Analysis of gene expression in prostate cancer epithelial and interstitial stromal cells using laser capture microdissection

All tissues were acquired and used with IRB approval from Kent State University and the appropriate institutions (see below). Frozen tissues in optimal cutting temperature media (OCT) were obtained for RNA isolation while formalin fixed paraffin embedded (FFPE) tissues were obtained for immunohistochemistry. Two types of OCT embedded tissues were obtained: 1) 5 micron sections for laser capture microscopy (LCM) and 2) OCT blocks for quantitative real-time PCR (QRT-PCR).

The serial frozen tissue sections for LCM were provided by The Ohio State University Prostate Cancer tissue Bank, part of the Human Tissue Resource Network (HTRN) in the Department of Pathology (Columbus, Ohio). The tumor samples were removed during radical prostatectomy and frozen in OCT. Tumors were categorized as intermediate grade (primarily Gleason grade 3). Two of three samples had a combined Gleason score of 6 and one had a GS 7. One of the serial sections from each tumor was stained with hematoxylin and eosin and the tumor areas marked for identification. Stromal tissue of all 3 samples appeared to contain a similar proportion of inflammatory cells.

For QRTPCR analysis twenty paired prostate tissues were provided by Dr. C. Magi-Galluzzi (Cleveland Clinic Foundation, Cleveland, OH). Tissues were obtained by radical prostatectomy, paired tumor and non-neoplastic tissues were selected from each prostate and frozen in OCT. All tumor samples were of T2 or T3 stage with combined Gleason score of 7 and were observed to have abundant epithelial tissue for RNA isolation.

Commercially available prostate tissue microarrays (TMAs) were purchased from Creative Biolabs (Fort Jefferson Station, NY). Tissue arrays consisted of cores of formalin-fixed, paraffin embedded prostatectomy cores in duplicate or triplicate from each prostate. Cores were arrayed in a rectangular fashion and were 1.0-1.5 mm in diameter and 5 μm in thickness. A total of 31 cases of carcinoma, 7 of benign hyperplasia, and 5 normal (non-neoplastic) controls were examined. Normal samples were obtained from cancer-free prostates from normal individuals. All tissues were selected and evaluated by an independent pathologist who determined Gleason grading and differentiation status. Nearly half of the cores were from high grade tumors with Gleason scores 8-10.

Non-neoplastic RWPE-1 cells were obtained from the American Type Culture Collection (Manassas, VA) and grown in K - SFM supplemented with 0.05 mg/mL bovine pituitary extract and 5 ng/mL EGF. Hormone responsive LNCaP tumor cells were grown in RPMI-1640 media supplemented with 10% FCS and antibiotics. Hormone insensitive LNCaP - C42, PC3, and DU145 tumor cells were grown in DME - F12 media supplemented with 10% FCS and antibiotics. All cells were maintained in 5% CO₂ at 37°C.

For LCM, the frozen sections were stained and dehydrated using the HistoGene LCM Frozen section staining kit as per manufacturer's recommendations. Cell capture and lysis was completed within 2 hours to assure quality RNA. The epithelial and interstitial stromal cells were isolated from ten slides containing 5 micron frozen tissue sections using an LCM microscope (Arcturus Bioscience, Mt View, CA). Neoplastic areas of the slide observed to have the most abundant cells of interest were identified and marked to direct the laser capture. Stromal cells were collected from areas adjacent to glandular epithelium and included inflammatory cells. Overall, 1000 to 2000 epithelial or stromal cells were captured per cap. To verify the accuracy of capture, tissue sections and caps were examined post-capture.

Biotin-labeled cRNA was hybridized to Affymetrix Human Genome U133A 2.0 chips (HG_U133A 2.0) for 16-hour at 45°C. The GeneChip^® Operating Software (GCOS) was used to run the Fluidics Station 400 and hybridized arrays were stained with the Midi_euk2v3 labeling kit for detection. The arrays were scanned using an Affymetrix^® GeneChip^® Scanner 3000. The signal intensities were normalized by Affymetrix software to the spike-controls located on the array chip. After chip normalizations, relative intensities were used to determine whether expression is absent (A), present (P), or marginal (M). Expression patterns between arrays were compared and raw signal strength was examined to verify that hybridization was effective.

