The genomic evolution of human prostate cancer.

Prostate cancers are highly prevalent in the developed world, with inheritable risk contributing appreciably to tumour development. Genomic heterogeneity within individual prostate glands and between patients derives predominantly from structural variants and copy-number aberrations. Subtypes of prostate cancers are being delineated through the increasing use of next-generation sequencing, but these subtypes are yet to be used to guide the prognosis or therapeutic strategy. Herein, we review our current knowledge of the mutational landscape of human prostate cancer, describing what is known of the common mutations underpinning its development. We evaluate recurrent prostate-specific mutations prior to discussing the mutational events that are shared both in prostate cancer and across multiple cancer types. From these data, we construct a putative overview of the genomic evolution of human prostate cancer.

Genomic epidemiology

Prostate cancer is the most common cancer affecting men in the developed world, accounting for 25% of all new cases of cancer in males (Cancer Research UK, 2014). Of men diagnosed at the current time, 84% are predicted to survive 10 or more years. The incidence is strongly correlated with age, rates rising sharply from 166 per 100 000 men at age 55–59 years to an overall peak of 800 per 100 000 in the 75–79 years age group. Age-specific mortality rates also rise sharply from age 55, with the highest mortality rates in the 85+ age group.

A landmark study examining the difference in the concordant occurrence of prostate cancer between monozygotic and dizygotic twins has revealed that 42% of prostate cancer may be explained by heritable risk—more than any other human cancer (Lichtenstein ). This analysis has been updated with an expanded study population and more comprehensive statistical modelling to reveal an average genetic heritability of 58% (Hjelmborg ). The first germline risk variant was discovered through the relationship of the BRCA2 gene and prostate cancer by the Breast Cancer Linkage Consortium. They estimated that heritable BRCA2 mutations confer a fivefold increased risk of prostate cancer. Similarly, BRCA1 mutations have also been shown to heavily predispose to prostate cancer, with both causing more aggressive disease and a worse prognosis (Eeles ). Genome-wide association studies have expanded the discovery of germline genetic variants, in particular common low-risk polymorphisms, such that ∼33% of familial risk in the European ancestry population is now accounted for (Al Olama ). Most susceptibility loci confer only a small increase in risk, with their effects acting multiplicatively.

Genomic heterogeneity

Multiple tumour foci are commonly detected within prostates from patients with prostate cancer, and comparison of the genomic landscape in both inter-related and geographically distinct regions within prostates has revealed independent tumour origins in several studies (Svensson ; Lindberg ; Cooper ). More recently, whole-genome sequencing of multiple metastatic sites from 10 tumours has revealed a common clonal origin containing 40–90% of total mutations and the majority of driver mutations (Gundem ). These data imply that metastases originate commonly from only one tumour foci. Once a cell population has successfully metastasised there is strong evidence of on-going clonal evolution that has enabled both ‘metastasis-to-primary' and ‘metastasis-to-metastasis' re-seeding (Gundem ; Hong ). It appears therefore that tumour heterogeneity increases as mutagenic processes continue to allow tumour clones to compete with one another within their micro-environment. This heterogeneity will only appear to decrease when an emergent tumour cell clone has mutated sufficiently to confer local or distant metastatic potential, or is able to survive cancer therapeutics. Phylogenetic trees can reconstruct the genomic archaeology of multi-focal tumours and may be described simplistically as linear, branched or independent as depicted in Figure 1. Many prostate cancers have been shown to have independent origins prior to development via a branching evolution stemming from a dominant clone.

Figure 1

Genomic heterogeneity in multi-focal prostate cancer. (A) Schematic prostatic section with three foci of genomically distinct prostate cancer. (B) Possible evolutionary trees from fertilised egg (FE) to the three foci of prostate cancer: (1) linear evolution; (2) branched evolution; (3) independent evolution.

