Abstract
Treatment of metastatic prostate cancer with androgen-ablation often elicits dramatic tumor regressions, but the response is rarely complete, making clinical recurrence inevitable with time. To gain insight into therapy-related progression, changes in gene expression that occurred following androgen-deprivation of an androgen-dependent prostate tumor xenograft, CWR22, and the emergence of an androgen-independent tumor, CWR22-R, were monitored using microarray analysis. Androgen-deprivation resulted in growth arrest of CWR22 cells, as evidenced by decreased expression of genes encoding cell cycle components and basal cell metabolism, respiration and transcription, and the induced expression of putative negative regulatory genes that may act to sustain cells in a nonproliferative state. Evolution of androgen-independent growth and proliferation, represented by CWR22-R, was associated with a reentry into active cell cycle and the up-regulation of several genes that were expressed at low levels or absent in the androgen-dependent tumor. Androgen repletion to mice bearing androgen-independent CWR22-R tumors induced, augmented, or repressed the expression of a number of genes. Expression of two of these genes, the calcium-binding protein S100P and the FK-506-binding protein FKBP51, was decreased following androgen-deprivation, subsequently reexpressed in CWR22-R at levels comparable with CWR22, and elevated further upon treatment with androgens. The dysregulated behavior of these genes is analogous to other androgen-dependent genes, e.g., prostate-specific antigen and human kallikrein 2, which are commonly reexpressed in androgen-independent disease in the absence of androgens. Other androgen-responsive genes whose expression decreased during androgen-deprivation and whose expression remained decreased in CWR22 were also identified in CWR22-R. These results imply that evolution to androgen-independence is due, in part, to reactivation of the androgen-response pathway in the absence of androgens, but that this reactivation is probably incomplete.
INTRODUCTION
In the clinical framework of prostate cancer, the evolution of disease can be considered a linear process from diagnosis to death that proceeds through a series of states. These states include localized disease, relapse after primary therapy, and metastatic androgendependent and -independent disease (1) . Prostate cancer is diagnosed at a median age of 66 years and, if organ-confined, may not impact survival until 10–15 years after the diagnosis is established (2) . Determining which localized prostate cancers are destined to impact on the quality or quantity of a patient’s survival, and as such, require treatment, is an area of controversy. In contrast, patients who present with metastatic disease are typically treated with androgen-ablation. Despite the fact that hormone blockade can result in significant regression of the tumor, it is rarely curative and most regressed tumors invariably relapse as androgen-independent lesions. At this point in the natural history, the majority of patients succumb to disease (2) . The identification of markers that correlate with molecular changes in tumors representing discrete points in the natural history, including androgen-independent status, could contribute substantially to the determination of prognosis, the selection of treatment, and improved outcomes. The ability to target specific antigens on the cell surface (3) , to block growth factor receptor signaling pathways (4) , or to use antisense molecules to decrease the expression of antiapoptotic proteins (5) , underscores the significance of molecular profiling of tumors representing the discrete clinical frameworks in which interventions are considered.
To begin to address this question, we used microarray analysis of 9792 genes to obtain a broad picture of the changes in transcription that correlate with androgen-ablation therapy of an androgen-dependent tumor and the emergence of androgen-independence. As a surrogate to the use of human tissue, we used the CWR22 xenograft model, which recapitulates the human condition in that it is dependent on androgens for growth and measured levels of prostate-specific antigen are directly proportional to tumor volume (6 , 7) . s.c. implantation of CWR22 results in exponentially proliferative tumors within approximately 3–4 weeks, which, when deprived of androgens, undergo rapid remission to a nonproliferative tumor mass, as indicated by dramatic decreases in Ki67 immunostaining (8 , 9) . This tumor mass, however, is viable, because readministration of androgens results in an increase in tumor size and an increase in the number of cells in active cell cycle (8) ,4 . From androgen-deprived and nonproliferative tumors, rapidly growing androgen-independent derivatives emanate some 3–12 months thereafter.
Here we show that the changes in gene expression following androgen-deprivation in CWR22 were indicative of growth arrest along with decreases in basal cell metabolism. Comparison of expression changes in androgen-deprived tumor cells to androgen-independent cells showed that the evolution of androgen-independent growth requires reentry into active cell cycle. Further comparison of the parent CWR22 tumor to the CWR22-R derivative in which androgens had been readministered for 2 days, led to the discovery of a small subset of androgen-responsive genes which were expressed in the absence of androgens.
