Molecular Features of the Transition from Prostatic Intraepithelial Neoplasia (PIN) to Prostate Cancer: Genome-wide Gene-expression Profiles of Prostate Cancers and PINs

Prostate cancer (PC) is the most common malignancy in males and the second-leading cause of cancer-related deaths in the United States and Europe (1). Despite surgery and radiation therapy that often cure localized disease, and the possibility of early diagnosis through testing for prostate-specific antigen in serum, up to 30% of treated PC patients suffer relapse (2, 3, 4). Most patients with relapsed or advanced disease initially respond to androgen-ablation therapy because early PC growth is androgen dependent, but they eventually progress to androgen-independent disease, which no longer responds to androgen ablation. The most important issue to be solved in this context is that this advanced stage of PC does not respond to other therapies either.

Investigations designed to clarify the molecular mechanisms that underlie the initiation and progression of PC have revealed altered expression of PTEN, p27KIPS, and Nkx3.1 tumor suppressor genes, the c-myc oncogene, and the Bcl-2 antiapoptotic gene in human and/or mouse PCs (5, 6). However, the overall molecular events are still largely unknown. High-grade PIN (prostatic intraepithelial neoplasia) has been considered to be a putative precursor of cancer (7), and some clinical and scientific evidence supports a relationship between high-grade PINs and PCs. However, that hypothesis is still controversial (8, 9). Added to the difficulties of proving a PIN-to-PC transition is that human PCs are histopathologically very heterogeneous, and PIN is generally a very small lesion. Most previous molecular-based studies have analyzed bulk cancer tissues that were contaminated by a large proportion of noncancerous cells including fibromascular, microvasculature, and inflammatory cells, but such strategies do not yield precise molecular profiles of the respective lesions.

To overcome that limitation, we obtained purified populations of PC and PIN cells, as well as normal prostatic epithelial cells as a control, by LMM (laser microbeam microdissection), before analyzing genome-wide gene-expression profiles of 20 PCs and 10 PINs on a cDNA microarray representing 23,040 genes. This is the first report to show precise expression profiles of PC and PIN tumors, and to disclose molecular features confirming a relationship between PINs and PCs.

Tissue samples were obtained with informed consent from 26 cancer patients undergoing radical prostatectomy. All of the surgical specimens were at clinical stages T_2a to T_3a with or without N₁, and their Gleason scores were 5–9. All of the samples were embedded in TissueTek OCT medium (Sakura, Tokyo, Japan) immediately after surgical resection and stored at −80°C until use. Histopathological diagnoses were made by a single pathologist (M. F.) before LMM. H&E-stained sections from adjacent frozen tissues were prepared to confirm the histological diagnosis. High-grade PINs included in the surgical specimen of PC were characterized and identified by the following criteria: (a) a basal cell layer consistently enveloped the intraductal/acinar proliferation; and (b) atypical cells that had large nuclei of relatively uniform size increased the chromatin content, which might be irregularly distributed, and had prominent nucleoli that were similar to those of carcinoma cells (7, 10). Once high-grade PIN cells, PC cells, and normal epithelium were identified by H&E study, we microdissected them by LMM from the adjacent frozen slides. Normal prostatic epithelial cells were microdissected from the prostatic lobe opposite to the prostate cancer, or the normal prostatic tissue apparently far from the cancer. From among the 26 resected tissues, 20 cancers and 10 high-grade PINs had sufficient amounts and quality of RNA for microarray analysis.

LMM and T7-based RNA amplification were performed as described previously (11). Prostate tumor cells and normal prostatic epithelial cells were isolated selectively by the EZ cut system with a pulsed UV narrow beam-focus laser (SL Microtest GmbH, Germany) in accordance with the manufacturer’s protocols. After DNase treatment, total RNAs were subjected to two rounds of T7-based amplification, which yielded 50 to 100 μg of amplified RNA (3) from each sample. Then 2.5-μg aliquots of amplified RNA from PC or PIN cells and from normal prostatic ductal epithelial cells were labeled by reverse transcription with Cy5-dCTP (tumor cells) or Cy3-dCTP (normal cells (Amersham Biosciences, Buckinghamshire, United Kingdom), as described previously (12).

