The Role of Genetic Markers in the Management of Prostate Cancer

EUROPEAN UROLOGY 62 (2012) 577 587 available at www.sciencedirect.com journal homepage: www.europeanurology.com Platinum Priority Review Prostate Cancer Editorial by Zoran Culig on pp. 588 589 of this issue The Role of Genetic Markers in the Management of Prostate Cancer Atish D. Choudhury a,b, Rosalind Eeles c, Stephen J. Freedland d, William B. Isaacs e, Mark M. Pomerantz a, *, Jack A. Schalken f, Teuvo L.J. Tammela g, Tapio Visakorpi h a Dana-Farber Cancer Institute, Boston, MA, USA; b Broad Institute of Harvard and the Massachusetts Institute of Technology, Cambridge, MA, USA; c Oncogenetics Team, The Institute of Cancer Research and Royal Marsden NHS Foundation Trust, Sutton, UK; d Durham VA Medical Center, Duke University School of Medicine, Durham, NC, USA; e Brady Urological Institute, Johns Hopkins Medical Institutions, Baltimore, MD, USA; f Department of Urology, Radboud University Nijmegen Medical Center, Nijmegen, The Netherlands; g Department of Urology, University of Tampere and Tampere University Hospital, Tampere, Finland; h Institute of Biomedical Technology and BioMediTech, University of Tampere and Tampere University Hospital, Tampere, Finland Article info Article history: Accepted May 28, 2012 Published online ahead of print on June 5, 2012 Keywords: Genetic markers Prostate cancer GWAS Genome-wide association study Screening Gene expression Prognostic markers Predictive markers Please visit www.eu-acme.org/ europeanurology to read and answer questions on-line. The EU-ACME credits will then be attributed automatically. Abstract Context: Despite widespread screening for prostate cancer (PCa) and major advances in the treatment of metastatic disease, PCa remains the second most common cause of cancer death for men in the Western world. In addition, the use of prostate-specific antigen testing has led to the diagnosis of many potentially indolent cancers, and aggressive treatment of these cancers has caused significant morbidity without clinical benefit in many cases. The recent discoveries of inherited and acquired genetic markers associated with PCa initiation and progression provide an opportunity to apply these findings to guide clinical decision making. Objective: In this review, we discuss the potential use of genetic markers to better define groups of men at high risk of developing PCa, to improve screening techniques, to discriminate indolent versus aggressive disease, and to improve therapeutic strategies in patients with advanced disease. Evidence acquisition: PubMed-based literature searches and abstracts through January 2012 provided the basis for this literature review. We also examined secondary sources from reference lists of retrieved articles and data presented at recent congresses. Cited review articles are only from the years 2007 2012, favoring more recent discussions because of the rapidly changing field. Original research articles were curated based on favoring large sample sizes, independent validation, frequent citations, and basic science directly related to potentially clinically relevant prognostic or predictive markers. In addition, all authors on the manuscript evaluated and interpreted the data acquired. Evidence synthesis: We address the use of inherited genetic variants to assess risk of PCa development, risk of advanced disease, and duration of response to hormonal therapies. The potential for using urine measurements such as prostate cancer antigen 3 (PCA3) RNA and the transmembrane protease, serine 2 v-ets erythroblastosis virus E26 oncogene homolog (avian) (TMPRSS2-ERG) gene fusion to aid screening is discussed. Multiple groups have developed gene expression signatures from primary prostate tumors correlating with poor prognosis, and attempts to improve and standardize these signatures as diagnostic tests are presented. Massive sequencing efforts are underway to define important somatic genetic alterations (amplifications, deletions, point mutations, translocations) in PCa, and these alterations hold great promise as prognostic markers and for predicting response to therapy. We provide a rationale for assessing genetic markers in metastatic disease for guiding choice of therapy and for stratifying patients in clinical trials, and discuss challenges in clinical trial design incorporating the use of these markers. * Corresponding author. Dana-Farber Cancer Institute, 450 Brookline Avenue, Boston, MA 02215, USA. Tel. +1 617 632 4524; Fax: +1 617 632 2165. E-mail address: mark_pomerantz@dfci.harvard.edu (M.M. Pomerantz). 0302-2838/$ see back matter # 2012 Published by Elsevier B.V. on behalf of European Association of Urology. http://dx.doi.org/10.1016/j.eururo.2012.05.054

578 EUROPEAN UROLOGY 62 (2012) 577 587 Conclusions: The use of genetic markers has the potential to aid disease screening, improve prognostic discrimination, and prediction of response to treatment. However, most markers have not been prospectively validated for providing useful prognostic or predictive information or improvement upon clinicopathologic parameters already in use. Significant efforts are underway to develop these research findings into clinically useful diagnostic tests in order to improve clinical decision making. # 2012 Published by Elsevier B.V. on behalf of European Association of Urology. 1. Introduction Despite widespread screening for prostate cancer (PCa) and major advances in the treatment of metastatic disease, PCa remains the second most common cause of cancer death for men in the Western world. In addition, the use of prostatespecific antigen (PSA) testing has led to the diagnosis of many potentially indolent cancers, and aggressive treatment of these cancers has caused significant morbidity without clinical benefit in many cases. The recent discoveries of inherited and acquired genetic markers associated with PCa initiation and progression provide an opportunity to apply these findings to guide clinical decision making. In this review, we discuss the potential use of genetic markers to better define groups of men at high risk of developing PCa, to improve screening techniques, to [(Fig._1)TD$FIG] discriminate indolent versus aggressive disease, and to improve therapeutic strategies in patients with advanced disease (see Fig. 1). 