Integrated Analysis Reveals Critical Genomic Regions in Prostate Tumor Microenvironment Associated with Clinicopathologic Phenotypes

Previous studies have mainly focused on changes in the epithelium, but recent investigations have highlighted the tumor microenvironment as equally important in carcinogenesis and progression. Genetic changes in tumor stroma have been reported by several independent groups working on a range of solid tumors (1–19), although typically, only selected markers have been used in most of these studies to assess LOH/allelic imbalance (AI).

In prostate cancer, it has also been suggested that tumor microenvironment plays an important role in the progression, acquisition of androgen independence, and distant metastases (20–22). Tumor–microenvironment interactions through diffusible soluble factors such as TGF-β, as well as the extracellular matrix, likely lead to the development of metastasis (20). Nonetheless, little is known about the genetic basis of the human prostate tumor microenvironment; one report described genotypic heterogeneity in mesenchymal cells with a small number of markers (9) and another showed allelic losses in the stromal component with a small number of samples and markers (23). These findings suggest that genetic alterations in prostate cancer stroma exist and may be biologically relevant.

To attain a global perspective of epithelial and stromal specific LOH/AI changes for prostate cancer and further refine our understanding of the LOH/AI regions, we conducted a genome-wide LOH/AI scan using 381microsatellite markers and subsequently integrated these structural data with publicly available and experimental expressional array analyses. We also sought to determine whether compartment-specific LOH/AI could be correlated with clinicopathologic features. Finally, we conducted two types of validation: functional genomic validation and immunohistochemistry in an independent series of prostate cancer.

We obtained 116 prostate cancer samples at any tumor classification with or without lymph node metastasis and analyzed them in accordance with the respective Institution's Human Subjects Protection Committees. All slides were rereviewed by a single genitourinary pathologist (P. Zheng). Clinical information such as Gleason score, tumor volume, and the status of lymph node metastasis was noted (Supplementary Table S1). An independent validation series of prostate cancer is described under Immunohistochemical Analysis.

Laser capture microdissection of nonneoplastic normal tissue, epithelial carcinoma cells, and tumor-associated stroma was conducted from formalin-fixed, paraffin-embedded tissue for all 116 cases, as previously described (7, 8, 17, 24, 25) and showed at the link http://www.nejm.org/doi/full/10.1056/NEJMoa071825. Stromal fibroblasts resided in between aggregations of epithelial tumor cells or no more than 0.5 cm distant from tumor nodule. While epithelial–stromal cell cross-contamination is a possibility, we used standard and well-established protocols to minimize this, as reported previously (8, 18, 24, 25). Corresponding normal DNA for each case was procured from normal tissue, located a distance away from the tumor or from a different tissue (formalin-fixed, paraffin-embedded) block containing only normal tissue (latter being the first choice). Genomic DNA was extracted as previously described, with the exception that incubation in proteinase K was done at 65°C for 2 days (7). Primer sets for multiplex PCR defined 381 microsatellite markers in 72 multiplex panels (Research Genetics, Invitrogen). Genotyping, analyses, and scoring of LOH/AI were done as reported (24, 26, 27). The methodologic veracity of LOH/AI scan using multiplex PCR on archived templates was extensively validated as published, that is, by a series of control experiments using chromogenic in situ hybridization (CISH), quantitative PCR, to eliminate the possibility of epithelial–stromal cell cross-contamination and to exclude any possibility of PCR artifact (24, 25).