We fabricated a genome-wide cDNA microarray with 23,040 cDNAs selected from the UniGene database (build no.131) of the National Center for Biotechnology Information (NCBI).⁷ Construction, hybridization, washing, and scanning were carried out according to methods described previously (11, 12). Signal intensities of Cy3 and Cy5 from the 23,040 spots were quantified and analyzed by substituting backgrounds with ArrayVision software (Imaging Research, Inc., St. Catharines, Ontario, Canada). Subsequently, the fluorescent intensities of Cy5 (tumor) and Cy3 (control) for each target spot were adjusted so that the mean Cy3/Cy5 ratio of 52 housekeeping genes was equal to one. Because data derived from low-signal intensities are less reliable, we determined a cutoff value on each slide (12), and we excluded genes from further analysis when both the Cy3 and the Cy5 dyes yielded signal intensities lower than that of the cutoff. For other genes, we calculated the Cy5/Cy3 ratio using the raw data of each sample.

We applied a hierarchical clustering method to both genes and tumors, excluding genes whose Cy3- and Cy5-fluorescence intensities were both below the cutoff value. To obtain reproducible clusters for classification of the 30 tumors, we selected 63 genes for which valid data were obtained in 80% of the experiments and whose expression ratios varied by SDs of more than 1.5. The analysis was performed with web-available software (Cluster and TreeView) written by Eisen.⁸ Before applying the clustering algorithm, we log-transformed the fluorescence ratio for each spot and then median-centered the data for each sample to remove experimental biases.

We identified genes the expression of which was altered from normal epithelium to 10 PINs and 20 PCs according to the following criteria: (a) genes for which we were able to obtain expression data in more than 50% of the cases examined; and (b) genes the expression ratio of which was more than 3.0 or less than 0.33 in more than 50% of informative cases.

We identified genes with changed expression in 20 PCs and 10 PINs according to the following criteria: (a) genes for which we were able to obtain expression data in more than 50% of the cases examined; and (b) genes the expression ratio of which was more than 3.0 in PCs and between 0.5 and 2.0 in PINs (defined as up-regulated genes) or genes whose expression ratio was less than 0.33 in cancers and between 0.5 and 2.0 in PINs (defined as down-regulated genes) in more than 50% of informative cases.

Formalin-fixed and paraffin-embedded prostatic tumor sections were immunostained with a mouse anti-APOD (apolipoprotein D) monoclonal antibody (NeoMarkers, Fremont, CA) or a rabbit anti-EPHA4 (EphA4) polyclonal antibody (Santa Cruz Biotechnology Inc., Santa Cruz, CA) for APOD or EPHA4 expression. PC tissues included PC cells, PIN cells, and normal prostatic epithelium heterogeneously. Deparaffinized tissue sections were placed in 10 mmol/L citrate buffer (pH 6.0) and were heated to 108°C in an autoclave for 15 minutes for antigen retrieval. Sections were incubated with a 1:10 dilution or a 1:100 dilution of primary antibody for APOD or EPHA4, respectively, in a humidity chamber for 1 hour at room temperature and were developed with peroxidase labeled-dextran polymer followed by diaminobenzidine (DAKO Envision Plus System; DAKO Corporation, Carpinteria, CA). Sections were counterstained with hematoxylin. For negative controls, primary antibody was omitted.

We used small interfering RNA (siRNA)-expression vector (psiU6BX) for RNA interference effect to the target genes. The U6 promoter was cloned upstream of the gene-specific sequence (19-nucleotide sequence from the target transcript, separated by a short spacer TTCAAGAGA from the reverse complement of the same sequence) and five thymidines as a termination signal as well as neocassette for selection with Geneticin (Sigma). The target sequences for EphA4 are 5′-TCCGAACCTACCAAGTGTG-3′ (198si), 5′-TCATGAAGCTGAACACCGA-3′ (468si) and 5′-GCAGCACCATCATCCATTG-3′ (1313si). The sequence of 5′-GAAGCAGCACGACTTCTTC-3′ (EGFPsi) corresponding to EGFP was as a negative control. PC3 PC cells were plated onto 10-cm dishes (5 × 10⁵ cells/dish) and transfected with psiU6BX containing EGFP target sequence (EGFP) or psiU6BX containing EPHA4 target sequence with LipofectAMINE 2000 (Invitrogen) according to the manufacturer’s instruction. Cells were selected by 500 μg/mL Geneticin for 1 week, harvested 48 hours after transfection, and analyzed by reverse transcription (RT)-PCR to validate knockdown effect on EPHA4. The primers of RT-PCR were 5′-GAAGGCGTGGTCACTAAATGTAA-3′ and 5′- TTTAATTTCAGAGGGCGAAGAC-3′ for EPHA4, and 5′-GAGAGAGAATGAAAAGTGGAGCA-3′ and 5′-GATTAACCACAACCATGCCTTAC-3′ for β2-MG, used to quantify the amount of cDNA input. These cells were also stained by Giemsa solution and applied for 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay to evaluate the colony formation and the cell number, respectively.