2. Evidence acquisition PubMed-based literature searches and abstracts through January 2012 provided the basis for this literature review. We also examined secondary sources from reference lists of retrieved articles and data presented at recent congresses. Cited review articles are only from the years 2007 2012, favoring more recent discussions because of the rapidly changing field. Original research articles were curated based on favoring large sample sizes, independent validation, frequent citations, and basic science directly related to potentially clinically relevant prognostic or predictive Risk calculation Screening Prognostic markers for recurrence risk Predictive markers for response to therapy Blood (germline genotype): BRCA2 mutation HOXB13 mutation Risk SNPs 8q24 17q12 (HNF1B), 17q24.3 10q11 (MSMB promoter) others Blood: PSA protein free PSA ppsa Post-DRE urine: PCA3 TMPRSS2-ERG fusion Blood (germline genotype): KLK2-KLK3 SNP rs2735839 17p12 SNP rs4054823 Primary tissue: IHC: p53 Ki67 PTEN loss (plus SMAD4, cyclin D1, SPP1) Blood (germline genotype): Androgen metabolism SNPs SLCO2B1,SLCO1B3 (androgen transport) SNPs Metastatic tissue: IHC: AR SPINK1 Her2/neu Phospho-Src FISH MYC amplification TMPRSS2-ETS fusions PTEN deletion FISH: AR amplification PTEN deletion TMPRSS2-ETS fusion BRAF,RAF1 translocation RNA from primary: CCP score MiRNA predictor Stem cell signatures High-grade signatures RNA from metastasis: AR splice variants AR activity signature Src family kinase signature DNA from metastasis: AR mutations Fig. 1 A sampling of genetic markers proposed to help guide clinical decision making for assessment of risk of developing prostate cancer, and for screening, prognosis, and prediction of response to therapy. AR = androgren receptor; BRAF = v-raf murine sarcoma viral oncogene homolog B1; BRCA2 = breast cancer 2, early onset; CCP = cell cycle progression; FISH = fluorescent in situ hybridization; Her2/neu = human epidermal growth factor receptor 2; HNF1B = HNF1 homeobox B; HOXB13 = homeobox B13; IHC = immunohistochemistry; KLK2-KLK3 = kallikrein-related peptidase 2 and 3; MSMB = b-microseminoprotein; mirna = microrna; MYC = v-myc myelocytomatosis viral oncogene homolog (avian); PCA3 = prostate cancer antigen 3; phospho-src = phosphorylated v-src sarcoma (Schmidt-Ruppin A-2) viral oncogene homolog (avian); PSA = prostate-specific antigen; PTEN = phosphatase and tensin homolog; RAF1 = v-raf-1 murine leukemia viral oncogene homolog 1; SLCO2B1 = solute carrier organic anion transporter family, member 2B1; SLCO1B3 = solute carrier organic anion transporter family, member 1B3; SNP = single nucleotide polymorphism; SPINK1 = serine peptidase inhibitor, Kazal type 1; TMPRSS2-ERG = transmembrane protease, serine 2 v-ets erythroblastosis virus E26 oncogene homolog (avian).

EUROPEAN UROLOGY 62 (2012) 577 587 579 markers. In addition, all authors on the manuscript evaluated and interpreted the data acquired. 3. Evidence synthesis 3.1. Risk alleles PCa has a strong heritable component, and family history has long been appreciated as a risk factor for developing the disease. Twin studies, in which monozygotic are compared with dizygotic twins, suggest that the inherited component of PCa risk is >40%, greater than the inherited component of risk for other common cancers, such as colon and breast [1]. However, until recently, little was known regarding the genetic underpinnings of this risk. With the completion of the Human Genome Project, the publication of the International Haplotype Map Project (a catalog of millions of common single nucleotide polymorphisms, or SNPs, in the human population), and a decrease in the cost of high-throughput genotyping, an unbiased genomewide search for inherited variants associated with PCa risk has become feasible. This approach, called a genomewide association study (GWAS), scans the entire genome, evaluating common inherited variants (minor allele frequency >1 5% in the population) in large numbers of cases and controls [2]. Up to 1 million genetic polymorphisms are evaluated in a typical GWAS. Given the large number of tests performed, case-control cohorts of several thousand are needed to achieve statistical power strong enough to confidently declare a positive association ( p < 1 10 7 is a common threshold). Requisite sample sizes have been attained through multi-institutional collaborations and international consortia devoted to this effort. As of this writing, >40 PCa susceptibility loci explaining approximately 25% of the familial risk have been identified (Table 1). The vast majority of GWASs reported to date have been conducted in populations of European ancestry, but GWASs in men of African American ancestry [3] and Japanese ancestry [4] have recently been performed, revealing risk markers apparently unique to these populations. Further work in these and other non-european ancestral groups is necessary. An up-to-date catalog of published GWASs is available at the US National Institutes of Health National Human Genome Research Institute Web site (http://www.genome.gov/gwastudies/). 