The data set contains LOH/AI status at 381 microsatellite markers for 116 different samples, each with neoplastic epithelium, tumor stroma, and normal tissue. The same 10 markers were never informative in at least one compartment and so excluded from further analyses. First, we sought to define regional LOH/AI hot/cold spots as the markers with significantly higher or lower frequency of LOH/AI compared with the average frequency of the markers along the same chromosome. We analyzed hot and cold spots of LOH/AI together with the concept that the genetic endpoint may be similar, for example, loss of a tumor suppressor gene and gain (cold spot) of an oncogenic event. Toward those ends, for each informative marker, the statistical significance of overall (across all samples) LOH/AI frequency compared with the respective chromosome average was analyzed using the exact binomial test (R base package binom.test; http://www.r-project.org). Second, the associations of LOH/AI across all informative markers (not just hot/cold spot markers) in epithelial and stromal samples with presenting clinicopathologic features (i.e., Gleason score, tumor volume, and the status of lymph node metastasis) were analyzed using a logistic model with LOH frequency as an outcome and each clinicopathologic feature as a predictor (28, 29). Multiple testing adjustment was applied by controlling false-positive report probability (FPRP; ref. 30) with a prior probability of 0.05 or 0.01. The FPRP indicates the probability that a statistically significant finding is a false positive by considering 3 factors: the P value magnitude, the statistical power, and the prior probability of true associations. Only those with P values <0.05 and estimated FPRP values less than 50%, indicating a small probability of being a false positive, are reported as statistically significant findings. An FPRP value with a prior probability of 0.05 and less than 50% is denoted as FPRP_0.05<0.5. We used more strict criteria, FPRP_0.01<0.1, for LOH/AI hot/cold spots expecting higher significant regions. For the association of LOH/AI with clinicopathologic features, we applied FPRP_0.05<0.5. For the positive correlation of LOH/AI with Gleason score, P value is obtained by one-side trend test (R, prop.trend.test) with false discovery rate (FDR) being controlled for correcting multiple testing.

Raw data from Gene Expression Omnibus (GEO) series GSE3325 at GEO platform GPL570 (Affymetrix HG-U133_Plus_2) were downloaded (31) and normalized using Robust Multi-array Averaging (RMA; ref. 32). The GSE3325 data included 13 individual benign, primary, and metastatic prostate cancers and 6 pooled samples from benign, primary, or metastatic prostate cancers. These specimens were grossly dissected maintaining at least 90% of the tissue of interest, based on examination of the frozen sections of each tissue block. We integrated microarray data of the genes residing within the regions that showed significant LOH/AI with our genetic data. We sought to use the expression microarray data to validate our LOH/AI hot/cold spots and to identify the significant genes and pathways in prostate carcinogenesis or progression. Throughout this study, we consistently selected all genes, that are located within 250 kb upstream and downstream of the LOH/AI regions of interest (significant hot/cold spots), that have significant expression values, and that are included in each microarray data set, for hierarchical clustering analyses. All hierarchical clustering analyses in this study were conducted using Cluster and TreeView softwares (http://rana.lbl.gov/EisenSoftware.htm; ref. 33). Genes belonging to hot/cold spots in both epithelium and stroma or in only epithelium were selected and hierarchical clustering was conducted for both genes and samples.

We generated stromal microarray expression data for functional validation of our LOH/AI hot/cold spots in stroma because we could not find appropriate public microarray data sets for prostatic stroma. Stromal cells cultured from normal peripheral zone tissues (F-PZ-64, F-PZ-79, F-PZ-82, F-PZ-102, F-PZ-105) and from tumors (F-CA-31, F-CA-39, F-CA-52, F-CA-67, F-CA-93) were established and grown as previously described (34). Total RNAs were extracted from semiconfluent cells (passages 4–5) one day after feeding fresh medium using RNeasy Mini Kit (Qiagen) according to the manufacturer's instructions and treated with TURBO DNA-free kit (Ambion). Hybridizations were conducted according to Illumina protocols by the Genomic Core, Lerner Research Institute, Cleveland Clinic, Cleveland, OH. All samples were hybridized to the Illumina Sentrix Human-6_v3 Expression BeadChip, which contains 48,000 distinct oligonucleotide probes. These transcriptome data are deposited in GEO [GSE34312 (not publicly released yet)]. After normalization using average normalization algorithm of BeadStudio Gene Expression Module v3.4 (Illumina), we selected all the genes within ±250 kb of LOH/AI hot/cold spots in only stroma and applied a hierarchical clustering method to both genes and samples.

We selected all the genes lying within and ±250 kb of the significant LOH/AI markers associated with Gleason score or tumor volume in both epithelium and stroma. With the genes associated with Gleason score or tumor volume, a hierarchical clustering was conducted for epithelium using public microarray data, GSE3325, and for stroma using our stromal microarray data (31).

