Molecular Features of Hormone-Refractory Prostate Cancer Cells by Genome-Wide Gene Expression Profiles

One of the most critical issues in prostate cancer clinic is emerging hormone-refractory prostate cancers (HRPCs) and their management. Prostate cancer is usually androgen dependent and responds well to androgen ablation therapy. However, at a certain stage, they eventually acquire androgenindependent and more aggressive phenotype and show poor response to any anticancer therapies. To characterize the molecular features of clinical HRPCs, we analyzed gene expression profiles of 25 clinical HRPCs and 10 hormonesensitive prostate cancers (HSPCs) by genome-wide cDNA microarrays combining with laser microbeam microdissection. An unsupervised hierarchical clustering analysis clearly distinguished expression patterns of HRPC cells from those of HSPC cells. In addition, primary and metastatic HRPCs from three patients were closely clustered regardless of metastatic organs. A supervised analysis and permutation test identified 36 up-regulated genes and 70 down-regulated genes in HRPCs compared with HSPCs (average fold difference > 1.5; P < 0.0001). We observed overexpression of AR, ANLN, and SNRPE and down-regulation of NR4A1, CYP27A1 , and HLA-A antigen in HRPC progression. AR overexpression is likely to play a central role of hormone-refractory phenotype, and other genes we identified were considered to be related to more aggressive phenotype of clinical HRPCs, and in fact, knockdown of these overexpressing genes by small interfering RNA resulted in drastic attenuation of prostate cancer cell viability. Our microarray analysis of HRPC cells should provide useful information to understand the molecular mechanism of HRPC progression and to identify molecular targets for development of HRPC treatment. [Cancer Res 2007;67(11):5117–25] Introduction Prostate cancer is the most commonmalignancy in males and the second leading cause of cancer-related death in the United States and Europe (1). The incidence of prostate cancer has been increasing significantly in most of developed countries due to prevalence of Western-style diet and explosion of the aging population (1, 2). The screening using serum prostate-specific antigen (PSA) lead to dramatic improvement of early detection of prostate cancer and resulted in an increase of the proportion of patients with a localized disease that could be curable by surgical and radiation therapies (1, 2). However, 20% to 30% of these prostate cancer patients still suffer from the relapse of the disease (3–5). Androgen/androgen receptor (AR) signaling pathway plays a central role in prostate cancer development, and the prostate cancer growth is usually androgen-dependent at a relatively early stage (3–5). Hence, most of the patients with relapsed or advanced disease respond well to androgen ablation therapy, which suppresses testicular androgen production by surgical castration or by administration of an agonist(s) to luteinizing hormone-releasing hormone (LH-RH) and antiandrogen drugs. Nonetheless, they eventually acquire androgen-independent and more aggressive phenotype that has been termed hormonerefractory prostate cancers (HRPCs). Recently, the combination of docetaxel and prednisone was established as the new standard of care for HRPC patients (6, 7), but they are not curable and their survival benefit on HRPC patients is very limited. Hence, many groups are now attempting various approaches to identify novel molecule targets or signaling pathways that contribute to growth of HRPC (8). Several studies using in vitro prostate cancer cell lines and mouse models have shown that the progression to HRPC could be associated with increased levels of AR expression, implicating that AR down-regulation by means of small interfering RNA (siRNA) or other methods should suppress tumor growth even in HRPCs (8–11). The AR gene was overexpressed in most of HRPCs, in 10% to 20% of which amplification of the AR gene was observed (12). In addition, a subset (<10%) of HRPCs was found to have somatic mutations in the AR gene, which could enhance ligand response (13). As consequence, expressions of several AR-regulated genes were reactivated even under androgen depletion (3, 14–16). Furthermore, the AR pathway in HRPCs was consider to rely on alterations in growth factors, such as insulin-like growth factor Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/). Accession codes: The complete microarray data set is available from the Gene Expression Omnibus (GSE6811). Requests for reprints: Hidewaki Nakagawa, Laboratory of Molecular Medicine, Human Genome Center, Institute of Medical Science, The University of Tokyo, 4-6-1 Shirokanedai, Minato-ku, Tokyo 108-8639, Japan. Phone: 81-3-5449-5375; Fax: 81-35449-5124; E-mail: hidewaki@ims.u-tokyo.ac.jp. I2007 American Association for Cancer Research. doi:10.1158/0008-5472.CAN-06-4040 www.aacrjournals.org 5117 Cancer Res 2007; 67: (11). June 1, 2007 Research Article Research. on August 15, 2017. © 2007 American Association for Cancer cancerres.aacrjournals.org Downloaded from (17), HER-2 (18), and cytokines, such as interleukin-6 (19), which could modify the AR activity. Overexpression of these growth factors or coactivators in HRPCs might change cancer cells to be independent of the AR signaling (8, 14, 20). Despite these latest