3.1.1. Clinical impact As a screening biomarker, germline genetic risk markers have appealing features: They do not fluctuate over time or in the setting of particular conditions and they are accessible at any age. The relative increased risk of developing the disease based on any single polymorphism discovered to date is small, generally <1.5-fold, but risk appears to increase with increasing number of risk alleles carried. A 2008 study involving >3500 PCa cases and controls correlated genotype at five risk SNPs known at that time with risk of developing the disease [5]. It was found that carriers of all five risk alleles had an odds ratio (OR) of 9.46 for developing the disease compared with men carrying no risk alleles. However, this represented a small subset of the population and, overall, genotype was not able to contribute significantly to prediction of PCa when added to established risk factors such as age and family history. Similarly, as newly discovered risk SNPs have been added to the model, they have not, in the large majority of subjects, appreciably added to the utility of more established risk factors [6]. An algorithm has been described to assess PCa risk based on the 26 risk SNPs known at the time, along with family history as a proxy for postulated, but unknown, risk variants [7]. GWASs reported to date have evaluated only common inherited variants (minor allele frequency >1 5% in the population studied) and describe only a minority of the genetic component of risk. Rarer variants (minor allele frequency in the population 1%) associated with risk may be more highly penetrant and carry higher relative risk, such as mutations in the breast cancer 2, early onset (BRCA2) gene[8]. For example, coding variants in the homeobox B13 (HOXB13) gene, which were recently discovered by targeted exonic sequencing of genes in a region of PCa linkage at chromosome 17q21-22, were found in <0.1% of controls, but 1.4% of patients with a very strong family history or early-onset PCa [9]. GWASs using rarer variants as cataloged in the 100 Genomes Project may reveal markers with higher ORs and explain a greater percentage of inherited risk. While the clinical utility of PCa risk models incorporating genetic variants is limited currently, performance of these models are expected to improve with the incorporation of variants associated with higher relative risk. However, the potential benefits of applying these risk models in clinical practice have yet to be demonstrated. Genotyping all men for up to 40 susceptibility loci would be a major health care expenditure, and it is unclear whether using genetic factors to define an early or intensive screening group would lead to improved outcomes in either the high- or low-risk groups: The former group could still be subject to morbidity from diagnosis and treatment from overscreening for potentially nonlethal PCa, and the diagnosis of aggressive cancer may be delayed in the latter group. Chemoprevention for PCa is also highly controversial. Although 5a-reductase inhibitors have consistently been demonstrated to reduce risk of PCa in men who have undergone PSA screening [10], the US Food and Drug Administration (FDA) has recommended against using these agents for PCa prevention because of potentially increasing the risk of high-grade tumors [11]. As risk models, preventive therapies, screening paradigms, and treatment recommendations improve over time, the incorporation of genetic markers may play an important clinical role in the future. It is hoped that men identified as being at high risk of developing PCa could have recommendations for screening and preventive therapy tailored to this increased risk, similar to women who inherit breast cancer-predisposing alleles of the BRCA1 and BRCA2 genes. 3.1.2. Functional characterization An intriguing aspect of GWAS findings is that the risk markers reside largely in noncoding regions of the genome. The mechanism of inherited risk is therefore not readily

580 EUROPEAN UROLOGY 62 (2012) 577 587 Table 1 Prostate cancer risk single nucleotide polymorphisms reaching statistical significance in genomewide association studies across multiple cohorts SNP Chromosomal locus Nearest known gene within 100 kb Region Odds ratio rs10187424 2p11 none Intergenic 1.06 1.12 a rs721048 2p15 EHBP1 Intronic 1.15 c rs1465618 2p21 THADA Intronic 1.16 1.20 c rs13385191 2p24 C2orf43 Intronic 1.10 1.21 d rs12621278 2q31 ITGA6 Intronic 1.32 1.47 c rs2292884 2q37 MLPH Intronic 1.02 1.19 a,b rs7629490 3p11 none Intergenic 1.04 1.09 b rs2660753 3p12 VGLL3 Intergenic 1.11 1.48 c rs10934853 3q21 EEFSEC Intronic 1.12 c rs6763931 3q23 ZBTB38 Intron 1.01 1.07 a rs10936632 3q26 CLDN11, SKIL Intergenic 1.08 1.14 a rs17021918 4q22 PDLIM5 Intronic 1.12 1.25 c rs7679673 4q24 TET2 Intergenic 1.15 1.37 c rs2121875 5p12 FGF10 Intron 1.02 1.08 a rs2242652 5p15 TERT Intron 1.11 1.19 a rs12653946 5p15 none Intergenic 1.20 1.33 d rs1983891 6p21 FOXP4 Intron 1.09 1.21 d rs130067 6p21 CCHCR1 Missense 1.02 1.09 a rs339331 6q22 RFX6 Intron 1.15 1.28 d rs651164 6q25 SLC22A1 Intergenic 1.10 1.20 b rs9364554 6q25 SLC22A3 Intronic 1.17 1.26 c rs10486567 7p15 JAZF1 Intronic 1.12 1.35 c rs6465657 7q21 LMTK2 Intronic 1.03 1.19 c rs2928679 8p21 SLC25A37 Intergenic 1.16 1.26 c rs1512268 8p21 NKX3-1 Intergenic 1.13 1.28 c rs10086908 8q24 None Intergenic 1.14 1.25 c rs7841060 8q24 None Intergenic 1.19 c rs13254738 8q24 None Intergenic 1.11 c rs13252298 8q24 None Intergenic 1.05 1.18 b rs16901979 8q24 None Intergenic 1.66 c rs16902094 8q24 None Intergenic 1.21 c rs445114 8q24 None Intergenic 1.14 c rs6983267 8q24 None Intergenic 1.13 1.42 c rs7000448 8q24 None Intergenic 1.14 c rs1447295 8q24 None Intergenic 1.29 1.72 c rs10993994 10q11 MSMB Intergenic 1.15 1.42 c rs4962416 10q26 CTBP2 Intronic 1.17 1.20 c rs7127900 11p15 TH Intergenic 1.29 1.40 c rs11228565 11q13 MYEOV Intergenic 1.05 1.23 b,c rs7931342 11q13 MYEOV Intergenic 1.19 1.25 c rs10875943 12q13 PRPH Intergenic 1.04 1.10 a rs902774 12q13 None Intergenic 1.11 1.24 b rs9600079 13q22 None Intergenic 1.12 1.24 d rs11649743 17q12 HNF1B Intronic 1.28 c rs4430796 17q12 HNF1B Intronic 1.16 1.38 c rs7210100 17q21 ZNF652 Intronic 1.35 1.69 e rs1859962 17q24 None Intergenic 1.20 c rs8102476 19q13 PPP1R14A Intergenic 1.12 c rs2735839 19q13 KLK3 Intergenic 1.25 1.72 c rs5759167 22q13 BIK Intergenic 1.14 1.20 c rs742134 22q13 BIK Intron 1.01 1.23 b rs5945619 Xp11 NUDT11 Intergenic 1.19 1.46 c rs5919432 Xq12 AR Intergenic 1.02 1.12 a SNP = single nucleotide polymorphism. All SNPs listed are independently correlated with risk. a Kote-Jarai et al. [88]. b Schumacher et al. [89]. c Pomerantz and Freedman [90]. d Takata et al. [4]. e Haiman et al. [3]. apparent and the risk alleles provide an opportunity to gain insight into prostate carcinogenesis. There are several potential mechanisms by which a genetic variant may be associated with altered cancer risk, including: (1) genetic linkage to a coding variant in a cancer-relevant gene (ie, the risk SNP is merely a proxy for the true causal exonic variant that was not tested in the GWAS), (2) alteration in promoter/ enhancer binding sites or chromatin structure affecting