Signal intensities for each gene were generated using the Microarray Suite 5.0 algorithm by Affymetrix GCOS software 1.1. In addition to the signal intensity, each gene was determined to be present, marginal, or absent using default software settings. Overlap in gene expression between epithelial and stromal cell samples was assessed by counting the number of probe sets with all three samples showing present calls. For analysis of differential expression between epithelial and stromal cell samples, a filter requiring a present call in at least 3 of the 6 arrays was applied. This reduced the total number of probe sets to be analyzed from 22,215 to 8,739. Signal intensities for the three epithelial and three stromal arrays were further analyzed using Cyber-T software http://cybert.microarray.ics.uci.edu/ using the default settings. This software generates p-values for each gene as a test of differences between groups using a Bayesian paired t-test [21]. A list of candidate differentially expressed genes was generated using genes with a posterior probability of differential expression [22] of 0.99 or higher, which corresponded roughly to a Bayes p-value of 0.001 or less.

Functional Gene Ontology (GO) annotation of genes of interest was performed using DAVID http://david.abcc.ncifcrf.gov/[23, 24] and Affymetrix databases. Gene functional classification and functional annotation clustering were performed to identify functional gene groups and ontology terms that are significantly overrepresented among genes of interest.

RNA samples were reverse transcribed using QuantiTect^® Reverse Transcription kit and DNase treatment was performed according to manufacturer's protocol (Qiagen Sciences, San Diego, CA). For LCM captured cells, pre-amplification of cDNA was done using TaqMan^® PreAmp Master Mix kit. Real-time PCR was performed using the TaqMan Universal Master Mix and optimized TaqMan probe sets (Table 1). Endogenous internal controls were run with every sample plate for comparisons and each sample was assayed in triplicate. Samples were amplified using the ABI 7000 thermocycler and Ct values were measured by the ABI Prism 7000 sequence detection system (Applied Biosystems, Foster City, CA). Amplification conditions were 95°C for 10 minutes, and 40 cycles of 95°C for 15 seconds and 60°C for 1 minute. The comparative Ct method (2^-ddCt) [25] was used to analyze gene expression differences between cell types for LCM captured cells and between tumor and non-neoplastic tissues for paired frozen prostate samples. For analysis of cell lines, gene expression in tumorigenic cell lines was compared to the non-tumorigenic cell line RWPE-1. Tests of significance were done using Dunnett's two-sided multiple comparison test.

Immunohistochemical staining of the prostate TMAs was performed using standard IHC techniques. Briefly, slides were deparaffinized using a sequential method of rehydration followed by antigen retrieval in citrate solution with heating. Endogenous peroxidase activity was blocked with a 3% hydrogen peroxide solution. Slides were probed with a rabbit polyclonal anti-WT1 antibody (Epitomics, Burlingame, CA). Staining was visualized using a biotinylated goat anti-rabbit IgG secondary antibody, streptavidin horseradish peroxidase solution, and DAB (Vector Laboratories, Burlingame, CA). Slides were counterstained with hematoxylin, mounted and examined by brightfield microscopy. Staining was visualized using an Olympus IX70 microscrope at 100× total magnification. Images were taken with a Diagnostic Instruments camera and analyzed using SPOT Advanced software. Immunoreactivity assessment was based on intensity of staining in epithelial cells relative to any nonspecific stromal reactivity. Slides were scored blindly by two different individuals. Relative staining intensity was scored using a 3 point scaling system, where 0 represents the absence of staining in any epithelial cells, 1 represents weak to moderate staining, and 2 represents strong staining in at least 25% of epithelial cells.

