Molecular Alterations in Primary Prostate Cancer After Androgen Ablation Therapy

Abstract

PURPOSE

After an initial response to androgen ablation, most prostate tumors recur, ultimately progressing to highly aggressive androgen independent (AI) cancer. The molecular mechanisms underlying progression are not well known, in part due to the rarity of AI samples from primary and metastatic sites.

EXPERIMENTAL DESIGN

We compared the gene expression profiles of ten AI primary prostate tumor biopsies with ten primary, untreated androgen-dependent (AD) tumors. Samples were laser capture microdissected, the RNA was amplified, and gene expression was assessed using Affymetrix Human Genome U133A Gene Chips. Differential expression was examined with principle component analysis (PCA) and Student t testing. Analysis of gene ontology was performed with Expression Analysis Systematic Explorer (EASE) and gene expression data were integrated with genomic alterations with DIfferential Gene locus MAPping (DIGMAP).

RESULTS

Unsupervised PCA showed that the AD and AI tumors segregated from one another. After filtering the data, 239 differentially expressed genes were identified. Two main gene ontologies were found discordant between AI and AD tumors: macromolecule biosynthesis was down-regulated and cell adhesion up-regulated in AI tumors. Other differentially expressed genes were related to IL-6 signaling, as well as angiogenesis, cell adhesion, apoptosis, oxidative stress, and hormone response. The DIGMAP analysis identified nine regions of potential chromosomal deletion in the AI tumors including 1p36, 3p21, 6p21, 8p21, 11p15, 11q12, 12q23, 16q12, and 16q21.

CONCLUSIONS

Taken together, these data identify several unique characteristics of AI prostate cancer that may hold potential for the development of targeted therapeutic intervention.

INTRODUCTION

Carcinoma of the prostate accounts for approximately one third of all cancers diagnosed in men in the United States and remains the second most common cause of cancer death in this group (ACS Cancer Facts and Figures 2004, http://www.cancer.org/docroot/STT/stt_0.asp). The survival and growth of prostate cancer cells is initially dependent on the presence of androgens, and virtually all prostate cancer patients respond when first treated with androgen ablation. However, resistance to hormone blockade ultimately results in the recurrence of highly aggressive and metastatic prostate cancer that is androgen-independent (1). Androgen-independent (AI) prostate cancer is therefore clinically defined as the progression of the disease under hormonal ablation.

Although several hypothesized mechanisms exist for the development of AI prostate cancer (reviewed in 2), and recent studies using prostate cancer models have shed additional light on the process (3), our understanding of the disease in patients remains incomplete at the molecular level, and the key genes involved are still largely unknown. Expression analysis of prostate cancer before and after hormone therapy may identify genes and pathways that are critical to its progression. Over the past few years, numerous studies have been published on the molecular profiles of human prostate cancer tissue. Several of these have included metastatic lesions (4–9); however, few have analyzed AI tumor cells from the site of the primary lesion. As an example, Holzbeierlein and colleagues compared the gene expression profiles of normal prostate, primary tumors before and during hormone therapy, and metastatic tumors, three of which were androgen-independent (8). Although the AI tumor group was small, they identified a pattern of gene expression, independent of treatment and metastatic status, unique to AI prostate cancer. Studies that delineate the molecular profile of AI tumors may be uniquely valuable in designing therapeutic interventions.

In the present study, we directly compared the gene expression profiles of ten AI tumor biopsies, taken from the primary site of prostate cancer, with ten primary, untreated prostate tumors. Each sample was microdissected to eliminate gene expression changes that could derive from cell types other than tumor. Expression patterns were evaluated with respect to metabolic pathways, gene ontologies and genomic alterations.

MATERIALS AND METHODS

Tissue Specimens

Androgen-dependent (AD) prostate carcinoma specimens were obtained from patients undergoing prostatectomy as first-line therapy at either Catholic University in Santiago, Chile (cases 1–9), or the University of North Carolina (case 10). The tumors were excised and one section was frozen while the remainder was processed for diagnosis. A total of 16 frozen cases were collected, anonymized, and transferred to the National Cancer Institute. Ten of the samples had sufficient tumor for inclusion in the study. The tumors were evaluated by two pathologists (J.W.G. and R.F.C.) and assigned Gleason scores of 5 (n=2), 6 (n=4), 7 (n=1), 8 (n=2), and 9 (n=1). For the comparison group of androgen-independent (AI) prostate cancer, we retrospectively obtained baseline, primary-site biopsies from patients who had participated in an IRB-approved National Cancer Institute Phase II trial studying docetaxel and thalidomide in metastatic androgen-independent prostate cancer. We examined a total of 82 snap-frozen biopsies from 30 cases, of which 23 cases contained cancer. We then selected the ten cases with the highest amount of tumor and RNA quality. All patients from whom the AI samples were obtained met at least one of the following parameters for clinically progressing, AI disease: two consecutively rising PSA levels (PSA ≥ 5.0), at least one new lesion on bone scan, or progressive measurable disease. In addition, in the absence of surgical castration, a serum testosterone of under 50 ng/ml and continuance on gonadotropin-releasing hormone (GnRH) antagonist was required. Clinical details, including treatment history prior to AI disease, for each of the AI cases are provided in Table 1.

Table 1.