EUROPEAN UROLOGY 62 (2012) 577 587 581 expression of adjacent or distant genes, or (3) change in the expression of noncoding RNAs. In fact, a recent report demonstrated that eight of the known PCa-risk SNPs and one novel risk SNP fall into the intervals of long noncoding RNAs, which are transcripts of 100 200 nucleotides or longer that make up most of the transcribed noncoding RNAs [12]. Determining the mechanism through which the risk alleles are acting is difficult because the variants that lead to a modestly increased risk in humans over a lifetime can be difficult to model in short-term laboratory experiments. For example, several independent risk loci have been discovered at chromosome 8q24, in a gene desert, with the closest annotated gene being v-myc myelocytomatosis viral oncogene homolog (avian) (MYC), located approximately 200 700 kb away. MYC has been demonstrated to be an important oncogene in PCa because the MYC locus is frequently amplified in human PCa and expression of human MYC in the mouse prostate leads to the development of prostate intraepithelial neoplasia followed by invasive adenocarcinoma [13]. While MYC RNA expression increases with castration in normal rat prostate [14], its protein expression decreases with antiandrogen treatment in androgen-dependent human PCa cell lines, with MYC overexpression allowing for androgen-independent proliferation of these cells [15]. SNPs in the MYC gene locus itself are not in linkage with the variants associated with PCa risk [16,17]. While there has been no convincing association between MYC RNA expression and risk allele status in either histologically normal or tumor tissue in humans [18], there is evidence that the risk loci contain functional transcriptional enhancers [19] that physically interact with the MYC locus in a tissue-specific manner [20,21]. Changes in these enhancers based on genotype may influence MYC regulation. Risk loci at 8q24 may also alter binding to the transcription factors TCF4 [22],FoxA1[19],or YY1 [23], which would be predicted to alter transcriptional responses to Wnt, androgen receptor, and other cellular signaling pathways. Candidate genes and pathways have been proposed at other PCa risk loci [24,25], and as mechanisms for inherited risk are elucidated, opportunities for developing preventive therapies will emerge. 3.2. Genetic markers in screening and diagnosis of prostate cancer The most common current screening test for PCa is measurement of the serum concentration of PSA. However, there is no single threshold value for PSA that can reliably distinguish patients with PCa from those without, and thus an unfortunate consequence of populationwide PSA screening is the cost of and morbidity from diagnostic biopsies in patients without cancer. There has been significant research into improving the performance of PSA itself, including measuring free PSA or truncated forms of PSA [26]. A GWAS has revealed significant association between PSA levels in patients without PCa and SNPs at six loci, suggesting that PSA thresholds for biopsy could be personalized based on genotype at these loci [27]. Novel markers with increased specificity for the detection of PCa are also of great interest. Among these markers is prostate cancer antigen 3 (PCA3), a noncoding RNA with expression confined to the prostate, and which is highly overexpressed in 95% of PCas compared with normal or benign hyperplastic prostate tissue [28]. Progensa PCA3 (Gen-Probe Inc., San Diego, CA, USA) is a commercially available diagnostic test that quantitatively detects PCA3 RNA expression in urine and prostatic fluid. A PCA3 score >35 in the urine has correlated with an average sensitivity and specificity of 66% and 76%, respectively, for the diagnosis of PCa (compared to a specificity of 47% for serum PSA at the cutoff for 65% sensitivity) [29]. Elevated PCA3 scores have also been demonstrated to increase the probability of a positive repeat biopsy in men with one or two prior negative biopsy results [30]. The transmembrane protease, serine 2 v-ets erythroblastosis virus E26 oncogene homolog (avian) (TMPRSS2- ERG) translocation is present in many PCas [31], and leads to the translation of a fusion protein of the androgenresponsive gene TMPRSS2 with ERG, a member of the v-ets erythroblastosis virus E26 oncogene homolog (ETS) family of transcription factors, leading to aberrant transcription of ETS targets in PCa cells. Detection of the TMPRSS2:ERG fusion in urine has been reported to yield >90% specificity and 94% positive predictive value for PCa detection [32], although a clinical diagnostic test is not yet available. The combination of urinary PCA3 and TMPRSS2- ERG with serum PSA levels has been reported to improve screening performance compared to PSA alone [33]. Other biomarkers being tested include urinary concentrations of b-microseminoprotein (MSMB), the expression of which is decreased by the rs10993994 risk SNP in the MSMB promoter. MSMB levels are decreased in PCa tissue compared to benign prostate disease, and decreased urinary levels have been shown to improve upon urinary PSA, but not serum PSA, for PCa diagnosis [34]. The discovery of novel biomarkers for the diagnosis of PCa in blood or urine is an active area of research. 3.3. Genetic markers and prediction of prostate cancer outcome Among factors currently used by clinicians to establish PCa prognosis are the D Amico criteria, which use tumor stage, Gleason grade, and PSA level to define risk groups. Other prognostic markers include PSA density and number of positive biopsy cores. Accurate prediction is critical as treatment decisions, which range from watchful waiting to multimodality therapy, are very much linked to prognosis. Thus, there is significant interest in using molecular markers to more precisely stratify patients to assess the need for therapy, the intensity of therapy, and the extent of surveillance required either before or after initial treatment. 3.3.1. Germline markers There is some evidence that PCa aggressiveness has a heritable component [35], and several groups have analyzed known germline risk markers for association with disease