There were significant differences in gene expression between the epithelial and stromal cell samples. A Venn diagram was created to determine the proportion of genes expressed in common between both tissue types (Figure 1). For this figure, we utilized very stringent criteria - to be considered present, the transcript had to receive a "Present" call on all three samples from that tissue type. 6946 of the 22215 probes on the array were present in all three sample pairs for at least one of the two cell groups and half of them (3452 genes) were significantly expressed in both epithelial and stromal tissue. Reducing the stringency by allowing there to be only two present calls instead of three produces higher numbers of present probes but the trends in the data are similar. About half of the expressed transcripts were in common between two tissues types, presumably required for functions shared between these cell types. The other half represents genes likely required for cell-specific functions that were the subject of further analysis. The cellular heterogeneity of stromal tissue was consistent with the observation that ~42% of expressed transcripts (2911 genes) were elevated in stromal tissues. Conversely, only ~8% of transcripts (583 genes) were elevated in prostate cancer epithelial cells and thus comprised unique gene expression patterns. Both prostate specific genes (e.g., PSA/Kallikrein 3 and Kallikrein 2) and epithelial marker genes (e.g., keratin 18 and desmoplakin) were expressed in tumor epithelial cells. Similarly, stromal marker genes, such as desmin and vinculin, were expressed in the cells collected from adjacent stromal tissue (see additional files 1 and 2). Expression of these genes verifies the specificity of the epithelia and stroma collected.

Bild vergrößern

Differential expression level comparisons of epithelial and stromal genes identified nearly 500 genes whose expression was significantly different between epithelial and stromal cells (Bayesian t-test, p < 0.001, posterior probability of differential expression > 0.99). Shown in Tables 2 and 3 are the genes with the highest probability of differential expression. Additional files 1 and 2 list 302 genes that were found to be significantly overexpressed in stromal tissue as compared to epithelial tissue, and 194 genes that were significantly overexpressed in epithelial tissue as compared to stromal tissue, respectively. To learn more about the types of gene expression differences between the two tissue types, the 496 differentially expressed genes (listed in additional files 1 and 2) were placed into functional categories using Gene Ontology (GO) Annotations.

Gene functional classification clustering analysis was performed with DAVID using two subsets of genes that were upregulated (a) in epithelial cells (n = 194) or (b) in stromal tissue (n = 302). Shown in Table 4 is a summary of these classifications (Entire list of GO_BP terms is found in additional file 3). Three gene clusters with enrichment scores greater than 1 were identified among epithelial genes (encompassing 11% of all genes), namely, membrane-associated glycoproteins (including proteases) and two groups of ion transport related genes (including metal ion and ATP dependent transporters). Notably, eleven gene clusters with enrichment scores greater than 1 were identified among stromal genes, encompassing about 28% of all stromal genes. The top three clusters included about 21 unique genes (24% of 86 grouped genes) and were comprised of collagen genes and muscle and organ development genes. Other clusters were composed of structural and intracellular matrix proteins (total of 12 genes, or 14%), immune and inflammation related genes (including MHC class II and complement components) (total of 23 genes, 27%), zinc finger transcription factors (10 genes, 12%), metal ion transporters and regulators (17 genes, 20%). The greater number of clusters identified by GO analysis of genes more highly expressed in stromal cells further demonstrates the broader diversity of gene expression patterns of the stromal tissue, due to the heterogeneous cell types that encompass the stromal compartment, supporting our earlier Venn analysis (Figure 1). We also found that both cell types shared some functional gene categories, such as ion transport and regulation related genes.

Overall, we observed that many of the differentially expressed genes identified in the "Significantly Higher" lists (additional files 1 and 2) fall into two categories that are important for cell signaling: transcription factors and growth control. Of the transcription factors identified, we examined three within the zinc finger family, namely Wilms' Tumor 1 (WT1), GATA2, and early growth receptor protein 1 (EGR1). Of the growth control genes identified, we examined those known to be important in prostate tumorigenesis such as the chemokines CCL5 and CXCL13 and members of the insulin-like growth factor (IGF) and fibroblast growth factor (FGF) signaling pathways including IGF-1, IGF-IR, IGFBP3, FGF-2, and FGFR-3 [25‐27]. Based on the microarray analysis, elevated expression of the zinc finger transcription factors (WT1, EGR1, and GATA2) and growth factor receptors (IGF-1R and FGF-R3) was observed in the epithelia, while expression of the chemokines (CCL5 and CXCL13) and growth factor ligands (IGF-1, FGF-2, and IGFBP3) was found in the stroma.

