Failure of hormone therapy in prostate cancer involves systematic restoration of androgen responsive genes and activation of rapamycin sensitive signaling

Article metrics


Androgen deprivation therapy for advanced prostate cancer is often effective, but not curative. Molecular pathways mediating the therapeutic response and those contributing to the subsequent hormone-refractory cell growth remain poorly understood. Here, cDNA microarray analysis of human CWR22 prostate cancer xenografts during the course of androgen deprivation therapy revealed distinct global gene expression profiles in primary, regressing and recurrent tumors. Elucidation of the genes involved in the transition between these states implicated specific molecular mechanisms in therapy failure and tumor progression. First, we identified a set of androgen-responsive genes whose expression decreased during the therapy response, but was then systematically restored in the recurrent tumors. In addition, altered expression of genes that encode known targets of rapamycin or that converge on the PI3K/AKT/FRAP pathway was observed in the recurrent tumors. Further suggestion for the involvement of these genes in hormone-refractory prostate cancer came from the observation that cells established from the recurrent xenografts were strongly inhibited in vitro by rapamycin. The results of this functional genomic analysis suggest that the combined effect of re-expression of androgen-responsive genes as well as the activation of rapamycin-sensitive signaling may drive prostate cancer progression, and contribute to the failure of androgen-deprivation therapy.


Analysis of the molecular effects of cancer therapies in vivo can help us to understand why certain therapies are effective and how therapy-failure and drug resistance may take place. Repeated sampling of specimens from human prostate cancer patients undergoing androgen-deprivation therapy is not possible. Therefore, we decided to explore the molecular events underlying human prostate cancer progression in the xenograft model system CWR22 (Pretlow et al., 1993; Wainstein et al., 1994; Cheng et al., 1996; Nagabhushan et al., 1996; Bubendorf et al., 1999) using functional genomics techniques. cDNA microarrays containing 6605 genes were applied to the analysis of CWR22 prostate cancer xenografts in nude mice before treatment (n=4), during androgen withdrawal (n=5), and after the development of recurrence (n=3). Multi-dimensional scaling (MDS) (Khan et al., 1998) was used to visualize global differences in gene expression profiles between the xenografts. MDS analysis revealed three distinct clusters of gene expression profiles corresponding to the three stages of tumor progression in the model system (Figure 1). This suggests that there are systematic global differences in gene expression patterns in androgen-dependent primary tumors, tumors that undergo regression, and those that grow independently of androgens. These differences should provide clues to the molecular mechanisms involved.

Figure 1

Multidimensional scaling (MDS) was used to generate a graphical representation of the relative similarity in global expression profiles between primary, regressing and recurrent xenografts. MDS quantitates differences in global gene expression patterns from one sample to another and visualizes these as distances in a three-dimensional space (Khan et al., 1998). Only genes that were expressed at measurable levels in each of the xenografts were included in the analysis. Samples from each of the three different groups tested showed distinct gene expression patterns as visualized by the presence of three distinct clusters in a three dimensional plot. The arrows indicate a hypothetical pathway for the transition of primary tumors to a regression phase and development into recurrent, hormone-independent tumors. The proportions of genes involved with the transitions from one stage of therapy to the next are indicated, and illustrated in more detail in Figures 2 and 3

We then applied statistical tools to define those genes and signaling pathways that may contribute to the observed global differences in gene expression. Gene clustering using temporal templates (Figure 2b) was used to identify genes that were systemically associated with therapy response. We identified 604 genes that changed by at least twofold during therapy (Figures 1 and 2a). A cluster of 59 genes was further identified based on a decrease of gene expression by at least threefold during androgen withdrawal therapy (Figure 2c). These highly castration-repressed transcripts included at least eight known androgen-responsive genes such as cell cycle regulators ODC1, PCNA, and c-fos (Figure 2d) as well as other proliferation-associated genes including BUB1, BUB3, CDC2 delta T, CDC20, CDC28, CDC45L, CDKN3, CENPF, CKS2, MAD2L1, NEK2, STK12, and UBCH10. These castration-repressed genes (subsequently referred to as androgen-responsive genes) define possible effector and signaling genes downstream of the androgen receptor that may be involved in mediating the cessation of cell proliferation and induction of tumor regression.

