Molecular substratification of bladder cancer: moving towards individualized patient management

Abstract

Despite advances in surgical techniques, perioperative therapies and postoperative management, outcomes for patients with bladder cancer have largely remained unchanged. Current management of bladder cancer still relies on pathologic staging that does not always reflect the risk for an individual patient. Studies assessing molecular alterations in individual tumors are offering insights into the myriad of cellular pathways that are deregulated in bladder tumorigenesis and progression. Alterations in pathways involved in cell-cycle regulation, apoptosis, cell signaling, angiogenesis and tumor-cell invasion have been shown to influence disease behavior. High-throughput assays are now allowing multiplexed assessment of biomarker alterations, thereby enabling characterization of novel molecular subtypes of bladder cancer. Such approaches have also been used for discovery and validation of robust prognostic molecular signatures. The future of bladder cancer management will rely on the use of validated multimarker panels for risk stratification, optimal surgical management, and theranostic strategies to identify and target specific alterations in individual tumors.

Introduction

The management of bladder cancer has witnessed several advances in the past decade. Improvements in visualization modalities now allow for more detailed localization of tumors within the bladder [Stenzl et al. 2010; Zlatev et al. 2015], and novel methods of urine-based cancer detection offer the opportunity for precise and noninvasive surveillance [Mitra and Cote, 2010; Birkhahn et al. 2013]. Administration of neoadjuvant chemotherapy has demonstrated oncologic benefit [Advanced Bladder Cancer (ABC) Meta-analysis Collaboration, 2005], and perioperative management protocols have improved patient recovery after surgery without increasing hospital readmissions [Daneshmand et al. 2014]. Despite these developments, however, survival outcomes for patients undergoing radical surgery for bladder cancer have remained fairly unchanged over the last 30 years [Zehnder et al. 2013]. Indeed, cancer of the urinary bladder remains the fifth most common malignancy in the United States and the eighth most frequent cause of cancer-related deaths [Siegel et al. 2016]. Worldwide, the disease accounts for over 165,000 deaths each year [Torre et al. 2015].

While the use of molecular correlates as a guide to treatment has become mainstay in several other cancer types, management of urothelial carcinoma of the bladder (UCB) is still largely based on tumor stage and other histopathological parameters. The genesis and progression of UCB is now known to involve alterations in several molecular pathways that are otherwise responsible for the maintenance of cellular homeostasis. These alterations often dictate the rate of tumor progression, and may therefore act as surrogates for identifying patients who have more aggressive disease. Subtyping patient populations based on the molecular alterations in their primary tumors may therefore permit risk stratification and administration of more personalized therapies.

Pathological subtypes of bladder cancer

While certain histopathological subtypes of bladder cancer are more aggressive than conventional forms, the most common methodology for substratifying UCB that reflects overall clinical risk is based on determination of tumor stage [Mitra et al. 2012b]. UCB can present as a noninvasive phenotype where malignant cells are restricted to the urothelial layer, and an invasive phenotype wherein tumor cells breach the basement membrane and may invade the subepithelial connective tissue and underlying muscle. Noninvasive UCB may present as two distinct forms. Papillary (Ta) tumors are generally exophytic, have a tendency to recur locally, but rarely invade the basement membrane or metastasize. However, carcinoma in situ (CIS) is a flat lesion with a high propensity for invasion and metastasis. Patients with only CIS lesions in their urinary tract may also have synchronous with or without development of metachronous tumors [Zehnder et al. 2014]. Ta tumors are suggested to develop due to molecular aberrations that are usually distinct from CIS and the invasive (T1–T4) cancers, although these pathways may not always be mutually exclusive (Figure 1) [Wu, 2005; Knowles, 2006]. Low-grade papillary tumors usually have a constitutively active receptor tyrosine kinase–Ras pathway, with activating mutations in HRAS and fibroblast growth factor receptor 3 (FGFR3) genes [Bakkar et al. 2003; Rieger-Christ et al. 2003; Van Rhijn et al. 2004]. High-grade Ta tumors are often characterized by homozygous deletion of p16^INK4a [Orlow et al. 1999]. CIS and invasive tumors show frequent alterations in the TP53 and retinoblastoma (RB) genes and pathways [Mitra et al. 2006b]. Loss of heterozygosity of chromosome 9q is more frequently noted in low-grade Ta tumors, although some investigators have found chromosome-9 deletions in both dysplastic urothelium and CIS lesions [Spruck et al. 1994; Hartmann et al. 2002]. When the occasional papillary tumor does transform to an invasive phenotype, it is usually due to accumulation of additional alterations in the p53 pathway. p16 alterations have also been identified in invasive tumors [Korkolopoulou et al. 2001]. Alterations in cadherins, matrix metalloproteinases (MMPs), vascular endothelial growth factors (VEGFs), and thrombospondin-1 (TSP-1), which remodel the extracellular matrix and promote tumor angiogenesis, are seen more commonly in muscle-invasive (T2–T4) tumors and also contribute to nodal metastasis [Wu, 2005]. Postcystectomy tumor recurrence rates are higher for patients with muscle-invasive cancers than those with nonmuscle-invasive tumors, and prognosis following such recurrence is generally poor [Mitra et al. 2012c].

Figure 1.

Model for urothelial tumorigenesis and progression.