advances in molecular analysis of AR pathways, the mechanisms by which prostate cancer cells survive and acquire their more aggressive phenotype after androgen ablation therapy are still not well understood. In this report, to characterize the molecular feature of clinical HRPCs, we did genome-wide cDNA microarray analysis of cancer cells purified from HRPC tissues by means of laser microbeam microdissection (LMM) and identified several deregulated genes in HRPCs, some of which might be involved in their androgenindependence and aggressive phenotype. These data should shed a light on a better understanding of the molecular mechanisms underlying clinical HRPCs and could suggest candidate genes whose products could serve as molecular targets for development of novel treatment for HRPC. Materials and Methods Patients and tissue samples. Tissue samples were obtained with informed consent from 43 HRPC patients undergoing prostatic needle biopsy, bone biopsy, transurethral resection of the prostate (TUR-P), and ‘‘warm’’ autopsy. Clinical HRPC was defined by elevation of serum PSA Figure 1. A, dendrogram of an unsupervised hierarchical clustering analysis of 254 genes (vertical columns ) across 35 prostate cancers (horizontal rows ). The unsupervised hierarchical clustering analysis clearly distinguished 25 HRPCs (red) from 10 HSPCs (blue ). Small subsets of cluster constituted by metastatic HRPC cells (-B , bone metastasis; -L , lymph node metastasis; -H, liver metastasis) and HRPC cells at the primary site (-P, prostate) from the same individuals (black boxes ). Tissue procurement methods for each HRPC specimen (A , autopsy; B , biopsy; T , TUR-P). All HSPC tissues were procured by radical prostatectomy. B, dendrogram of a supervised analysis of 106 genes (vertical columns ) across 13 HRPCs at the prostate and 10 HSPCs (horizontal rows ). Each cell in the matrix represents the expression level of a single transcript in a single sample. Red and green, transcript levels, above and below the median for that gene across all samples. Black, unchanged expression; gray, no detectable expression. The 36 up-regulated genes and 70 down-regulated genes that can distinguish HRPC cells from HSPC cells are listed in Tables 1 and 2, respectively. Cancer Research Cancer Res 2007; 67: (11). June 1, 2007 5118 www.aacrjournals.org Research. on August 15, 2017. © 2007 American Association for Cancer cancerres.aacrjournals.org Downloaded from levels at three consecutive times and/or enlargement of tumor in spite of androgen ablation therapy. All of the samples were embedded in OTC Compound (Tissue-Tek) immediately after tissue procurement and stored at 80jC until their use. Histopathologic diagnoses were made by a single pathologist (M.F.) before LMM, and H&E-stained sections from adjacent frozen tissues were prepared to confirm the histologic diagnosis. Among the 43 HRPC patients we obtained, 25 primary or metastatic tumor specimens from 18 HRPC patients had sufficient amounts and good quality of RNAs for our microarray analysis. Tissue procurement method and therapeutic treatments for these 18 HRPC patients are shown in Supplementary Table S1. We microdissected cancer cells from these frozen slides by means of LMM (EZ cut system with a pulsed UV narrow beam-focus laser, SL Microtest GmbH). Simultaneously, 10 hormone-sensitive prostate cancers (HSPCs) or hormone-naive prostate cancers were also microdissected from 10 untreated operable cases undergoing radical prostatectomy, and normal prostatic (NP) epithelial cells were also microdissected from one benign prostatic hyperplasia patient and four bladder cancer patients, where we confirmed no apparent prostate cancers or prostatic intraepithelial neoplasias histopathologically. Genome-wide cDNA microarray analysis and acquisition of data. LMM and T7-based RNA amplification were done as described previously (21). Amplified RNAs of 2.5 Ag each were labeled by reverse transcription with Cy5-dCTP for cancer cells or Cy3-dCTP for normal cells (Amersham Biosciences) as described previously (21). We fabricated a genome-wide cDNA microarray with 36,864 cDNAs selected from the UniGene database (build no. 131) of the National Center for Biotechnology Information. Construction, hybridization, washing, and scanning were carried out according to methods described previously (21). Signal intensities of Cy3 and Cy5 from the 36,864 spots were quantified and analyzed by substituting backgrounds with ArrayVision software (Imaging Research, Inc.). Subsequently, the fluorescent intensities of Cy5 (cancer) and Cy3 (normal control) for each target spot were adjusted so


Introduction
Prostate cancer is the most common malignancy in males and the second leading cause of cancer-related death in the United States and Europe (1).The incidence of prostate cancer has been increasing significantly in most of developed countries due to prevalence of Western-style diet and explosion of the aging population (1,2).The screening using serum prostate-specific antigen (PSA) lead to dramatic improvement of early detection of prostate cancer and resulted in an increase of the proportion of patients with a localized disease that could be curable by surgical and radiation therapies (1,2).However, 20% to 30% of these prostate cancer patients still suffer from the relapse of the disease (3)(4)(5).