Clinical Data for Androgen-independent Prostate Cancer Specimens

Case	Race¥	Age at Diagnosis	Gleason at Diagnosis	Stage at Diagnosis	Time from Diagnosis to Biopsy (years)	Treatment§ History	Radiation Therapy	Survival After Biopsy (years)
AI-1	C	69	7	T2–T3	4	Z, F	Y	0.8
AI-2	C	57	8	D2	6.5	L, F	N	1
AI-3	C	40	8	D2	1.3	L, C	N	1.3
AI-4	AA	69	8	T2–T3	7	L, F	Y	2.8
AI-5	C	73	9	T2–T3	7.2	L, F	Y	0.8
AI-6	C	55	7	T2–T3	12.7	L, F	Y	n/a
AI-7	C	66	8	T2–T3	10.8	O	Y	2.4
AI-8	AA	63	8	T2–T3	7.8	L, O	Y	4.8
AI-9	C	54	7	N1-2, D2	17	O, F	Y	n/a
AI-10	C	62	9	T2–T3	7.9	Z, C	Y	0.5

^¥C = Caucasian, AA = African American

^§Z = Zoladex, F = Flutamide, L = Lupron, C = Casodex, O = orchiectomy

Laser Capture Microdissection and RNA Isolation

Each of the twenty frozen tissue blocks was recut into 8 μm thick sections onto glass slides and stored at −80ºC. Each section was individually removed from storage and immediately stained as follows; 70% ethanol for 15 seconds, de-ionized water for 10 seconds, Mayer’s hemotoxylin (Sigma-Aldrich, St. Louis, MO) for 15 seconds, de-ionized water and bluing solution (Sigma-Aldrich) for 10 seconds each, and then eosin (Sigma-Aldrich) for 5 seconds, followed by dehydration for 10 seconds each in increasing concentrations of ethanol. Finally, the tissue was completely dehydrated in xylenes for 20 seconds. Cells from each case were microdissected by laser capture microdissection (LCM) (10) with the PixCell IIe according to the manufacturer’s protocol (Arcturus Engineering, Inc., Mountain View, CA), and total RNA was isolated with the PicoPure RNA Isolation Kit (Arcturus). The samples were subjected to DNAse treatment for 15 minutes, and RNA quality and quantity were assessed with the Bioanalyzer 2100 (Agilent Technologies, Inc., Palo Alto, CA) and Degradometer software, version 1.2 (http://www.dnaarrays.org) (11).

RNA Amplification, Microarray Sample Synthesis, and Hybridization

RNA was amplified by modifying a previously established protocol that combines the RiboAmp (Arcturus) and Affymetrix (Affymetrix, Inc., Santa Clara, CA) systems (12), resulting in biotin-labeled, antisense cRNA. Total RNA was used for amplification, since this approach has been shown to introduce less bias than when mRNA is used (13). One to ten nanograms of total RNA from each sample were subjected to two rounds of linear amplification with the RiboAmp HS Kit (Arcturus). Two micrograms of antisense RNA (aRNA) from the second round of amplification were then used to synthesize double-stranded cDNA with the regular RiboAmp kit (Arcturus), since the RiboAmp HS kit has a maximum RNA input capacity of only 250 nanograms. The resulting cDNA was then used as template for the synthesis of antisense cRNA labeled with biotinylated UTP and CTP by in vitro transcription using the BioArray High Yield RNA Transcript Labeling Kit (Enzo Life Sciences, Inc., Farmingdale, NY). In total, each sample underwent three rounds of amplification, which has previously been shown to decrease the average distribution size of aRNA without affecting reproducibility on oligonucleotide arrays (14). Fifteen micrograms of each labeled sample were then fragmented according to protocol (Affymetrix) and hybridized to Human Genome U133A GeneChip arrays for 16 hours. Microarrays were washed and stained using the “EuKGE-WS2v4” protocol and then scanned using the Affymetrix GeneChip Scanner 3000. The raw microarray data were uploaded to the Gene Expression Omnibus (GEO) public repository (www.ncbi.nlm.nih.gov/geo/), GEO series number GSE2443.

Data Filtering and Normalization, Clustering and Statistical Analysis

Expression profile data were prepared for analysis using Microarray Analysis Suite v5.0 software (MAS 5.0, Affymetrix), setting the scaling of all probe sets on all chips to a constant value of 1000. The data were then filtered to include only those probe sets having “present” or “marginal” calls (detection p-value < 0.065) in at least 10% of the samples. The global expression patterns were studied by principal component analysis (PCA) considering one dimension for each gene on the array (Partekpro 5.0 software, Partek Inc., St. Charles, Missouri). PCA determines a set of principal components as linear combinations of original dimensions such that the first principal component (PC) is in the direction of highest variance of the distribution, the next PC is in the direction of highest of remaining variance and so on (15). Eigen analysis of correlation matrix was used. A projection on the first three PC’s covering highest variance permits dimension reduction of multidimensional data for graphical visualization. In this 3D plot, a point represents a tissue sample whereas the close clustering of points indicates similar gene expression patterns.

Analysis of Gene Ontology Representation

Genes showing significant differential expression were categorized by their ontologies using Expression Analysis Systematic Explorer (EASE) software (16). The number of genes assigned to each ontology term was compared to the total population on the microarray to identify the probability of over-representation of each ontology. Over-representation analysis calculated for each of the gene ontology terms provided an EASE score, which was the upper boundary of the distribution of Jackknife Fisher exact probabilities. The gene ontologies having an EASE score less than 0.05 were considered significantly over-represented.

Identification of Potential Chromosomal Deletion Regions

The normalized and filtered data set was subjected to DIfferential Gene locus MAPping (DIGMAP) analysis as previously described (17). Briefly, the gene locations were first mapped through the Gene Annotation Project database. UniGene clusters and their genomic locations in the data set were annotated, and information from this step was used to generate a DIGMAP source file that was used in the subsequent analysis. Next, a viewer program (DIGMAPviewer) read the DIGMAP source file, and DIGMAP partitioned the microarray data into subsets by chromosome number and sub-chromosomal locations. A graphical presentation was generated using a heat map to represent each data point with a colored cell that quantitatively reflected the original differential expression value. Genomic regions exhibiting differential gene expression were marked as Differential Flag Regions (DFRs) by visual inspection of the graphical displays.