In order to more precisely quantify the expression of genes in the LCM-derived samples used in the oligonucleotide microarray, we analyzed the selected genes described above using quantitative real-time PCR and the 2^-ddCt method [28]. Due to limitations in the quantity of RNA obtained from the laser captured samples, expression of only seven of the ten genes examined was confirmed in at least two of the three samples, namely WT1, GATA2, CCL5, CXCL13, IGF-1, IGF-1R and FGF-2. Of those genes analyzed, fold difference values were at least 1.2-fold or greater relative to the paired cell type, i.e. epithelia relative to stroma, or vice versa (data not shown). Interestingly, elevated EGR1 expression was not confirmed by real-time analysis of epithelial cells.

Once tissue-specific expression patterns were established, we then asked whether those genes were also expressed in normal prostate tissue. We quantified expression of the genes described above in ten paired frozen tumor and non-neoplastic prostate tissues. Expression in tumor tissue was normalized to paired non-neoplastic tissue obtained from the same prostate. The genes highly expressed in microdissected tumor epithelial cells were expected to be abundant in surgically dissected tumor tissue enriched with tumor epithelial cells. For the genes WT1, GATA2, and FGFR3, the expression pattern in surgically dissected tumor tissue was consistent with that in the microdissected epithelial cells (Figure 2 Panel A). In contrast, paired tissue analysis showed IGF1R levels in tumor tissues were similar to those in non-neoplastic samples. Another exception to the pattern of elevated expression in tumor tissues of genes identified in epithelial cells, was the significantly higher EGR1 expression in non-neoplastic compared to tumor tissues (p < 0.05, paired t-test). This lack of elevated EGR1 expression in tumor tissue was consistent with the results of the real-time PCR quantitation of expression in microdissected tissue. Overall, the expression patterns in non-neoplastic tissues from paired samples were consistent with those in stroma cells obtained from laser capture microscopy (Figure 2 Panel B). These results can be attributed to relatively fewer epithelial cells in the normal tissue samples. Thus, as a reflection of stromal cell prevalence, the stromal genes are more highly expressed in non-neoplastic tissue than in the paired tumor tissue; all mean fold difference values were 1.7-fold or greater in non-neoplastic tissues.

Bild vergrößern

Additional analysis of zinc finger transcription factor expression was expanded to include another ten paired frozen tumor and non-neoplastic tissue samples. Data was analyzed by clinical stage to determine whether there was any relationship between clinical stage, especially invasiveness, and gene expression. Expression in invasive T3 stage tumors was compared to that of non-invasive stage T2 tumors. As seen in Table 5 top, in the majority of invasive stage T3 tumors, WT1 expression levels are higher (2.0 fold or greater) in tumor than in non-neoplastic tissues. Conversely, in the majority of localized stage T2 tumors WT1 expression levels are lower in tumor than non-neoplastic tissues (Table 5 bottom). Surprisingly, EGR1 expression was consistently lower in tumor tissues relative to non-neoplastic tissues for both stage T2 and T3 tumors. GATA2 expression was also reduced in tumor tissues in the majority of stage T2 samples, although in stage T3 tumors, expression was not consistent. To focus on the inverse relationship between WT1 and EGR1, we also examined five established prostate cells lines for their expression levels of each gene. Expression in the prostate tumor cell lines LNCaP, LNCaP-C42, PC3 and DU145 was normalized to expression in the non-tumorigenic cell line RWPE-1 (Figure 3). With the exception of the DU145 cells, WT1 and EGR1 expression levels were consistent with the frozen paired tissue samples examined; that is, WT1 expression was elevated (p < 0.05) and EGR1 expression reduced (p < 0.001) in tumor cell lines relative to RWPE-1 prostate epithelial cells.

Bild vergrößern

Because elevated levels of WT1 mRNA expression were observed in laser capture samples, frozen tissues, and tumorigenic cells, we examined WT1 protein expression by immunohistochemical analyses of prostate tissue microarrays. These results demonstrated the presence of WT1 protein in 65% of tumor samples examined (Figure 4). In contrast, WT1 protein was not detected in both normal prostate and benign prostatic hyperplasia samples (Figure 4 Top). Notably, in those samples with WT1 expression, the majority of staining was cytoplasmic with only a few samples demonstrating nuclear expression (not shown). Both cytoplasmic and nuclear WT1 staining has been shown in other tumor types [29, 30].

Bild vergrößern