Recent advances in prostate cancer research: large-scale genomic analyses reveal novel driver mutations and DNA repair defects

Prostate cancer (PCa) is a disease of mutated and misregulated genes. However, primary prostate tumors have relatively few mutations, and only three genes ( ERG, PTEN, and SPOP) are recurrently mutated in more than 10% of primary tumors. On the other hand, metastatic castration-resistant tumors have more mutations, but, with the exception of the androgen receptor gene ( AR), no single gene is altered in more than half of tumors. Structural genomic rearrangements are common, including ERG fusions, copy gains involving the MYC locus, and copy losses containing PTEN. Overall, instead of being associated with a single dominant driver event, prostate tumors display various combinations of modifications in oncogenes and tumor suppressors. This review takes a broad look at the recent advances in PCa research, including understanding the genetic alterations that drive the disease and how specific mutations can sensitize tumors to potential therapies. We begin with an overview of the genomic landscape of primary and metastatic PCa, enabled by recent large-scale sequencing efforts. Advances in three-dimensional cell culture techniques and mouse models for PCa are also discussed, and particular emphasis is placed on the benefits of patient-derived xenograft models. We also review research into understanding how ETS fusions (in particular, TMPRSS2-ERG) and SPOP mutations contribute to tumor initiation. Next, we examine the recent findings on the prevalence of germline DNA repair mutations in about 12% of patients with metastatic disease and their potential benefit from the use of poly(ADP-ribose) polymerase (PARP) inhibitors and immune modulation. Lastly, we discuss the recent increased prevalence of AR-negative tumors (neuroendocrine and double-negative) and the current state of immunotherapy in PCa. AR remains the primary clinical target for PCa therapies; however, it does not act alone, and better understanding of supporting mutations may help guide the development of novel therapeutic strategies.


Introduction
Prostate cancer (PCa) is the most commonly diagnosed non-skin cancer in American men and is estimated to account for about 30,000 deaths this year in the USA and at least 10 times as many worldwide 1,2 . The disease is curable when locally confined, but treatment options are limited for metastatic disease. First recognized in the 1940s as an effective therapy for metastatic PCa 3 , androgen deprivation remains the primary option for patients with advanced disease; however, tumors invariably relapse into incurable metastatic castration-resistant PCa (mCRPC) 4 . Further targeting of the androgen receptor (AR) axis with more effective drugs has extended survival by a few months but leads to resistance, including an increase in once-rare neuroendocrine and non-neuroendocrine/AR-negative tumors 5 . Owing to recent large-scale sequencing efforts, there is now a better understanding of the genomic landscape of PCa, including characterization of lower-frequency but nonetheless important mutations in SPOP and DNA repair genes (for example, BRCA2).
This review covers some of the recent advances in understanding PCa, including identification and targeting of key genetic aberrations (ERG, SPOP, and DNA repair defects), improvements in disease models, the emergence of AR-negative disease, and current immunotherapy strategies. Although AR signaling remains the ultimate driver of most PCa, tumors show an assortment of additional alterations that help promote disease progression and at the same time provide new opportunities for targeting this resilient disease. Although we sought to cover a wide range of topics, many fell beyond the scope of this report. However, several of those important themes can be found in previous reviews, including epigentics 6,7 , diet 8 , tumor metabolism 9,10 , biomarkers 11 , microRNAs 12,13 , the role of the microenvironment 14,15 , and racial disparities 16 .