MATERIALS AND METHODS
Propagation of CWR22 Xenograft Tumors in Athymic Nude Mice.
Four- to 6-week old nude athymic BALB/c male mice were obtained from the National Cancer Institute-Frederick Cancer Center and maintained in pressurized, ventilated caging at the Memorial Sloan-Kettering Cancer Center. The CWR22 tumor xenograft line was propagated in the animals by s.c. injection of minced tumor tissue from an established tumor together with Matrigel (Collaborative Research, Bedford, MA) as described in Refs. 6 and 7 . Sustained-release 5-α-dihydrotestosterone pellets (12.5 mg; Innovative Research of America, Sarasota, FL) were placed in the s.c. flanks of the mice prior to receiving tumor material for the purpose of maintaining constant amounts of circulating androgens. Tumors of approximately 1.5 × 1.0 × 1.0 cm were evident in 3–4 weeks. Androgen withdrawal was achieved both by surgical castration under pentobarbital anesthesia and removal of the testosterone pellets. Tumor sizes were determined by caliper measurements of height, weight, and depth. Androgen-independent CWR22-R xenograft tumors were propagated in nude athymic female BALB/c mice. Addition of androgens to CWR22-R was accomplished by implantation of sustained-release 5-α-dihydrotestosterone pellets into the s.c. flanks of mice bearing viable, rapidly growing tumors. Xenograft tissues from CWR22 tumors in the presence of androgens, 4-, 8 and 20 days after androgen withdrawal, and from the CWR22-R tumors in the presence or absence of androgens were collected after sacrificing the mice and flash frozen in liquid nitrogen. Tumor fragments were isolated from areas that showed microscopic evidence of neoplastic enrichment (at least >90%) and a lack of necrosis. These tumor specimens represent a subset of those described recently (8) in which extensive histological and immunohistochemical analyses were carried out.
Preparation of mRNA and Microarray Hybridization.
Total RNA was isolated from flash-frozen tumor xenograft tissue using TRIzol reagent (Life Technologies, Inc., Rockville, MD) and treated with amplification-grade DNaseô reagent (Life Technologies, Inc.). Poly(A) mRNA was isolated using Oligotex columns (Qiagen, Valencia, CA) and quantified on a spectrophotometer. Two hundred ng of each mRNA was visualized on a 0.8% denaturing agarose gel to assess quantity and quality. Purity as defined by the 260/280-nm spectrophotometer was typically >1.9. Cohybridizations on the UniGEM 1.0 microarray were performed by Incyte Microarray Operations (Incyte Pharmaceuticals Incorporated, Palo Alto, CA) using 200 ng of each test mRNA labeled with Cy3 and 200 ng of mRNA from CWR22 labeled with Cy5. The UniGEM 1.0 array used in these studies contains 10,000 spotted cDNA elements, 9,704 test elements, and 296 control elements. Of the 9,704 test elements, 4,864 represent fragments of known human transcripts. The remaining 4,840 elements are anonymous ESTs, 5 of which 923 are annotated as weakly, moderately, or highly similar to known genes.
Northern Hybridization.
Northern blots were prepared according to standard procedures. Briefly, 10 μg of total RNA from each sample was electrophoresed in denaturing formaldehyde agarose gels. RNA was transferred to Hybond N+ membrane. Probes were prepared by PCR amplification of the relevant I.M.A.G.E. cDNA clone, purified on Qiaquick columns (Qiagen) and labeled using the Prime-It II kit (Stratagene, La Jolla, CA) according to the manufacturer’s instructions. Blots were hybridized using a sodium phosphate/SDS buffer, and posthybridization washes were carried out as described (10) . Blots were imaged and signals quantified by a PhosphorImager (Molecular Dynamics, Sunnyvale, CA). Filters were striped in boiling 0.5% SDS solution. Hybridization signals from an 18S-specific rRNA probe (Ambion, Inc., Austin, TX) were used as a measure of RNA loading across the blot.
Sequence Verification of cDNA Clones.
I.M.A.G.E. cDNA clones representing EST sequences of interest were obtained from Genome Systems (St. Louis, MO) and the inserts amplified by PCR. Sequences were generated using an ABI-377 automated fluorescent sequencer and queried against the public databases using BLAST (BLASTn or X, as appropriate). Annotations were compared with those reported by Incyte, Inc., and UniGene.