Mutational processes shaping the cancer genome

Next-generation sequencing of somatic variants has enabled us to catalogue the mutations that have arisen within cancers. Roughly half of all prostatic tumours contain a fusion of E26 transformation-specific (ETS) family transcription factor genes with androgen-responsive promoters, most commonly transmembrane protease, serine 2 (TMPRSS2). At present this event defines the main molecular subtype of prostate cancer. Point mutations are believed to be less contributory in prostate carcinogenesis, with exome sequencing discovering a relatively low mutational frequency of 0.3–5 per Mb (Taylor ; Barbieri ; Grasso ). In a saturation analysis of point mutations, small mutations across multiple tumour types, only four significantly mutated genes were discovered from the prostate cancer data set; SPOP (found in 10.1% of samples), TP53 (3.6%), ATM (2.2%) and MED12 (3.6% Lawrence ). When the prostatic data set was compared with cancer genes from the other 20 cancer types an additional two significantly mutated genes were detected, FOXA1 (2.9%) and COL5A1 (2.2%).

The frequency of copy-number aberrations (CNAs) in prostate cancer is significantly higher than that of point mutations, suggesting carcinogenesis and progression is primarily the result of chromosomal re-arrangements. CNA burden, defined as the percentage of the genome affected by CNAs, has been shown to correlate with tumour grade, biochemical recurrence and metastasis of prostate cancer, with metastatic samples containing an average CNA burden of 32%, compared with 5% in primary disease (Baca ; Hieronymus ). CNA burden was also shown to be an independent prognostic biomarker for biochemical recurrence and metastasis after surgery and could further stratify the probability of recurrence in intermediate Gleason 7 prostate cancers (Hieronymus ).

Curating CNAs to determine mechanistically which genes are driving oncogenesis remains challenging, and is especially true when large regions have been gained or lost. Whereas homozygous deletions often occur focally, loss of heterozygosity (LOH) and gains commonly affect large regions of the genome, and inferring which target gene within that region has conferred cellular growth advantage remains challenging. Pan-cancer analyses of somatic CNAs, examining the similarities and differences between diverse tumour types have added power to detect recurrent focal regions within the genome, many of which are thought not to contain known oncogenes or tumour suppressor genes (TSGs; Zack ).

A recent pan-cancer analysis delineated mutational signatures contributing to the genomic landscape (Alexandrov ). Furthermore, many of these signatures could be attributed to distinct mutational processes that are complicit in oncogenesis. For prostate cancer, signatures corresponding to aging and DNA mismatch repair (MMR) deficiency were detected, with predominant NpCpG to NpTpG substitutions. The contribution of an aging signature in prostate cancer comes is unsurprising given its age-related epidemiology. The DNA MMR deficiency signature is comprised of very large numbers of substitutions, together with small insertions/deletions of bases that are characteristic of cancers with defective DNA MMR and termed ‘microsatellite instability'. Although germline variants of MMR genes are thought to predispose to prostate cancer (Raymond ), somatic mutations have rarely been detected (Taylor ). Despite these findings, decreased expression of MMR genes appears common (Chen ) leading to the hypotheses that defects in as yet unknown genes are deregulating the MMR pathway.

The contribution of germline variants and mutational processes in the development of prostate cancer are summarised in the upper portion of Figure 2. Prostate cancer can be delineated according to ETS fusion status and their associated genomic aberrations, which are discussed in the subsequent section. Non-ETS-specific mutations may represent a convergent pathway to later stage disease and subsequent castrate resistance. Some of these mutations occur in genes closely involved in androgen receptor signalling and prostatic growth, whereas others occur in oncogenic pathways common to other cancer types. These pathways are discussed in turn in subsequent sections of this review.

Figure 2

The putative genomic evolution of prostate cancer from normal epithelium to castrate-resistant, metastatic cancer. The pathological stages of prostate cancer are depicted on the left hand side of the figure, with corresponding genomic mutations that equate to cancer progression on the right. Cytobands are annotated according to which driver genes are most strongly implicated by the corresponding aberration.