Formalin-fixed and paraffin-embedded prostate cancer sections from 49 independent cases were immunostained with a rabbit anti-STIM2 polyclonal antibody (Abcam Inc) for STIM2 expression in prostate cancer stroma. Both Gleason grade III (G3) and Gleason grade IV and/or V (GIV/V) lesions were examined for each case. Deparaffinized tissue sections were placed in 10 mmol/L citrate buffer (pH = 6.0) and were heated to 125°C in a Decloaking Chamber (BioCore Medical) for 30 seconds for antigen retrieval. These sections were then incubated with a 1:300 dilution of primary antibody overnight at 4°C, secondary antibody, and the avidin-biotin complex (VECTASTAIN Elite ABC Kit; Vector Laboratories, Inc.) and developed with diaminobenzidine. Sections were counterstained with hematoxylin. For negative controls, primary antibody was omitted. Stromal expression of STIM2 was determined by counting sum of intensities of positive stromal cells divided by area of stroma, using Image-Pro Plus 7.0 (Media Cybernetics, Inc.). Five random fields were analyzed with a magnification of 40×. STIM2 expression in tumor stroma was divided by those in normal stroma for normalization. The statistical significance of STIM2 expression levels between GIII and GIV/V lesions was determined with Wilcoxon sign-rank test.

Overall, 371 markers across all chromosomes, ranging from 7 on chromosome 22 to 31 on chromosome 1, were analyzed: 38,460 PCR reactions (19,639 for epithelium, 18,821 stroma) were informative for LOH/AI evaluation. We identified a total of 43 hot/cold spots, 17 occurring in both epithelium and stroma, 18 only in the epithelium, and 8 only in the stroma (Table 1). No known genes were contained at 7 loci (D1S2134, D3S2427, D4S2417, D6S1277, D9S922, D12S2078, and D14S606), however, we found expressed sequence tags (EST) at these loci (Table 1). Genes belonging to hot/cold spots common to both epithelium and stroma include HOXB1-HOXB9 cluster, HOXB13, ZNF554, ZNF555, ZNF556, ZNF57, and ZNF77, as well as 7 miRNA (Table 1). FGF12, ACVRIB, the KRT cluster, and APOL1-APOL4, as well as 6 miRNAs, were included in the genes lying in hot/cold spots of LOH/AI in epithelium only (Table 1), and TGFβ1I1, WFDC5, WFDC12, and TP53TG5, as well as 5 noncoding RNAs, notably miR-125b-1, were among the genes of hot/cold spots in stroma only (Table 1).

LOH/AI markers in association with clinicopathologic features (Supplementary Table S1) may reveal clues to clinical behavior and biologic diversity and/or serve as biomarkers of prognosis or therapeutic options. Using logistic regression, we analyzed our genome-wide LOH/AI scan across all informative markers to identify compartment-specific loci that were significantly associated with aggressiveness of disease as reflected by Gleason score, tumor volume, and regional nodal status. We identified 15 LOH/AI regions, 10 in epithelium and 5 in stroma, associated with Gleason score (Table 2), 11 (8 in epithelium and 3 in stroma) of which showed a positive correlation with increasing Gleason score (one-sided trend test, FDR < 0.05; Supplementary Table S2). Among these 11 markers, 8 (6 in epithelium and 2 in stroma) remained significantly positively correlated with Gleason score even when a more rigorous method was applied (Bonferroni-adjusted P < 0.05). In association with tumor volume, 40 markers were identified in epithelium and 35 in stroma (Supplementary Table S3). Interestingly, there were no significant LOH/AI markers correlated with regional nodal metastases.

We focused on genes of LOH/AI hot/cold spot regions and analyzed their expression profiles using publicly available microarray data (see Methods). This served 2 purposes. First, this acts as a functional (transcript expressional) validation of the genetic data. Second, integration of the genetic and expressional data would lend clues to the compartment-specific functions of the genes within the hot/cold spot regions as they relate to progression and prognosis. Hierarchical cluster analysis supervised by 87 genes found in LOH/AI hot/cold spot regions ±250 kb resulted in 3 distinct groups: normal epithelium, cancer, and metastatic cancer when considering genetic information from both epithelium and stroma (Fig. 1A). When we used 61 genes residing within the LOH/AI hot/cold spot ±250 kb regions occurring specifically in epithelium, only 2 groups could be clearly distinguished, namely, normal epithelium and metastatic cancer groups (Fig. 1B). Interestingly, the primary cancer samples were equally clustered with normal epithelium or with metastatic disease.