Figure 2

cDNA microarray data analysis using template based clustering reveals genes implicated in androgen-signaling and tumor regression. (a) Column 1 through 6 represents color-coded relative gene expression levels for 604 genes (9.1% of total) that underwent a significant change in gene expression from primary (column 1), 0.5 days (column 2), 2 days (3), 4 days (4), 8 days (5), and 16 days (6) after castration. Template based clustering of genes with similar expression profiles are plotted and ordered based on the gene expression kinetics. (b) The 12 templates used for clustering are shown. (c) A small subset of 59 castration-repressed genes was identified and is shown color-coded during the tumor regression in columns 2–6. The last column (no. 7) represents the recurrent CWR22R tumors and provides visualization of the restoration of the castration-repressed transcripts in the recurrent tumors. (d) A list of the 59 castration-repressed genes. Known androgen receptor (AR) targets are indicated in red color. The 604 genes were defined based on selecting expression profiles that had a maximum correlation coefficient greater than 0.7 for one of the 12 temporal templates. More stringent data filtering was applied to identify 59 genes, which were repressed by at least threefold in response to castration

The level of expression of these strongly castration-repressed transcripts was then analysed in the recurrent tumors. Virtually all of these genes (57/59, 96.6%) were re-expressed in the recurrent tumors, as defined by a level of expression reaching at least 70% of that seen in primary untreated tumors. In a few cases, expression levels in the recurrent tumors exceeded those observed in the primary tumors. For example, the FKBP5 gene, which encodes for a steroid receptor interacting immunophilin (Nair et al., 1997), was one of the most strongly repressed (ratio <0.2) genes during therapy (Figures 2c and 3c). In contrast, FKBP5 expression in the recurrent tumors was twofold higher than in the primary tumors (Figures 2c and 3c), as also suggested by another study (Amler et al., 2000). The restoration of androgen-responsive genes was validated in duplicate experiments on an independent cDNA microarray containing 8152 genes. In this data set, mRNA levels of the 465 most abundant 3495 transcripts (13.3%) changed during therapy. Of these 465 androgen-responsive genes, 94.8% had expression levels that were similar in the recurrent and primary tumors. These results therefore indicate that the restoration of expression of androgen-repressed transcripts is a systematic feature of recurrent prostate cancers. This provides unbiased, genomic-level support to the hypothesis that the reactivation of androgen-responsive genes in the absence of a ligand is involved in the growth of androgen-independent tumors (Culig et al., 1994; Gregory et al., 1998; Craft and Sawyers, 1998; Craft et al., 1999; Abreu-Martin et al., 1999; Peterziel et al., 1999). The gene expression profiles do not indicate at which level of androgen regulation this reactivation takes place and what the specific molecular mechanisms may be. The virtually universal restoration of androgen-responsive genes suggests that the reactivation could be taking place at the level of the androgen receptor itself. The reactivation of androgen-responsive genes may therefore explain part of the global gene expression changes seen in the recurrent tumors. However, in the MDS analysis (Figure 1), the recurrent tumors assumed a unique gene expression state that was different from the primary tumors and that could not be solely explained by the restoration of androgen-responsive genes.

Figure 3

A second group of genes associated with recurrent tumors was identified by an analysis of genes that were differentially expressed between primary and recurrent xenografts. (a) 32 clones representing 30 different genes that were most consistently differentially expressed (out of a total of 164 genes that changed twofold or more) in two independent experiments are shown (PFKP was represented by more than one cDNA clone and illustrates the reproducibility of the analysis). The color-coding reflects the relative gene expression ratio (normalized to the mean ratio for four primary tumors) for each of six different recurrent xenografts tumors. For the six recurrent tumors, the mean expression ratio relative to the mean expression levels of the primary tumors is indicated in the ‘‘Mean’’ column. Additionally, the maximum ratio (Max.) for the upregulated genes, and the minimum ratio (Min.) for the downregulated genes is also indicated (left column). The ‘pool’ column depicts the ratios of a direct cDNA microarray experiment where four primary tumors were pooled and compared to four recurrent tumors. (b) Selected PI3/AKT/FRAP pathway related genes that were associated with hormone-refractory cell growth (based on >twofold induction of the recurrent tumors relative to the primary level) are shown. Color-coded gene expression ratios as well as the mean is also shown as in a. The criteria for selecting these genes were (i) a >twofold change in the average ratio between primary and recurrent tumors (or during therapy) and (ii) evidence from the literature suggesting the interaction of these gene products with macrolide drugs or their involvement in a rapamycin-sensitive pathway. (c) Four FK506-binding protein genes are also shown which were associated with hormone refractory tumor growth (based on at least a twofold response to therapy and were restored to greater than 80% of primary levels in the recurrent tumors). Color-coded gene expression ratios are shown for each of four primary (P) tumors, four tumors regressing following therapy (T), and six recurrent tumors (R). The mean ratio of gene expression (relative to the primary tumors) are shown for tumors undergoing therapy. (d) Rapamycin (solid lines) and FK506 (dotted lines) were tested for their effects on the viability of the hormone refractory CWR22R cell line (red lines) in vitro. The recurrent CWR22R cell line was highly sensitive to rapamycin (IC50 0.1 nM) and underwent cell death. In contrast, LNCap (blue lines) showed partial growth arrest without cell death, even at higher doses. FK506 did not have an effect on either cell line