RESULTS

Technical Parameters

To assess the reliability and reproducibility of the protocol, each sample was evaluated following each step (Table 2). The amount of dissection was similar for most samples; however, the needle biopsies provided less RNA than the whole tumors, an average of 8.7 versus 38.6 nanograms respectively. This may have been due to a lower density of cells per shot in the biopsies compared to the whole tumors. The Bioanalyzer electropherograms for all samples in the study showed dominant 18S and 28S ribosomal peaks and no obvious degradation peaks. Two recent studies showed that consistency in RNA quality, rather than undiminished integrity, is the critical determinant for avoiding artifactual differential expression (11, 18). However, these studies also showed that the 28S/18S ratio (provided by the Bioanalyzer software) is not a reliable indicator of RNA quality. Thus, we chose to additionally employ a quantitative measurement of RNA degradation to ensure that the RNA from the two tumor groups was of similar quality. The Degradometer software (11) utilizes the ratio of the average value of all degradation peak signals to the 18S peak signal multiplied by 100 to calculate an objective, quantitative degradation factor, and showed no significant difference between the two sample groups (an average of 20.0 for the biopsies and 21.8 for the whole tumors), although six of the needle biopsies had insufficient concentrations for the calculation of a degradation factor.

Table 2.

Microdissection, RNA Yield, RNA Amplification, and Microarray Performance

Case	LCM Shots	RNA Quality Assessment*	RNA Quantity (ng)	Round 1 Template (ng)	Round 1 aRNA (ng)	Round 2 aRNA (ug)	Round 3 Template (ug)	Final cRNA (ug)	Background Signal	Noise	Mean P Call Signal	Scale Factor	GAPDH 3′/5′	Actin 3′/5′	% Present
AD Tumors:
AD-1	3100	67	34.5	10.0	61	104	2	44	50.7	2.7	2306.7	13.9	6.5	22.2	29.1
AD-2	3000	29	18.1	10.0	72	87	2	37	60.3	3.6	1871.6	8.1	7.1	53.3	33.4
AD-3	3100	17	77.8	10.0	61	105	2	43	46.7	2.8	2318.4	12.2	9.0	37.6	29.5
AD-4	3100	18	32.1	10.0	208	77	2	51	45.1	2.5	2690.2	17.8	7.8	23.5	28.8
AD-5	3000	15	26.4	10.0	70	65	2	81	50.2	2.7	2258.5	14.1	4.6	25.7	29.1
AD-6	3100	15	18.8	10.0	536	77	2	45	56.1	3.4	1940.9	6.1	5.8	34.8	36.5
AD-7	3000	14	48.1	10.0	89	103	2	71	58.4	3.2	2440.6	12.2	10.0	25.1	32.5
AD-8	3100	17	52.2	10.0	55	85	2	94	54.7	3.1	2299.6	11.3	9.1	75.0	31.8
AD-9	3200	21	28.2	10.0	35	64	2	104	131.7	10.1	2353.8	7.7	8.0	15.4	29.2
AD-10	3000	5	50.0	10.0	64	79	2	67	68.7	4.4	2203.9	6.6	7.0	8.9	33.1
AD Average		21.8	38.6	10.0	125.1	84.6	2	63.7	62.3	3.9	2268.4	11.0	7.5	32.2	31.3
AI Biopsies:
AI-1	3100	n/a	2.6	2.6	12	24	2	48	49.0	2.7	2135.3	10.5	14.7	10.7	31.4
AI-2	3050	n/a	14.0	10.0	44	39	2	52	56.5	3.8	1756.6	5.2	10.5	22.2	38.1
AI-3	3150	n/a	12.8	10.0	80	62	2	51	52.1	3.4	1984.8	8.7	6.2	29.6	33.6
AI-4	3100	n/a	5.3	5.3	27	88	2	57	62.7	4.3	1889.3	6.1	8.6	20.5	34.1
AI-5	1400	n/a	1.0	1.0	10	41	2	51	61.2	3.5	2060.1	9.2	8.5	17.8	32.8
AI-6	3000	n/a	13.0	10.0	58	73	2	43	47.9	2.6	2064.1	17.9	6.0	76.9	26.7
AI-7	3200	17	15.0	10.0	31	33	2	70	46.9	2.9	1917.1	5.3	10.9	30.9	36.3
AI-8	3050	12	9.3	9.3	242	44	2	76	50.7	2.9	1970.5	9.1	6.5	10.1	32.9
AI-9	2200	9	5.9	5.9	39	81	2	57	61.1	3.5	2261.7	9.2	9.6	18.5	31.5
AI-10	3050	42	7.8	7.8	57	79	2	59	149.3	12.1	2086.5	7.6	14.0	31.8	26.4
AI Average		20.0	8.7	7.2	60.0	56.4	2	56.4	63.7	4.2	2012.6	8.9	9.6	26.9	32.4

^*Based on Degradometer calculations, lower values indicate higher quality, i.e. less degradation. Samples with RNA concentrations below the level needed to employ the degradometer were designated “n/a” for none available, although the Bioanalyzer electropherograms for these samples were similar to the other biopsies.

The technical parameters for microarray hybridization showed high consistency amongst all samples with respect to background and noise, with all values well within the manufacturer’s recommended maximums of 200 and 5, respectively. Further, the scale factors, which can indicate skewing of the data between groups, potentially introducing error into differential expression comparisons, showed no statistically significant difference. The manufacturer’s recommended maximum value for the 3′/5′ ratios for housekeeping genes GAPDH and β-actin is 3.0, which typically assumes high quality, unamplified RNA. The protocol employed here resulted in much higher ratios averaging 7.5 and 9.6 for GAPDH and 32.2 and 26.9 for β-actin. Similar observations have been reported in the literature by others using amplified RNA (12), as each round of amplification shortens RNA transcript lengths, eventually resulting in the loss of 5′ regions and increasing the ratio of signal between the 3′ and 5′ probe sets. The percent of probe sets called present was not statistically different between the two tumor groups and was within the manufacturer’s recommended range expected for human tissue.