Computational Analysis of Microarray Data.
Raw data from the UniGEM microarray hybridizations were received from Incyte Pharmaceuticals Incorporated (Palo Alto, CA) and exported from GEMTools (Incyte Pharmaceuticals, Inc.) into a relational database. These data were subjected to an error-correction procedure to minimize a systematic bias commonly seen with two-color microarray hybridizations. Briefly, ratios of all Cy3 and Cy5 signals were transformed using a moving average correction procedure. The corrected ratios were then transformed using a power function to minimize skewness and kurtosis and standardized to yield a distribution with a mean of zero and a variance of one 6 . After processing, data with a signal-to-background ratio of 3:1 or greater and a spot area coverage of at least 40% were up-loaded into a relational database for storage, retrieval, and report generation. Data were divided up into flat lists of increased and decreased expression for each experiment and queried across multiple experiments to identify the behavior of sub-sets of genes relevant to the questions outlined in the text. Functional classification of differentially expressed genes was carried out using the “hierarchical” function in GEMTools, which was constructed from functional information contained in UniGene (and associated MedLine links) and the LifeSeq database (Incyte Pharmaceuticals, Inc.).
RESULTS
Initial Results Using the UniGEM 1.0 cDNA Microarray.
As a surrogate to the direct use of human tissue, we used the serially transplantable CWR22 prostate tumor xenograft model (6 , 7) to study profiles of gene expression in therapy-related prostate cancer progression. We used the UniGEM 1.0 cDNA microarrays from Incyte Microarray Operations (Incyte, Inc.; see “Materials and Methods”), which are 10,000-element arrays of control elements, and I.M.A.G.E cDNA clones representing a substantial fraction of known human genes (see“ Materials and Methods”). RNAs were cohybridized on each array to provide a measure of the relative expression levels (expressed as a fold difference, rather than as absolute levels) between two cell populations. The response to androgen-deprivation was measured by cohybridization of mRNAs prepared from xenografts grown in the presence of androgens (day 0) with mRNAs prepared from tumors starved of androgens for 4, 8 and 20 days (days 4, 8 and 20). Changes in gene expression in the independent tumor, CWR22-R, were measured by cohybridization with mRNA from the androgen-dependent tumor CWR22 grown in the presence of androgens. Induction or augmentation of androgen-responsive genes was measured by comparing gene expression in CWR22 versus CWR22-R replete with androgens for a period of 2 days (see below).
Incyte Microarray Operations have reported the identification of differentially expressed genes at ratios as low as 1.75-fold 7 However, before the experiments outlined above were performed, we chose to measure this threshold empirically on the UniGEM 1.0 array. Identical mRNA populations from CWR22 were co-hybridized to the array. No differences in gene expression were expected (Fig. 1) ⇓ . At 1.75-fold, 106 ESTs were identified as differentially expressed as compared with 16 differences at >2-fold and only three differences> 2.5-fold. Thus, empirically, at a level of 2.5-fold, we would expect three differences to occur by chance. We have recently performed statistical analyses on these and other UniGEM 1.0 experiments 6 which show that 2.5-fold is conservative, corresponding to a P of approximately 10−4 on normalized data. However, because we hybridized each sample on only a single UniGEM 1.0 array, the degree of error below 2.5-fold was thought to be too high for reliable for inter-GEM comparisons (see “Discussion”).
Graphical representation of UniGEM 1.0 microarray cohybridization data. Shown are log-scale plots of Cy3 and Cy5 processed data for two experiments. A, the“ self-versus-self” cohybridization; B, cohybridization of mRNAs from CWR22 and CWR22 starved of androgens for 20 days. X and Y axes are fluorescent intensities.
An initial test of the reliability and reproducibility of the microarray method was the fortuitous identification of multiple, differentially expressed ESTs that comprised component parts of the same gene. Fig. 2 ⇓ illustrates the two differentially expressed genes (the putative M-phase phosphoprotein 2, MMP-2, and the hepatoma transmembrane kinase ligand, ephrinB5 or lerk-5) for which more than one EST was associated with each gene. As illustrated, the profiles for each set of two ESTs were remarkably similar.