Androgen/androgen receptor (AR) signaling pathway plays a central role in prostate cancer development, and the prostate cancer growth is usually androgen-dependent at a relatively early stage (3)(4)(5).Hence, most of the patients with relapsed or advanced disease respond well to androgen ablation therapy, which suppresses testicular androgen production by surgical castration or by administration of an agonist(s) to luteinizing hormone-releasing hormone (LH-RH) and antiandrogen drugs.Nonetheless, they eventually acquire androgen-independent and more aggressive phenotype that has been termed hormonerefractory prostate cancers (HRPCs).Recently, the combination of docetaxel and prednisone was established as the new standard of care for HRPC patients (6,7), but they are not curable and their survival benefit on HRPC patients is very limited.Hence, many groups are now attempting various approaches to identify novel molecule targets or signaling pathways that contribute to growth of HRPC (8).
Several studies using in vitro prostate cancer cell lines and mouse models have shown that the progression to HRPC could be associated with increased levels of AR expression, implicating that AR down-regulation by means of small interfering RNA (siRNA) or other methods should suppress tumor growth even in HRPCs (8)(9)(10)(11).The AR gene was overexpressed in most of HRPCs, in 10% to 20% of which amplification of the AR gene was observed (12).In addition, a subset (<10%) of HRPCs was found to have somatic mutations in the AR gene, which could enhance ligand response (13).As consequence, expressions of several AR-regulated genes were reactivated even under androgen depletion (3,(14)(15)(16).Furthermore, the AR pathway in HRPCs was consider to rely on alterations in growth factors, such as insulin-like growth factor (17), HER-2 (18), and cytokines, such as interleukin-6 (19), which could modify the AR activity.Overexpression of these growth factors or coactivators in HRPCs might change cancer cells to be independent of the AR signaling (8,14,20).Despite these latest advances in molecular analysis of AR pathways, the mechanisms by which prostate cancer cells survive and acquire their more aggressive phenotype after androgen ablation therapy are still not well understood.
In this report, to characterize the molecular feature of clinical HRPCs, we did genome-wide cDNA microarray analysis of cancer cells purified from HRPC tissues by means of laser microbeam microdissection (LMM) and identified several deregulated genes in HRPCs, some of which might be involved in their androgenindependence and aggressive phenotype.These data should shed a light on a better understanding of the molecular mechanisms underlying clinical HRPCs and could suggest candidate genes whose products could serve as molecular targets for development of novel treatment for HRPC.