Noninvasive and invasive tumors have distinct molecular alterations. Noninvasive tumors have constitutive activation of the Ras–mitogen-activated protein kinase pathway, while flat carcinoma in situ (CIS) and invasive lesions have alterations in p53 and other cell-cycle-regulatory molecules. Loss of heterozygosity of chromosome 9q is more common in low-grade papillary (Ta) carcinomas, although chromosome 9 deletions are also noted in progressive CIS. Development of muscle-invasive disease (T2–3 tumors) involves additional alterations in Rb and p16. Locations of arrow tails and heads correspond to the disease stage(s) before and after the noted alteration(s), respectively. Molecules indicate alteration events that pose a risk for progression of a particular phenotype.

Molecular alterations and their prognostic impact

Bladder tumorigenesis involves alterations in multiple homeostatic pathways with profound deregulations within a complex molecular circuitry. Distinct molecular alterations have been documented in noninvasive and invasive UCB. The net effect of such deregulations on key cellular processes establishes the ultimate fate of the tumor (Table 1). Therefore, these alterations often serve as predictors of outcome, and they may also act as therapeutic targets [Mitra and Cote, 2007, 2009; Youssef et al. 2009].

Table 1.

Molecular alterations in urothelial carcinoma of the bladder and their prognostic impact.

Marker	Cellular function	Association of alteration with prognosis
		Recurrence probability	Survival probability	Other
Cell cycle
p53a	Inhibits G1-S progression	↑	↓
p21b	Cyclin-dependent kinase inhibitor	↑	↓
Mdm2c	Mediates the proteasomal degradation of p53			Advanced stage/grade
p14b	Inhibits MDM2		↓
p16b	Cyclin-dependent kinase inhibitor	↑	↓
Rbd	Sequesters E2F; inhibits cell-cycle progression	↑	↓
p27b	Cyclin-dependent kinase inhibitor		↓
Apoptosis
Bcl-2c	Inhibits caspase activation		↓
Baxb	Releases cytochrome c from mitochondria; promotes apoptosis		↓
Apaf-1b	Promotes apoptosis		↓
Caspase-3b	Promotes apoptosis	↑
Cell signaling
HRASc	Activates Raf and PI3K			Nonprogressing Ta tumors
FGFR3e	Receptor for fibroblast growth factor; transmits growth signals	↑
ErbB-1c	Receptor for epidermal growth factor; transmits growth signals	↑	↓
ErbB-2c	Receptor for epidermal growth factor; transmits growth signals		↓
STAT3c	Regulates gene expression; increases Bcl-2 expression	↑	↓
Tumor angiogenesis
VEGFc	Promotes angiogenesis through nitric oxide synthase	↑	↓
VEGFR2c	Receptor for VEGF; transmits angiogenic signals			Advanced stage, nodal metastasis
uPAc	Degrades extracellular matrix	↑	↓
bFGFc	Growth factor stimulating angiogenesis			Local recurrence
aFGFc	Growth factor stimulating angiogenesis			Advanced stage
TSP-1b	Inhibits angiogenesis	↑	↓
Invasion
E-cadherinb	Mediates intercellular adhesion	↑	↓
Thymidine phosphorylasec	Promotes VEGF and interleukin-8 secretion; induces MMP	↑
MMP-2c	Degrades extracellular matrix	↑	↓
MMP-9c	Degrades extracellular matrix
ICAM-1c	Binds integrins			Advanced grade, large tumors, nodal metastasis
α6β4 integrind	Links collagen VII to cytoskeleton; transduces regulatory signals		↓

^aAltered  ↑ Increased

^bUnderexpressed/lost  ↓ Decreased

^cOverexpressed

^dLost/hyperphosphorylated

^eOveractivated

aFGF, acidic fibroblast growth factor; bFGF, basic fibroblast growth factor; FGFR3, fibroblast growth factor receptor 3; HRAS, protein of the Harvey rat sarcoma viral oncogene homolog gene; MMP, matrix metalloproteinase; PI3K, phosphatidylinositol 3-kinase; Rb, retinoblastoma protein; STAT, signal transducer and activator of transcription; TSP-1, thrombospondin-1; uPA, urokinase-type plasminogen activator; VEGF, vascular endothelial growth factor; VEGFR2, VEGF receptor 2; ICAM, intracellular adhesion molecule; ErbB, epidermal growth factor receptor family member; Ta, papillary type.

Cell-cycle alterations

The most extensively characterized cellular process in UCB involves the pathways that control cell-cycle progression [Mitra et al. 2012a]. The cell cycle is primarily controlled by the p53 and Rb pathways, which closely interact with mediators of apoptosis, signal transduction, and DNA repair.

The p53 protein is encoded by the TP53 tumor-suppressor gene that is located on chromosome 17p13.1 [Mitra et al. 2007]. The protein inhibits cell-cycle progression at the G₁-S transition by transcriptionally activating p21^WAF1/CIP1. While UCB generally exhibits loss of a single 17p allele, mutation in the remaining allele can lead to TP53 inactivation and loss of its tumor-suppressor function [Mitra et al. 2005a]. Loss of heterozygosity on chromosome 17 generally occurs in advanced UCB stages and is associated with an aggressive phenotype. Wild-type p53 has a half-life of up to 30 min, which prevents its accumulation in the cell nucleus [Mitra et al. 2005b]. However, TP53 mutations result in an altered protein that is resistant to normal ubiquitin-mediated degradation. This causes increased intranuclear p53 accumulation that can be detected by immunohistochemistry.