ETS fusion-positive tumours

Recurrent gene fusions involving the oncogenic ETS transcription factors are found in roughly half of prostate-specific albumin-screened prostate cancers. Balanced structural re-arrangements, with specific abundance of ETS transcription factor gene fusions correlate with early onset prostate cancer as opposed to ‘classical' elderly onset prostate cancer (Weischenfeldt ; Steurer ). The most common fusion links the TMPRSS2 androgen-responsive promoter and the transcription factor gene ERG (Tomlins ). Many other androgen-related genes have since been discovered with fusion to other members of the ETS family including ets variant 1 (ETV1), ets variant 4 (ETV4), ets variant 5 (ETV5), and Friend leukaemia virus integration 1 (FLI1; Tomlins ).

The inter-dependence of chained chromosomal re-arrangements, termed ‘chromoplexy' has been commonly observed in ETS family fusion-positive prostate cancer and may disrupt multiple genomically distant cancer genes co-ordinately (Baca ). The precise mechanism by which this occurs has yet to be elucidated fully, but it appears to represent an early or initiating step (Svensson ; Weischenfeldt ). Significantly, primary tumours may contain hundreds of rearrangements including translocations, deletions, insertions and inversions (Berger ), contributing significantly to the overall CNA burden.

Whole-genome chromatin immunoprecipitation analyses have shown that ERG can bind to AR downstream target genes, and have suggested that ETS activation promotes epithelial-mesenchyme transition and tumour-invasive properties (Tomlins ; Massie ). This is supported through concordance analysis of TMPRSS2-ERG gene fusions in prostate adenocarcinoma and its precursor, prostatic intraepithelial neoplasia (PIN). Within the same tumour samples an association of ETS gene family fusions is seen with progression from PIN to cancer (Carver ). The function of ETS fusions in cancer has been further explored in mouse models where ERG expression results in the development of PIN only in the context of the phosphoinositide-3-kinase (PI3K) pathway activation (King ), and combination with other lesions such as AR overexpression or PTEN loss leads to invasive adenocarcinoma (Carver ). Multiple studies have also confirmed the correlation of ETS re-arrangements with PTEN inactivation as synergistic steps in the development of prostate cancer (Chen ; Carver ; King ; Steurer ).

Other aberrations associated with positive ETS status include focal deletions at 3p14, representing an aggressive phenotype with early PSA recurrence (Krohn ). This focal deletion is detected in ∼20% of primary and 30% of advanced prostate cancers (Williams ). Three putative TSGs have been implicated in the deletion: FOXP1, RYBP and SHQ1, though the precise mechanism by which these deletions act is not yet clear (Taylor ; Krohn ).

The clinical significance of ETS re-arrangements in prostate cancer is still not fully resolved. Data are conflicting; ETS fusions have been reported as associated with both more aggressive and more indolent disease; most likely reflected by heterogeneity in study cohorts (Tomlins ).

ETS fusion-negative tumours

Speckle-type POZ protein (SPOP) mutations and deletions in the q arm of chromosome 5 and 6 anti-correlated with ETS fusion-positive tumours (Figure 2). SPOP is an E3 ubiquitin ligase substrate-binding protein. Increased prevalence of these deletions with age has been shown to be strictly limited to ETS-negative cancers (Weischenfeldt ). We discuss here known mechanisms of oncogenesis and the clinical implication of these mutations.

The most common SPOP point mutation in prostate cancer involves the substrate-binding cleft of the gene. This is mutated in 6–15% of tumours across multiple independent cohorts, and may define a distinct molecular subclass of ETS-negative prostate cancer (Barbieri ; Grasso ). Functional studies are now required to determine how these mutations relate to known and possibly new pathways of oncogenesis. A recent study of the changes in the ubiquitin landscape induced by prostate cancer-associated mutations in SPOP highlighted stabilisation of the oncogene DEK and subsequent promotion of prostate epithelial cell invasion (Theurillat ).

Deletions involving the chromodomain helicase DNA-binding protein 1 gene (CHD1) locus at 5q21 are associated with ETS fusion-negative tumours and occur in 10–25% of both primary and metastatic tumours (Barbieri ; Grasso ). Point mutations and rearrangements involving CHD1 have also been identified (Berger ; Grasso ). CHD1 alters gene expression possibly by modification of the chromatin structure. Prostate tumours with CHD1 deletion have been shown to contain an excess of both CNAs and intra-chromosomal re-arrangements (Baca ).