Genomic analysis of primary tumors
Analysis of PCa at the genome level began around the turn of the century with a wide range of studies using a combination of techniques, such as comparative genomic hybridization, DNA microarray, and targeted sequencing 17,18 . Whole genome sequencing (WGS) efforts began around 2011, including a project that performed WGS of seven primary tumors (Figure 1) 19 . Over the next 2 years, whole exome sequencing (WES) efforts expanded to analyze over a hundred primary tumors 20,21 . A major leap came in 2015 with publication of the data from the PCa branch of The Cancer Genome Atlas (TCGA), a landmark study that published molecular characterization (genomic, epigenomic, and proteomic) of 333 primary prostate tumors 22 . Another large-scale study (published in 2017) is the Genomic Hallmarks of Prostate Cancer, which includes WGS for 200 primary tumors and WES for an additional 277 23 . Two 2018 studies performed WGS on 92 and 93 additional primary tumors, generating more useful data for analysis 24,25 . One issue that can arise when comparing datasets from different studies is a lack of uniform pipeline analysis (that is, data standardization, normalization, and statistical cutoffs). A 2018 report sought to tackle this problem by re-analyzing 1,013 available WES datasets (680 primary and 333 metastatic) using a common analysis pipeline 26 . As a direct result of these transformative studies, researchers One of these studies included 10 hormone treatment-naïve metastatic tumors (metastatic prostate cancer) 24 . *A total of 293 primary tumors were analyzed in this study, but 200 were included in a previous study and not included here 23 . The study by Armenia et al. 26 is a uniform re-analysis of previous whole exome sequencing studies, including many of those listed here. mCRPC, metastatic castration-resistant prostate cancer; mPC, metastatic prostate cancer; TCGA, The Cancer Genome Atlas; WES, whole exome sequencing; WGS, whole genome sequencing.
finally have an encompassing view of the genetic landscape of primary PCa that provides an important point of reference for understanding this complex disease.
At a broad glance, prostate tumors have, on average, fewer mutations (0.7 per Mb) than other common cancers, such as breast (1.2 per Mb), bladder (7.1 per Mb), colorectal (3.1 per Mb), and melanoma (12.1 per Mb) 31 . Despite having relatively few point mutations, PCa is characterized by a high rate of genomic instability and chromosomal rearrangments 32 . The most frequent genomic aberration in primary tumors is a chromosomal rearrangement fusing strong AR-regulated promoters with ETS family genes (62%), resulting in their prominent overexpression (Table 1) 22 . Although multiple ETS fusions have been identified, the most common is TMPRSS2-ERG, which arises from an approximately 3 Mb deletion on chromosome 21 that brings the androgen-regulated TMPRSS2 promoter upstream of ERG 22,33 . In addition, about 3% of primary tumors show mutation/deletion of ERF, an ETS repressive cofactor, providing another mechanism for increasing ETS activity without

Genomic analysis of metastatic tumors
Understanding the genomic landscape of primary tumors has many benefits (for example, understanding tumor origin, aiding prognosis, and revealing therapy options), but there is also a more practical need to understand the lethal form of disease, mCRPC. Androgen deprivation and AR signaling inhibitors (ARSis) (for example, abiraterone and enzalutamide) are initially quite effective, but tumors eventually develop resistance via various mechanisms, including (but not limited to) intra-tumoral androgen synthesis, AR amplification, AR ligand-binding domain mutations, or expression of constitutively active AR splice variants 4,39 . Sequencing efforts with metastatic tumors (Figure 1) identified enrichment of some mutations seen in primary disease, including amplification/mutation of AR (61%), amplification of MYC (20%), and deletion/mutation of TP53 (47%) and PTEN (41%) ( Other recent advances in analyzing mCRPC include circulating tumor cell (CTC) isolation and single-cell sequencing [40][41][42] . In a 2015 report, the authors used single-cell RNA sequencing on 76 CTCs from 12 patients with mCRPC and found enrichment in expression of stem cell genes, non-canonical WNT signaling, and a range of AR splice variants, sometimes even within the same cell 41 . In patients with multiple metastases, tumors usually share common driver mutations and appear to either be clonal or show convergent selection for therapy resistance 29,43 . Moreover, one report analyzed sequencing of multiple metastases within patients and observed that many seeded from an earlier metastasis 43 . The authors also found that metastases within a patient are likely to share tumor suppressor (also known as CD49f), which aids invasion and survival 50,51 . It is not clear why primary tumors and normal luminal cells take so poorly to tissue culture conditions. It is likely that something within the in vivo microenvironment has not been properly replicated in culture (paracrine factors, cell-cell interactions, and so on). As a way to better mimic the in vivo microenvironment, research has expanded into three-dimensional (3D) culture systems.
Prostate cell culture in 3D (that is, spheroids, prostaspheres, and organoids) has aided research by providing more physiologically relevant conditions and allowing more complex cultures 52-54 . Organoids can be derived from tumors or normal prostate cells, which can recapitulate basal, intermediate, and luminal cells as well as tumor initiation events such as prostatic intra-epithelial neoplasia (PIN) 55-59 . 3D culture can also incorporate different cell types, such as combining epithelial cells plus stroma or cancer cells plus osteoblasts 60-62 . Growing primary human tumors in 3D remains a difficult task, but metastatic tumors have been cultured with some success 52,58,63 . Innovative studies using organoid cultures have also improved our understanding of prostate tumor initiation and cell of origin 64,65 . A 2016 report provides a protocol for growing prostate organoids using a fairly complex serum-free medium with a variety of growth factors and inhibitors 52,58,66 . Interestingly, many cell lines behave differently in 2D versus 3D culture; for example, LNCaP cells have higher docetaxel resistance in 3D 67,68 . Though technically challenging, these new culture methods allow better modeling of normal and tumor epithelial structure. However, better understanding of prostate cell biology is needed so we can more efficiently culture prostate tissues, especially primary tumors.