Materials and Methods
Patients and tissue samples.Tissue samples were obtained with informed consent from 43 HRPC patients undergoing prostatic needle biopsy, bone biopsy, transurethral resection of the prostate (TUR-P), and ''warm'' autopsy.Clinical HRPC was defined by elevation of serum PSA The unsupervised hierarchical clustering analysis clearly distinguished 25 HRPCs (red) from 10 HSPCs (blue ).Small subsets of cluster constituted by metastatic HRPC cells (-B , bone metastasis; -L, lymph node metastasis; -H, liver metastasis) and HRPC cells at the primary site (-P, prostate) from the same individuals (black boxes ).Tissue procurement methods for each HRPC specimen (A , autopsy; B , biopsy; T , TUR-P).All HSPC tissues were procured by radical prostatectomy.B, dendrogram of a supervised analysis of 106 genes (vertical columns ) across 13 HRPCs at the prostate and 10 HSPCs (horizontal rows ).Each cell in the matrix represents the expression level of a single transcript in a single sample.Red and green, transcript levels, above and below the median for that gene across all samples.Black, unchanged expression; gray, no detectable expression.The 36 up-regulated genes and 70 down-regulated genes that can distinguish HRPC cells from HSPC cells are listed in Tables 1 and 2, respectively.levels at three consecutive times and/or enlargement of tumor in spite of androgen ablation therapy.All of the samples were embedded in OTC Compound (Tissue-Tek) immediately after tissue procurement and stored at À80jC until their use.Histopathologic diagnoses were made by a single pathologist (M.F.) before LMM, and H&E-stained sections from adjacent frozen tissues were prepared to confirm the histologic diagnosis.Among the 43 HRPC patients we obtained, 25 primary or metastatic tumor specimens from 18 HRPC patients had sufficient amounts and good quality of RNAs for our microarray analysis.Tissue procurement method and therapeutic treatments for these 18 HRPC patients are shown in Supplementary Table S1.We microdissected cancer cells from these frozen slides by means of LMM (EZ cut system with a pulsed UV narrow beam-focus laser, SL Microtest GmbH).Simultaneously, 10 hormone-sensitive prostate cancers (HSPCs) or hormone-naive prostate cancers were also microdissected from 10 untreated operable cases undergoing radical prostatectomy, and normal prostatic (NP) epithelial cells were also microdissected from one benign prostatic hyperplasia patient and four bladder cancer patients, where we confirmed no apparent prostate cancers or prostatic intraepithelial neoplasias histopathologically.
Genome-wide cDNA microarray analysis and acquisition of data.LMM and T7-based RNA amplification were done as described previously (21).Amplified RNAs of 2.5 Ag each were labeled by reverse transcription with Cy5-dCTP for cancer cells or Cy3-dCTP for normal cells (Amersham Biosciences) as described previously (21).We fabricated a genome-wide cDNA microarray with 36,864 cDNAs selected from the UniGene database (build no.131) of the National Center for Biotechnology Information.Construction, hybridization, washing, and scanning were carried out according to methods described previously (21).Signal intensities of Cy3 and Cy5 from the 36,864 spots were quantified and analyzed by substituting backgrounds with ArrayVision software (Imaging Research, Inc.).Subsequently, the fluorescent intensities of Cy5 (cancer) and Cy3 (normal control) for each target spot were adjusted so that the mean Cy3/Cy5 ratio of 52 housekeeping genes was equal to one.Because data with low-signal intensities are less reliable, we determined a cutoff value on each slide, and we excluded genes from further analysis when both the Cy3 and the Cy5 dyes yielded signal intensities lower than that of the cutoff value.For other genes, we calculated the Cy5/Cy3 ratio using the raw data of each sample.
Hierarchical clustering and statistical analysis for genome-wide gene expression profiles.We applied an unsupervised hierarchical  (11).June 1, 2007 clustering method to both genes and tumors, excluding genes, for which both Cy3 and Cy5 fluorescence intensities were below the cutoff value.To obtain reproducible clusters for classification of the 35 tumors, we selected 254 genes (about one percentile of the whole human transcripts) for which valid data were obtained in 80% of the experiments and whose expression ratios varied by SDs of >1.75.We log transformed the fluorescence ratio for each spot and then median centered the data for each sample to remove experimental biases.The unsupervised clustering analysis was done with web-available software (Cluster version 3.0 and TreeView version 1.0.12)written by Eisen. 12As a supervised analysis, we applied a random permutation test to identify genes that were expressed at a significantly different level between the two groups (HRPC: 1 versus HSPC: 2).For each gene (g ), we used a measure of correlation P (g, c) = [A 1 (g) À A 2 (g )] / [SD 1 (g ) + SD 2 (g )], which reflects the difference of average (A) between the groups relative to the SD within the groups (22).The results are compared with the corresponding distribution obtained for random idealized expression patterns c*, obtained by randomly permuting the coordinates of c = (c 1 , c 2 , . .., c n ), where c i = +1 or 0 according to whether the i-th sample belongs to group 1 or group 2 (22).The random permutations were applied so that each group had a constant number of samples.Using thus obtained empirical distribution of null hypothesis, we calculated P value for each gene.We selected 106 genes for which valid data were obtained by P value of <0.0001, average fold difference of >1.5, and one group present of >60%.
