activity and protein blocking was performed and the first and second primary antibodies were detected using tyramide signal amplification for AlexaFluor488 and AlexaFluor555, respectively (PerkinElmer, Waltham, MA). The third and fourth primary antibodies raised in different species were detected using AlexaFluor647 and AlexaFluor750 fluorochrome-conjugated secondary antibodies. Nuclei were counterstained using 4′,6-diamidino-2-phenylindole (DAPI).
The intensity of each stained marker was normalized across TMA cores of all patients. The average of the normalized marker intensity per patient was used as the final metric. The relative area for each stromal cell class (fibroblast, myofibroblast, and smooth muscle) was measured as the area of positive pixels or cell counts within stroma in each TMA core.
For a given cell class, the average relative area or relative cell count in the stroma in all TMA cores per patient was used as the final metric.
Cell class variables were categorized in quartiles when appropriate.

| Statistical analysis
Demographics for patients with high-quality image data are presented in Table S2 were calculated using the two-tailed Pearson product moment correlation coefficient function, and P value using two-tailed Student t test. Unsupervised hierarchical clustering and data plotting were performed using ClustVis 21 online tool.
For survival analyses, we used Cox proportional hazard regression model and Kaplan-Meier plots with the Wald test and log-rank, respectively. Proportional hazard assumption was tested for given variables using the Schoenfeld test. The percentage proportions of the stromal cell classes were multiplied by 10 for the Cox proportional hazard regression in order to yield hazard ratios for a ten percent change in the proportion.
If multiple tests were performed, P values were controlled for 20% false discovery rate using the Benjamini-Hochberg step-up procedure. 22 P < .05 were considered statistically significant. All statistical analyses and data plotting were performed using the IBM SPSS 24 (SPSS Inc, Armonk, NY).   Figure 3D). Specifically, patients in Cluster 6 had 83% more fibroblasts but 77% less smooth muscle than in Clusters 1-5.

| The proportion of fibroblasts in stroma is an independent predictor of clinical outcome
As the "fibroblast-high, smooth muscle-low" phenotype of Cluster 6 showed clear association with poor survival and aggressive cancer,  Abbreviation: CI, confidence interval Bolded P values remain significant after Benjamini-Hochberg procedure (P < 0.05).
(P < .001). We observed an inverse association between the proportion of fibroblast and the total stromal area, whereas the proportion of smooth muscle cells weakly correlated with the total stromal area ( Figure S5).
As Cluster 6 was enriched in CRPC, we studied the non-CRPC and CRPC patients separately. The proportions of fibroblasts were significantly higher and those of smooth muscle cells lower in CRPC cases compared with non-CRPC cases ( Figure 5A). Despite the difference, the proportion of fibroblasts predicted cancer-specific patient survival also in non-CRPC cases in Kaplan-Meier analysis (P = .027) and in Cox regression univariate (HR 2.28; P = .001) and multivariate (HR 2.18; P = .037) analyses ( Figure 5B; Table S3).

| Fibroblasts predict cancer-specific outcome in an independent prostatectomy cohort
The prognostic effect of fibroblasts was validated in Cohort II which comprised radical prostatectomy samples from primary PCa. In Cohort II, 340 patients (97%) had high-quality image data and the distribution of the proportion of fibroblasts was similar to that of Cohort I ( Figure   S6). Univariate Cox regression analysis of the proportion of fibroblasts in the stroma showed a significantly worse cancer-specific outcome with increasing proportion of fibroblasts (HR 1.89; P = .009; Table 1).
Moreover, a high proportion of fibroblasts was validated as an independent predictor of PCSM in a multivariate Cox regression analysis when adjusted for Gleason score and age at prostatectomy (HR 1.73; P = .034). Interestingly, the proportion of fibroblasts in matched benign cores (from the same patients adjacent to tumor) also predicted PCSM when comparing the highest and lowest quartiles in Kaplan-Meier analysis ( Figure S7). However, the Cox regression analysis did not reach statistical significance (HR 1.51; P = .093) for the proportion of fibroblasts in the benign TMA cores. Benign cores were available only in Cohort II. We found no significant associations between the proportion of fibroblasts and the expression of AR or PTEN proteins ( Figure S8).
Although VIM is also expressed in endothelium and in immune

| DISCUSSION
We quantified the cell subtype composition of human PCa stroma by using mIHC, as well as systematic and quantitative digital image prostatectomy was not pre-determined and therefore may introduce confounding bias in patient survival. However, the fact that the same conclusions were obtained from both cohorts suggest that our key observations are robust and validated. Importantly, the fibroblastrich stroma was prognostic in both cohorts independent of tumor grade and patient age in a Cox regression model.

| CONCLUSION
Comprehensive multiplexed mIHC and quantitative digital image analysis facilitates high-resolution and unbiased analysis of the cell subtypes in the cancer-associated stroma in human PCa. We conclude that a fibroblast-rich stroma associates with aggressive disease and adverse clinical outcome. The study provides evidence for the clinical significance of fibroblast-rich stroma in human PCa that could have implications for prevention, diagnosis, prognostication and therapy of prostate cancer.

ACKNOWLEDGMENTS
The

CONFLICT OF INTERESTS
Sami Blom is an employee of Aiforia Technologies Oy, a company providing image analysis products and services. Other authors declare no conflict of interests.