Genetically modified mouse models
Mouse models have been extremely useful for studying disease initiation and progression in vivo and can broadly be separated into two categories: genetically engineered mouse (GEM) models and xenograft models (Figure 2) 69 . GEM models rely  on engineering the mouse genome to knockout or express specific genes, which can be done globally (classic) or in specific tissues (conditional) via tissue-specific promoter-driven expression of Cre recombinase paired with floxed (flanked-loxP) target alleles ( Figure 2A). For prostate-specific expression, the most commonly used promoter is the rat Probasin promoter (Pb), the Large Pb (LPB), or the related ARR 2 Pb, which contains Pb plus enhancer elements for higher expression 72,88 . Another common driver for conditional models is a tamoxifeninducible knock-in Cre (CreERT2) at the Nkx3.1 locus, which is more specific for prostate luminal cells but carries the caveat of losing one functional copy of the gene 74 . With CreERT2 models, Cre is still made only in the promoter-specified tissues but must be activated by the addition of tamoxifen (which causes nuclear localization) and thus grants far greater temporal control of recombination. Other valuable CreERT2 drivers include the basal keratins 5 and 14 (K5 and K14) and the luminal keratin 8 (K8) 75,76,89,90 . These keratin promoters provide basal/luminal specificity in the prostate but are also expressed in many other epithelial tissues. These models have been especially useful for lineage-tracing experiments, which use a brief pulse of tamoxifen to tag epithelial cells with fluorescent proteins and follow them as they divide and differentiate over time 76 .
There are GEM models matching many of the common alterations observed in human prostate tumors, including MYC overexpression 70,91 , Pten loss 79,92 , ERG overexpression 81,86,93 , and SPOP mutation 82 ( Figure 2A). The first-generation mouse models of PCa (TRAMP and LADY) used Pb-driven SV40 T-antigen (Tag), which promotes massive proliferation and creates tumors displaying partial neuroendocrine differentiation 72,73 . These models have regained some popularity recently, as neuroendocrine tumors are becoming more common in human PCa. Some prostate GEM models give rise to metastasis, but only one, the LPB-Tag/Pb-Hepsin model expressing SV40 and cell-surface protease Hepsin, reliably metastasizes to bone (up to 40% by 23 weeks of age), the most frequent site in human patients 77,94-96 . Other GEM models used for PCa are included in Figure 2A and have been previously reviewed 97-99 . In summary, recent advancements in GEM models enable the study of autochthonously developing PCa in immune-competent animals, but they lack the complexity of human genetics and human prostate biology and rely on contrived genetic manipulations. To study human tumors in vivo, experiments rely on xenograft models.

Xenograft mouse models
With xenograft models, human samples (tissue or cell line) are implanted into immune-compromised mice. Samples can be engrafted via a variety of routes, and the most common is subcutaneous, orthotopic, renal capsule, or tail vein ( Figure  2B) 100 . Subcutaneous grafts allow easy injection and monitoring of tumor growth, while orthotopic injections benefit from a proper microenvironment at the cost of more difficult injection and monitoring. Orthotopic injections can be made into the prostate (for primary tumors) or the metastatic site, including intratibial injections for studying bone metastasis ( Figure 2B) (10), BRCA2 homozygous loss (one), and neuroendocrine subtype (four) 63 . PDX models continue to evolve and enable researchers to test a variety of therapeutic strategies against a range of genomic tumor backgrounds and to better understand tumor resistance mechanisms.