Semiquantitative reverse transcription-PCR and real-time quantitative reverse transcription-PCR.Total RNA was extracted using RNeasy kit (Qiagen) according to the manufacturer's instruction, treated with DNase I (Roche Diagnostics), and reversely transcribed to single-stranded cDNA using random hexamer or oligo(dT) 12-18 primer with SuperScript reverse transcriptase II (Invitrogen).We prepared appropriate dilutions of each single-strand cDNA followed by normalizing cDNA content using b-actin (ACTB) as a quantitative control, showing PCR using single-strand cDNA as PCR templates.The primers of each transcripts were the following: ACTB ACTB was used to normalize each expression and the primer sequences of ACTB were described above.
Immunohistochemistry. Paraffin-embedded tissue sections were deparaffinized, subjected to treatment with microwave at 360 W for 1 min four times in antigen retrieval solution, high pH (DAKO), and then treated with peroxidase blocking reagent (DAKO) followed by protein block reagent (DAKO).Immunohistochemical study was carried out using the Ventana automated IHC systems (DiscoveryTM, Ventana Medical systems, Inc.).Sections were incubated with a 1:100 diluted solution of a mouse monoclonal antibody (mAb; NCL-AR-318, Novocastra) against the NH 2terminal portion of the human AR overnight at 4jC.The automated protocol is based on an indirect biotin-avidin system using a biotinylated universal secondary antibody and diaminobenzidine substrate with hematoxylin counterstaining.

Results
Sample collecting and hierarchical clustering analysis of expression profiles.For this study, we collected 77 frozen specimens from 43 HRPC patients through prostatic needle biopsy, bone biopsy, TUR-P, or ''warm'' autopsy.Nearly two thirds of specimens were not qualified for LMM and microarray analysis through the pathologic evaluation by H&E staining or after evaluation of their RNA quality, and finally, RNAs of 25 HRPC specimens from 18 HRPC patients were available for further microarray analysis.All of these 18 patients had been treated with maximum androgen blockade with LH-RH agonist treatment or surgical castration as shown in Supplementary Table S1.Tissue procurements from six autopsies were done immediately after patient death.Twenty-five HRPC specimens from 18 patients included 13 HRPCs at the primary site (prostate), 8 bone metastases, 3 lymph node metastases, and 1 liver metastasis.Simultaneously, HSPC cells were also microdissected from 10 untreated operable patients undergoing radical prostatectomy, and NP epithelial cells were also microdissected from five non-prostate cancer patients.These NP cells from five males were used as a normal mixture control for our cDNA microarray analysis.We successfully microdissected HRPC cells, HSPC cells, and NP cells from each clinical sample to exclude the contamination of stromal cells and host organ cells at the metastatic sites, such as bone marrow cells, lymphocytes, and hepatocytes (Supplementary Fig. ).
An unsupervised hierarchical clustering analysis using expression patterns of 254 genes that we selected based on strict conditions (i.e., valid data obtained in 80% of the experiments and expression ratios that varied by >1.75 SDs) clearly classified the 35 tumors into two major groups, the HRPC and HSPC groups (Fig. 1A).This unsupervised hierarchical clustering analysis also classified multiple tumors from the same individuals to small subgroups regardless to the metastatic organs (Fig. 1A, black boxes), suggesting little influence on expression patterns by host organs of their metastatic sites due to the precise microdissection technique in our laboratory.