Finally, we examined the possibility that gene expression differences could result from the genetic variation between the patients, since the samples were derived from patients in two separate countries (i.e. the United States and Chile). One of the samples in the AD group originated within the U.S. (AD-10) while the rest derived from Chile, yet this tumor clustered well within the AD tumor group when analyzed by hierarchical clustering (data not shown), indicating no obvious difference on this account.

Differential Gene Expression Between Tumor Groups

Unsupervised principle component analysis based on the largest three principal components revealed separate clustering of the AD and AI tumor groups along PC #2 (9.35% variance) as shown in Figure 1. This indicates the presence of a large number of genes (about 1000 if the variance is equal for all genes) distinguishing the two tumor groups. In general, the AD tumors clustered more tightly together while the AI tumors, although predominantly separate from the AD tumors, clustered more loosely. This likely represents the relative clinical similarity of the tumors within each group, with the AD tumors being from newly diagnosed, untreated patients, and the AI tumors being from patients having undergone various treatments (i.e. radiation, orchiectomy, flutamide, or combinations) while progressing to AI disease status. We found no correlation between treatment histories and gene expression profiles in this study.

FIGURE 1.

Principal component analysis of AD (red) and AI (green) prostate cancer. The probe sets were filtered to include only the 11,663 transcripts detected in at least 10% of all the samples. The projection on three principal components of greatest variation covering 34.7% of the total variance is shown.

To identify the specific genes that were differentially expressed between the two tumor groups, the normalized data were filtered to remove all probe sets not called present in at least 20% of the samples, which resulted in 10,041 probe sets remaining. Two-sample Student-t testing identified 256 probe sets showing differential expression at p<0.005 between the two groups (please see the supplemental material for the unabridged list). These probe sets represented 239 genes, as several genes were identified by multiple probe sets. Approximately 61.3% of the genes showed down-regulation in the AI tumor cells, while 38.7% showed up-regulation. Many of the genes are involved in processes of carcinogenesis such as angiogenesis, cell adhesion and the microenvironment, cell death including apoptosis, hormone response, oxidative stress and cancer cell metabolism, key signaling pathways, and metastasis. A list of selected differentially expressed genes is presented in Table 3.

Table 3.

Selected^** Differentially Expressed Genes Between AD and AI Tumor Groups

Gene Symbol	Probe Set	Fold Change AI/AD	Parametric p-value	Description	Map	Significant Reference
Angiogenesis:
LMO2	204249_s_at	1.67	0.0011	LIM domain only 2 (rhombotin-like 1)	11p13	(49)
VWF	202112_at	4.63	0.0014	von Willebrand factor	12p13.3	(50)
PECAM1	208982_at	2.62	0.0001	platelet/endothelial cell adhesion molecule (CD31)	17q23	(50)
THSP1	201108_s_at	2.18	0.0038	thrombospondin 1	15q15	(51)
EDG4	206723_s_at	0.57	0.0043	endothelial differentiation, lysophosphatidic acid G-protein-coupled receptor, 4	19p12	(52)
SDC2	212158_at	3.13	0.0008	syndecan 2 (heparan sulfate proteoglycan 1, cell surface-associated, fibroglycan)	8q22–q23	(53)
TEK	206702_at	2.82	0.0015	TEK tyrosine kinase, endothelial	9p21	(25)
Cell adhesion:
BPAG1	212254_s_at	2.78	0.0001	bullous pemphigoid antigen 1, 230/240kDa	6p12–p11	(39, 54)
CDH11	207173_x_at	3.10	0.0004	cadherin 11, type 2, OB-cadherin (osteoblast)	16q22.1	(38)
FN1	211719_x_at	2.72	0.0007	fibronectin 1	2q34	(55)
Apoptosis/Cell death:
TRAF5	204352_at	1.85	0.0019	TNF receptor-associated factor 5	1q32	(56)
GRIM19	220864_s_at	0.21	0.0020	cell death-regulatory protein GRIM19	19p13.2	(57)
NMP200	203103_s_at	0.45	0.0001	nuclear matrix protein NMP200 related to splicing factor PRP19	11q12.2	(58)
MCL1	200797_s_at	0.48	0.0005	myeloid cell leukemia sequence 1 (BCL2-related)	1q21	(59, 60)
GADD45B	207574_s_at	0.26	0.0031	growth arrest and DNA-damage-inducible, beta	19p13.3	(61)
GADD45G	204121_at	0.23	0.0001	growth arrest and DNA-damage-inducible, gamma	9q22.1–q22.2	(61)
Hormone response:
REA	201600_at	0.67	0.0042	repressor of estrogen receptor activity	12p13	(62)
KLK2	210339_s_at	0.53	0.0011	kallikrein 2, prostatic	19q13.41	(63)
KLK3	204582_s_at	0.29	0.0006	kallikrein 3, (prostate specific antigen)	19q13.41	(63)
GREB1	205862_at	0.18	0.0019	GREB1 protein	2p25.1	(64)
Oxidative stress:
COX8	201119_s_at	0.57	0.0024	cytochrome c oxidase subunit VIII	11q12–q13	(65)
COX7C	217491_x_at	0.51	0.0041	cytochrome c oxidase subunit VIIc	5q14	(65)
SOD2	215078_at	0.11	0.0005	superoxide dismutase 2, mitochondrial	6q25.3	(36, 37)
Metastasis:
EIF4EL3	213571_s_at	0.73	0.0027	eukaryotic translation initiation factor 4E-like 3	2q37.1	(66)
Prostate cancer-associated:
NOV	214321_at	3.80	0.0016	nephroblastoma overexpressed gene	8q24.1	(46)
MIF	217871_s_at	0.58	0.0037	macrophage migration inhibitory factor (glycosylation-inhibiting factor)	22q11.23	(47)
TACSTD2	202286_s_at	0.51	0.0049	tumor-associated calcium signal transducer 2	1p32–p31	(29)
STAT5B	212550_at	0.71	0.0001	signal transducer and activator of transcription 5B	17q11.2	(29)

^**For genes where multiple probe sets were identified as differentially expressed, only one probe set was included in the table. Only genes previously identified in the cancer literature are included in the table. Most genes relating to the over-represented genes ontologies, including genes relating to IL-6, are presented only in Table 4.