Progress in understanding the role of ETS factors
One of the unique genomic alterations in PCa is the recurrent fusions involving strong AR-regulated promotors to ETS family transcription factors (most frequently, ERG) 27,33 . In normal prostate tissue, ERG is expressed at very low levels, but it is overexpressed in PIN and adenocarcinoma 106,107 . The most frequent fusion (caused by a 3 Mb deletion) joins the TMPRSS2 promoter upstream of ERG ( Figure 3A)   , and cross-talk between these pathways is still being investigated.

Targeting of ETS factors
Direct targeting of transcription factors is notoriously difficult but not impossible 126 . One route is to target a downstream effector of ERG that is more amenable to inhibitors, such as YAP, which can be inhibited by verteporfin ( Figure 3B). Verteporfin treatment decreases VCaP (ERG + ) xenograft tumor growth in mice 93 . In efforts to target ERG directly, phage-display library screens have been used to identify 12 ERG inhibitory peptides 110 . Two of the peptides were modified (to improve cell permeability and localize to the nucleus) and tested in vitro and in vivo. The peptides disrupted DNA binding, destabilized ERG ( Figure 3B), decreased VCaP and PC3-ERG invasion, and reduced VCaP xenograft growth 110 .
Important lessons may be learned from another cancer type with recurrent ETS fusions, Ewing's sarcoma, in which more than 90% of tumors are driven by a rearrangement fusing the EWSR1 gene to FLI1, a paralog of ERG 127,128 . ETS fusions do occur in some other cancers, but it is not a frequent event 129 . YK-4-279, a FLI1 inhibitor initially developed for Ewing's sarcoma, was reported to shrink PCa tumors in ERG + mouse xenograft models ( Figure 3B) 111,112 . Another potential candidate from the Ewing's sarcoma field is trabectedin (and its secondgeneration analogue lurbinectedin), which works in part by binding DNA minor grooves in ETS binding sites and disrupting EWS-FLI1 binding at target promoters ( Figure 3B) 113,130,131 . Very few studies (and only two phase II clinical trials) have investigated trabectedin in PCa, but the results were disappointing 132,133 . However, patients were not initially stratified by ERG status and the study did not use the newer drug lurbinectedin. These studies demonstrate that there are multiple ways to target ERG, directly or indirectly, and these therapies may be an effective option for patients with ETS + prostate tumors.

DNA repair mutations in prostate cancer
Despite having a low burden of point mutations compared with other cancers, PCa has a high rate of genomic instability (amplifications, deletions, and chromosomal rearrangements) 32 . Genomic instability is a result of DNA damage, which can arise from many sources, including (but certainly not limited to) DNA replication stress, alkylating agents, mitotic chromosome segregation errors, and radiation 23,134,135 . DNA can also be damaged as a result of transcriptional stress; AR has been reported to recruit topoisomerase enzymes to counter DNA torsional stress caused by transcription and enhancer looping 136,137 . An extreme form of genomic instability is chromothripsis, which occurs in about 20% to 30% of primary prostate tumors and involves acute chromosome shattering and reassembly, causing deletions and rearrangements 23,138 . Damage that breaks the phosphate backbone or requires repair via base excision repair, mismatch repair, or nucleotide excision repair will lead to single-strand breaks (SSBs) 134 . If SSBs occur close together on opposite strands, double-strand breaks (DSBs) can occur, which are more severe and must be repaired by homologous recombination (HR) or non-homologous end joining (NHEJ) (Figure 4). HR can occur only if a sister chromatid is present (late S or G 2 phase) and uses the non-damaged DNA as a template for error-free repair of the damaged chromatid, whereas NHEJ can repair DSBs at any cell cycle stage (predominantly G 0 /G 1 ) but has the possibility of introducing deletions or insertions 139,140 . Further information on DNA damage-sensing and repair mechanisms can be found in several recent reviews 134,141,142 .
Germline mutations in DNA repair genes are responsible for a variety of human hereditary diseases, many of which include a predisposition to cancer 143,144 . Two such genes are the key signaling kinases in the DNA damage response-ATM (primarily activated by DSB) and ATR (primarily activated by SSB)-which can activate other signaling proteins, including CHK1 (CHEK1) and CHK2 (CHEK2) (Figure 4) 142 . Two other important DNA repair genes are BRCA1 and BRCA2, for which single-copy germline mutations increase the risk of multiple cancers, most significantly breast and ovarian cancer 145-147 . Mechanistically, BRCA1/BRCA2 are crucial for recruiting RAD51, which is required for HR. BRCA1 has multiple roles, including promoting loose-end resection and aiding RAD51 loading onto DNA 148 . A recent study suggests that BRCA1 competes with 53BP1 for binding at DSBs and helps determine whether repair is shunted toward NHEJ or HR (Figure 4) 149 .
In 2016, a multi-institutional study sequenced the germline DNA of nearly 700 men with mCRPC and observed that 11.8% of patients carried a germline mutation in a DNA repair gene, most frequently BRCA2 (5.3%), CHEK2 (1.9%), or ATM (1.6%) ( Table 3)    Although most studies are ongoing, a 2018 report used retrospective analysis of an earlier study 150 to examine whether PCa patients with germline DNA repair defects had different responses to ARSi, docetaxel, or PARPi 158 . Of the previously treated patients, 60 out of 390 had germline mutations and 36% of those were treated with PARPi (some with platinum chemotherapeutic as well). Germline status had no statistically significant correlations with response to docetaxel, ARSi (abiraterone/enzalutamide), or PARPi 158 . Being a retrospective analysis, this study comes with multiple caveats, including the lack of initial patient stratification, inconsistent treatment methods, and a relatively small number of patients with germline mutations who received PARPi treatment (total of 22, of which 16 were BRCA2 mutant). In addition, the study did not have information about the tumor mutation landscape, so there could yet be a correlation between DNA repair mutations and PARPi response that was masked by confounding factors (for example, ETS fusion, AR, or PTEN status) 158 . Ongoing trials may yet prove to be beneficial for patients with germline DNA repair defects, and further research is needed to better understand how these mutations affect response to androgen deprivation and other PCa therapy resistance.

