Co-regulation analysis of closely linked genes identifies a highly recurrent gain on chromosome 17q25.3 in prostate cancer

Tissues were procured from untreated patients undergoing radical prostatectomy for clinically localized prostate adenocarcinoma at the Hospital Clínic of Barcelona. Samples were obtained after informed consent by the patients and approval by the Institutional Ethics Committee. Tissue fragments from tumoral and non-tumoral areas were embedded in OCT, snap-frozen, and stored at -80°C at the tumor bank of this institution (Table 1). Non-tumoral samples were completely devoid of cells or glands with neoplastic appearance upon histologic examination of serial sections corresponding to the processed tissues, and are hereafter called normal samples. The rest of the specimen was routinely formalin-fixed and paraffin-embedded. HeLa and RWPE1 cells (ATCC, Manassas, VA) were grown in DMEM (PAA, Ontario, Canada) supplemented with 10% FBS or keratinocyte serum-free medium (KSFM; Gibco, Carlsbad, CA), respectively. Primary cultures (PC1 and PC2) were derived from prostatectomies in which the adenocarcinoma was macroscopically detected. Fibroblast-depleted explants were grown for 4–5 weeks in KSFM supplemented with 10^-11 M 5-α-dihydrotestosterone.

Microscopical examination of hematoxylin-eosin (H&E) stained sections from frozen tissues was used to assess the percentage of each histological compartment (stroma, total epithelium, normal and neoplastic epithelium) prior to RNA extraction. Normal samples contained an average of 45% epithelium, stromal samples, obtained from non-tumoral areas of the specimens, contained less than 1% epithelium and tumoral samples contained an average of 70% epithelium, being 90% of it neoplastic glands. At least twenty 20-μm cryosections were used for RNA isolation, and the first and the last sections were H&E-stained to monitor for tumoral and normal gland status. RNA from tissues and cell lines was isolated using RNeasy kits with a DNase I digestion step (Qiagen, Valencia, CA), and analyzed on a 2100 Bioanalyzer instrument (Agilent Technologies, Palo Alto, CA) to assess quality and quantity.

The samples used for microarray analyses were grouped in four biological groups: adenocarcinomas (n = 20), normal samples (7 normal samples from 7 prostate cancer patients and a pool of normal prostate tissues, n = 8), and normal-associated stromal samples (n = 3), all of them from the peripheral zone of the prostate, and the cell lines described above (n = 4). Biotinylated cRNA (10 μg) from the 35 samples enumerated above was processed and hybridized to Affymetrix Human Genome Focus Arrays (Affymetrix, Santa Clara, CA). Arrays were scanned at 3 μm resolution in an Agilent HP G2500A GeneArray scanner (Agilent Technologies). Data were RMA-normalized [12], followed by quantile-quantile probe-level between-array normalization [13]. Normalized expression data were analyzed by FADA [14], which applies Q-mode Factor Analysis. Genes were considered differentially expressed between the tumoral and normal groups when their associated q-value was below 2.5 × 10^-4. Q-values were computed from t-test P-values using the Benjamini-Hochberg step-down false-discovery rate (FDR) algorithm [15]. Normalized expression data were standardized and submitted to UPGMA hierarchical clustering [16]. Microarray data sets from this study are deposited in the Array Express repository under accession number E-MEXP-1331.

Tissue samples from seven of the patients whose samples had been used in microarray analysis were selected for laser microdissection (Table 1). Eight-μm cryosections were mounted onto plastic membrane slides (PALM, Bernried, Germany), fixed in cold 70% ethanol, stained with hematoxylin, dehydrated, air-dried and stored at -80°C until use. Laser microdissection was performed using the PALM MicroBeam System, with the Laser Microdissection and Pressure Catapulting technology. Approximately 1.2 mm² of both normal and tumoral epithelium were collected separately from each sample and RNA isolation was performed as above.