Identification of Overly Represented Gene Ontologies

The list of 256 probe sets representing the genes showing significant differential expression was analyzed for over-representation of specific gene ontologies using the EASE software. Two major biological groups were identified (Table 4, top). The first group included genes that are involved in ribonucleoprotein complexes, are structural constituents of ribosomes, or are otherwise involved in protein biosynthesis. The genes in this group predominantly (20 of 31) showed lower expression in androgen-independent prostate cancer (AIPC). The second group included genes involved in the extracellular matrix and cell adhesion molecule activity and showed predominantly (15 of 19) increased expression. In addition, by a review of literature for the differentially expressed genes, we found the signaling pathway of the proinflammatory cytokine interleukin-6 (IL-6) and genes whose expression related to IL-6 to be over-represented (Table 4, bottom).

Table 4.

Over-represented Ontologies of Genes Differentially Expressed in AIPC

Gene Symbol	Fold Change	Description	Parametric p-value	Probe set	Gene Ontology†or Relationship to IL-6
ST3GALVI	3.44	alpha2,3-sialyltransferase	0.0045	213355_at	PB, MB
LUC7A	2.75	cisplatin resistance-associated overexpressed protein	0.0007	220044_x_at	RC
SF3B1	2.04	splicing factor 3b, subunit 1, 155kDa	0.0013	201071_x_at	RC
KIAA0970	1.92	KIAA0970 protein	0.0031	202304_at	RC, SCR
PNAS-4	1.92	CGI-146 protein	0.0003	212371_at	RC, R
SFRS11	1.91	splicing factor, arginine/serine-rich 11	0.0035	200685_at	RC
HNRPH3	1.73	heterogeneous nuclear ribonucleoprotein H3 (2H9)	0.0011	208990_s_at	RC
PTMA	1.59	prothymosin, alpha (gene sequence 28)	0.0017	200773_x_at	RC, SCR
UBE3A	1.57	ubiquitin protein ligase E3A (human papilloma virus E6-associated protein, Angelman syndrome)	0.0014	211575_s_at	PB
SFRS7	1.56	splicing factor, arginine/serine-rich 7, 35kDa	0.0044	201129_at	RC
FLJ10283	1.51	hypothetical protein FLJ10283	0.0022	218534_s_at	RC
EIF4EL3	0.73	eukaryotic translation initiation factor 4E-like 3	0.0027	213571_s_at	PB
NMT2	0.72	N-myristoyltransferase 2	0.0043	215069_at	PB
COPS6	0.72	COP9 subunit 6 (MOV34 homolog, 34 kD)	0.0032	213504_at	PB
KIAA0759	0.71	KIAA0759 protein	0.0022	36865_at	PB, RC
MRP63	0.66	mitochondrial ribosomal protein 63	0.0039	221995_s_at	SCR
MRPL20	0.65	mitochondrial ribosomal protein L20	0.0002	220526_s_at	RC, SCR, PB
HNRPA0	0.65	heterogeneous nuclear ribonucleoprotein A0	0.0007	201055_s_at	RC
EIF3S9	0.65	eukaryotic translation initiation factor 3, subunit 9 eta, 116kDa	0.0001	203462_x_at	PB
RPL34	0.65	ribosomal protein L34	0.0010	200026_at	RC, SCR, PB
RPL39	0.63	ribosomal protein L39	0.0015	208695_s_at	RC, SCR, PB
RPL36	0.60	ribosomal protein L36	0.0045	219762_s_at	RC. SCR, PB
RPS21	0.55	ribosomal protein S21	0.0028	200834_s_at	RC, SCR, PB
RPS14	0.53	ribosomal protein S14	0.0006	208646_at	RC, SCR, PB
RPL35A	0.53	ribosomal protein L35a	0.0029	213687_s_at	RC, SCR, PB
MRP63	0.51	mitochondrial ribosomal protein 63	0.0009	204386_s_at	SCR
	0.50	similar to 40S ribosomal protein S18	0.0023	201049_s_at	RC, SCR, PB
RPS16	0.45	ribosomal protein S16	0.0046	213890_x_at	RC, SCR, PB
RPS29	0.40	ribosomal protein S29	0.0018	201094_at	RC, SCR, PB
GADD45B	0.28	growth arrest and DNA-damage-inducible, beta	0.0031	207574_s_at	RC, SCR, PB
GADD45G	0.23	growth arrest and DNA-damage-inducible, gamma	0.0001	204121_at	RC, SCR, PB
VWF	4.63	von Willebrand factor	0.0014	202112_at	ECM, CAMA
CSPG2	3.10	chondroitin sulfate proteoglycan 2 (versican)	0.0010	204619_s_at	ECM
CDH11	3.10	cadherin 11, type 2, OB-cadherin (osteoblast)	0.0004	207173_x_at	CAMA
LAMB1	3.01	laminin, beta 1	0.0047	201505_at	ECM, CAMA
ASPN	2.87	asporin (LRR class 1)	0.0017	219087_at	ECM
FN1	2.72	fibronectin 1	0.0007	211719_x_at	ECM, CAMA
COL15A1	2.67	collagen, type XV, alpha 1	0.0006	203477_at	ECM, CAMA
PECAM1	2.62	platelet/endothelial cell adhesion molecule (CD31)	0.0001	208982_at	CAMA
FLJ20736	2.46	hypothetical protein FLJ20736	0.0022	218244_at	ECM
COL5A2	2.30	collagen, type V, alpha 2	0.0043	221729_at	ECM
SPARC	2.29	secreted protein, acidic, cysteine-rich (osteonectin)	0.0045	212667_at	ECM
ECM2	2.25	extracellular matrix protein 2, female organ and adipocyte specific	0.0040	206101_at	ECM
THBS1	2.18	thrombospondin 1	0.0038	201108_s_at	ECM, CAMA
LAMA4	2.13	laminin, alpha 4	0.0040	202202_s_at	ECM, CAMA
LTBP2	1.61	latent transforming growth factor beta binding protein 2	0.0050	204682_at	ECM
GNE	0.67	UDP-N-acetylglucosamine-2-epimerase/N- acetylmannosamine kinase	0.0027	205042_at	CAMA
ICAM3	0.60	intercellular adhesion molecule 3	0.0043	204949_at	CAMA
CD84	0.58	CD84 antigen (leukocyte antigen)	0.0032	211189_x_at	CAMA
BAIAP2	0.40	BAI1-associated protein 2	0.0012	209502_s_at	CAMA
IL6	0.62	interleukin 6 (interferon, beta 2)	0.0034	205207_at
JAK1	2.30	Janus kinase 1 (a protein tyrosine kinase)	0.0009	201648_at	Signal transduction (27)
STAT5B	0.71	signal transducer and activator of transcription 5B	0.0001	212550_at	Signal transduction (27)
JUNB	0.45	jun B proto-oncogene	0.0035	201473_at	Gene expression (31)
JUND	0.39	jun D proto-oncogene	0.0002	203752_s_at	Gene expression (31)
MCL1	0.48	myeloid cell leukemia sequence 1 (BCL2-related)	0.0005	200797_s_at	Gene expression (30)
SDC2	3.13	syndecan 2 (heparan sulfate proteoglycan 1, cell surface-associated, fibroglycan)	0.0008	212158_at	Gene expression (32)
LUC7A	2.75	cisplatin resistance-associated overexpressed protein	0.0007	220044_x_at	Chemosensitivity (67)
KLK2	0.53	kallikrein 2, prostatic	0.0011	210339_s_at	AR activation (33)
KLK3	0.29	kallikrein 3, (prostate specific antigen)	0.0006	204582_s_at	AR activation (68)
GADD45B	0.26	growth arrest and DNA-damage-inducible, beta	0.0031	207574_s_at	AR activation (33)
GADD45G	0.23	growth arrest and DNA-damage-inducible, gamma	0.0001	204121_at	AR activation (33)