Several retrospective studies have reported that nuclear accumulation of p53 is prognostic in UCB, especially in patients treated with radical cystectomy [Sarkis et al. 1993; Esrig et al. 1994; Serth et al. 1995; Chatterjee et al. 2004; Shariat et al. 2004, 2009b]. The rate of altered p53 expression in tumors has been shown to increase progressively from normal urothelium to nonmuscle-invasive UCB, to muscle-invasive disease and metastatic lymph nodes [Shariat et al. 2007, 2010a, 2010b]. Despite such evidence, controversy still exists on the prognostic role of p53 in bladder tumorigenesis and progression. Indeed, discordance in p53 nuclear accumulation and TP53 mutations have been documented [George et al. 2007]. A meta-analysis of the role of p53 in UCB that examined data from 117 studies noted that observational discrepancies may be related to the choice of p53 antibody used in immunohistochemical assays, variability in interpretation and stratification criteria, and other technical and specimen-handling inconsistencies [Malats et al. 2005]. The authors concluded that current evidence is not sufficient to suggest that p53 alterations may be used as a prognostic marker in UCB. A phase III trial designed to evaluate the benefit of stratifying organ-confined invasive UCB patients based on their p53 status for adjuvant cisplatin-containing chemotherapy could not confirm the prognostic value of the protein alteration or any association with chemotherapeutic response [Stadler et al. 2011]. However, this trial had several limitations, including high patient refusal rate, lower than expected event rate, and failures to receive assigned therapy that compromised the study’s power.

The p21^WAF1/CIP1 gene encodes for the p21 cyclin-dependent kinase inhibitor (CDKI). This is transcriptionally regulated by p53, and loss of p21 expression is a potential mechanism by which p53 alterations influence tumor progression [Mitra et al. 2006b]. Loss of p21 expression has been reported to be an independent predictor of UCB progression and its maintenance of expression appears to abrogate the deleterious effects of altered p53 [Stein et al. 1998].

Mdm2 is involved in an autoregulatory feedback loop with p53, thus controlling its activity. Increased p53 levels upregulate MDM2 by transactivating its promoter, and the translated protein mediates proteasomal degradation of p53. The resultant lowered p53 levels then reduce the levels of Mdm2. MDM2 amplification has been observed in UCB, and its frequency increases with increasing tumor stage and grade [Simon et al. 2002]. MDM2 is transcriptionally inhibited by p14. The protein is encoded by p14^ARF, one of the two splice variants derived from the CDKN2A locus that is situated on chromosome 9p21. Because p14^ARF is induced by the E2F transcription factor, it forms the link between the Rb and p53 pathways [Bates et al. 1998]. p14^ARF may be inactivated by homozygous deletion or by varying degrees of methylation of the promoter region [Mitra and Cote, 2009]. The other splice variant, p16^INK4a, encodes for p16 that is a CDKI. Reports suggest that homozygous p16^INK4a deletions in nonmuscle-invasive UCB have higher recurrence rates, but deletions that affect both p16 and p14, which deregulate both Rb and p53 pathways, correlate with the worst prognosis [Orlow et al. 1999]. Hemizygous and homozygous deletions of the CDKN2A locus have been found in 40–60% and 10–30% of cases, respectively [Rebouissou et al. 2012].

Encoded on chromosome 13q14, the Rb protein interacts with regulatory proteins involved in the G₁-S transition. Dephosphorylated Rb sequesters the transcription factor E2F. Upon phosphorylation of Rb by cyclin-dependent kinases, E2F is released leading to transcription of genes required for DNA synthesis. Inactivating RB mutations resulting in loss of protein expression have been noted in UCB [Miyamoto et al. 1995]. In conjunction with other cell-cycle regulatory proteins, Rb has also been shown to be prognostic in UCB [Chatterjee et al. 2004; Shariat et al. 2004]. Rb phosphorylation is facilitated by cyclin or cyclin-dependent kinase complexes. Negative regulation of cyclin-dependent kinases is achieved by CDKIs such as p21, p16 and p27, which act as tumor suppressors. Low p27 levels have been associated with advanced-stage adenocarcinomas of the bladder [Kapur et al. 2011]. p27 alterations have also been linked with shortened disease-free and overall survival in UCB [Kamai et al. 2001]. In the case of patients with pT1 tumors treated with radical cystectomy, p27 alterations in combination with other immunohistochemical markers improved the predictive value of a nomogram based on standard clinicopathological variables [Shariat et al. 2009a]. Combined immunohistochemical assessment of p53, p21, Rb, cyclin E1 and p27 has been shown to yield predictive accuracies superior to that of any single molecular marker in patients with UCB treated with radical cystectomy, and can improve risk stratification [Shariat et al. 2008, 2012].

Alterations in apoptotic pathways

Apoptosis is a complex and highly regulated process comprising a series of coordinated steps that occur throughout normal development and in response to a variety of initiation stimuli resulting in programmed cell death. Apoptosis can be initiated by two pathways. The extrinsic pathway involves activation of cell-surface death receptors, whereas the intrinsic pathway is mediated by mitochondria. Both pathways activate caspases that cleave cellular substrates and lead to the characteristic apoptotic changes. In vitro tumor-specific caspase-8 expression has been shown to induce apoptosis in urothelial carcinoma cell lines [Koga et al. 2000]. Decreased caspase-3 expression has also been associated with a higher probability of disease recurrence and cancer-specific mortality [Karam et al. 2007].