Heterogeneous deletions of 6q12–q22 confer poor prognosis across multiple outcome parameters (Kluth ; Williams ). The mitogen-activated protein kinase kinase kinase 7 (MAP3K7) gene encodes TAK1 and is located at the peak of this broad region. As TAK1 has a role in the signalling transduction induced by TGF beta and morphogenetic protein, and controls a variety of cell functions including transcription regulation and apoptosis, it is reported as a strong candidate for the driver TSG within this region.

Prostate development and androgen signalling

The homeodomain-containing transcription factor Nkx3.1 is a putative tumour suppressor that has been shown to be a critical regulator of prostate epithelial differentiation and stem cell function in mouse models (Shen and Abate-Shen, 2010). Its gene, NKX3.1 often undergoes LOH, often a consequence of whole-arm allelic deletion of 8p, and is observed at high frequency in both ETS fusion-positive and -negative prostate cancers. Downregulation of Nkx3.1 appears early in prostate cancer and may act as a ‘gatekeeper' event in cancer initiation (Baca ). Although the incidence of LOH increases with tumour grade, the other allele remains unmutated with low levels of gene expression, leading to the speculation that epigenetic modification has a significant role in its downregulation (Shen and Abate-Shen, 2010).

The androgen receptor (AR) is a nuclear hormone receptor whose signalling is central to normal and cancerous prostate development (Shen and Abate-Shen, 2010). Blockade of the androgen pathway remains the mainstay of non-surgical treatment but only postpones the inevitable progression of disease. AR binding is implicated in tumour initiation through close proximity of AR-binding sites with re-arrangement break points (Berger ), raising the possibility that distant genomic loci, bought together in close physical contact by AR complexes are re-arranged through transcriptional stress.

Genes believed to modulate AR activity and that are mutated in prostate cancer include the nuclear receptor co-activator 2 gene (NCOA2) and forkhead-box A1 (FOXA1). Mutations, as well as focal and non-focal gains have been detected in NCOA2 that were significantly correlated with elevated NCOA2 transcript levels (Taylor ). Non-castrate patients with primary tumours harbouring NCOA2 mutation, overexpression or high-level amplification had significantly higher rates of recurrence (Taylor ). FOXA1 is an AR cofactor that is recurrently mutated in both primary and metastatic tumours (Barbieri ). FOXA1 expression has been demonstrated to increase cellular proliferation in the presence of androgen (Grasso ), and may have a role in the progression of castrate-resistant prostate cancer. Other members of the forkhead-box family are located within regions that are recurrently deleted and a putative role as tumour suppressors has been suggested (Taylor ).

There is a complex interaction between the AR and other signalling pathways, for instance the PI3K pathway. These interactions may help explain some mechanisms behind castrate resistance. The AR itself undergoes gene amplification, point mutations and alteration in splicing leading to constitutively active variants. Amplification (46% of samples) and point mutations (10% of samples) are reported in hormone refractory metastatic tumours but these are infrequently found in localised prostate cancer prior to implementation of therapy (Barbieri ; Grasso ). These data infer that AR aberrations are important as a mechanism for resistance to hormonal therapy (Visakorpi ). Recent data have concluded that different tumour cell subclones within metastatic sites can carry independent mutations associated with castrate resistance (Gundem ). Commonly, AR copy number was demonstrated to have increased at separate time points implicating continuing selective pressure on the AR pathway. Furthermore, metastasis from half of the patients studied underwent polyclonal seeding with the transfer of multiple tumour clones between metastatic sites. In the majority of those patients, subclones carrying mechanisms associated with castration resistance were found to have re-seeded multiple sites.

Generic pathways

PI3K

PI3Ks are a family of enzymes involved in cellular functions such as cell growth, proliferation, differentiation, motility, survival and intracellular trafficking. Recurrent aberrations in the PI3K pathway, in particular the PTEN gene, reinforce its central importance in the pathogenesis of prostate cancer and confirm interest in its potential for targeted therapy.