Expression profiles of ESTs from the same gene. Fold changes (Y axis) relative to CWR22 for four ESTs in each of the cohybridizations (time point/state) are illustrated. ESTs designated by accession nos. N46841 and N32560 are homologous to the putative M-phase phosphoprotein, type 2; those with accession nos. W69112 and N39570 are homologous to Lerk-5 (ephrinB5). A value of ≥0.4 is equivalent to a fold-change of −2.5.
Because the cDNA fragments spotted on the UniGEM 1.0 array were not sequence-validated prior to spotting, a selection of 68 differentially expressed cDNA clones were obtained and sequenced. Of the 68 clones, identical sequences to those reported by Incyte, Inc., were identified in 62 cases (92%). To validate differential expression, Northern analysis was performed for 67 of these clones. In all but one case in which readable and quantifiable signals could be obtained, differential expression was confirmed. Examples of Northern blots relevant to the text below are illustrated in Fig. 3 ⇓ . It should be noted that several genes associated with prostate cancer progression, particularly the AR, prostate specific antigen, and human kallikrien 2 genes, among others, are not represented by I.M.A.G.E clones on the chip. This, however, was beyond our control, because the microarrays were commercial products and not fabricated in our laboratories.
Results of Northern analysis for six differentially expressed genes in CWR22 (d0), CWR22 starved of androgens for 20 days (d20), and CWR22-R and CWR22-R replete with androgens (CWR22-R+A). FKBP51/54 and S100P are androgen-dependent (Northern panels 1 and 2), but expressed in the absence of androgens in CWR22-R. Four genes identified as “stable up-regulated” (Northern panels 3–6), i.e., where expression is low in CWR22 (and CWR22 following androgen-deprivation), higher in CWR22-R, and not significantly affected by the readdition of androgens. Relative expression levels for each EST probe were compared with signals generated by an 18S rRNA probe (marked 18S). Panels on the far left and right of the Northerns depicts the relative intensity (X axis) of quantifiable gene expression in the different tumors (Y axis) for each EST after correcting for loading using signals from the 18S probe.
Differences in Gene Expression after Androgen-deprivation.
At a nominal cutoff of >2.5-fold, changes in the levels of transcription were identified in 7, 58, and 144 ESTs at the day-4, -8 and -20 time points after androgen withdrawal, respectively. Striking overlap was identified between the sets of genes that decreased in expression with increasing time of androgen deprivation. For example, 33 of 43 genes (71%) identified as decreased in expression at day 8 (Table 1) ⇓ were also significantly decreased at day 20, often with increasing magnitude. In contrast, genes with increased expression following androgen deprivation were less uniform over the three time points. Of the four genes with increased expression at day 4, only one gene, ceruloplasmin, was elevated at day 8 and day 20. Likewise, only 4 of 15 genes elevated at day 8 were also elevated at day 20 (Table 1) ⇓ .
Differentially expressed genes in androgen-deprived and androgen-independent tumors
A functional classification tool from Incyte Pharmaceuticals, Inc., called “Hierarchy,” (see “Materials and Methods”), was used to obtain broad, functionally thematic information on these large numbers of genes. We were specifically interested in any functional theme that might be over-represented among the differentially expressed genes and assist in characterizing the cellular consequences of androgen-starvation. Of the 122 genes that were decreased in expression, 51 (22%) were annotated. Nine of these genes (18%) were designated as involved in cell cycle or mitosis and meiosis, i.e., cdc2, cdc28 protein kinase 2 (Cshk2), cyclin B1, cyclin D2, α- and β-tubulin, M-phase phosphoprotein 2 (two occurrences; see Fig. 2 ⇓ ), and protein phosphatase type 1. Six of the 51 genes were annotated as involved in protein degradation and cleavage, i.e., cathepsin K, ubiquitin-specific protease 11 and peptidase β (mitochondrial processing), and six genes involved in metabolism and respiration, i.e., glyceraldehyde-3-phosphate dehydrogenase, lactate dehydrogenase A, α-enolase, phosphogylcerate mutase, NAPDH dehydrogenase and cytochrome c oxidaseVIic. Other genes with decreased expression were identified as transcription and translation factors, i.e., the TFIID component, TAFII20, and the TIF2 component, TIF2β. These collectively suggest that CWR22 cells exit active cell cycle, concomitant with down-regulation of metabolism, respiration, transcription, and translation. Of the 38 genes that were elevated after androgen-withdrawal, 22 had annotations associated with them. No obvious functional theme was apparent. However, we noted two genes, BTG-1 and EXT1, whose increased expression suggested a dominant cell-cycle arrest, because both are putative tumor suppressor genes (11 , 12) . Other genes, such as prostate carcinoma tumor antigen, whose expression is modulated by androgens (13) , were also identified. We did not see any clear evidence for the induction of proapoptotic genes. However, that these cells arrest after androgen starvation does not presuppose that the tumor mass is then “de-bulked” by apoptosis, consistent with our observations of little or no increase in terminal deoxynucleotidyl transferase-mediated nick end labeling-positive cells following deprivation (8) .