The Bcl-2 family of proteins is involved in the intrinsic apoptotic pathway; it includes antiapoptotic members such as Bcl-2 as well as proapoptotic members such as Bax and Bad. Increased Bcl-2 expression has been associated with poor prognosis in UCB patients treated with radiotherapy or synchronous chemoradiotherapy [Ong et al. 2001; Hussain et al. 2003]. Bcl-2 may also serve as a marker in patients with advanced UCB undergoing radiotherapy who may benefit from neoadjuvant chemotherapy [Cooke et al. 2000]. Bcl-2 expression has been associated with decreased tumor-free survival in high-grade T1 disease, and may serve as a good prognostic indicator in nonmuscle-invasive UCB in combination with p53 [Wolf et al. 2001; Gonzalez-Campora et al. 2007]. A prognostic index using Mdm2, p53, and Bcl-2 has also been proposed where aberrations in all three markers corresponded to the worst survival probability in UCB [Maluf et al. 2006]. On the other hand, Bax expression is an independent predictor of a more favorable prognosis in invasive UCB [Giannopoulou et al. 2002; Korkolopoulou et al. 2002; Gonzalez-Campora et al. 2007]. Bax mediates its proapoptotic role through the activation of Apaf-1 [Mitra et al. 2006c]. Decreased Apaf-1 expression has been associated with higher mortality in UCB patients [Mitra et al. 2013a].

Alterations in cell-signaling mechanisms

Several cell-surface receptors modulate signals from external cues and transmit them via transduction pathways to the nuclei of urothelial cells. Aberrations in these receptors with or without the transmitted signals can lead to uncontrolled cellular proliferation and tumor formation.

Of the FGFR family members, activating mutations of FGFR3 are the most extensively studied alterations in UCB. Nearly 60–70% of low-grade papillary Ta tumors harbor FGFR3 mutations [Pasin et al. 2008; Van Rhijn et al. 2010]. FGFR3 activation results in downstream signaling through the Ras–mitogen-activated protein kinase (MAPK) pathway. FGFR3 and Ras mutations may be mutually exclusive; nearly 82% of grade 1 tumors and Ta tumors have mutations in either a Ras gene or FGFR3, suggesting that MAPK pathway activation may be an obligate event in most of these cases [Jebar et al. 2005]. HRAS expression has also been associated with noninvasive UCB recurrence at initial presentation [Birkhahn et al. 2010].

Epidermal growth factor receptor (EGFR) family members include ErbB-1 and ErbB-2 (Her2/neu), which are overexpressed in invasive UCB [Wright et al. 1991; Mellon et al. 1996; Korkolopoulou et al. 1997]. Increased ErbB-1 expression has been associated with higher probability of progression and mortality [Liukkonen et al. 1999; Kramer et al. 2007]. Similarly, increased ErbB-2 expression has been associated with aggressive UCB and poor disease-specific survival [Lipponen and Eskelinen, 1994; Kruger et al. 2002a, 2002b; Bolenz et al. 2010]. However, other reports have indicated that ErbB-2 expression is not correlated with prognosis [Jimenez et al. 2001; Kassouf et al. 2008]. While the combined expression profile of ErbB-1 and ErbB-2 has been suggested as a better outcome predictor than each marker alone, this finding has also not been corroborated [Chow et al. 2001; Memon et al. 2006].

Variable expression of sex-steroid hormone receptors has been postulated as a potential cause for differential behavior of UCB between sexes, although direct evidence to this effect is lacking [Mitra et al. 2014b]. Across both sexes, decreased estrogen receptor-β expression has been associated with better progression-free survival rates in patients with noninvasive UCB [Tuygun et al. 2011]. Androgen receptor expression has been noted in 75% and 21.4% of patients with nonmuscle-invasive and muscle-invasive UCB, respectively [Boorjian et al. 2004].

Janus kinase (JAK) constitutes a family of tyrosine kinases that is activated by cytokine and growth receptors and mediates multiple signaling pathways. Increased preoperative plasma levels of interleukin-6, a ligand for the corresponding cytokine receptor, presumably increase JAK signaling and are an independent predictor of UCB recurrence and survival [Andrews et al. 2002]. Following JAK activation, the most well characterized molecular events include activation of the signal transducer and activator of transcription (STAT) pathway, which control transcription of several important genes. STAT1 can reduce Bcl-2 expression and STAT3 has the opposite effect [Stephanou et al. 2000]. Results from our group suggest that STAT3 expression, in combination with other markers, can predict risk of recurrence and survival in UCB patients [Mitra et al. 2009b].

Modulation of tumor angiogenesis

Angiogenesis involves the production of tumor cell-derived factors that interact with stromal elements to recruit endothelial cells to the site of malignancy and establish a vascular supply, which provides the required nutrients for growth of the cancer cells. Angiogenesis is histologically measured by microvessel density, which may be associated with disease-free and overall survival in UCB [Bochner et al. 1995]. Microvessel density quantification may also provide additional prognostic information in UCB patients with p53-altered tumors [Bochner et al. 1997].