Another set of genes was identified which showed differential expression between the primary and recurrent tumors. Based on the mean gene expression ratios from six recurrent and four primary tumors, expression levels of 104 of the 3495 informative genes (3.0%) were significantly (twofold or more) increased, and those of 60 genes (1.7%) decreased in the recurrent tumors (Figure 1). Figure 3a shows 30 genes (out of a total of 164 differentially expressed genes) that were most systematically altered in the recurrent tumors. Among these 164 genes were several genes coding for proteins that either converged on the PI3K/AKT/FRAP pathway or represented direct targets of macrolide drugs (such as rapamycin and FK506). Also, as highlighted in Figure 3b, several genes that were androgen-responsive and re-expressed in the recurrent tumors met these criteria and suggested involved of rapamycin-sensitive signaling in hormone-refractory tumors.

These findings led us to hypothesize that rapamycin-sensitive gene products and signaling pathways may play a role in androgen-independent growth in the recurrent tumors. To further evaluate this hypothesis, we studied the effects of rapamycin and FK506 on the growth and viability of a cell line established from the recurrent CWR22R xenografts. Rapamycin is a known inhibitor of the PI3K/AKT/FRAP pathway (Kunz et al., 1993; Brunn et al., 1996; Sekulic et al., 2000). Death of the hormone-independent CWR22R cells was observed at very low doses of rapamycin (IC50 0.1 nM) (Figure 3d), whereas hormone-responsive LNCap prostate cancer cell lines exhibited partial inhibition, even at high doses (Figure 3d). The results are based on two different cell lines that are not isogenic and may have other differences contributing to the observed effects. However, taken together with the global-scale gene expression studies, the data from this in vitro sensitivity testing indicate that further studies are warranted to explore rapamycin as a candidate drug for the treatment of hormone refractory prostate cancers. Cancer cells exhibit greater than a 1000-fold (IC50 ranging from <1 nM to >10 mM) variability in their sensitivity to rapamycin, possibly reflecting mechanisms of intrinsic resistance (Hosoi et al., 1998). Cancer cells that have activated genes and pathways that signal through the PI3K/AKT/FRAP pathway may be particularly sensitive. For example, IGF-1 receptor activation is associated with the efficacy of rapamycin treatment in childhood sarcomas (Dilling et al., 1994). Several growth factors and related genes that we observed to be overexpressed in the recurrent prostate cancers relative to the primary tumors (such as HGF, VEGFC, FGF2, IGFBP3, PDGFA, LTBP4, GFR, PGF, IPTKB, CDS1, and FKHL13, data not shown) could have similarly contributed to the activation of the PI3K/AKT/FRAP pathway and alterations in the rapamycin target expression.

We found that FK506 treatment did not have any effect on the growth of either the CWR22R or LNCaP cells (Figure 3d). Since FK506 and rapamycin bind many of the same intracellular targets, their different biological effects may be informative in elucidating those molecular pathways that are most critical for progression of prostate cancer. Rapamycin and FK506 both bind to FKBP12 (FK506-binding protein 12) (Sabers et al., 1995; Liu et al., 1991). Rapamycin-FKBP12, but not the FK-506-FKBP12 complex inhibits FRAP (FKBP-Rapamycin Associated Protein), a member of the phosphoinositide-3-kinase related kinases regulating translation following mitogenic activation of the PI3K/AKT/FRAP pathway. In contrast, FK506, but not rapamycin, inhibits calcineurin activity (Liu et al., 1991). This suggests that of the many known and unknown targets of rapamycin and FK506, FRAP and the activity of the PI3K/AKT pathway is a more likely candidate than calcineurin as a drug target in hormone-refractory prostate cancer. However, elucidation of the specific pathways inhibited by the rapamycin is important to uncover the molecular mechanisms of the observed effects and to identify critical targets of growth and survival signaling in hormone-refectory prostate cancer.