We conducted a hierarchical clustering of stromal cells, supervised by 28 genes of LOH/AI hot/cold spot ± 250 kb regions from stroma only (Fig. 2). This accurately separated normal stromal cells and cancer-derived stromal cells.

We further focused on genes residing in proximity to our LOH/AI regions associated with Gleason score and tumor volume and analyzed their expression profiles using publicly available microarray data (see Methods) for epithelial analysis as well as our experimentally generated expression data for stromal analysis. With only one exception, P3, in neoplastic epithelial analysis, samples were clearly separated into normal, primary cancer, and metastatic disease by 31 genes within the LOH/AI (± 250 kb) regions in neoplastic epithelium associated with Gleason score (Supplementary Fig. S1A). Our stromal cell cultures derived from normal tissues and from prostate cancer were accurately clustered supervised by the 18 genes belonging to our stroma-only LOH/AI (± 250 kb) regions associated with Gleason score (Supplementary Fig. S1B). As for tumor volume, epithelial prostate cancer samples were also successfully separated into normal, primary cancer, and metastatic disease in epithelial analysis (Supplementary Fig. S2A). In contrast, our stromal cell cultures did not cluster into tumor-derived stroma and normal stroma groups in stromal analysis for tumor volume (Supplementary Fig. S2B).

To further validate our genetic data, we selected our strongest candidate gene, STIM2, lying in D4S2397, which is the most significant stroma-associated LOH marker correlated with Gleason score (Supplementary Table S2). We examined STIM2 expression in 49 independent primary prostate cancer samples that had both Gleason grade III and Gleason grade IV and/or V lesions by immunohistochemical analysis (Fig. 3). Stromal protein expression of STIM2 (y-axis) showed significant downregulation in the stroma of GIV/V lesions compared with that of GIII lesions (P < 0.001; Fig. 3A).

Recent molecular studies have provided a myriad of information about genetic changes in prostate cancer and accumulating data over the last 2 decades have shown the importance of the tumor stroma in prostate carcinogenesis, with the latter based on mainly functional and cell biologic studies. Here, we have completed a genome-wide LOH/AI scan and integrated these data with global expression array data for prostate cancer epithelium and stroma and correlated these with presenting clinicopathologic features and validated our observations via functional genomics approach and immunohistochemistry of a newly identified stroma-relevant gene associated with Gleason score.

First, genome-wide LOH/AI scan revealed LOH/AI hot/cold spots that were important in each compartment alone and in both compartments. Among the 225 genes in proximity to the 43 LOH/AI hot/cold spot regions germane to epithelium and/or stroma, several candidate tumor suppressor genes, such as HOXB13 (epithelium and stroma), the APOL1-4 cluster and ACVR1B (epithelium), and TP53TG5 (stroma), are notable. The APOL family of genes (APOL1-6) encodes programmed cell death proteins that can initiate apoptosis or autophagic cell death, as well as belong to a multicomponent innate immunity complex–building platform (35). Thus, loss of function of the APOL1-4 region will lead to loss of apoptosis and autophagic cell death regulation as well as potential decreased innate immunity locally. Our observations here, that is, loss of APOL1-4 function leading to loss of autophagic cell death and consequent loss of essential nutrients and energy to tumor epithelial cells in a paracrine fashion are consistent with the autophagic tumor stroma model of cancer metabolism underscored by the work of Witkiewicz and colleagues (36). Furthermore, the relevance to innate immunity lends credence to early hypotheses of metabiomic relationships in initiating various neoplasias (35, 37), including that of the prostate. Of note, the activin/TGF-β signaling subnetwork has already been acknowledged to be important in prostate carcinogenesis (38). We have shown that D12S297 harbors the human activin receptor, ACVR1B, which has been described as having somatic mutations and deletions in at least 4 other types of cancers, and which is expressed in prostate epithelium (39). ACVR1B encodes an activin receptor that signals via the TGF-β pathway as well.