PTEN acts by dephosphorylating lipid-signalling intermediates to deactivate PI3K-dependent signalling and its loss is associated with age and ETS-positive status (Steurer ). Deletions at the PTEN gene locus occur in ∼40% of primary prostate cancers, and inactivating mutations occur in another 5–10% (Barbieri ; Grasso ; Weischenfeldt ). These events are more common in advanced disease (Taylor ; Grasso ), and a striking correlation between homozygous PTEN deletion and survival has been documented (Reid ).

The membrane-associated guanylate kinase, WW and PDZ domain containing 2 (MAGI2) gene and Phosphatidylinositol-4,5-bisphosphate 3-kinase, catalytic subunit alpha (PIK3CA) gene potentially also subvert PTEN and therefore PI3K activity. MAGI2 encodes a PTEN scaffolding protein that is recurrently disrupted by balanced re-arrangements without any evidence of copy-number loss (Berger ). PIK3CA has been reported to undergo amplifications and activating point mutations in ∼25% and 5% of prostate cancers, respectively (Barbieri ). These events appear anti-correlated with PTEN deletions, supporting the notion of functional redundancy due to similar mechanisms of action.

Cell cycle

Aberrations that drive uncontrolled cell cycling are central to oncogenesis. Here we briefly discuss the effects of mutations in genes that encode the retinoblastoma protein (RB1), the p53 protein (TP53) and the v-myc myelocytomatosis viral oncogene (MYC) in prostate cancer.

RB1 is a tumour suppressor that checks cell cycle progression from the G1 to S cell cycle phase, and is dysfunctional in many cancers. The RB1 gene is more commonly deleted or mutated in castration-resistant prostate cancer (up to 45% of patients) than in clinically localised prostate cancer, and is coincident with the emergence of castrate-resistant disease (Taylor ; Grasso ).

The TP53 protein activates expression of the p21WAF1 cyclin-dependent kinase inhibitor, regulating the cell cycle and acting as a classic tumour suppressor. In mouse models of prostate cancer, inactivation of TP53 is necessary to bypass the cellular senescence mechanisms that are activated upon the loss of PTEN (Chen ). Aberrations in TP53 are recurrently seen in both clinically localised, as well as advanced cancer.

The MYC proto-oncogene encodes a transcription factor that causes oncogenesis through cycle progression and cell survival. Amplification and to a lesser extent, mutation of MYC are common in prostate cancer, although MYC is also often differentially expressed in the absence of any mutation (Shen and Abate-Shen, 2010; Taylor ; Barbieri ; Grasso ). Often amplification is non-specific, involving a gain of the entire arm of chromosome 8, such that this mutation may increase expression of other oncogenes to confer a growth advantage.

MAPK/ERK pathway

The MAPK/ERK pathway has a central role in many cancers, though its role in prostate cancer is less well established. The pathway is frequently perturbed in advanced prostate cancers (Taylor ) and may enhance transcriptional activity of the AR. Activating mutations in KRAS and BRAF occur in roughly 10% of Asian patients but are rare in Caucasian men (Taylor ; Barbieri ; Grasso ).

Conclusions

We have summarised common aberrations that contribute to the mutational landscape of human prostate cancer. The landscape encompasses inherited variants and mutational processes, differences in ETS family fusion-positive and -negative tumours, and the aberrations that may go on to cause metastatic and castrate-resistant disease.

Significant challenges impeding the application of genomic medicine in prostate cancer include high levels of intra-tumoural heterogeneity and multifocality in primary tumours, and the often long natural history from diagnosis to metastasis or lethality. Both of these challenges hinder the generation of risk-stratification tools that correlate clinical outcomes with the genomic landscape. As structural variants contribute greatly to the genomic landscape, we envisage that multi-region whole-genome sequencing of hundreds of tumours will be required to better understand the natural evolution of prostate cancer. These methods may then pave the way to the generation of an affordable, prompt, genomic-based screening and treatment strategy from the clinic.