To avoid contamination of PC or PIN tissues with stromal cells as much as possible, we performed LMM of 26 surgical specimens of PCs. Because PCs and PINs originate from prostatic epithelial ductal cells, we also purified normal ductal epithelial cells as controls for our microarray study. Normal prostatic ductal cells, PC cells (PC), and PIN cells (PIN) were clearly microdissected from each clinical specimen as shown (N) in Fig. 1.

RNA from each microdissected specimen was amplified for analysis on a genome-wide cDNA microarray. We generated the expression profiles of 30 prostatic tumors (20 PCs and 10 PINs) from 26 patients (cancer and PIN coexisted in four patients, in which cases both lesions were examined). On the basis of expression patterns of 63 genes that we selected applying strict conditions (i.e., valid data obtained in 80% of the experiments and expression ratios that varied by SDs of more than 1.5), all of the tumors fell into two major groups according to their status of malignancy in an unsupervised way, except for three cases (Fig. 2,A) in which two PCs (12T and 20T) with low Gleason scores (score 5) were included in the branch corresponding to the PIN group, and one PIN (23PIN) belonged to the PC group. Histological re-examination of this misplaced PIN diagnosed it as a very-high-grade PIN (Fig. 2 B) that pathologically resembled PC and probably represented a transitional lesion to invasive PC. These experiments showed that, although it was generally possible to distinguish PC from PIN by molecular classification on the basis of expression profiles, the distinction was not as obvious as we expected, and some molecular features were commonly shared. These results offered considerable validity to the concept that high-grade PINs can become invasive PCs.

Because we have obtained supportive evidence for the concept that PIN is a precursor of invasive PC through our hierarchical-clustering analysis of molecular profiles, it seemed likely that genes involved in an early stage of prostatic carcinogenesis would be deregulated from normal epithelium to PIN and also their deregulation should be retained in PCs (Fig. 3,A). On the basis of this hypothesis of prostatic carcinogenesis, we identified 21 genes that were commonly up-regulated in PINs and PCs according to the criteria mentioned in Materials and Methods (Fig. 3,A; Table 1); they included AMACR, OR51E2, RODH and SMS. On the other hand, 63 genes were selected as commonly down-regulated in PINs and PCs (Fig. 3,A; Table 2); that list included several proteinases such as SERPINB1, SERPING1, and MMP7.

We then focused on differential expression patterns between PINs and PCs to search for genes likely to be involved in the transition from non-invasive precursor PINs to malignant cancers (Fig. 3B). Comparing the expression profiles of 20 PCs with those of 10 PINs, we identified 41 up-regulated genes (Table 3) and 98 down-regulated genes (Table 4). The list included POV1, CDKN2C, EPHA4, APOD, FASN, and VWF as up-regulated, and ITGB2, LAMB2, PLAU, and TIMP1 as down-regulated; the altered genes might be involved with cell adhesion or motility in invasive PC cells.

To validate the differential expression pattern in the transition from PIN to PC, we performed immunohistochemical analysis of apolipoprotein D (APOD) and EPHA4, listed in Table 3 as up-regulated genes in the PIN-to-PC transition. In general, PC tissues include PC cells, PIN cells, and normal prostatic epithelium heterogeneously, and we compared the staining pattern of each kind of cells associated with prostatic carcinogenesis on the same tissues from the same patient. As shown in Fig. 3 C, APOD was abundantly expressed in PC cells, whereas PINs and normal prostatic epithelium from the same patient had no expression of APOD protein. The EPHA4 protein was also strongly expressed in PC cells while PINs and normal prostatic epithelium from the same patient had no, or very weak, expression of EPHA4 protein.