VEGFs are angiogenesis-promoting signaling proteins that stimulate cellular responses by binding to VEGF receptors (VEGFRs). VEGFR2 (KDR/Flk-1) mediates most of the known cellular responses to VEGF. VEGFR2 expression has been associated with increasing disease stage and muscle invasion of UCB [Xia et al. 2006]. Our studies have also shown that VEGFR2 expression may be an important determinant for nodal metastasis in UCB patients [Mitra et al. 2006a]. VEGF stimulates nitric oxide synthase, which in turn stimulates nitric oxide formation and tumor vascularization. Increased expression of VEGF in nonmuscle-invasive UCB is associated with early recurrence and progression [Crew et al. 1997]. High serum levels of VEGF are associated with high UCB stage and grade, vascular invasion, CIS, metastases, and poor disease-free survival [Bernardini et al. 2001].

VEGF also induces the formation of urokinase-type plasminogen activator (uPA), which degrades the extracellular matrix, thereby facilitating endothelial cell migration and invasion. uPA generates plasmin that stimulates production of basic and acidic fibroblast growth factors (bFGF and aFGF, respectively). Preoperative plasma uPA levels have been associated with disease progression and death from UCB [Shariat et al. 2003]. Urine bFGF levels have been correlated with UCB stage and local disease recurrence [Nguyen et al. 1993; Gazzaniga et al. 1999]. Urinary aFGF levels in invasive UCB patients also show correlation with disease stage [Chopin et al. 1993].

In addition to regulating the cell cycle, p53 plays an important role in angiogenesis by upregulating TSP-1, a potent inhibitor of angiogenesis. Tumors with p53 alterations are associated with decreased TSP-1 expression, and such tumors demonstrate higher microvessel density [Grossfeld et al. 1997]. Decreased expression of TSP-1 has been associated with lower probabilities of recurrence-free and overall survival in UCB. A combination of angiogenesis-related biomarkers including VEGF, bFGF, and TSP-1 has also been associated with established clinicopathologic features of biologically aggressive disease in patients who underwent radical cystectomy for muscle-invasive UCB [Shariat et al. 2010c]. On multivariable analyses that adjusted for standard pathological features, bFGF and TSP-1 were identified as independent predictors of disease recurrence and cancer-specific mortality.

Invasive potential of tumor cells

The potential of urothelial carcinoma cells to invade the vasculature and lymphatics dictates their ability to spread to adjacent structures and metastasize to distant sites. Ubiquitous to all tissues, cadherins are prime mediators of intercellular adhesion. E-cadherin is the prototypic member of the cadherin family, and it plays a critical role in epithelial cell–cell adhesion. Decreased E-cadherin expression has been significantly correlated with higher risk of tumor recurrence and progression, as well as with shorter survival in UCB patients [Bringuier et al. 1993; Byrne et al. 2001; Mahnken et al. 2005; Mhawech-Fauceglia et al. 2006; Mitra et al. 2013a].

A tumor’s ability to degrade the matrix and invade the basement membrane is facilitated by the actions of several protease families including uPAs and MMPs. Expression levels of transcripts encoding thymidine phosphorylase, an enzyme that promotes MMP production, is 33-fold higher in muscle-invasive UCB than in nonmuscle-invasive tumors, and 260-fold higher than in the normal bladder [O’Brien et al. 1995]. The corresponding protein levels in muscle-invasive tumors are eightfold higher than in nonmuscle-invasive tumors and 15-fold higher than in normal bladder tissue [O’Brien et al. 1996]. Increased nuclear reactivity of thymidine phosphorylase has been associated with a higher risk of nonmuscle-invasive UCB recurrence [Aoki et al. 2006; Nonomura et al. 2006]. Increased MMP-2 and MMP-9 expression have been associated with higher UCB stage and grade [Davies et al. 1993; Gerhards et al. 2001]. Increased expression of MMP-2 can also predict poor relapse-free and disease-specific survival [Vasala et al. 2003]. The MMP-9:E-cadherin ratio has also been reported to be prognostic for disease-specific survival in UCB patients [Slaton et al. 2004].

Integrins are transmembrane glycoproteins which, when altered, can promote tumor progression, invasion and metastasis. They are receptors for proteins such as adhesion molecules and collagen. Intercellular adhesion molecule 1 (ICAM1) is a member of the immunoglobulin superfamily that binds to certain integrin classes. Immunohistochemical studies in UCB have revealed that ICAM1 expression is closely associated with an infiltrative histological phenotype [Roche et al. 2003]. Serum ICAM1 levels have also been correlated with the presence, grade and size of bladder tumors [Ozer et al. 2003]. ICAM1 is a member of multimarker models that can predict nodal status in patients with UCB [Mitra and Cote, 2011]. In normal urothelial cells, the α6β4 integrin is in close relationship with collagen VII, and it restricts cell migration. Loss of polarity of α6β4 integrin expression has been noted in nonmuscle-invasive UCB, and muscle-invasive tumors show either a loss of α6β4 integrin with or without collagen VII expression or a lack of colocalization of the two proteins [Liebert et al. 1994]. Patients with tumors that exhibit weak α6β4 integrin immunoreactivity have better outcomes than those with either no expression or strong overexpression [Grossman et al. 2000]. Overall, molecular markers of invasion are therefore relatively reliable predictors of patient outcome in UCB.

Molecular characterization of bladder cancer subtypes

Alterations in several molecular pathways can, in tandem, influence the pathogenesis of bladder tumors and their ultimate clinical behavior. Analyzing these alterations in combination may therefore provide deeper insight into the pathobiology of the disease, while also generating panels of markers that may be able to better predict patient outcome and treatment response. Such panels can be generated across all strata of functional cellular processing – at the epigenetic, genetic, transcriptomic, proteomic and metabolomic levels.