SPOP mutation
One of the novel PCa alterations elucidated by genomic sequencing efforts is mutation of SPOP 21 . Heterozygous point mutations in SPOP occur in about 10% of primary and metastatic tumors (Table 1 and Table 2) and are mutually exclusive to PTEN loss and ETS rearrangements. SPOP is an adapter component of the CUL3 E3-ligase complex, which has multiple degradation targets, including AR and its co-activators SRC-3 (NCOA3) and TRIM24 ( Figure 5) [169][170][171][172] . Other recently identified direct targets of SPOP are the BET family proteins: BRD2/3/4 [173][174][175] . BET proteins are transcriptional co-activators Other studies have confirmed similar rates (about 8-12%) of germline DNA repair defects in patients with mCRPC but had conflicting results as to whether germline mutant patients respond better to anti-androgen therapy 157-160 . Providing evidence against better outcomes are reports that germline mutant patients did not respond any better to initial androgen deprivation 157 and patients with mCRPC saw no additional benefit from abiraterone or enzalutamide 158 . However, other studies observed that DNA repair-deficient mCRPC tumors responded better to abiraterone 160 and mCRPC patients with germline BRCA1/ BRCA2/ATM mutations showed a better rate of greater than 90% prostate-specific antigen (PSA) reduction (78% versus 28%) and overall survival at 4 years (~75% versus ~25%) on ARSi therapy 159 . Although germline DNA repair mutations account for only about 12% of patients with mCRPC, it will be important to better understand how these patients will respond to anti-androgen therapies. Furthermore, the identification of DNA repair mutations (whether germline or somatic) may open a window to new therapeutic options for thousands of the roughly 30,000 men who succumb to metastatic disease every year in the US and also identify family members at increased risk for cancer 1 .