Microdissected tissues and 4 paired normal-tumoral samples included in the microarray analysis were used for Q-PCR analyses. Custom-designed TaqMan Low Density Arrays (TLDA; Applied Biosystems, Foster City, CA) contained primers and probes for 45 genes, and the RPS18 gene as the calibrator. One ng of total RNA was used as template for reverse transcription for each replicate of non-microdissected (triplicates) and microdissected (quadruplicates) samples. Q-PCR was performed on an ABI PRISM 7900 HT instrument (Applied Biosystems), and relative quantitation determined by the ΔΔCt method.

Tissue microarrays (TMA), were built using a Manual Tissue Arrayer 1 (Beecher Instruments, Sun Prairie, WI), and contained a total of 52 paraffin-embedded tumors, 21 PIN and 40 normal samples, each in duplicated or triplicated cores. Two μm thickness TMA sections were mounted on xylaned glass slides (DAKO, Glostrup, Denmark) and used for immunohistochemistry. Mouse monoclonal antibodies specific for myosin VI (MYO6, clone MUD19; 1/100 dilution) and ephrin type-A receptor 2 precursor (EPHA2, clone D7; 1/50 dilution) were purchased from Sigma (Madrid, Spain) and a rat monoclonal antibody specific for multidrug resistance-associated protein 4 (ABCC4, clone M4I-10; 1/50 dilution) was purchased from Abcam (Cambridge, MA). Immunohistochemistry was performed with the Envision system (DAKO) and developed with diaminobenzidine, after antigen retrieval in a pressure cooker with citrate buffer pH 6 (MYO6) and EDTA pH 9 (ABCC4) or no retrieval (EPHA2). The staining was scored as a percentage of positive cells and its intensity as null (0), weak (1), moderate (2) or strong (3). Differences over 20% in the percentage of positive cells and/or one or more than one degree of intensity were considered as a significant change in expression. Images were obtained in an Olympus BHT microscope (Olympus, Germany) with an Olympus Camedia C-3030 camera. Immunohistochemistry information is MISFISHIE compliant [17].

To determine putative non-random colocalizations of coexpressed genes, we determined the precise genomic localizations of all the genes present on the array (ENSEMBL-NCBI 36 assembly of the consensus human genome sequence). We then determined the groups of four or more consecutive FADA-selected genes that were all either over- or underexpressed in tumoral samples and simultaneously colocalized within a distance of 4 Mb, or less. We modelled the random distribution of over- or underexpressed genes as a Poisson distribution, taking as lambda parameter the product of the number of genes present in the array within the tested region multiplied by the total number of over- or underexpressed genes and divided by the number of genes in the array.

Tumoral samples from the same 20 adenocarcinoma cases used in the microarray analysis and 8 lymph node metastases were used. From each paraffin-embedded sample, a 2-μm section was obtained for FISH analysis and a consecutive tissue section was H&E-stained in order to identify the tumoral and normal regions to be analyzed. Sections were deparaffinized, washed in 2 × SSC, dehydrated, denatured and hybridized with a BAC probe corresponding to the segment of interest on 17q25.3 (RP11-165M24) and a chromosome 17 centromeric probe (CEP17, Vysis, Des Plaines, IL). Slides were then washed in 0.4 × SSC and in 2 × SSC, counterstained with DAPI II (Vysis) and imaged using an Olympus BX60 fluorescence microscope (Olympus) with the MetaSystems software (MetaSystems, Germany). Signals corresponding to both the RP11-165M24 and the chromosome 17 centromeric probes were scored in 200 non-overlapping nuclei of the tumoral zone and gains were defined in samples where ≥ 10% of the analyzed nuclei presented a ratio between the RP11-165M24 and the reference probe ≥ 1.5. In all cases, a non-tumoral area was analyzed to assess its normal chromosomic status.