To investigate the growth or survival effect of EPHA4 on PC cells, we knocked down their endogenous expression by mammalian vector-based RNA interference technique by using the PC cell line. The transfection of one of the siRNA-expressing vectors, 1313si, clearly reduced the endogenous expression of EPHA4 (Fig. 4,A). This knocking-down effect by the siRNA on EPHA4 mRNA resulted in drastic growth suppression in colony formation assay as well as in MTT assay (Fig. 4 B and C). These findings strongly suggested that overexpression of EPHA4 in PC cells was associated with the enhanced growth of cancer cells.

PC is one of the most common malignant diseases in Western countries. A number of research groups have investigated gene-expression profiles of PCs (13, 14, 15, 16), but the results presented here are markedly different from theirs. Compared with a meta-analysis of four representative microarray datasets (17), only three (7.5%) of the 40 genes listed as being commonly up-regulated in PCs, and seven (17.5%) of the 40 down-regulated genes, coincided with our data. We attribute these discrepancies to the fact that prostate tumors are histopathologically very heterogeneous, containing a large proportion of stromal cells, and expression profiles from bulk tissues could be substantially influenced by contamination with surrounding noncancerous cells and by the proportions of various cell types. For example, genes that are highly expressed in stromal cells, but not in epithelial cells, may be selected as down-regulated elements in microarray analyses of bulk tissues, because stromal components are less abundant in cancer tissues than in normal prostate (∼17 versus 60%; ref. 15). SDF1 (stromal cell-derived factor 1), which is highly expressed in stroma, showed low signal (under the cutoff) in our experiments (data not shown). However, Stamey et al. (15) reported that gene as down-regulated in PCs, probably because stromal components were present in the normal control. Similarly, genes that are expressed in normal epithelial cells and cancer cells, but not in stromal cells, may be selected as up-regulated elements during microarray analysis of bulk tissue if noncancerous epithelial cells are enriched in the tumor tissue. For instance, PRSS8 (prostasin) was reportedly up-regulated in many microarray studies that used bulk tissues (17), but in our experiments, expression of this gene was either unchanged or down-regulated in some PCs as compared with normal epithelium. PRSS8 is predominantly expressed in normal prostatic epithelial cells; several studies using in situ techniques have revealed its down-regulation in PC cells (18, 19). These observations make clear that contamination of cancer tissues with stromal cells or normal prostate epithelium can mislead efforts to understand the molecular pathology of PC; therefore, microdissection is definitely required for precision in microarray analyses of PCs. In this study, we applied the LMM system (11) to purify populations of cancer cells, PIN cells, and normal prostatic epithelial cells from surgical specimens, in an effort to obviate influences from stromal components and normal prostate.

Several lines of clinical and pathological evidence have linked PINs to PCs, and high-grade PIN has been widely considered the precursor of invasive PC (8, 10). PINs and PCs can show certain genetic alterations in common (20). However, some data have argued against this hypothesis; for example, early-stage PCs are not always accompanied by PIN (9, 10) and some genetic markers of PINs are not always observed in PCs (21). Hence, considerable controversy has arisen about the natural history of high-grade PIN and the mechanism of PIN development, and the putative transition from PIN to invasive PC has remained unsettled.

We analyzed the precise molecular profiles of purified populations of PC and PIN cells to investigate the molecular mechanism of prostate carcinogenesis and to examine the molecular linkage between PINs and PCs. Our unsupervised hierarchical clustering analysis was able to distinguish most PCs from PINs on the basis of their expression profiles, although two low-grade PCs with Gleason scores of 5 fell in the PIN cluster and one high-grade PIN was classified into the PC group. The latter turned out to be a very-high-grade PIN in a retrospective histological review and probably represented a transitional lesion to invasive PC. These results implied that the gene-expression profile of very-high-grade PIN is very similar to that of PC, supporting the concept that at least some high-grade PINs are precursors of PC. It should also be added that prostate pathologists have long recognized a phenomenon in which PC spreads along prostatic ducts to mimic PIN (22). In fact, an early controversy that arose when the concept of PIN was first introduced was that the lesions merely reflected intraductal spread of invasive cancer. Except for the one PIN case, our microarray data comparing PIN cells with PC cells rejects this criticism and clearly supports the concept that PINs are precursor lesions of PC, not ductal spread of PC cells.