582 EUROPEAN UROLOGY 62 (2012) 577 587 aggressiveness. While there have been studies describing an association between risk allele status and prognosis, few of these findings have been validated in independent cohorts [36 38]. One allele associated with risk of developing PCa has also been associated with aggressive PCa and PCa-specific mortality in more than one study [39,40]: rs2735839, located in the kallikrein-related peptidase 2 and kallikrein-related peptidase 3 (KLK2-KLK3) intergenic region. KLK3 encodes PSA, and the PCa-risk SNP rs2735839 (G) was one of the six loci identified by GWAS that were associated with higher PSA levels in patients without PCa [27]. Interestingly, it is the nonrisk allele of rs2735839 (A) that has been correlated with worse PCa-specific survival. One hypothesis explaining these findings is that elevated PSA levels associated with higher risk, but more indolent disease, may allow cancers in these patients to be detected at an earlier stage and thus protect from the eventual mortality associated with PCa. Yet the locus does appear to influence prostate biology as well [41]. Further work is needed to determine how variation at this locus influences outcome. No matter what the mechanism, a potential clinical use for this SNP would be to help determine PSA cutoff for biopsy; those with the A allele may benefit from a lower cutoff. The majority of studies for genetic variants that affect aggressiveness and prognosis of PCa performed to this point have focused on alleles associated with risk of developing cancer. However, it is possible that a different set of variants influences prognosis of cancer after it develops, such as genes involved in steroid metabolism, cancer progression, immune surveillance, or drug metabolism. While candidate gene approaches have yielded results that warrant further follow-up [42], it will be important to perform true GWASs rather than studying selected sets of SNPs. One difficulty in these studies has been heterogeneity in defining disease aggressiveness and outcome because different clinicopathologic criteria have been used to define more aggressive disease across sites even within the same study [43]. Thus, future studies should include large numbers of patients across diverse ethnic backgrounds, with standardized definitions of aggressiveness and, ideally, standardized pathologic review. For correlation with mortality, it would be important for the screening parameters, thresholds for intervention, and treatment modalities to be similar across sites to reduce biases that could result from patients with similar genetic composition having different outcomes from patients with other compositions simply due to disparate standards of care in different parts of the world. 3.3.2. Somatic structural genetic changes Massive efforts are currently in progress to describe the mutational landscape of PCa [44,45]. Among the most common genetic alterations in PCa are amplification of the 8q24 locus containing the MYC oncogene and deletions in the loci containing the phosphatase and tensin homolog (PTEN) tumor suppressor at 10q23 and the NK3 homeobox 1 (NKX3.1) gene at 8p21. In addition to TMPRSS2-ERG and other translocation events involving ETS family members, the solute carrier family 45, member 3 v-raf murine sarcoma viral oncogene homolog B1 (SLC45A3-BRAF) and epithelial splicing regulatory protein-1 v-raf-1 murine leukemia viral oncogene homolog-1 (ESRP1-RAF1) gene fusions have also been found in PCa [46]. Oncogenes that are frequently mutated in other cancer types, such as phosphoinositide- 3-kinase, catalytic, alpha polypeptide (PIK3CA); v-ki-ras2 Kirsten rat sarcoma viral oncogene homolog (KRAS); and BRAF, are not commonly mutated in PCa. However, multiple other events targeting the RB signaling pathway, PI3K pathway, RAS/RAF pathway, and androgen receptor pathway, among others, have been detected [44]. Advances in technology and decreases in costs have increased the feasibility of whole-genome sequencing, which has the potential to discover novel translocation events and mutations, including noncoding mutations. Such analysis has already revealed recurrent translocations involving the cell adhesion molecule CADM2 or the PTEN-interacting protein MAGI2, in addition to recurrent mutations of the Speckle-type POZ Protein (SPOP), which was not previously implicated in prostate tumorigenesis [45]. More studies involving larger samples sizes and cancers at various stages of progression will allow for better definition of the genetic alterations in PCa and how these events can alter clinical behavior. Certain genetic alterations have been consistently correlated with poor prognosis in the literature; for example, amplification of the MYC locus at 8q24 [47] and P53 overexpression as a proxy for inactivation [48]. The impact of other known genetic alterations on clinical behavior of tumors remains unclear and somewhat controversial; for example, TMPRSS2-ERG has been implicated as both a negative and positive prognostic marker in the literature [49]. While ETS gene fusions seem to drive PCa development, they are early events in tumorigenesis and so their contribution to progression and the behavior of advanced cancers remains unclear [50]. Several groups have assessed somatic markers in combination in an attempt to improve their prognostic capability. In assessing the relationship of PTEN deletion with the TMPRSS2-ERG fusion, two independent groups found that patients with neither lesion had a favorable prognosis [51,52]. However, no poor-risk group has been identified consistently across studies. These discrepancies may be related to the fact that PTEN loss is not defined consistently (genetic deletion of the PTEN locus vs loss of protein expression), ETS fusion-positive groups are not defined consistently (TMPRSS2-ERG only vs inclusion of other ETS fusions), different outcomes are measured across studies (biochemical recurrence after local therapy vs PCa-specific mortality in patients managed conservatively), and small numbers of patients are represented in each subgroup. Other markers tested in combination with PTEN loss for prognostic information include tumor protein p27 gene loss [53], hemoxygenase-1 overexpression [54], and HER2/3 overexpression [55]. A four-protein signature, as assessed by immunohistochemical staining for PTEN in combination with a subset of proteins involved in tumor growth factor-b signaling: SMAD4, cyclin D1, and SPP1, was found to predict