Identification of deregulated genes in the progression from HSPC to HRPC.To extract genes that showed significantly differential expression levels in HRPCs and HSPCs, we carried out a random permutation test using the expression profiles of 13 HRPCs at the prostate and 10 HSPCs.We selected 13 HRPCs at the primary site (prostate) among 25 HRPC specimens for this random permutation test because multiple HRPC samples from one individual showed quite similar patterns in the unsupervised hierarchical clustering analysis (Fig. 1A) and also because potential microenvironmental influence from host organs should be excluded as much as possible.The supervised analysis and this random permutation test (average fold difference > 1.5; P < 0.0001) identified 36 up-regulated genes and 70 down-regulated genes in HRPCs compared with HSPCs, which were considered to be involved in the presumably HRPC progression, that is to say, their androgen-independent growth and more aggressive or malignant phenotype (Fig. 1B).Table 1 listed 36 up-regulated genes in HRPC, including AR, small nuclear ribonucleoprotein peptide E (SNRPE), and anillin, actin binding protein (ANLN).Notably, the expression level of AR in HRPC cells was much higher than that in HSPCs, which was concordant with several previous reports studying the cell line models (9)(10)(11) and clinical samples (9).As shown in Fig. 2, semiquantitative RT-PCR validated overexpression of 11 genes in HRPC cells.On the other hand, Table 2 listed 70 down-regulated genes in the progression to HRPCs, including NR4A1, CYP27A1, and HLA-A.
AR expression and activity in clinical HRPC cells.We further analyzed AR protein expression in clinical HRPCs by immunohistochemistry and the transcriptional level of AR-regulated genes by real-time quantitative PCR, which should reflect the actual AR activity as a transcriptional factor in the nucleus.Immunohistochemical analysis for AR using 6 HRPCs and 16 HSPCs showed the positive staining in the nuclei of all HRPC, HSPC, and normal prostate.The staining intensity or patterns in HRPC cells (Fig. 3A) were similar to those in HSPC (Fig. 3B) and NP cells (Fig. 3B), although the transcript levels of AR in HRPC cells were much higher that in HSPC and NP cells, which were analyzed by semiquantitative PCR (Fig. 2) and also quantified by real-time PCR (Fig. 3D).Furthermore, real-time quantitative PCR showed that transcriptional levels of AR-regulated or downstream genes, PSA and NKX3.1 (24), in HRPC cells were also similar to those in HSPC and NP cells (Fig. 3D), in spite of >10 times overexpression of the AR transcript in HRPC cells.These findings implicated that clinical HRPC cells are likely to maintain AR activity in the cell by overexpressing AR mRNA, under very low level of circulating testicular androgen, but the stabilized AR protein and the actual AR activity levels as a transcriptional factor in the nuclei of HRPC cells were similar to that in HSPC cells and NP cells.Knockdown effect of ANLN and SNRPE on HRPC cell growth.To investigate the contribution of non-AR pathways or genes to HRPC phenotype, we selected ANLN and SNRPE, whose expressions were most significantly high in HRPCs (Fig. 2; Table 1).We constructed several vectors, designed to express siRNA specifically to ANLN (siANLN) and SNRPE (si1-3), and transfected each of them into prostate cancer cell line 22Rv1, which expressed ANLN and SNRPE at high level.The transfection with siANLN showed the significant knockdown effect on the ANLN transcript (Fig. 4A) and resulted in drastic reduction of the numbers of colonies (Fig. 4B, left) as well as those of the viable cells measured by MTT assay (Fig. 4B, right), whereas the transfection of a negative control (siEGFP) did not show any effect.Among three siRNA constructs to SNRPE, SNRPE-si1 and SNRPE-si3 significantly knocked down SNRPE expression (Fig. 4C) and caused drastic reduction of the numbers of colonies (Fig. 4D, left) as well as those of the viable cells measured by MTT assay (Fig. 4D, right), whereas the transfection of the other plasmid (si2) or a negative control (siEGFP) showed no or little knockdown effect on SNRPE expression and did not affect prostate cancer cell viability.These findings suggested that non-AR pathways represented by overexpressing genes in HRPC, such as SNRPE and ANLN, could play some important roles in the prostate cancer cell viability as well as the AR pathway.