We identified genes that were up- or down-regulated (Tables 1 and 2) commonly in PINs and PCs comparing with normal epithelium; those genes may be involved in an early step of prostatic carcinogenesis. The up-regulated elements in PINs and PCs included AMACR (α-methyl acyl-CoA racemase), OR51E2 (olfactory receptor, family 51, subfamily E, member 2), RDOH (3-hydroxysteroid epimerase), and SMS (spermine synthase). AMACR has a key role in β-oxidation of dietary branched-chain fatty acids and is often overexpressed in PCs (23, 24). OR51E2 is also overexpressed in PCs (25), but its function is unknown. RDOH functions in androgen catabolism and is likely to be associated with androgen-dependent prostate tumorigenesis. SMS is a polyamine-biosynthesis enzyme; polyamines such as spermine have been implicated in the growth of PC cells and protection from apoptosis (26). On the other hand, SERPIN-B1 [serine (or cysteine) proteinase inhibitor, clade B (ovalbumin) member 1] and SERPIN-G1 [serine (or cysteine) proteinase inhibitor, clade G (ovalbumin), member 1] were down-regulated from normal prostate epithelium to PIN. The pigment epithelium-derived factor encoded by SERPIN-F1 was recently reported to be a key inhibitor of stromal vasculature and growth of epithelial tissue in mouse prostate (27); in pigment epithelium-derived factor-deficient mice, stromal vessels were increased and associated with epithelial-cell hyperplasia. Those observations suggest to us that SERPIN-B1 and SERPIN-G1 may also be key regulators of prostate growth. Almost all genes that were up- or down-regulated in PINs were also up- or down-regulated, respectively, in PCs compared with normal prostate epithelium, implying that PCs retain the molecular features of their putative precursor PINs. These data supported the notion of PIN-to-PC transition.

Next, we focused on genes showing differential expression after the transition from PINs to PCs (see Fig. 3,B; Tables 3 and 4). The list included POV1, CDKN2C, EPHA4, APOD, and FAM as up-regulated genes and LAMB2, ITGB2, PLAU, and TIMP1 as down-regulated genes. EPHA4 is one of the receptor tyrosine kinase receptors and is likely to play a critical role in neuronal circuit development and angiogenesis by regulating cell shape and motility (28). However, EPHA4 overexpression in PCs is novel and may involve PC cell viability and motility. Some of the latter are associated with cell adhesion and proteinase activity, suggesting that their expression changes may contribute to the invasive phenotype by abolishing ductal structures during the transition from PIN to PC. We validated the differential expression of APOD and EPHA4 by the immunohistochemical analysis on several PC tissues (Fig. 3 C), which was consistent with our precise RNA profiles and indicated that our profile analysis was highly reliable.

Furthermore, we investigated by the small interfering RNA strategy whether overexpression of EPHA4 was associated with enhanced growth of PC cells. As shown in Fig. 4, down-regulation of EPHA4 in PC cell line resulted in a drastic reduction in PC cell viability, suggesting that the overexpression of EPHA4 in the transition of PIN to PC cells is likely to be essential in PC cell growth, indicating that targeting to EPHA4 may be a promising approach to develop novel PC treatments.

Overall, it is interesting that our lists of differentially expressed genes in prostate carcinogenesis via PINs to invasive PCs included many genes associated with lipid metabolism, including AMACR, FASN (fatty acid synthase), MFGE8 (milk fat globule-EGF factor 8 protein), APOD (apolipoprotein D), APOL1 (apolipoprotein L1), PLA2G2A (phospholipase A2, group IIA), and SORL1 (sortilin-related receptor L). Epidemiological aspects suggest that the development of PC is strongly associated with high fat intake (1); our data concerning these genes may help to explain the apparent association between fat intake and prostate carcinogenesis.

In conclusion, our extensive list of genes derived from precise expression profiles of PCs and PINs should provide useful information for identifying molecular targets for the prevention and treatment of PC, as well as for understanding the molecular mechanism of prostatic carcinogenesis, especially the transition from PIN to PC.

We thank Noriko Sudo, Saori Osawa, and Miwako Ando for fabrication of the cDNA microarray; Emi Ichihashi for analysis of the data; Tae Makino and Noriko Ikawa for preparation of samples by cryostat; and Drs. Soji Kakiuchi and Takefumi Kikuchi for helpful discussions.