EUROPEAN UROLOGY 62 (2012) 577 587 583 biochemical recurrence significantly better than Gleason score alone [56]. It remains to be determined which combinations of events will provide the most reliable prognostic information to guide clinical decision making. 3.3.3. Gene expression changes Gene expression is an attractive marker for understanding the behavior of a particular cancer because the measurement of messenger (mrna) levels of many genes at once may reflect the phenotype or state of a cancer better than specific genetic variants or alterations alone. For example, gene expression signatures have been successfully applied to define subclasses of breast cancer with different biologic behavior and response to therapy. Gene expression-based diagnostic tests, under the commercial names of Oncotype Dx (Genomic Health, Inc., Redwood City, CA, USA) and MammaPrint (Agendia, Irvine, CA, USA), are currently in clinical use for assessing recurrence risk and for predicting benefit from adjuvant chemotherapy in patients with localized estrogen receptor-positive, lymph node-negative breast cancer [57]. There are multiple challenges in the interpretation, standardization, and reproducibility of assessing gene expression in PCa. The disease itself can be highly heterogeneous and potentially multifocal, even within the same patient, as suggested by different Gleason grades often identified within an individual prostate specimen. Moreover, the tissue used in these studies comes from diverse sources, including transrectal ultrasound (TRUS) biopsies, transurethral resection of the prostate samples, and radical prostatectomy samples. Further driving heterogeneity are the different preservation techniques and time to fixation. Finally, heterogeneity is also influenced by the admixture of other tissue elements within a given sample (eg, less aggressive tumor, normal prostate epithelium, and normal stromal elements). There is evidence that much of the disease progression seen in patients with low-risk PCa is due to inadequate biopsy sampling [58]. In such patients, gene expression profiling of the Gleason 6 tumor sampled in a TRUS biopsy would not reflect the clinical behavior of the nonsampled aggressive tumor. Several groups have published reports of gene expression signatures in primary prostate tumors that correlate with poorer prognosis in retrospective analysis. However, the gene lists generated in these signatures generally have not overlapped across studies and no subset has yet been validated for clinical use [59]. Among more recent attempts to generate useful biologic information using gene expression, Penney et al. assessed gene expression differences in Gleason 6 versus Gleason 8 cancers to help predict clinical behavior in patients with Gleason 7 cancer by similarity to more or less aggressive disease [60]. Other groups have assessed the expression levels of gene sets corresponding to stem cell-like states [61,62] or markers of increased cell-cycle progression (CCP) [63], hypothesizing that these states would reflect more aggressive disease. In addition, expression profiles of micrornas, which are short noncoding RNAs that can bind mrnas to regulate gene expression, have been used to generate a diagnostic microrna predictor that can independently forecast postoperative outcome [64]. No consensus has arisen regarding which markers are most useful for assessing prognosis, but a number of companies are interested in developing subsets of these markers as diagnostic tests. For example, a CCP score assessing 46 genes involved in CCP is being commercially developed by Myriad Genetics (Salt Lake City, UT, USA) with the brand name Prolaris. Validation studies of prognostic biomarkers in a standardized fashion across large numbers of patients are needed to determine whether they can help guide clinical decision making. 3.4. Markers of response to therapy An ongoing topic of significant interest is the identification of genetic markers that predict response or resistance to therapy. Such markers could prove valuable for selecting a therapy with greatest likelihood of efficacy in an individual patient and for stratifying patients in clinical trials based on likelihood of response to the tested therapies. 3.4.1. Germline markers The most common initial systemic therapy for metastatic PCa is androgen deprivation therapy (ADT), but the time to progression with this therapy is highly variable. It has been hypothesized that inherited genetic factors may account for some of this variability, and SNPs in genes involved in androgen metabolism [65] and androgen transport [66,67] have been studied for influence on response to ADT. While certain SNPs in these pathways were correlated to time to progression on ADT in these studies, these require validation in larger independent cohorts. 3.4.2. Somatic markers ADT remains the first-line therapy of choice in metastatic disease for most patients with PCa, and thus interest in markers predicting response has focused on choice of treatment in patients who have developed castrationresistant PCa (CRPC). The topic of biomarkers in the diagnosis and treatment of men with CRPC has recently been reviewed in this journal [68]. Given that there are no clinically validated predictive genetic biomarkers to this point, this discussion will focus on potential directions given our current understanding of the biology of PCa. The androgen receptor (AR) signaling pathway remains critical in CRPC, as demonstrated by clinical responses to novel agents decreasing circulating androgens to below castration levels (abiraterone) [69] and more potent antagonists of the AR (eg, MDV3100) [70]. Theoretically, tumors with AR amplification/overexpression would be expected to respond well to either class of agent, while those with loss of AR expression (through transdifferentiation or another mechanism) might be resistant to both. Patients whose predominant mechanism of resistance to ADT is mediated through intratumoral androgen synthesis would be predicted to respond well to abiraterone, which can inhibit the C17,20-lyase enzyme required for androgen synthesis in both the adrenal glands and the tumor. Conversely, patients