The specific features of our transcriptional profiling included a stringent selection of normal and tumoral prostate tissue samples, based on epithelial representations, the use of pure stromal samples, and the inclusion of four epithelial cell cultures. These cell lines were HeLa cells as a non-prostate epithelial cell line, a normal prostate epithelial cell line (RWPE1) and two prostate primary cultures, the last three with features of basal epithelial cells, as discussed below. The microarray data analysis by FADA permits to identify those genes that most significantly contribute to a given phenotype, and to group the samples according to their degree of biological relatedness [14]. FADA identified 318 genes as those most significantly differentially expressed between normal and tumoral tissues. These genes clustered samples into four groups according to their origin (tumoral, normal and stromal samples, and a group comprising all cultured cells, Figures 1A and 1B), and clustered genes into five functional groups (non-tumoral, basal, normal luminal and stromal, tumoral and tumoral-proliferative, Figure 1B). Of these genes, 134 were overexpressed and 184 underexpressed in tumoral vs. normal prostate tissues (Figure 1B, see Additional files 1 and 2). The genes overexpressed by tumors could be further subdivided into two groups according to their coexpression in the different sample groups: a tumoral gene set (99 genes), most expressed in tumoral samples, and a tumoral and proliferative gene set (35 genes) most expressed in both tumoral samples and cultured cells. Similarly, the genes underexpressed in tumors could be subdivided into three subgroups of coexpressed genes: a non-tumoral set (51 genes), expressed at high levels in normal prostate, pure stromal and cultured cell samples, and at significantly lower levels in tumoral samples; a normal luminal epithelium and stromal set (105 genes), highly expressed in normal and stromal samples, but expressed at low levels in tumoral samples or in cultured cells; and a basal epithelial set (28 genes), strongly expressed in normal prostate samples and in cultured prostate epithelial cells, but weakly expressed in tumoral or stromal samples or in non-prostate HeLa cells. The consideration of the latter gene set as characteristic of prostatic basal epithelium relies on the facts that many of its member genes are known markers of basal cells of prostate glands and other stratified epithelia, such as those for tumor protein p63 (TP73L), keratin 5 (KRT5), keratin 7 (KRT7) and keratin 14 (KRT14), or laminin beta 3 (LAMB3) [18, 19], and that they lack expression of markers of prostate malignancy such as alpha-methylacyl-coenzyme A racemase (AMACR, [20]) and hepsin (HPN, [21, 22]).

Bild vergrößern

As a validation of the microarray expression data, 45 out of the 318 tumoral-normal discriminant genes selected by FADA were further validated by Q-PCR on tumoral-normal paired samples, including whole tissue samples as a technical validation and microdissected tissues in order to assess whether the over- or underexpression of these genes was associated with the epithelial component of the samples. These genes were selected for Q-PCR validation according to their ranking by FADA for discrimination between tumor vs. normal prostate samples, and their capacity to correctly cluster samples into the above four categories obtained by microarray analysis (Figure 2A).

Bild vergrößern

For the four non-microdissected paired samples, 1 of the 45 genes analyzed could not be evaluated due to failure of the TaqMan probe to yield reproducible quantitation. Of the 44 remaining genes, 31 (31/44, 70%) presented tumoral vs. normal Q-PCR ratios that were similar to, or in the same direction as (upregulated or downregulated) those determined by microarray analysis for the same samples (Figure 2B and 2C). For the 7 microdissected pairs, 43 of the 45 genes could be evaluated, 28 of which (28/43, 65%) yielded tumor vs. normal Q-PCR ratios that roughly corresponded to those determined by microarrays for the same samples, 26 of these 28 genes being coincident with the 31 genes validated in non-microdissected samples (Figure 2C). These results reflect the usefulness of selecting samples with special attention to their histological characteristics, but also the necessity of validating microarray data results by complementary approaches such as Q-PCR.