Gene- and transcriptome-level profiling are the most commonly used approaches in UCB, and these have been used to identify markers that characterize various bladder cancer subsets [Bartsch et al. 2010]. Efforts to profile the entire coding region of the bladder cancer genome by interrogating thousands of genes using high-throughput array-based technologies has led to deeper understanding of the molecular alterations that are associated with various UCB stages. Early studies profiling noninvasive primary UCB documented 230 genes that were differentially expressed in Ta tumors compared with normal bladder, and 86 genes that could distinguish high-grade Ta tumors from their low-grade counterparts [Aaboe et al. 2005]. Supervised learning approaches have been used to build a 16-gene molecular classifier that could identify the presence of CIS in nonmuscle-invasive UCB [Dyrskjøt et al. 2004]. Similar methodologies have also been adopted to distinguish between nonmuscle-invasive and muscle-invasive disease based on their genomic signatures [Dyrskjøt et al. 2003; Modlich et al. 2004; Blaveri et al. 2005]. Using hierarchical clustering of gene expression data, two intrinsic molecular subtypes of UCB have also been proposed [Lindgren et al. 2010]. Analysis of these subtypes indicated that genomic instability was the most distinguishing feature of tumors in the second subtype, and that this trait was not dependent on TP53/MDM2 alterations. While several signatures have been reported across various UCB stages, none have thus far been adopted in clinical practice, and the clinicopathologic determination of disease stage remains the gold standard. This is attributable, in large part, to the potential for false discovery that hampered early efforts, resulting in validation problems [Schultz et al. 2006; Dyrskjøt et al. 2007]. Given the relatively limited number of samples in these studies compared with the number of probe sets interrogated, there remained a possibility for discovery of spurious elements. Furthermore, imperfect genomic coverage across older generations of microarrays and other customized platforms impeded data reproducibility [Kuo et al. 2002]. This often led to different molecular classifiers constructed in comparable clinical populations that showed few common markers.

The advent of newer oligonucleotide microarray technology that can interrogate the entire coding region of the human genome, while also accounting for splice variants and nonprotein-coding transcripts, has broadened the realm of transcriptomic profiling in UCB [Mitra et al. 2009a]. More recent efforts using whole-genome, exome and transcriptome sequencing have identified specific mutational signatures in distinct patient subsets [Nordentoft et al. 2014]. Using primary tumor gene-expression datasets, novel UCB subtypes have been proposed, based on molecular determinants of tumor differentiation states: basal, intermediate and differentiated [Volkmer et al. 2012]. Volkmer and colleagues noted that each subtype harbored a unique tumor-initiating population, and keratin 14 (KRT14) marked the most primitive differentiation state that preceded KRT5 and KRT20 expression. The basal UCB differentiation subtype (defined by KRT14⁺KRT5⁺KRT20⁻) was associated with significantly worse overall survival compared with intermediate and differentiated subtypes. Using whole genome mRNA expression profiling, another report documented three unique molecular subtypes of muscle-invasive UCB that shared some genetic features with established subtypes of breast cancer [Choi et al. 2014]. The study authors designated these subtypes as basal, luminal and p53-like muscle-invasive tumors. Although this basal subset had a distinct molecular signature from that described by Volkmer and colleagues, both these subsets were characterized by increased expression of high-molecular-weight keratins (KRT14 and KRT5). In addition, the basal subset of muscle-invasive bladder tumors described by Choi and colleagues shared biomarkers with basal breast cancers and was characterized by p63 activation and more aggressive disease at presentation. Tumors in the luminal subset contained features of active PPARγ and estrogen receptor transcription and were enriched with activating FGFR3 mutations and potential FGFR inhibitor sensitivity. This subtype also exhibited KRT20 upregulation that was associated with the differentiated subtype described by Volkmer and colleagues. The p53-like tumors were consistently resistant to neoadjuvant methotrexate, vinblastine, doxorubicin and cisplatin chemotherapy, and all chemoresistant tumors adopted a p53-like phenotype after therapy.

Gene expression profiles from 308 tumor cases were used in another effort to define molecular subtypes of UCB: urobasal A and B, genomically unstable, squamous cell carcinoma-like, and an infiltrated class of tumors [Sjödahl et al. 2012]. Urobasal A subtype tumors were characterized by a close-to-normal basal cytokeratin (KRT5) expression, and significantly higher frequency of FGFR3 and PIK3CA mutations, as noted in the luminal subset of muscle-invasive UCB described by Choi and colleagues. In contrast, urobasal B and squamous cell carcinoma-like tumors showed elevated expression of KRT5, KRT6A-C, KRT14 and KRT16, indicating a keratinized or squamous phenotype. Cytokeratin expression patterns for squamous cell carcinoma-like tumors were comparable with that of the basal subset of muscle-invasive UCB described by Choi and colleagues. The urobasal B subset also had a higher frequency of muscle-invasive tumors with TP53 mutations, reminiscent of the p53-like muscle-invasive tumors described by Choi and colleagues. The genomically unstable subset was characterized by TP53 mutations and grossly rearranged genomes. Whereas the urobasal B cases maintained expression of the FGFR3-associated signature (including TP63 and CCND1 expression), FGFR3-mutated genomically unstable and squamous cell carcinoma-like cases showed a considerable drop in the FGFR3-associated signature. The heterogeneous infiltrated class of tumors was characterized by the presence of tumor-infiltrating cells including T lymphocytes, myofibroblasts and endothelial cells. Patients with urobasal A tumors had good prognoses; those with genomically unstable and infiltrated-group tumors had an intermediate prognosis, and patients with urobasal B and squamous cell carcinoma-like tumors had the worst prognosis. The authors have subsequently demonstrated that the major molecular subtypes may be identified using a combination of immunohistochemical markers, histologic findings and pathologic grade [Sjödahl et al. 2013].