^†RC = ribonucleoprotein complex, PB = protein biosynthesis, MB = macromolecule biosynthesis, R = ribosome, SCR = structural constituent of ribosome, ECM = extracellular matrix; CAMA = cell adhesion molecule activity

Identification of Potential Chromosomal Deletion Regions

A total number of 7,002 distinct UniGene clusters and their genomic locations were annotated from the data set. A graphical presentation was generated using a heat map to show the quantitative differential expression for each probe set at its chromosomal location. The twenty samples were clustered according to similarity in differential expression patterns. Five of the AI tumors (AI-1, 2, 5, 6, and 7) and six of the AD tumors (AD-1, 3, 4, 5, 7, and 8) were included for subsequent analysis due to the similar expression pattern within the samples of each group, and because chromosomal deletions typically occur in only a subset of any given tumor type. Based on these remaining samples, nine DFRs showed concordant down regulation in the AI samples (Table 5), representing regions of potential chromosomal deletion. To estimate the known significance of each region in prostate cancer, literature searches using PubMed (http://www.ncbi.nlm.nih.gov/entrez/query.fcgi) were performed to identify the prevalence of each region in the cancer research literature and those specific to prostate. Regions 1p36, 8p21, and 16q21 demonstrated the highest degree of known significance in prostate cancer. The 16q21–16q24.3 region typified the DFRs in this study and is presented in Figure 2.

Table 5.

Chromosomal Regions of Down-regulation in AIPC

Chromosome Region	Number of Genes	Citations in Prostate	Citations in All Cancers	Prostate/All Cancers (%)
1p36.1–1p36.33	26	18	56	32
3p21.2–3p21.31	81	4	443	1
6p21.31–6p21.33	160	2	188	1
8p21.2–8p23.3	221	92	356	26
11p15.3–11p15.5	247	4	460	1
11q12.3–11q13.5	350	3	51	6
12q23–12q24.31	9	2	56	4
16q12.1–16q13	50	3	52	6
16q21–16q24.3	333	46	354	13

FIGURE 2.

Chromosomal view of differential gene expression in AI and AD prostate cancer. Microarray data from 5 AI samples and 6 AD samples are displayed in columns. Rows represent ordered mapped chromosome locations derived from part of chromosome 16 (16q21–16q24.3 or 56~89 Mb). Fluorescence ratios were calculated at a specific gene level across all samples (tumor/tumor sample mean) and plotted on a log₂ scale. The red color represents an expression level above the mean expression of a gene across all samples; the black color represents mean expression, and the green color represents expression lower than the mean.

The gene lists for each region were compared with the list of 239 genes identified in the initial analysis of differential expression. Each region contained one or more genes that were identified by both analyses as follows: 3p, 1 gene; 4q, 1 gene; 6p, 2 genes; 8p, 1 gene; 11p, 3 genes; 11q, 2 genes; and 16q, 8 genes. The common genes in region 16q21–24.3 are indicated in Figure 2.