Discussion
Most of the patients with relapsed or advanced prostate cancer respond well to androgen ablation therapy.However, the tumors eventually acquire androgen-independent and more aggressive phenotype for which most anticancer drugs or therapies are not effective, finally leading prostate cancer patients to death.(8).In this study, we approached to the molecular mechanism of acquirement of this more aggressive phenotype by analyzing human HRPC clinical samples but not by the use of cell lines or mouse models.The gene expression studies for clinical HRPCs have been very limited thus far, largely due to difficulties to obtain Figure 3. Immunohistochemical analysis of prostate cancer tissues by anti-AR mAb.Immunoreactivity with anti-AR antibody exhibited positive staining in the nuclei of HRPC cells (A ), HSPC cells (B), and NP epithelial cells (C ).Their staining intensity or patterns in HRPC cells is similar to those in HSPC and NP cells, although the mRNA level of AR in HRPC cells was much higher than in HSPC and NP cells (Figs. 2 and 3D ).D, left, real-time quantitative PCR showed >10 times overexpression of AR transcript in HRPC cells (samples 2-8 ) comparing with that of HSPC cells (samples 9-15 ) and NP cells (sample 1).On the other hand, transcriptional levels of PSA (middle ) and NKX3.1 (right ) of HRPC cells, which reflect AR activity, were similar to those of HSPC cells and NP cells.ACTB was used to quantify each of the cDNA contents, and the relative quantity (Y-axis ) was calculated so that the expression in NP cells was one.Real-time PCR was done duplicated for each sample (white and black columns ).appropriate frozen HRPC samples (25,26).Hence, we are confident that our precise genome-wide expression profiles of clinical HRPC cells are very valuable.The random permutation test comparing the expression profiles of 13 HRPCs with those of 10 HSPCs identified 36 up-regulated genes and 70 down-regulated genes in HRPCs (Tables 1 and 2).Some of such genes were considered to be associated with their androgen-independent growth and more aggressive phenotypes of clinical HRPCs.Among the 36 upregulated genes in HRPCs, at first, we focused on AR overexpression.In spite of AR transactivation of mRNA in HRPC cells, the amount of stabilized AR protein in the nucleus and AR activity measured by the transcriptional levels of its downstream target genes (PSA and NLX3.1) in HRPC cells were similar to those in HSPC and normal prostate epithelial cells.Several reports suggested that even under low level of circulating testicular androgen, HRPCs still maintain some level of dependency to the AR pathway (10,11,26,27) and our data also support this concept.However, of course, the retention of AR activity itself does not explain the more aggressive phenotype of clinical HRPCs, and apparently, the non-AR pathways should contribute to this clinical HRPC phenotype.
The list of up-regulated genes in HRPC (Table 1) included ANLN and SNRPE as well as AR.ANLN interacts with and activated RhoA and that this complex is likely to be essential for the growthpromoting pathway and aggressive features of lung cancers through phosphatidylinositol 3-kinase/Akt signaling (28), indicating that its overexpression in HRPCs can be involved with aggressive phenotype of clinical HRPCs.SNRPE may be involved with RNA splicing, but its function is unknown.Our siRNA experiments showed that overexpression of ANLN and SNRPE could play some important roles in prostate cancer cell viability and aggressive phenotype of HRPCs.
The list of the down-regulated genes in HRPC (Table 2) included NR4A1, CYP27A1, and HLA-A antigen.NR4A1 belongs to the steroid nuclear hormone receptor superfamily and its expression can cause apoptosis (29).NR4A1 expression is regulated by LH (30) and its down-regulation in HRPCs can reflect LH depletion in the patients under the treatment of LH-RH antagonist.CYP27A1 catalyses hydroxylations in the bioactivation of vitamin D3 (31).Epidemiologic evidence suggests an inverse relationship between prostate cancer and serum vitamin D levels (32), and active vitamin D3 inhibits growth and invasion of human prostate cancer cells (31).Downregulation of NR4A1 and CYP27A1 can provide HRPC cells with some advantages for their survival and growth.Notably, HLA-A antigen, one of the MHC molecules, and many other HLA antigens (which were not listed in Table 2 because of their P value of 0.001-0.0001)were significantly down-regulated in clinical HRPCs, implicating that HRPC cells could acquire immunotolerance (33).