Differences in Gene Expression Between Androgen-dependent and -independent Xenografts.
Changes in the levels of transcription between the androgen-dependent xenograft, CWR22, and the androgen-independent xenograft, CWR22-R, were measured by cohybridization of mRNA populations derived from each tumor-type. Above 2.5-fold, 57 genes were identified as differentially expressed, 13 and 44 with increased and decreased expression, respectively (Table 1) ⇓ . Attempts to functionally classify these genes were curbed by the large percentage of the genes without functional annotation (only 4 of 12 that were increased and 3 of 16 unique genes that were decreased; see below). Nonetheless, of the 13 increased genes, three were associated with thyroid hormone receptor signaling, the thyroid hormone receptor, α2 (c-erb-A or αTR), the thyroid hormone receptor-associated cofactor, TRAC1 (Ref. 14 ; also known as SMRT), and the αTR corepressor, SUN-CoR (15 , 16) . We also identified two cell surface receptors, syndecan I and α-integrin-6, whose dysregulated expression is associated with cancer progression (17 , 18) .
Relationship Between Androgen-deprivation and the Emergence of Androgen Independence.
Because androgen-independent tumors are derived from androgen-dependent cells deprived of androgens, we reasoned that an examination of expression changes in the androgen-independent derivative, CWR22-R, should take into account how these genes had behaved during deprivation and vice versa. Of a total of 34 unique genes that were found to be elevated >2.5-fold at any time point after androgen-deprivation, all were found to be expressed at similar levels in both CWR22 and CWR22-R. This implied that the elevated expression of all of the genes in androgen-deprived cells was subsequently reduced to CWR22-like levels in CWR22-R. In contrast, of the 122 genes that were reduced >2.5-fold in expression following deprivation, 28 remained significantly decreased in the androgen-independent tumor (i.e., were reduced in expression in CWR22-R versus CWR22; Table 2 ⇓ ).
Response of genes in androgen-deprived tumors to androgen exposurea
One explanation for the continually reduced expression of the 28 genes in androgen-deprived and androgen-independent tumor cells is that these genes are androgen-dependent, i.e., that their expression is either directly or indirectly dependent on transcription by AR. To test this, proliferative, androgen-independent CWR22-R tumors were exposed to androgens for 2 days (hereafter referred to as CWR22-RA) and the mRNA from these androgen-replete cells cohybridized with mRNA from the primary androgen-dependent CWR22 tumor on the UniGEM array. Whereas addition of androgens to androgen-deprived tumors would result in the reexpression of androgen-responsive genes (19) , androgens would also lead to regrowth of the tumor 8 resulting in reentry into the cell cycle and increases in basal metabolism, in addition to other, as yet uncharacterized, processes. Thus, the identification of AR target genes would be masked by changes in other aspects of cell physiology expected in cycling versus noncycling cells. In contrast, the addition of androgens to a proliferating cell population (CWR22-R) would induce or augment the expression of androgen-responsive genes without necessarily inducing a phenotypic change. This strategy would “normalize” for the effects of proliferation and enhance the ability to identify hormone-specific changes, providing an “androgen-responsiveness test” for genes with interesting profiles of expression. It is important to note that the readdition of androgens to mice bearing proliferative CWR22-R tumors did not change the growth characteristics of these tumors, at least as measured by tumor volume. Other androgen-independent CWR22 tumor derivatives that some of us have established, 9 in fact, are suppressed by the readdition of androgens. 8
The decreased expression of the 28 genes seen in both androgen-deprived CWR22 and androgen-independent CWR22-R cells was not observed in CWR22-RA, suggesting that, indeed, expression had increased as a result of androgen exposure (Table 2) ⇓ . Eleven of these 28 genes had been annotated and at least one of them, α-tubulin, has been shown to be androgen-responsive (19) , suggesting that the other genes not previously thought to be androgen-responsive (e.g., HSP40, HSP70-2) may, in fact, be responsive.