In conclusion, large-scale gene expression data as a function of therapy response and tumor progression in vivo may provide insights to the signaling pathways that could be exploited for therapeutic intervention in hormone-refractory prostate cancer, an often incurable and chemotherapy-resistant cancer. Our functional genomics study revealed two gene expression profiles that are operational during prostate cancer progression and endocrine therapy failure in vivo. These are the virtually uniform restoration of androgren-responsive signaling and the activation of genes associated with rapamycin sensitive signaling. Based on the results of this study, rapamycin is a candidate drug for pre-clinical and clinical evaluation in the treatment of hormone-refractory human prostate cancer.

Materials and methods

Xenografts and cell lines

CWR22 xenografts were derived from a primary human prostate carcinoma and can be serially transplanted in nude mice. The primary CWR22 xenografts undergo growth arrest and partial tumor regression after castration of the mouse host. After 3 to 10 months, androgen-independent tumor growth resumes in about half of the cases, reflecting the clinical course in patients. Fresh frozen tissue from CWR22 human prostate cancer xenografts (Pretlow et al., 1993) was obtained from 13 different mice at different stages of hormonal therapy and tumor progression (four primary untreated CWR22, five CWR22 therapy time points after 0.5, 2, 4, 8 and 16 days after castration, and four independent hormone-refractory CWR22R strains). LNCaP (ATCC) and CWR22R cell lines (kindly provided by Dr Jim Jacobberger's Laboratory at Case Western University) were cultured in RPMI1640 10% fetal bovine serum (Life Technologies, Rockville, MD, USA) at 37°C and 5% CO2. mRNA was extracted with the FastTrack 2.0 Kit (Invitrogen Corporation, Carlsbad, CA, USA).

Analysis of mRNA expression by cDNA microarrays

Custom cDNA microarrays were constructed consisting of 6605 to 8000 elements (sequence verified clones from Research Genetics, Huntsville, AL, USA), representing different (non-redundant) transcripts including 4032 to 7700 known (named) genes (Mousses et al., 2000). All xenografts were analysed at least twice. Either LNCap or CWR22R were used as a reference and labeled with Cy5. The reference cDNA was simultaneously hybridized with Cy3 labeled test specimens on a cDNA microarray as previously described (Mousses et al., 2000). Details of the fabrication of the microarray slides, image generation, and the software used for the ratio analysis, and bioinformatics, and supplemental methods are available at (Mousses et al., 2000) (

Template-based clustering algorithm

Template based gene clustering is a supervised form of clustering which we developed to identify the genes that respond to time course treatments based on their correlation to templates that represent the temporal kinetics of the expression profiles. Complete descriptions of template based gene clustering including methods, theory, and equations used, including the algorithm is available at (

Multidimensional scaling

Multidimensional scaling analysis of gene expression data has been described elsewhere (Khan et al., 1998; Mousses et al., 2000). Briefly, MDS is a method that represents measurements of similarity (between pairs of objects), as Euclidean distance between points in a dimension-reduced space. The dimension reduction procedure is to minimize the difference between the distance in higher-dimensional space and the approximate distance in the dimension-reduced (2-D or 3-D) space.

Drug treatment and cell viability

Exponentially growing LNCaP or CWR22R cells were trypsinized and plated at 0.5×105 or 1×105 cells/ml respectively in 96-well culture plates. After 24 h, cells were treated for 72 h with serial twofold dilutions of either FK-506 (Tacrolimus, Calbiochem Inc., San Diego, CA, USA) or rapamycin (Sirolimus) Sigma, St. Louis, MO, USA). DMSO was added to the control wells. The experiments were performed in triplicate. Cell viability was measured by the WST-8 assay (Dojindo Molecular Technologies Inc., Gaithersburg, MD, USA).