Second, we found 15 specific loci of LOH/AI in the epithelium or stroma associated with Gleason score (Table 2). We suspect that these regions and genes are relevant to invasion and progression and focused especially on stroma. Two loci (4p15.2 and 19p13.3) harbored significant LOH/AI in the stroma, correlated with Gleason score. One of the genes located in the 4p15.2, which is the most significant region, is STIM2 (stromal interaction molecule 2). Our immunohistochemical analysis of 49 independent prostate cancer samples has given us an in-principle validation of our genetic and functional genomic studies. STIM1 was originally proposed as a candidate tumor suppressor and regulates intracellular calcium (40, 41). STIM2 is also a transmembrane protein that, in limited studies, has been shown to mediate cellular proliferation (42). In view of our observations, further studies on the role of STIM2 in different cancers, especially in different compartments, are required. The 19p13.3 region contains AES which is known to be associated with Wnt signaling pathway and mitogen—activated protein kinase (MAPK)/extracellular signal—regulated kinase 1 (ERK1), which is an upstream kinase for NF-κB activation (43, 44). The cluster of ZNFs including ZNF77 also at the 19p13.3 region was previously associated with different cancers, and ZNF77 status has been used as a prognostic marker in early-stage renal cell carcinomas and stage I breast cancers to predict recurrence (45, 46).

We subsequently integrated global transcriptome data from public archives and those we experimentally derived with our LOH/AI hot/cold spots. Importantly, these provided functional validation of our genomic data. The gene expression profile using neoplastic epithelium from public sources with only genes from our epithelium-related LOH hot/cold spots, which is not the entire epithelium gene expression profile, could not accurately cluster the samples. Similarly, we also attempted the cluster analysis supervised by the genes from only stromal hot/cold spots, and as expected, it resulted in failure to cluster the samples (data not shown). In contrast, these samples could be accurately clustered when taking into account the genes from the stromal hot/cold spots, indicating that genomic alterations with consequent expressional alterations in the stroma are necessary and play an important role in prostate carcinogenesis and progression. Equally important is a notable negative. The genes within the LOH hot/cold spot regions identified as important in correlating with tumor volume were germane only in the epithelium. This was validated when expression data supervised by the genes within the LOH hot/cold spot regions relevant to tumor volume could not accurately classify normal stroma from tumor stroma (Supplementary Fig. S2B). In other words, at least in prostate cancer, tumor volume does not appear to be influenced by stromal genomic or expressional alterations of genes located within tumor volume–associated stromal LOH regions in this study. This makes teleology logical given that the genes playing roles in tumor stroma are almost certainly relevant to invasion, progression, and metastasis, as noted above.

By integrating genomic and global expression data for the epithelium and stroma of prostate cancer, we were able to detect significant regions and candidate genes important in the tumor-microenvironment cross-talk that leads to carcinogenesis and progression. Importantly, we have conducted functional genomic validation of the genes in the regions associated with Gleason score as well as tumor volume, in a compartment-specific manner, as well as validated a stromal gene associated with Gleason score in an independent series of tumors using a different technique, namely, immunohistochemistry for STIM2, as proof of principle. Parenthetically, we also noted that the epithelial expression of STIM2 (D4S2397 is also a significant epithelium-associated LOH marker correlated with Gleason score as well) was downregulated in G4/5 over G3. However, we did not formally analyze the epithelial expression in this study because the effect size did not appear as marked as for stroma. These data provide fundamental insights into compartment-specific roles in prostate carcinogenesis or progression and may also reveal the solid tumor microenvironment as a novel compartment for important biomarkers of prognosis as well as targeted therapy.

No potential conflicts of interests were disclosed.

The authors thank the members of the laboratory of C. Eng for technical assistance and helpful discussion over the years.

This work was funded, in part, by the Department of Defense US Army Prostate Cancer Research Program (C. Eng). C. Eng holds the Sondra J. and Stephen R. Hardis Endowed Chair of Cancer Genomic Medicine at the Cleveland Clinic, and the ACS Clinical Research Program, funded, in part, by the F.M. Kirby Foundation.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.