DISCUSSION

In the present study, the gene expression profiles of newly diagnosed, androgen-dependent primary prostate tumors were compared with those of prostate biopsies from patients who progressed to develop metastases and were treated with hormone ablation therapy. This latter group of clinical specimens represents a unique and precious resource, as very few patients undergo surgical procedures after establishment of advanced disease. The study examined one step in the overall progression of prostate cancer, specifically the effect of androgen ablation therapy on primary tumor cells. This differs from an analysis of metastatic, rapidly growing tumor cells, and it is important to consider primary AI tumor cells and metastatic AI tumors as separate entities with regards to hormone therapy. Both are AI; however, the primary lesions grow slowly and “persist” without androgens, whereas the metastatic lesions grow rapidly and significantly expand the tumor burden of patients. This distinction is consequential as the AI primary tumor expression data set may reveal molecular changes more closely associated with “effective androgen ablation therapy,” as opposed to those related with treatment failure and subsequent clinical breakthrough.

The expression array analysis generated a large amount of interesting data, and identified many individual genes that were differentially regulated between AI and AD primary-site prostate tumor cells. These included genes involved in angiogenesis, apoptosis, oxidative stress, and hormone response. All of the differentially expressed genes are of potential interest for follow-up studies, and some have previously been identified to have potential as clinical biomarkers. However, to better understand the functional themes related to androgen withdrawal, we searched for patterns of gene expression using gene ontology analysis. Two main groups that differed between AI and AD tumors were identified: those associated with ribosomes and protein synthesis and those associated with cell adhesion and the extracellular matrix (ECM). Although aggressive tumors are generally expected to have higher expression and activity of the protein synthesis machinery, we found the converse, with the majority of these genes showing lower expression in AI cells. The second dominant ontological group, genes associated with cell adhesion and ECM, showed a nearly uniform increased expression in the AI cells. This was also unexpected, since the literature shows a mixture of up-and down-regulation of ECM and adhesion molecules during prostate cancer progression. Thus, androgen blockade initially appears to act in clinical prostate samples, at least in part, by facilitating a more normal phenotype through the reversal of two critical cancer-related activities, increased protein synthesis (19) and decreased adhesion. Review of the literature indicates that this is not without precedent. In a recent study, Patriarca and colleagues described up-regulation of E-cadherin and α/β-catenin in prostate tumors after hormonal ablation, and suggested that a more differentiated phenotype results after the treatment (20). It has also been recently suggested that telomerase expression patterns are reverted towards a normal phenotype after hormone ablation, particularly in high-grade tumors (21). Moreover, a review of protein expression changes after androgen deprivation therapy showed decreased proliferation markers (22), which appears to agree with our findings of a generalized decrease in protein synthesis.

It is important to note, however, that several of the up-regulated adhesion and ECM genes are associated with endothelial cells, namely VWF, PECAM1 (CD31), COL5A2, COL15A1, LAMB1, FN1, THBS1, and SPARC (23–25). In addition, other genes that we found to be up-regulated in AIPC, including JAK1, cadherin 11, and TIE2/TEK, have previously been found to be over-expressed in microvascular endothelial cells (24) or endothelial morphogenesis (23, 25). The origin of this gene expression is puzzling, since endothelial cells were excluded during microdissection, and since tumors treated with androgen deprivation have been shown to display decreased microvessel density (22). Prostate tumor cells may participate in vasculogenic mimicry, whereby tumor cells themselves express endothelial-associated markers and form vasculogenic networks both in vitro and in vivo (26), which could account for the higher expression of these genes in AIPC.

Gene expression related to IL-6 and its signaling pathway was also a central theme represented in the data. There is an increasing body of evidence suggesting that IL-6 is involved in the progression of prostate cancer (27) and may even have utility as a diagnostic marker for predicting progression (28). IL-6 signaling involves activation of signal transducer and activator of transcription (STAT) proteins by the Janus kinase (JAK). Both JAK1 and STAT5B were differentially regulated in this study, indicating perturbation of this pathway in AIPC. We previously found that STAT5B was down-regulated in high grade AD tumors compared to moderate grade (29). Since STAT5B was even further down-regulated in the AIPC cells, it may have potential as both a marker for progression and a therapeutic target. In addition to IL-6 signaling, the down-regulation of IL-6 is likely related to the differential regulation of several of the other genes including MCL-1 (30), junD and junB (31), and syndecan-2 (32). A potentially critical facet of IL-6 in prostate cancer is its ability to independently activate the androgen receptor (reviewed in 33). It seems logical that the expression of the androgen-responsive genes kallkreins 2 and 3, and GADD45 β and γ, was lower in the AIPC cells in this study, since IL-6 was also decreased. Finally, neuroendocrine differentiation (NED) may be involved in the development of AIPC (28), and IL-6 has been shown to promote NED in prostate cancer cells (34). Since the AIPC tumors showed differential expression of genes in the IL-6 pathway, we examined the data for the neuroendocrine markers synaptophysin and chromogranin A. We found a trend of increased chromogranin A expression (2.4-fold, p<0.07) in the AI tumors, which concurs with others showing more significant increases in NED in AI disease (35).

It is also important to compare the genes identified in this study with those hypothesized to be involved in the potential mechanisms for the development of androgen-independence. Chen et al. recently showed that, in seven pairs of xenograft tumors before and after the development of androgen-independence, only the androgen receptor gene (AR) was differentially expressed. In the study presented here, AR expression was not significantly different between the two groups. However, since the data were analyzed for gene expression changes consistent amongst the tumors in each group, it is possible that increased AR expression was present in a subset of the tumors, as a non-significant trend of AR over-expression was present in the AI group. However, although the model presented by Chen et al. showed that over-expression of AR alone was sufficient for the development of androgen-independence, there are numerous genes identified in the literature to be involved in this process. Feldman and Feldman (2) present a concise review of the potential mechanisms for the development of AIPC, and several of the genes identified in the study presented here may fit these postulated mechanisms. For example, the antioxidant enzyme superoxide dismutase 2 was down-regulated ninefold in the AI tumors, and has been shown to inversely correlate with prostate cancer progression in cell models (36) and in tissue (37), fitting the model of a decrease in protective enzymes that may give rise to an increase in the frequency of mutation. Another potential mechanism for the development of AIPC is the “outlaw pathway,” whereby the androgen receptor is stimulated by non-androgen growth factors, and IL-6, discussed above, fits this model, as do several genes downstream of AR activation, for example kallikreins 2 and 3 and GADD45, which were differentially expressed. In addition, cadherin 11 has previously been shown to be up-regulated in hormone-refractory prostate cancer cell lines (38), and showed threefold higher expression in the AIPC cells in this study. Bullous pemphigoid antigen 1 (BPAG1) also showed increased expression of nearly threefold in the AIPC cells. BPAG1 is a hemi-desmosome protein whose expression becomes up-regulated with the onset of invasive growth (39). Further studies examining the specific roles of the genes identified here in the development or maintenance of the AI phenotype are necessary.