Comprehensive characterization of the genomic landscape of UCB through the efforts of The Cancer Genome Atlas Research Network resulted in the identification of four expression clusters of high-grade muscle-invasive UCB (Figure 2) [The Cancer Genome Atlas Research Network, 2014]. Tumors in cluster I had papillary-like morphology with increased FGFR3 expression, mutations and copy-number gain, thereby suggesting that these patients may respond to inhibitors of FGFR or its downstream targets. These tumors also showed decreased miR-99a and miR-100 expression, which in turn downregulate FGFR3 expression [Oneyama et al. 2011]. Tumors in clusters I and II also showed features similar to those of luminal A breast cancer, with high expression of luminal breast differentiation markers, including GATA3 and FOXA1. These tumors also showed increased expression of uroplakins, E-cadherin and members of the miR-200 family. Increased expression of ERBB2 and estrogen receptor-β (ESR2) by these tumors also suggested that they may serve as potential targets for hormonal therapies. Expression signature of tumors in cluster III (‘basal or squamous-like’) were similar to that of basal-like breast cancers and squamous cell cancers of the head and neck, and lung. These were characterized by increased expression of epithelial lineage genes, including KRT14, KRT5, KRT6A and EGFR. Taken together, the above findings suggest the presence of several distinct molecular subtypes of UCB with characteristic expression signatures may impact prognosis and can be candidates for unique therapeutic approaches.

Figure 2.

Characterization of novel molecular subtypes of bladder cancer.

Integrated mRNA, miRNA and protein data analysis by The Cancer Genome Atlas Research Network resulted in the identification of four distinct patient clusters. Expressions of select markers are compared side by side to subtypes described by other investigators [Volkmer et al. 2012; Choi et al. 2014; Sjödahl et al. 2012]. (a) Papillary histology, FGFR3 alterations, FGFR3 expression and reduced FGFR3-related miRNA expression were enriched in cluster I. (b) Expression of epithelial lineage genes and stem/progenitor cytokeratins were high in cluster III, some of which exhibited variant squamous histology. Increased KRT5 and KRT14 expression also characterized the basal subtypes described by Volkmer and colleagues and Choi and colleagues, and the squamous cell carcinoma (SCC)-like subtype described by Sjödahl and colleagues. (c) Luminal breast and urothelial differentiation factors were enriched in clusters I and II. (d) ERBB2 mutation and ESR2 expression were enriched in clusters I and II. Partly adapted with permission from The Cancer Genome Atlas Research Network [2014].

Prognostic value of multimarker assessment

Beyond identifying novel bladder cancer subtypes, molecular-expression profiling has also been used to identify UCB patient subsets that differ in their clinical outcomes. The advent of technologies that can assess multiple markers in a reliable, efficient and cost-effective way have led to their adoption for development of prognostic panels. Several studies have quantified finite numbers of molecular targets across several UCB-associated cellular pathways in an attempt to define prognostic signatures [Birkhahn et al. 2007].

A strategy such as this was used to develop an objective method for predicting recurrence and progression in noninvasive tumors at first presentation, to potentially allow treatment individualization for these patients [Birkhahn et al. 2010]. A 24-gene panel spanning across relevant cancer pathways was used to profile patients initially presenting with Ta grade 2–3 tumors who belonged to one of three outcome-based groups: those with no recurrence, recurrence or progression within 5 years of follow-up. A multivariable model based on CCND3 expression showed 97% sensitivity and 63% specificity for identifying patients who recurred. A similar model based on HRAS, VEGFR2 and VEGF identified patients who progressed with 81% sensitivity and 94% specificity.

We have also used this approach to identify molecular alterations associated with progression across all UCB stages, which could potentially supplement disease staging in predicting clinical outcome [Mitra et al. 2009b]. The expressions of 69 genes involved in different cancer pathways were assessed on primary UCB specimens to identify a panel of four markers (JUN, MAP2K6, STAT3 and ICAM1) that were associated with disease recurrence and overall survival. Differences in 5-year probabilities for recurrence and survival based on a favorable versus unfavorable profile using this panel were 41% versus 88%, and 61% versus 5%, respectively (both, p < 0.001). The prognostic potential of this panel was confirmed on an independent external dataset (disease-specific survival, p = 0.039).

As with efforts to characterize bladder cancer subtypes, early studies employing broad-transcriptomic profiling resulted in the identification of large prognostic panels. In one effort, 105 bladder tumors were analyzed using oligonucleotide arrays, and support vector machine algorithms were utilized to test the prognostic abilities of the profiled genes [Sanchez-Carbayo et al. 2006]. For predicting overall survival, resulting accuracies were 82% and 90% when considering all UCB patients or only those with muscle-invasive disease, respectively. A 174-probe signature was also attributed to patients with node-positive disease and poor survival.