584 EUROPEAN UROLOGY 62 (2012) 577 587 Table 2 Potential markers requiring validation for use in stratified medical decision making Genetic event Existing treatment Reference AR overexpression, mutation, splice variants, activity PTEN deletion AR antagonist C17,20 lyase inhibitor PI3K (with or without mtor) inhibitor plus AR antagonist De Bono et al. [26] Tran et al. [70] Carver et al. [79] Mulholland et al. [80] MYC amplification Unknown Gustafson and Weiss [81] Delmore et al. [82] BRCA2 mutation TMPSRSS2-ERG PARP1 inhibitor Fong et al. [83] Brenner et al. [77] BRAF/RAF1 mutation or translocation BRAF or MEK inhibitor Palanisamy [46] MSMB risk SNP or MSMB levels Unknown Whitaker et al. [84] SPINK1 overexpression (ETS fusion-negative tumors) EGFR inhibitor Ateeq et al. [85] Her2/neu overexpression HER2 inhibitor Gregory et al. [86] Activated MET Cabozantinib Torres et al. [76] Low AR activity Elevated SRC-family kinase activity AR = androgen receptor. SRC inhibitor Mendiratta et al. [72] Tatarov et al. [87] with AR mutations leading to promiscuity to nonandrogen substrates might be predicted to respond better to MDV3100, which prevents AR nuclear import. CRPC primarily mediated through constitutively active splice variants lacking the ligand-binding domain at the carboxy-terminal of the AR protein might require novel agents targeting the aminoterminal of the protein [71]. In future clinical trials of these types of agents, it may prove useful to stratify patients by AR copy number, expression, mutational profile, splice variants, and markers of AR activity (such as a transcription-based AR activity signature) [72], along with intratumoral androgen concentrations and C17,20 lyase activity to assess these as predictive markers. In addition to hormonal agents, other novel therapies have recently demonstrated improvements in overall survival in patients with CRPC in phase 3 studies, including the taxane cabazitaxel (Sanofi SA, Paris, France), the tumor vaccine Sipuleucel-T (Dendreon Corp., Seattle, WA, USA), and the radiopharmaceutical radium-223 (Algeta ASA, Oslo, Norway). A promising molecularly targeted agent in CRPC that does not target the AR is cabozantinib (XL184) (Exelixis, South San Francisco, CA, USA), a met protooncogene (hepatocyte growth factor receptor) (MET) and vascular endothelial growth factor receptor 2 (VEGFR2) inhibitor that led to significant improvements in bone scans and reduction in bone pain in a majority of treated patients, although with considerable toxicity [73]. Multiple markers have been implicated as corresponding to taxane resistance in CRPC [74], including elevated class III b-tubulin expression [75]; however, these markers have yet to be prospectively validated for predicting resistance to either docetaxel or cabazitaxel. Likewise, markers of activated MET signaling have been proposed as a prognostic indicator in other cancer types responsive to cabozantinib [76] but also have yet to be prospectively validated. Future directions in management of CRPC involve defining the activated signaling pathways in each patient s cancer to allow direct targeting of the altered pathways essential for tumor cell survival (oncogene addiction) or targeting vulnerabilities engendered by aberrant pathway activation (synthetic lethality). Both of these concepts have been illustrated in preclinical model systems. For example, the neoplastic phenotype conferred to prostate cells by expression of the SLC45A3-BRAF or ESRP1-RAF1 fusion proteins is sensitive to RAF or mitogen-activated protein kinase (MEK) inhibition [46]. In the case of ETS translocations, inhibitors of poly(adp-ribose) polymerase-1 (PARP1) potentiate DNA damage induced by the expression of the TMPRSS2-ERG fusion protein, and PARP1 inhibitors also inhibit ETS-positive, but not ETS-negative, PCa xenograft growth [77]. These findings are reminiscent of the synthetic lethality of PARP1 inhibition with BRCA1/BRCA2 mutations seen in breast and ovarian cancer. A sampling of genetic markers with potential for use in stratified medicine decision making, along with agents suggested to be useful in their treatment in recent clinical or preclinical studies, is presented in Table 2. Novel drugs directly targeting many other genetic alterations in PCa (ERG; mutated SPOP; serine peptidase inhibitor, Kazal type 1 [SPINK1], which is overexpressed in a subset of ETS fusion-negative PCas) are in preclinical development, and novel signaling pathways involved in CR are an ongoing topic of laboratory research. Several challenges remain in developing genetic markers to predict clinical responses. The first issue is acquisition of metastatic tissue for molecular analysis prior to initiating a new treatment. This is likely essential in PCa due to the heterogeneity of primary tumors and the molecular changes that occur with metastasis and prior treatments. For example, it has been demonstrated in breast cancer that there is significant discordance in staining for HER2 in metastatic versus primary lesions [78]. One approach for tissue acquisition is mandatory biopsy of metastatic lesions for participation in clinical trials, although there are morbidity and costs associated with these procedures, and patients may decline participation based on this requirement. Molecular analysis of circulating tumor cells (CTCs) has been posited as a solution to this issue to allow for a liquid biopsy, and this has been accepted by the FDA as a marker of drug-trial response. However, the relationship of CTCs to bulk metastatic lesions remains unclear, and genetic and molecular changes in prostate CTCs have yet to be demonstrated to predict response to therapy in metastatic lesions.