Standard approaches to the analysis of microarray data, including our own as discussed above, cluster genes based on transcriptional profiles and thus overlook gene expression patterns of contiguous chromosomal regions. Using the newly developed DIGMAP approach, we identified nine genomic regions of interest in AIPC. Most of the regions appear from the literature to be frequent deletions in a variety of human cancers, and regions 1p36, 8p21, and 16q21 have additional significance in prostate cancer. These regions are generally hypothesized to include tumor suppressor genes or other genes required for maintenance of a normal or less aggressive phenotype. For example, chromosomal deletions at 16q have been correlated with more malignant grade tumors (40) and with tumors with poor clinical outcomes (41), which agrees with our findings that this region may be increasingly affected during progression to the AI state.

Overall, comparison of the expression data relative to the genome is intriguing. The AI tumor cells show decreased expression of several genes that map to distinct genomic regions, including known hot spots for prostate cancer. Mechanistically, this could occur either via DNA deletions and/or epigenetic phenomenon such as gene promoter methylation. There are two possible implications of this finding. First, even though androgen withdrawal therapy is effective at slowing the progression of prostate cancer clinically, it does not stop the continued progression of expression changes related to genomic alterations. Alternatively, there may be inherent differences in genomic status between patients where the majority will not recur after treatment (the AD group in this study) and those that are clearly aggressive (the AI group). This is an enticing possibility, as it would suggest that, for prognostic purposes, the two patient groups could be stratified based on genome-related expression data.

However, an important caveat is that the AI and AD tumor groups in this study differ in that, in addition to the status of androgen-dependence, one group became clinically aggressive and the other may not. Consequently, the expression differences could be due to this clinical behavior, rather than to hormonal therapy and androgen-independence. There are numerous studies in the prostate cancer literature that compare recurrent tumors to tumors prior to recurrence, without separating out the tumors that would never recur, and this complicates the conclusions that can be drawn. Identifying gene expression profiles that segregate the androgen-dependent tumor that will recur from those that will not is an area of great interest (42) and will make an important contribution to prostate cancer prognostics. Approaches that emphasize the identification of key groups of genes, such as the gene ontology analysis we present here, may shed light on the identification of recurrent tumors and how they respond to therapy.

These data provide a significant contribution to our knowledge of the molecular characteristics of androgen-independent prostate cancer; however, the potential weaknesses of the study warrant discussion. First, since the AIPC biopsy specimens contained relatively few tumor cells and most were completely exhausted in obtaining the RNA for microarray analysis, it was not possible to conduct traditional validation experiments such as quantitative RT-PCR. However, the use of RNA amplification in conjunction with microarray analysis has been repeatedly validated in the literature (43–45), and the protocol we used provided high efficiency and little technical variation between the samples both within and between the tumor groups. In addition, in silico validation showed that several of the genes identified in this study, for example NOV (46) and MIF (47), have been shown previously to be differentially expressed in AI cells or with increasing prostate tumor grade, and the “Significant Reference” in Table 3 frequently cites a report validating the findings in this study. A second concern was that the differentially expressed genes could have resulted from the treatment that the AIPC patients had undergone, rather than being a characteristic of AIPC itself. However, since the AIPC biopsies included in this study derived from patients who had undergone a diverse array of treatment, it was unlikely that treatment alone produced the consistent differential expression we identified. A third concern was that the differentially expressed genes could have resulted from the different processing that biopsies undergo as compared to whole prostatectomies. However, the RNA isolated from all samples was of similar quality. Finally, the ratios of differential expression are modest in comparison to studies using little or no RNA amplification. However, RNA amplification dampens the variation of gene expression (13), which likely reduced the dynamic range of the data while still allowing for the statistically significant separation of gene expression between the two groups.

The various potential biases in the study, although addressed to the best of our ability, remain a source of concern. We were not able to obtain patient-matched AD and AI specimens, and processing of whole prostatectomies and core biopsies, although both frozen, are intrinsically different, as we have studied previously (48). Thus, although the resulting data show significant corroboration in the literature regarding differential expression of individual genes, and the genes could be grouped by gene ontology analysis and regions of potential chromosomal deletion, it is still possible that the results were influenced by bias error. In spite of this, understanding androgen-independence as it develops in patients is of critical importance, and must be moved forward even with its inherent challenges. Thus, this initial study is a screening effort to identify potential biomarker and therapeutic candidates, some of which may be false positives. The design of subsequent validation studies will benefit by including such sample sets as those with larger numbers of cases, most of which will be formalin-fixed, patient-matched series, and autopsy specimens.

In conclusion, this study defines the effects of androgen ablation therapy on the gene expression profile of primary prostate cancer cells that are resistant to treatment. These data establish the state of the transcriptome of a discrete and important step in the process of prostate cancer progression, beyond an untreated high-grade lesion yet prior to an androgen-independent metastatic lesion, and may be critical to developing intervention strategies for this advanced disease.