  1. Abreu-Martin MT, Chari A, Palladino AA, Craft NA, Sawyers CL . 1999 Mol. Cell Biol. 19: 5143–5154

  2. Amler LC, Agus DB, LeDuc C, Sapinoso ML, Fox WD, Kern S, Lee D, Wang V, Leysens M, Higgins B, Martin J, Gerald W, Dracopoli N, Cordon-Cardo C, Scher HI, Hampton GM . 2000 Cancer Res. 60: 6134–6141

  3. Brunn GJ, Williams J, Sabers C, Wiederrecht G, Lawrence Jr JC, Abraham RT . 1996 EMBO J. 15: 5256–5267

  4. Bubendorf L, Kolmer M, Kononen J, Koivisto P, Mousses S, Chen Y, Mahlamaki E, Schraml P, Moch H, Willi N, Elkahloun AG, Pretlow TG, Gasser TC, Mihatsch MJ, Sauter G, Kallioniemi OP . 1999 J. Natl. Cancer Inst. 91: 1758–1764

  5. Cheng L, Sun J, Pretlow TG, Culp J, Yang NS . 1996 J. Natl. Cancer Inst. 88: 607–611

  6. Craft N, Sawyers CL . 1998 Cancer Metastasis Rev. 17: 421–427

  7. Craft N, Shostak Y, Carey M, Sawyers CL . 1999 Nat. Med. 5: 280–285

  8. Culig Z, Hobisch A, Cronauer MV, Radmayr C, Trapman J, Hittmair A, Bartsch G, Klocker H . 1994 Cancer Res. 54: 5474–5478

  9. Dilling MB, Dias P, Shapiro DN, Germain GS, Johnson RK, Houghton PJ . 1994 Cancer Res. 54: 903–907

  10. Gregory CW, Hamil KG, Kim D, Hall SH, Pretlow TG, Mohler JL, French FS . 1998 Cancer Res. 58: 5718–5724

  11. Hosoi H, Dilling MB, Liu LN, Danks MK, Shikata T, Sekulic A, Abraham RT, Lawrence Jr JC, Houghton PJ . 1998 Mol. Pharmacol. 54: 815–824

  12. Khan J, Simon R, Bittner M, Chen Y, Leighton SB, Pohida T, Smith PD, Jiang Y, Gooden GC, Trent JM, Meltzer PS . 1998 Cancer Res. 58: 5009–5013

  13. Kunz J, Henriquez R, Schneider U, Deuter-Reinhard M, Movva NR, Hall MN . 1993 Cell 73: 585–596

  14. Liu J, Farmer Jr JD, Lane WS, Friedman J, Weissman I, Schreiber SL . 1991 Cell 66: 807–815

  15. Mousses S, Bittner ML, Chen Y, Doughtery ER, Baxevanis A, Meltzer PS, Trent JM . 2000 Functional Genomics: Gene Expression Analysis by cDNA Microarrays Livesey FJ and Hunt SP (eds) Oxford University Press: Oxford pp 113–137

  16. Nagabhushan M, Miller CM, Pretlow TP, Giaconia JM, Edgehouse NL, Schwartz S, Kung HJ, de Vere White RW, Gumerlock PH, Resnick MI, Amini SB, Pretlow TG . 1996 Cancer Res. 56: 3042–3046

  17. Nair SC, Rimerman RA, Toran EJ, Chen S, Prapapanich V, Butts RN, Smith DF . 1997 Mol. Cell.Biol. 17: 594–603

  18. Peterziel H, Mink S, Schonert A, Becker M, Klocker H, Cato AC . 1999 Oncogene 18: 6322–6329

  19. Pretlow TG, Wolman SR, Micale MA, Pelley RJ, Kursh ED, Resnick MI, Bodner DR, Jacobberger JW, Delmoro CM, Giaconia JM, et al. 1993 J. Natl. Cancer Inst. 85: 394–398

  20. Sabers CJ, Martin MM, Brunn GJ, Williams JM, Dumont FJ, Wiederrecht G, Abraham RT . 1995 J. Biol. Chem. 270: 815–822

  21. Sekulic A, Hudson CC, Homme JL, Yin P, Otterness DM, Karnitz LM, Abraham RT . 2000 Cancer Res. 60: 3504–3513

  22. Wainstein MA, He F, Robinson D, Kung HJ, Schwartz S, Giaconia JM, Edgehouse NL, Pretlow TP, Bodner DR, Kursh ED, Pretlow TG . 1994 Cancer Res. 54: 6049–6052

Download references


Supported in part by NIH grant CA57179 (to T Pretlow), CaPCURE (L Bubendorf) and Swiss National Science Foundation (L Bubendorf and U Wagner) and the Krebsliga beider Basel and Novartis Foundation (U Wagner).

Author information

Correspondence to Olli-P Kallioniemi.

Rights and permissions

Reprints and Permissions

About this article


  • microarray
  • prostate cancer
  • hormone-refractory
  • androgen-independent
  • CWR22

Further reading