We attempted to identify the genes that were differentially expressed between HRPCs in metastatic site and those in the primary site (prostate).Because prostate cancer can preferentially metastasize to bone, several reports (34,35) indicated that the microenvironment in bone marrow could promote prostate cancer growth and change their phenotype more aggressive.In comparing the gene expression patterns between metastatic tumors and primary tumors, it is critical to exclude the cells of the host organs of the metastatic tumors, and the expression profiles of bone metastasis of prostate cancer was very vulnerable

Cancer Research
Cancer Res 2007; 67: (11).June 1, 2007 to contamination of bone marrow cells (26).In our study, the expression profiles of the microdissected cancer cells in bone metastasis was expected to reflect such inferences with the microenvironment in bone marrow and we did the supervised analysis using the expression profiles of 8 HRPC cells at bone metastatic lesions and 10 HRPC cells at the prostate.However, our analysis unexpectedly failed to distinguish them, and our supervised analysis did not clearly separate HRPC cells at the bone metastasis from those at the prostate.Taken together with the findings from the unsupervised hierarchical clustering analysis, our data indicated that the differences in expression patterns among the multiple metastatic loci derived from the individual patients were much smaller than the interindividual differences in the expression patterns.
In conclusion, our precise microarray analysis of clinical HRPC cells should provide useful information to understand the molecular mechanism of the hormone-refractory and more aggressive phenotypes of clinical HRPC and to identify molecular targets for the treatment of HRPCs.

Figure 1 .
Figure 1.A, dendrogram of an unsupervised hierarchical clustering analysis of 254 genes (vertical columns ) across 35 prostate cancers (horizontal rows ).The unsupervised hierarchical clustering analysis clearly distinguished 25 HRPCs (red) from 10 HSPCs (blue ).Small subsets of cluster constituted by metastatic HRPC cells (-B , bone metastasis; -L, lymph node metastasis; -H, liver metastasis) and HRPC cells at the primary site (-P, prostate) from the same individuals (black boxes ).Tissue procurement methods for each HRPC specimen (A , autopsy; B , biopsy; T , TUR-P).All HSPC tissues were procured by radical prostatectomy.B, dendrogram of a supervised analysis of 106 genes (vertical columns ) across 13 HRPCs at the prostate and 10 HSPCs (horizontal rows ).Each cell in the matrix represents the expression level of a single transcript in a single sample.Red and green, transcript levels, above and below the median for that gene across all samples.Black, unchanged expression; gray, no detectable expression.The 36 up-regulated genes and 70 down-regulated genes that can distinguish HRPC cells from HSPC cells are listed in Tables1 and 2, respectively.

Figure 2 .
Figure 2. Semiquantitative RT-PCR confirmed the elevated expression of 11 genes that could distinguish HRPC cells from HSPC cells (7 HRPCs and 7 HRPCs microdissected from prostrate cancer tissues).ACTB was used to quantify the each of cDNA contents.

Figure 4 .
Figure 4. Knockdown of ANLN and SNRPE by siRNA in prostate cancer cells attenuated their growth and viability.A and C, knockdown effect of siRNA on ANLN and SNRPE in prostate cancer cell line 22Rv1 was evaluated by semiquantitative RT-PCR using cells transfected with each of siRNA-expressing vectors to ANLN (siANLN), SNRPE (si1-3), and a negative control vector (siEGFP).ACTB was used to quantify RNAs.Colony formation assay was assessed on 22Rv1 cells (B and D, left) transfected with each of indicated siRNA-expressing vectors to ANLN (siANLN), SNRPE (si1-si3), and a negative control vector (siEGFP).Cells were visualized with 0.1% crystal violet staining after 14-d incubation with geneticin.MTT assay was done for each of 22Rv1 (B and D, right ) transfected with indicated siRNA-expressing vectors to ANLN (siANLN), SNRPE (si1-si3), or a negative control vector (siEGFP).Columns, average after 14-d incubation with geneticin; bars, SD.ABS, absorbance at 490 nm, and at 630 nm as reference, measured with a microplate reader (Y-axis ).These experiments were carried out in triplicate.*, P < 0.01, Student's t test.

Table 1 .
Up-regulated genes in the progression to HRPC NOTE: Difference: the average fold difference in expression level between HRPCs and HSPCs.Molecular Features of HRPC www.aacrjournals.org5119 Cancer Res 2007; 67:

Table 2 .
Down-regulated genes in the progression to HRPC (Cont'd) Difference: the average fold difference in expression level between HRPCs and HSPCs.