Targeting poly(ADP-ribose) polymerase
For patients with tumors deficient in DSB repair, there is strong rationale for targeting poly(ADP-ribose) polymerase (PARP), a family of proteins that are required for sensing and repairing SSBs (Figure 4) 161 . Without PARP, SSBs will cause stalling of Figure 5. SPOP mutations and prostate cancer. SPOP adapts the CUL3 E3-ligase complex to directly target and degrade AR and the AR co-activators TRIM24, SRC-3, and BRD2/3/4 [170][171][172][173][174][175][176] . Thus, loss of SPOP increases both AR and its cofactors, leading to increased AR signaling. (*) ERG has also been reported as a direct SPOP target, although this is a debated topic [176][177][178] . SPOP has been shown to target PD-L1, a key target of checkpoint inhibitor immunotherapies 179 . For indirect mechanisms, SPOP mutation strongly correlates with deletion of CHD1, which leads to 53BP1 stabilization and preference for error-prone NHEJ DSB repair 58,180 . SPOP mutation also upregulates PI3K/mTOR signaling via an unknown mechanism, which aids tumor growth and survival 83 . AR, androgen receptor; DSB, double-strand break; NHEJ, non-homologous end joining.
that upregulate a variety of oncogenes, including AR, MYC, and ERG 173 . BET inhibitors are being investigated for use in PCa 181 , but SPOP mutant tumors with elevated BET expression may be more resistant to such therapies 173,181 . Earlier this year, PD-L1 was also identified as an SPOP target, which has implications for immunotherapy and will be discussed later 179 .
ERG has also been reported as an SPOP target, although there are conflicting data in the literature for this relationship [176][177][178] . In support of the connection, two reports from 2015 identified ERG as a direct SPOP degradation target. One report found that knockdown of SPOP or CUL3 in PC3 and DU145 cells increased ERG (but not ETV1 or ERF) protein expression by increasing protein half-life and that this could be blocked by inhibition of casein kinase I 177 . The authors also used a tissue microarray with immunohistochemistry (IHC) (for ERG expression) and in situ hybridization (for TMPRSS2-ERG fusions) to identify 14 ERG-positive/fusion-negative primary tumors, of which five (36%) had SPOP mutations, suggesting that SPOP mutation may upregulate ERG without fusion events 177 . The second study found that SPOP knockdown increased ERG in LNCaP, C4-2, PC3, and 22RV1 lines 176 . Both studies reported that SPOP targets a recognition site coded by the fourth exon of ERG that is usually disrupted in TMPRSS2-ERG fusions (Figure 3), thus suggesting that fusions not only cause transcriptional upregulation but also increase protein stability 176,177 .
In 2017, a study using a GEM model of SPOP mutation (F133V) observed high-grade PIN at 6 months in Pten −/+ mice and prostate tumors in Pten −/− mice by 12 months 82 . Additionally, the authors found that SPOP mutation alone increases PI3K/ mTOR signaling. Furthermore, while PI3K signaling has negative feedback on AR, Pten −/− /SPOP F133V tumors maintain Ar and express high levels of its transcriptional targets (Fkbp5, Psca, and Nkx3.1) 82 . Thus, SPOP appears to have key intersections between AR and PI3K pathways which may help explain its importance in PCa ( Figure 5). These findings, however, do not provide a reasonable explanation for the mutual exclusivity between PTEN loss and SPOP mutation in human PCa. In a 2018 report from the same group, the authors investigated potential effects on ERG with their GEM model. Unlike the investigators in the 2015 studies 176,177 , who observed that SPOP mutation or knockdown in PCa lines increased ERG protein, this group did not observe upregulated Erg or an Erg transcriptional signature in their SPOP F133V mice 178 . In addition, they detected ERG (by IHC) in only one out of 22 SPOP-mutant human tumor samples, leading them to conclude that mutant SPOP was not regulating ERG 178 . One important lesson from these conflicting studies is that loss of SPOP protein does not appear to have the same functional consequence as overexpression of a mutant. Importantly, SPOP is rarely deleted in PCa tumors and the vast majority of mutations are in-frame missense point mutations. Clearly, additional research is needed to fully characterize SPOP mutants and understand how they differ in their interactions with wild-type SPOP targets.
SPOP has also been linked to DNA repair. SPOP mutant tumors have especially high rates of chromosomal rearrangements and share a transcriptional signature with BRCA1 loss, implicating SPOP in genomic instability 182 . Furthermore, one of the common associations with SPOP mutations is deletion of CHD1, which is involved in DSB repair and whose loss correlates with poor survival 24,180,183 . A query on cBioPortal using all available PCa datasets shows that CHD1 loss and SPOP mutation are significantly correlated (p <0.01). Specifically, 57% of tumors with CHD1 deletion (68/119) have an SPOP mutation and 29% of tumors with SPOP mutations (68/233) have CHD1 deletion. A 2017 report demonstrated that CHD1-null cells (mouse stem cells and 22RV1) are more sensitive to olaparib (PARPi) and carboplatin and have increased 53BP1 protein stability, which promotes error-prone NHEJ repair (Figure 4) 180 . Furthermore, PTEN deletion is mutually exclusive to CHD1 deletion, and PTENnull tumors require CHD1 for proliferation and survival 184 . Thus, CHD1 loss may partly explain why SPOP mutations are exclusive of PTEN loss. In summary, SPOP has been implicated in many key PCa pathways (AR, MYC, ERG, PI3K, and DNA repair) and work has only recently begun to uncover specific targets and oncogenic mechanisms. Better understanding of SPOP, including its connections with ERG and CHD1, will be needed to help choose successful targeting strategies.