Dysregulated Expression of Androgen-responsive Genes.
We extended this strategy to ask whether any of the genes that were increased or decreased following androgen-deprivation and subsequently reexpressed in CWR22-R at similar levels to CWR22 were also androgen-responsive (i.e., androgen-dependent, but expressed in the absence of androgens). Of the 94 genes that were decreased following deprivation and subsequently reexpressed in CWR22-R, only one gene, the calcium binding protein S100P, was elevated (12.4-fold) in CWR22-RA as compared with CWR22 (Fig. 3 ⇓ ; Table 3 ⇓ ). We next asked whether any of the 13 genes that were found to be elevated in CWR22-R as compared with CWR22 (and not changed as a consequence of androgen-deprivation) responded to androgen exposure, i.e., whether they were elevated, remained the same, or were repressed. Seven of 13 genes were elevated in CWR22-RA and CWR22-R, whereas the expression of the other six genes was reduced in CWR22-RA as compared with CWR22-R. Although the levels of expression found in CWR22-RA were suggestive of an additional increase following exposure of androgens to CWR22-R, Northern analysis showed that only one of the seven genes, FKBP51/54 was significantly increased. Expression was elevated from 3.4-fold in CWR22-R (versus CWR22) to 5.9-fold in CWR22-RA (versus CWR22-R), a composite increase of approximately 1.73-fold. The other six genes were increased or decreased <0.7-fold (Fig. 3) ⇓ . Reexamination of the androgen-deprivation data for FKBP51/54 showed decreases of 2.2-, 1.7-, and 2.1-fold at the day-4, -8, and -20 time points, respectively, suggesting in fact that this gene might exhibit the same behavior as the calcium binding protein, S100P, during androgen-deprivation, which we confirmed by Northern analysis (Fig. 3) ⇓ . Although similar in behavior to S100P, FKBP51/54 was unique in that it was the only gene of 122 unique genes that was decreased following androgen-withdrawal and reexpressed at higher levels in the androgen-independent derivative, CWR22-R.
Response of over-expressed genes in CWR22-R to androgensa
“Stable Up-Regulation” of Genes in CWR22-R.
Overall, 6 of 13 genes whose expression was increased in CWR22-R relative to CWR22 were found to be “stable up-regulated,” in that their expression was not significantly modulated either by androgen-deprivation or androgen reexposure. These genes are therefore analogous, in part, to the continuous up-regulation of oncogenes in primary tumors (Fig. 3) ⇓ . Providing that the levels of expression are low or absent in CWR22 as well as in androgen-deprived cells, these genes represent candidates for the molecular diagnosis of androgen-independent cells even in the presence of androgen-dependent cells. However, because the microarray experiments described here only measure differential gene expression, one cannot predict the absolute levels of expression in androgen-dependent tumors. Thus, Northern analysis was carried out to assess expression in CWR22 and 20 days after castration. These genes showed variable but generally low levels of expression in the parent tumor and during androgen-deprivation, but they were increased in expression in CWR22-R. Examples of some of these genes are shown in Fig. 3 ⇓ .
DISCUSSION
Androgen-ablation or blockade of androgen receptor-mediated signaling can result in the regression of human prostate cancer. Relapse is inevitable, in large part because the majority of cells within the tumor mass do not undergo apoptosis; rather, they enter a nonproliferative state in which the cells are still viable (2) . Thus, we chose to examine therapy-related prostate cancer regression and progression in the CWR22 model by microarray analysis. This allowed an investigation into clinically relevant points in the natural and treated history of prostate cancer, notably, regression after androgen-ablation therapy and androgen-independent tumor growth.
The parallel expression analysis of thousands of genes affords unique opportunities to visualize such biological processes, and a growing number of formats exist with which one can achieve these goals (20, 21, 22) . Microarray hybridization is perhaps the most straightforward approach, but for laboratories without an“ in-house” array system, “hybridization-by-contract” (e.g., through Incyte Pharmaceuticals, Incorporated) is expensive. Thus, our approach in this study was to use a minimum number of UniGEM arrays to gain insight into the process under study. Empirical determination of differential expression was facilitated through a “self-versus-self” cohybridization where a fold difference of >2.5 did not detect a significant number of changes. Clearly, rehybridization of the samples two to three times each would help identify cognate differences at lower fold thresholds (e.g., 1.8–2.0-fold), but would have been prohibitively expensive. Thus, we consider the current data to be indicative, rather than definitive, for the process under study, and that follow-up by other methods such as Northern analysis or quantitative real-time reverse transcription-PCR essential for those genes that can be defined as interesting in the present context.