The expression of proteins corresponding to several of the above defined functional gene clusters was additionally validated by immunohistochemistry on tissue microarrays. A monoclonal antibody to myosin VI (MYO6) stained luminal cells with a homogeneous cytoplasmic pattern, with tumor cells showing frequent apical reinforcement (Figure 3A and 3D). This protein was overexpressed in a majority of prostate tumors (41 of 49 samples in which both tumoral and normal glands were represented in the same cores, 83.6%), and also in a majority of PIN lesions (17 of 21 samples, 80.6%). The multidrug resistance-associated protein 4 (ABCC4) showed a predominant membrane staining pattern (Figure 3B and 3E), and was overexpressed in 82.9% (39 of 47 samples) of tumor cases and in 81.25% (13 of 16 samples) of PIN lesions. Finally, the ephrin type-A receptor 2 precursor (EPHA2), whose transcript was underexpressed in tumoral samples, was undetectable in luminal cells from normal and tumoral glands or stroma, while it was strongly expressed in a subpopulation of basal cells in normal glands (Figure 3C and 3F), consistent with the inclusion of this gene in the basal cell compartment signature.

Bild vergrößern

The analysis of colocalization of genes in adjacent positions in the genome can provide insights into higher levels of regulation that determine the selective coexpression of genes in cancer or putative regions of chromosomic instability [23]. The criteria used in this analysis included a significant association of coexpression for 4 or more FADA-selected genes colocalized within 4 Mb or less in the genome. Our analysis found one cluster of contiguous overexpressed genes with a significant bias in distribution (p = 6.85 × 10^-4), located on chromosome 17q25.3 (Figure 4A), which included five FADA-selected genes: synaptogyrin 2 (SYNGR2), solute carrier family 25 member 10 (SLC25A10), procollagen-proline, 2-oxoglutarate 4-dioxygenase (proline 4-hydroxylase), beta polypeptide (P4HB), pyrroline-5-carboxylate reductase 1 (PYCR1) and dihydrouridine synthase 1-like (DUS1L). The last four genes colocalize within a segment of less than 0.4 Mb. We re-analyzed the expression profiles of the genes in the 17q25.3 region present in the microarrays used in this study, and found that many of the genes in this region were in fact overexpressed in a significant proportion of tumors compared to both normal and stromal tissues, even if they had not been selected in our original FADA analysis because of their lower discriminant power when taken individually (Figure 4B).

Bild vergrößern

The relatively coordinate overexpression of so many genes, with a tight genomic colocalization to 17q25.3, could be due to genomic amplifications or gains of this region or, alternatively, to the co-regulation by factors acting in trans, or by a locus control region effect. In order to address this issue, we performed fluorescent in situ hybridization (FISH) on paraffin-embedded samples, corresponding to the same 20 cases analyzed for their transcriptional profiles. The results showed a gain in 17q25.3 (Figure 4C) in a majority of tumors (13 out of 20, 65%), which was never observed in matched normal glands. The median gain observed was 3 copies of the region, which correlated well with the overall levels of overexpression of the genes in this chromosomal region, with ratios generally not above 2-fold (tumoral vs. normal tissue). Genetic and phenotypic heterogeneity within tumors is a recurring theme in prostate cancer [24], a possible reflection either of multiclonality or of clonal drift, and it was also observed in this analysis. This heterogeneity was found amongst samples from different cases carrying the alteration, where the percentage of cells with a gain in 17q25.3 varied from 14% to 80%, with a mean of 47.4% nuclei containing the abnormality. Furthermore, the 17q25.3 gain showed a heterogeneous distribution in any given sample, with tumoral zones of the same sample with and without this alteration. Since we could only analyze some regions of the samples, this heterogeneity suggests that the observed prevalence of this segmental copy number gain is most likely an underestimation.

Prostate intraepithelial neoplasia (PIN) lesions could be analyzed near the tumoral zone in four of the tumors that presented this gain. Although the number of PIN nuclei were usually not sufficient for statistic significance, all cases showed a gain in 17q25.3 (Figure 4D), suggesting that this alteration is an early event in prostate cancer. Finally, the frequent finding of this gain in more than 60% (5 of 8) of metastatic samples (Figure 4E), with a mean of 69.4% of nuclei containing the alteration (ranging from 47.5% to 98.5%), indicates that this alteration is conserved during all steps of tumor progression.