Androgen receptor-negative prostate cancer
The vast majority of prostate tumors depend on the AR pathway for survival 28 . However, there are small subsets of PCa whose frequency appears to be increasing, including neuroendocrine PCa (NEPC) 185 and double-negative PCa (DNPC) 186 , which lack AR expression and therefore are not sensitive to androgen deprivation or ARSi.
Primary prostate tumors often show regions of neuroendocrine foci, but predominantly NEPC tumors (also referred to as 'small cell') are rare at initial diagnosis (<2%) 5,187,188 . However, since the advent of new ARSi therapies (abiraterone in 2011 and enzalutamide in 2012) 189,190 , there has been an increase in NEPC, which now accounts for about 15% of mCRPC and has become a mechanism of ARSi resistance 185,186,191 . These tumors are typically more aggressive and are characterized by their lack of AR and expression of neuroendocrine-associated genes, such as chromogranin A (CHGA) and synaptophysin (SYP) 5,185 . NEPC tumors often show upregulation of stemassociated genes (for example, SOX2 and MYCN) [192][193][194] and upregulation of genes associated with epithelial-mesenchymal transition (for example, SNAI1 and VIM) 5 and frequently lose expression of TP53 and RB1 195 . Moreover, these molecular changes are also directly implicated in resistance to 195 . Interestingly, NEPC tumors have TMPRSS2-ERG fusions at about the same rate as adenocarcinomas; however, owing to the lack of AR signaling, the expression levels of ERG are low in these tumors. These findings suggest that NEPC represents trans-differentiation from an androgenresponsive, epithelial-derived precursor, as opposed to the possibility that they originate from normal neuroendocrine cells, which make up less than 1% of cells in the normal prostate 38,185 .
In addition to NEPC, there is a recently identified subtype called DNPC that lacks AR and NEPC markers 186,196 . For example, recent RNA sequencing and pathway analysis with mCRPC samples from 96 patients treated before or after 2012 (the advent of abiraterone/enzalutamide) identified an increase in patients with NEPC (6.3% to 13.3%) as well as an even larger increase in DNPC (5.4% to 23.3%) 186 . It is also possible that DNPC is not an entirely distinct subset of PCa but rather an intermediate step on the way from adenocarcinoma to NEPC. An AR-negative LNCaP line (LNCaP APIPC ), which lacks CHGA and SYP, was used to investigate the mechanisms driving DNPC. This line has diminished AKT signaling and relies on an upregulated autocrine FGF8 → FGFR → ERK signaling pathway for survival 186 . Analysis of human tumor data and PDX lines with DNPC confirmed a pattern of upregulated FGFs (FGF1/8/9), FGFRs (FGFR1/2/3/4), and an ERK signature. Furthermore, LNCaP APIPC xenografts are sensitive to FGFR inhibitors (CH-5183284, PD173074) 186 . Thus, targeting the FGFR/ERK signaling axis may be beneficial for patients with DNPC, although it has yet to be tested in clinical trials. Thus, while AR-negative PCa accounts for a minority of prostate tumors, they are becoming more common and will require a different therapeutic strategy than classic AR-positive PCa.