Attempts to functionally classify differentially expressed genes were successful for sets of genes in which the majority has associated annotations. In this regard, we were able to characterize the sets of genes depressed following androgen deprivation. Collectively, the decrease in expression of multiple cell cycle components, as well decreases in metabolism, respiration, and transcription/translation suggest that androgen-deprivation elicits growth arrest in CWR22 cells. Because readministration of androgens can elicit tumor regrowth, we consider the androgen-deprived cells to be in an arrested, but viable, state. We also identified increases in the expression of several genes following androgen-deprivation that have not been associated with prostate cancer progression, but which may act to suppress to tumor growth or whose expression is positively correlated with growth inhibitory signals (e.g., androgen-deprivation). These included BTG-1, which can suppress the growth of NIH3T3 cells, and which may act directly or indirectly to modulate the transcriptional regulation of cell cycle genes (11) and EXT-1, a putative tumor suppressor gene identified through mutation analysis in hereditary multiple exostoses (12) . Between 2–2.5-fold, we also identified several other genes whose functions were provocative in this regard: (a) mac25, whose expression is decreased in many cancer cells, including breast, from an initially high level in senescent epithelial cells (23 , 24) , (b) tumor growth inhibitory factor, which has been identified as a smad-2-binding protein and repressor of transcription (25) and, (c) TSC-22, which is thought to act as a tumor suppressor and repressor of transcription and whose down-regulated expression is correlated with tumor cell line growth (26 , 27) .
Simple comparison of CWR22 relative to androgen-deprived cells and CWR22-R showed that the vast majority of genes modulated between CWR22 and androgen-deprived cells returned to CWR22-like levels in CWR22-R. Because all of the cell cycle components (positive and negative) returned to “normal,” we conclude that evolution to androgen-independence in CWR22-R required reentry into active cell cycle.
The genes that we identified in androgen-starved and androgen-independent cells did not show any evidence of expression expected from infiltrating immune cells, consistent with our histological analysis that the tumor fragments used for expression analysis were highly enriched for the neoplastic cell populations.
We and others have shown previously that the levels of AR mRNA (for which ESTs were not represented on the UniGEM 1.0 chip) and AR protein in CWR22 decrease following androgen-deprivation, but are subsequently reexpressed at some point before the emergence of hormone-refractory disease (8 , 19) . It should follow, then, that the compendium of AR-responsive genes would subsequently increase in expression in CWR22-R. However, we identified at least 28 genes that were not reexpressed and two, S100P and FKBP51, that were. These results suggest that the AR-signaling pathway is partly reactivated in the absence of androgens. However, the mechanism of AR gene reactivation (i.e., whether by ligand-independent means or other mechanisms) is not clear from the present study. It should be noted that administration of androgens to CWR22-R (i.e., CWR22-RA) may only have resulted in the up- or down-regulation of genes that were most strongly androgen-dependent, and that other genes with the characteristics of FKBP51 and S100P may yet exist. However, the relevance of identifying androgen-responsive genes that are reexpressed in the absence of androgens is supported by the observation that one of the two elevated genes, S100P, is androgen-responsive and has been shown to be dysregulated in androgen-independent derivatives of LNCaP cells and in androgen-independent PC-3 and DU145 cells (28) . This underscores the usefulness of the strategy documented here, which sought to identify such genes by comparing proliferative tumors under different androgen-stimulated conditions (i.e., CWR22 versus CWR22-R versus CWR22-R replete with androgens). It is worth noting that had we simply compared CWR22 versus CWR22-R, we would not have identified S100P as an interesting candidate gene in this context.
One of the other genes identified by this approach, FKBP51, an FK-506 binding protein with isopropyl peptidyl isomerase activity (29) , was unique in being the only gene in which expression levels were changed after androgen-deprivation and subsequently expressed at higher levels in CWR22-R versus CWR22. It is of note that this gene was also overexpressed 2.3-fold in another androgen-independent xenograft, MSKPC-6, which some of us have derived from CWR22.10Although this gene has not been reported as androgen-responsive, it is known to be a component part of the protein complex (which includes the heat shock proteins HSP90, HSP70, and HSP40), responsible for binding and folding native nuclear steroid hormone receptor proteins (e.g., estrogen, progesterone, and mineralocorticoid receptors; Ref. 29 ). As far as we are aware, an association with native AR has not been explored nor reported.