EUROPEAN UROLOGY 62 (2012) 577 587 585 The second issue is the difficulty of using our current tools in analyzing genetic and molecular changes in the tumor to assess the dependencies/vulnerabilities of the cancer cells therein. It is possible that single markers are excellent predictors of therapeutic response, but it is also possible that a combination of multiple markers (genetic or epigenetic alterations, gene expression changes, or phosphoproteomic profiles) may better reflect the phenotypic state of a cancer, and thus likelihood of response to a therapy. There is significant effort both experimentally and computationally to derive multicomponent signatures reflecting activated signaling pathways and thus predicting responses. The third challenge is incorporating genetic markers into clinical trial design to assess their predictive capability. Several institutions have begun clinical initiatives for the large-scale molecular profiling of patient cancers, although the details regarding which patients are tested for which lesions using what platforms vary among the sites. Ideally, patients would be stratified by lesions such as those listed in Table 2 and treated with the indicated therapy. Options would expand as new somatic mutations are discovered and treatments developed. However, the design of clinical trials assessing these markers and the corresponding agents is complicated by defining a clinically meaningful end point (overall survival is difficult given the multiplicity of available treatments in CRPC and the likelihood of future crossover), defining what the control group in such a study would be, the availability of agents in such trials, and the large number of patients required in each arm to have the power to assess the efficacy of each individual drug. As in any clinical trial, a negative result with a drug could be related to an invalid target, an ineffective drug, or incorrect patient selection. There will need to be discussions and agreements between investigators, multiple institutions, pharmaceutical companies, and regulatory agencies to adapt clinical trial design to the era of molecularly targeted therapies. 4. Conclusions With advances in genotyping and sequencing technologies, and greater mechanistic understanding of PCa through experimental and computational techniques, continued discovery of novel genetic markers associated with disease initiation, progression, and response/resistance to treatment is expected. Translation of these discoveries to the development of clinical diagnostic tests and to patient stratification in clinical trials is critical. Large clinical trials involving multi-institutional collaborations will be required to prospectively validate the utility of these markers for clinical decision making. In the meantime, institutional and regulatory infrastructure must be in place to most efficiently test the appropriate hypotheses for maximal scientific benefit and benefit to patients. Author contributions: Mark M. Pomerantz had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Choudhury, Pomerantz. Acquisition of data: None. Analysis and interpretation of data: Choudhury, Pomerantz, Eeles, Freedland, Isaacs, Schalken, Tammela, Visakorpi. Drafting of the manuscript: Choudhury, Pomerantz. Critical revision of the manuscript for important intellectual content: Choudhury, Pomerantz, Eeles, Freedland, Isaacs, Schalken, Tammela, Visakorpi. Statistical analysis: None. Obtaining funding: None. Administrative, technical, or material support: None. Supervision: Eeles, Freedland, Isaacs, Schalken, Tammela, Visakorpi. Other (specify): None. Financial disclosures: Dean A. Tripp certifies that all conflicts of interest, including specific financial interests and relationships and affiliations relevant to the subject matter or materials discussed in the manuscript (eg, employment/affiliation, grants or funding, consultancies, honoraria, stock ownership or options, expert testimony, royalties, or patents filed, received, or pending), are the following: Atish D. Choudhury is supported by a US National Institutes of Health (NIH) T-32 Training Grant, the Anne Huber Foster Fellowship, the Prostate Cancer Foundation Young Investigator Award and the US Department of Defense Physician Scientist Training Award. Rosalind Eeles is supported by Cancer Research UK Grant C5047/A7357, the US NIH Cancer Post-GWAS initiative grant number 1 U19 CA 148537-01 (GAME_ON initiative), The European Commission s Seventh Framework Programme grant agreement number 223175 (HEALTH-F2-2009-223175), and support from the US NIH National Institute for Health Research to the Biomedical Research Centre at The Institute of Cancer Research and Royal Marsden Foundation NHS Trust, The Institute of Cancer Research and The Everyman Campaign, Prostate Action, The Orchid Cancer Appeal, The National Cancer Research Network UK, and The National Cancer Research Institute UK. Mark M. Pomerantz is supported by the Prostate Cancer Foundation Young Investigator Award. Tapio Visakorpi is supported by the Academy of Finland, the Cancer Society of Finland, the Reino Lahtikari Foundation, the Sigrid Juselius Foundation, European Union, TEKES, and the Medical Research Fund of Tampere University Hospital. Funding/Support and role of the sponsor: None. References [1] Lichtenstein P, Holm NV, Verkasalo PK, et al. Environmental and heritable factors in the causation of cancer analyses of cohorts of twins from Sweden, Denmark, and Finland. N Engl J Med 2000;343: 78 85. [2] Manolio TA. Genomewide association studies and assessment of the risk of disease. N Engl J Med 2010;363:166 76. [3] Haiman CA, Chen GK, Blot WJ, et al. Genome-wide association study of prostate cancer in men of African ancestry identifies a susceptibility locus at 17q21. Nat Genet 2011;43:570 3. [4] Takata R, Akamatsu S, Kubo M, et al. Genome-wide association study identifies five new susceptibility loci for prostate cancer in the Japanese population. Nat Genet 2010;42:751 4. [5] Zheng SL, Sun J, Wiklund F, etal. Cumulative association of five genetic variants with prostate cancer. N Engl J Med 2008;358:910 9. [6] Sun J, Kader AK, Hsu FC, et al. Inherited genetic markers discovered to date are able to identify a significant number of men at considerably elevated risk for prostate cancer. Prostate 2011;71:421 30. [7] Macinnis RJ, Antoniou AC, Eeles RA, et al. A risk prediction algorithm based on family history and common genetic variants: application to prostate cancer with potential clinical impact. Genet Epidemiol 2011;35:549 56. [8] Kote-Jarai Z, Leongamornlert D, Saunders E, et al. BRCA2 is a moderate penetrance gene contributing to young-onset prostate cancer: impli-