Researchers from Chungbuk National University (Chungbuk, South Korea) have also employed high-throughput profiling strategies to identify several markers associated with progression of nonmuscle-invasive bladder cancer. The group initially identified an eight-gene signature (comprising S100A8, CELSR3, PFKFB4, HMOX1, MTAP, MGC17624, KIF1A and COCH) that was associated with disease progression in this patient subgroup [Kim et al. 2010b]. Interestingly, S100A8, in combination with IL1B, S100A9 and EGFR, were also identified as important mediators of progression for muscle-invasive bladder cancer in a separate analysis [Kim et al. 2011]. The group also documented an expression signature of S100A8-correlated genes being a strong predictor of progression in patients with nonmuscle-invasive disease [Kim et al. 2010a]. A multivariable Cox regression model using a subset of three genes from the original signature (CELSR3, KIF1A and COCH) was also shown to be an independent predictor of nonmuscle-invasive bladder tumor progression [Jeong et al. 2011]. Decreased MGC17624 expression was correlated with disease progression in the original analysis, and its association with RUNX3 promoter methylation was shown to represent a poor prognostic combination in patients with nonmuscle-invasive tumors [Ha et al. 2012]. Hypermethylation of three other genes (HOXA9, ISL1 and ALDH1A3) was also shown to be an independent predictor of nonmuscle-invasive disease recurrence and progression [Kim et al. 2013].

Decision models based on clinicopathologic metrics can provide reasonable prognostic value to influence clinical management [Ahmadi et al. 2013; Mitra et al. 2013b]. Recent studies have focused on combining such clinical models with biomarkers to improve prognostic performance. The largest effort to discover and validate a prognostic genomic signature for clinically high-risk bladder cancer to date performed transcriptome-wide profiling of patients with muscle-invasive with or without node-positive UCB, resulting in the identification of a 15-feature genomic classifier that had a prognostic value of 77% on blinded independent validation [Mitra et al. 2014a]. The genomic classifier also uniquely reported on the prognostic potential of certain nonprotein-coding transcripts, which have recently been shown to play important regulatory roles in cancer development [Mitra et al. 2012d]. While the prognostic accuracy of a model that comprised clinical variables alone was 78% in the validation set, it improved to 86% when the genomic classifier was added (Figure 3a). Performance of the 15-feature genomic classifier was also validated on four independent datasets that confirmed its prognostic potential.

Figure 3.

Comparative prognostic performance of clinical risk models and biomarker alterations.

(a) Prognostic accuracy of a multivariable clinical model (black curve), as measured by area under a receiver-operating characteristic curve, was 78%. This improved to 86% when a genomic classifier was added (red curve). (b) Prognostic accuracy of another clinical model (red curve) was noted at 76%, which improved to 81% when smoking information was added (blue curve). Additional information on marker alterations increased the prognostic accuracy to 85% (green curve).

Reproduced with permission from Mitra et al. [2013a, 2014a].

We have also examined the prognostic importance of a panel of nine biomarkers across all UCB stages [Mitra et al. 2013a]. In this study, the addition of smoking history to a clinical model improved its prognostic accuracy from 76% to 81% (Figure 3b). The prognostic accuracy increased to 85% when information from the biomarker panel was added, which was significantly higher than the clinical model alone (p < 0.001) or when combined with clinical and patient smoking variables (p = 0.018). Subsequent studies have confirmed that combining smoking information with molecular markers can improve prognostication in UCB patients [Wang et al. 2014]. These data suggest that multimarker assessment can yield robust validated prognostic biomarker panels that can identify subsets of UCB patients with varying outcomes. Their performance may be enhanced in combination with clinical and epidemiologic variables, thereby identifying candidates who may need more aggressive management.

Conclusion

Bladder cancer is increasingly being recognized as a disease that cannot be treated exclusively on the basis of pathologic staging; therapeutic strategies need to focus on the molecular alterations in individual tumors. The availability of sophisticated genomic, proteomic, computational and statistical tools has enabled an increased understanding of the molecular events that lead to urothelial tumorigenesis and progression. Future UCB management will employ consensus marker panels that can provide accurate predictions of prognosis and therapeutic response in individual patients. This will require collaborative multi-institutional efforts that focus on discovery, thorough validation and clinical standardization.

While targeted therapeutic strategies are now being developed for UCB, it is critical to recognize that the multistep process of bladder carcinogenesis may likely not be treated using single-agent regimens. Synergism among agents targeting various pathways is the next step towards rational UCB management, with a goal of achieving optimal therapeutic response. Indeed, recent efforts towards characterizing the bladder cancer genome have laid the roadmap towards identifying the potential therapeutic roles for several targeted agents [Mitra et al. 2015]. This represents a scientific leap where clinical medicine is now poised to translate bench-level findings to the bedside. However, use of prognostic markers in clinical decision-making algorithms has thus far not gained universal traction in UCB management. Barriers to incorporation of biomarkers in clinical practice include inadequate independent validation, lack of consensus on physical reference standards, limited evidence of analytic and clinical validity of standardized assays, and limited validation in prospective randomized trial settings. Efforts are now underway to define best practice standards for prognostic and therapeutic biomarker reporting, and development of quality systems for theranostic implementation [Srivastava et al. 2008; Khleif et al. 2010; Kattan et al. 2016]. Risk-stratifying patients based on validated standardized prognostic-marker panels, followed by optimal surgical treatment and interrupting crucial pathway checkpoints through employment of therapeutic agents that target multiple molecular pathways, will be crucial towards effective management of this disease.