Androgen-exposure of CWR22-R also led to the identification of a set of genes whose expression was not influenced by androgens, which we refer to as “stable up-regulated.” Northern analysis of some of these genes showed low or undetectable expression in CWR22 and CWR22 starved of androgens, but increased expression in CWR22-R. Strikingly, three of the up-regulated genes were found to be involved in thyroid hormone receptor signaling: (a) the thyroid hormone receptorα (αTR); (b) the ligand-independent corepressor, SMRT or TRAC1; and (c) the unliganded corepressor, SUN-CoR. One of these, SUN-CoR, whose expression was elevated in CWR22-R, was found to be repressed following readditions of androgens, showing that expression was increased only in the absence of androgens. These results suggest that imbalances of T3 signaling components might play a role in androgen-independent growth, which is plausible based on reports showing that T3 can potentiate, albeit indirectly, the transcription of some, but not other, androgen-responsive genes (30 , 31) .
Aside from the mechanisms responsible for independent growth, these up-regulated genes represent putative molecular diagnostics for androgen-independent disease, providing that these observations are recapitulated at the protein level. In this regard, Bubendorf et al. (32) have recently reported the combined use of cDNA microarrays and tissue multiblocks for the identification of markers in CWR22-R derivatives, which were then tested by immunohistochemistry on appropriate tumor samples. Their work identified two genes, IGFBP2 and HSP27, which were over-expressed in 100% and 31% of hormone-refractory tumors, respectively. Presumably, the application of the methods outlined by Bubendorf et al. (32) to the sets of genes identified here would identify those relevant to human prostate cancer. That we did not identify either of these genes as overexpressed in the CWR22-R derivative used in the current study underscores the potential tumor heterogeneity and, perhaps, multiple mechanisms through which prostate tumors may escape the growth arrest induced by androgen-deprivation.
It is important to note that CWR22 and CWR22-R represents a single model of the evolution of hormone-autonomous tumor growth derived from a single patient, and that other models exist [e.g., LNCaP and LNCaP-C4 (33) ; LAPC-4 and LAPC-9 (34 , 35)] . We further recognize that androgen-independent tumors are heterogeneous, whether growing as xenografts or in humans, and probably include both androgen-dependent as well as androgen-independent tumor clones (35) . Nevertheless, the identification of changes in gene expression that are consistent with the results of other model systems and humans is encouraging (e.g., BTG-1, PCTA-1, S100P and the human selenium binding protein in LNCaP cells; α-tubulin and α-enolase in CWR22 and prostate specific membrane antigen in man, among many others) and suggest that the genes identified in the current study provide a useful guide-post for the molecular analysis of hormone-refractory prostate cancer.
Footnotes
-
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
-
↵1 Supported by CaPCURE (D. B. A. and H. I. S.), the American Cancer Society (D. B. A.), The PepsiCo Foundation, and the Eleanor and Paul Stevens Foundation. D. B. A. is supported by a CaPCURE Young Investigator Award.
-
10 Unpublished observations.
-
↵2 These authors contributed equally to this work.
-
↵3 To whom requests for reprints should be addressed, at Genomics Institute of the Novartis Research Foundation, 3115 Merryfield Road, San Diego, CA 92121. Phone: (858) 812-1522; Fax: (619) 812-1584; E-mail: garret_hampton{at}yahoo.com
-
↵4 Unpublished data.
-
↵5 The abbreviations used are: EST, expressed sequence tag; AR, androgen receptor; FKBP51/54, 54-kDa progesterone receptor associated protein; BTG-1, B-cell translocation gene-1.
-
↵6 Unpublished data.
-
↵7 Incyte Technical Survey, “GEM Microarray reproducibility study,” 1999. Internet address: www.incyte.com.
-
↵8 Unpublished observations.
-
↵9 D. B. A., W. D. F., and B. H.
- Received February 8, 2000.
- Accepted September 1, 2000.
- ©2000 American Association for Cancer Research.
References
Volume 60, Issue 21