Immunotherapy
Recently, cancer immunotherapy has received significant attention and has demonstrated great potential across different types of cancer. Immunotherapies can broadly be grouped into three strategies: cancer vaccines, immune checkpoint inhibitors, and engineered live immune cell components. Cancer vaccines use tumor-specific proteins to generate a targeted immune response or tag tumors with a lethal, targetable protein.
There are a handful of vaccine-based trials for PCa, although the only currently FDA-approved therapy is Sipuleucel-T (also known as Provenge ® ) 197 . Sipuleucel-T uses prostatic acid phosphatase as a tumor antigen and has shown about 4-to 5-month extended survival for patients with mCRPC, while patients with lower baseline PSA level had even greater (about 13-month) survival 198,199 . Several additional PCa vaccines are being tested, and detailed information about those studies can be obtained in the cited reviews 197,200 .
A second branch of immunotherapy is checkpoint inhibition, which attempts to re-activate cancer-targeting T cells that have been disarmed by tumors. The primary targets for checkpoint inhibition are CTLA-4, PD1, and the PD1 ligands PD-L1/ PD-L2. An early checkpoint inhibitor is ipilimumab, which targets CTLA-4 and was approved by the FDA in 2011 for the treatment of melanoma. Ipilimumab has been tested in PCa trials with mixed results; it was able to delay progression but did not extend overall survival 197,201  Another recent report focused on targeting myeloid-derived suppressor cells (MDSCs), which can shield tumors from T cells 209 . A GEM model with Pb-driven knockout of Pten, Trp53, and Smad4 was developed, and combinations of checkpoint inhibitors (anti-CTLA4 and anti-PD1 antibodies) plus inhibitors against multiple tyrosine kinases (dasatinib and cabozantinib) and PI3K (dactolisib/BEZ235) were evaluated. The authors found that combination therapy (checkpoint + tyrosine kinase + PI3K inhibition) had a major effect on decreasing tumor burden.
Furthermore, they went on to discover that a key mechanism of this therapy was decreased cytokine production and MDSC tumor infiltration caused by the inhibition of tyrosine kinases and PI3K, which in turn sensitized the tumors to the checkpoint inhibition 209 .
The third branch of immunotherapy uses engineered immune cells, including chimeric antigen receptor T-cell (CAR-T) therapy. CAR-T involves isolating patient immune cells and genetically engineering them to express a chimeric protein fusing a tumor-recognizing antibody region with a T-cell activation domain 210 . The engineered cells are then grafted back into the patient. CAR-T cells can directly recognize tumors and trigger activation. This therapy has been extremely successful for treating B-cell acute lymphoblastic leukemia and B-cell lymphoma 210,211 . Trials with CAR-T therapies for PCa are under way and primarily involve using PSMA and PSCA as targeting antigens 212,213 . The multiple immunotherapy strategies of cancer vaccines, checkpoint inhibitors, and CAR-T continue to improve. Meanwhile, patient stratification based on tumor mutational burden, PD1/PD-L expression, and tumor-enriched antigens such as PSMA and PSCA will help direct these therapies to the patients most likely to benefit.

Conclusions
Owing to large-scale sequencing efforts (Figure 1), the PCa field now has a near-comprehensive view of the mutational landscape of human PCa. These studies revealed that ETS fusions are the most frequent mutation in primary tumors, occurring in a little over half of cases ( Figure 3). Likewise, the second most common alteration, PTEN loss, occurs in about a quarter of primary tumors. PCa shows great resilience in evading androgen deprivation therapy and finding ways to maintain the AR pathway, but there is a growing number of tumors that do not express AR and rely on alternate survival mechanisms 4 .
Perhaps the key takeaway is that human prostate tumors are driven by a combination of alterations in a handful of signaling pathways. Understanding the role of these pathways in tumor initiation, progression, and therapeutic resistance will be critical in the future. In order to functionally test the role and mechanisms of these signaling pathways, there is a need to continually improve existing cell culture and animal models. Recently developed GEM models and organoid culture conditions can provide great opportunities for studying disease initiation, metastasis, and testing therapies against patient-derived tumors.
As has become clear with other cancers, there will almost certainly be no single effective therapy for PCa. While most tumors will likely still benefit from improved AR-targeting therapies, it will be important to recognize subsets of tumors that may benefit from targeting other supporting mutations.

Competing interests
The authors declare that they have no competing interests.

Grant information
This work was funded by the National Institutes of Health: P30CA015704, P50CA097186, and R01CA176844.
The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Homology Directed DNA Repair Frequency of Double-Stranded Breaks
Open Peer Review Current Referee Status: Editorial Note on the Review Process are commissioned from members of the prestigious and are edited as a F1000 Faculty Reviews F1000 Faculty service to readers. In order to make these reviews as comprehensive and accessible as possible, the referees provide input before publication and only the final, revised version is published. The referees who approved the final version are listed with their names and affiliations but without their reports on earlier versions (any comments will already have been addressed in the published version).
The referees who approved this article are:

Version 1
The benefits of publishing with F1000Research: Your article is published within days, with no editorial bias You can publish traditional articles, null/negative results, case reports, data notes and more The peer review process is transparent and collaborative